Open Access
How to translate text using browser tools
Author Affiliations +

The known diversity of dart-poison frog species has grown from 70 in the 1960s to 247 at present, with no sign that the discovery of new species will wane in the foreseeable future. Although this growth in knowledge of the diversity of this group has been accompanied by detailed investigations of many aspects of the biology of dendrobatids, their phylogenetic relationships remain poorly understood. This study was designed to test hypotheses of dendrobatid diversification by combining new and prior genotypic and phenotypic evidence in a total evidence analysis. DNA sequences were sampled for five mitochondrial and six nuclear loci (approximately 6,100 base pairs [bp];  = 3,740 bp per terminal; total dataset composed of approximately 1.55 million bp), and 174 phenotypic characters were scored from adult and larval morphology, alkaloid profiles, and behavior. These data were combined with relevant published DNA sequences. Ingroup sampling targeted several previously unsampled species, including Aromobates nocturnus, which was hypothesized previously to be the sister of all other dendrobatids. Undescribed and problematic species were sampled from multiple localities when possible. The final dataset consisted of 414 terminals: 367 ingroup terminals of 156 species and 47 outgroup terminals of 46 species.

Direct optimization parsimony analysis of the equally weighted evidence resulted in 25,872 optimal trees. Forty nodes collapse in the strict consensus, with all conflict restricted to conspecific terminals. Dendrobatids were recovered as monophyletic, and their sister group consisted of Crossodactylus, Hylodes, and Megaelosia, recognized herein as Hylodidae. Among outgroup taxa, Centrolenidae was found to be the sister group of all athesphatanurans except Hylidae, Leptodactyidae was polyphyletic, Thoropa was nested within Cycloramphidae, and Ceratophryinae was paraphyletic with respect to Telmatobiinae. Among dendrobatids, the monophyly and content of Mannophryne and Phyllobates were corroborated. Aromobates nocturnus and Colostethus saltuensis were found to be nested within Nephelobates, and Minyobates was paraphyletic and nested within Dendrobates. Colostethus was shown to be rampantly nonmonophyletic, with most species falling into two unrelated cis- and trans-Andean clades. A morphologically and behaviorally diverse clade of median lingual process-possessing species was discovered.

In light of these findings and the growth in knowledge of the diversity of this large clade over the past 40 years, we propose a new, monophyletic taxonomy for dendrobatids, recognizing the inclusive clade as a superfamily (Dendrobatoidea) composed of two families (one of which is new), six subfamilies (three new), and 16 genera (four new). Although poisonous frogs did not form a monophyletic group, the three poisonous lineages are all confined to the revised family Dendrobatidae, in keeping with the traditional application of this name. We also propose changes to achieve a monophyletic higher-level taxonomy for the athesphatanuran outgroup taxa.

Analysis of character evolution revealed multiple origins of phytotelm-breeding, parental provisioning of nutritive oocytes for larval consumption (larval oophagy), and endotrophy. Available evidence indicates that transport of tadpoles on the dorsum of parent nurse frogs—a dendrobatid synapomorphy—is carr


The past four decades have witnessed a dramatic increase in scientific knowledge of dendrobatid frogs, known commonly as dart-poison frogs. Extensive field and collection studies have more than tripled the number of recognized species from 70 in 1960 to 247 at the time of manuscript completion. Dendrobatid species occupy streams, dense forests, open fields, lowland rainforests, cloud forests, páramos, and aquatic, terrestrial, and arboreal habitats from Nicaragua to Bolivia and the Atlantic forest of Brazil and from the Pacific coast of South America to Martinique in the French Antilles. All species but one are diurnal. Insfar as is known, all dendrobatids lay terrestrial eggs, either on the ground or in phytotelmata, and many are characterized by elaborate reproductive behaviors, including transport of tadpoles on the dorsum of parent frogs and provisioning of nutritive oocytes for larval consumption.

Approximately one-third of the known species of dendrobatids secrete powerful toxins from dermal granular (poison) glands. Three of these poisonous species were used traditionally by the Emberá people of the Chocó region of western Colombia to poison their blow-gun darts for hunting (Myers et al., 1978), earning the group its common name. The pioneering work initiated by John W. Daly and Charles W. Myers more than 30 years ago has led to the discovery in dendrobatids of over 450 lipophilic alkaloids of at least 24 major structural classes (Daly et al., 1999), with novel alkaloids being discovered continuously. Many of these so-called dendrobatid alkaloids have proven to be invaluable research tools outside systematics. For example, batrachotoxins are used extensively in research on sodium channels, epibatidine is a powerful tool in the study of nicotinic receptors and functions, and histrionicotoxins are important for studying the neuromuscular subtype of nicotinic receptors (Daly et al., 1997, 2000; Daly, 1998). It has become clear that some kind of sequestration mechanism is responsible for obtaining alkaloids from the diet and incorporating them into the skin (Daly et al., 1994a), but the details of the mechanism are unknown, as are the dietary sources of the vast majority of dendrobatid alkaloids. Formicine ants, a siphonotid millipede, melyrid beetles, and scheloribatid mites have been identified as likely dietary sources for certain alkaloids (Saporito et al., 2003; Dumbacher et al., 2004; Saporito et al., 2004; Takada et al., 2005), but the remaining alkaloids are still unknown elsewhere in nature. The hydrophilic alkaloid tetrodotoxin has also been detected in one species of dendrobatid (Daly et al., 1994b), and it is unknown if its occurrence is of symbiotic or dietary origin. Dendrobatid toxicology continues to be a highly active area of investigation.

In addition to studies of dendrobatid toxins, the conspicuous, diurnal activity of many species of dendrobatids has given rise to a large and growing literature in many areas of evolutionary biology. Among the diverse studies are many investigations of breeding biology and territoriality (e.g., Silverstone, 1973; Wells, 1978, 1980a, 1980b, 1980c; Weygoldt, 1987; Zimmermann and Zimmermann, 1988; Summers, 1989; Aichinger, 1991; Caldwell, 1997; Fandiño et al., 1997; Juncá, 1998; Caldwell and de Oliveira, 1999; Summers et al., 1999a, 1999b; Lüddecke, 2000 “1999”; Bourne et al., 2001; Pröhl and Berke, 2001; Pröhl, 2003; Narins et al., 2003, 2005; Summers and McKeon, 2004), diet specialization (Silverstone, 1975a, 1976; Toft, 1980,1995; Donnelly, 1991; Caldwell, 1996; Parmelee, 1999; Darst et al., 2005), predation (Test et al., 1966), resource use and partitioning (Crump, 1971; Donnelly, 1989b; Caldwell, 1993; Lima and Moreira, 1993; Lima and Magnusson, 1998; Wild, 1996), learning (Lüddecke, 2003), population dynamics (e.g., Toft et al., 1982; Aichinger, 1987; Donnelly, 1989a, 1989c; Duellman, 1995), phonotaxis (Gerhardt and Rheinlaender, 1980), energetics (Navas, 1996a, 1996b), and correlates of ecology and physiology (Pough and Taigen, 1990). Similarly, investigations in comparative and developmental morphology have revealed bizarre and fascinating structures (Haas, 1995; Grant et al., 1997; de Sá, 1998; Myers and Donnelly, 2001). Ongoing research in these and related fields continues to generate novel discoveries with far-reaching implications in evolutionary biology.

In contrast to the major advances achieved in many aspects of their biology, the phylogeny of dendrobatid frogs remains poorly understood. Detailed knowledge of phylogeny is necessary to explain the evolutionary origins of the behaviors and other features that have been studied and provides an essential predictive framework to guide future research. Some progress has been made in recent years, as several workers have incorporated phylogenetic analysis into their research programs (e.g., Summers et al., 1999b; Santos et al., 2003; Vences et al., 2003a; Graham et al., 2004; Darst et al., 2005), but they have looked at only a small portion of the diversity of dendrobatids and have not incorporated all available evidence. Many questions remain unaddressed or unsatisfactorily answered because of a lack of understanding of dendrobatid phylogeny.

Dendrobatid monophyly has been upheld consistently (e.g., Myers and Ford, 1986; Ford and Cannatella, 1993; Haas, 2003; Vences et al., 2003a) since it was first proposed by Noble (1926; see Grant et al., 1997), but the relationships among dendrobatids remain largely unresolved. Recently, studies of DNA sequences (e.g., Clough and Summers, 2000; Vences et al., 2000, 2003a; Santos et al., 2003) have provided some insights, but limitations in both taxon and character sampling have restricted their impact on the understanding of dendrobatid phylogeny, and few taxonomic changes have resulted. Generally, dendrobatid systematics has been based on few characters, few rigorous tests, and no comprehensive analysis of available evidence.

Difficulties in understanding the phylogeny of dendrobatid frogs are compounded by the taxonomic problems that surround many nominal species and lack of appreciation of species diversity (Grant and Rodríguez, 2001). Sixty-six recognized species were named over the decade 1996–2005, 51 of which are currently referred to Colostethus. Many nominal species throughout Dendrobatidae are likely composed of multiple cryptic species awaiting diagnosis (e.g., Caldwell and Myers, 1990; Grant and Ardila-Robayo, 2002), but the rapid increase in recognized diversity is not unaccompanied by error, and critical evaluation of the limits of nominal taxa will undoubtedly result in some number of these being placed in synonymy (e.g., Coloma, 1995; Grant, 2004).

The most generally accepted view of dendrobatid systematics is based primarily on the work of Savage (1968), Silverstone (1975a, 1976), and Myers and colleagues (e.g., Myers and Daly, 1976b; Myers et al., 1978, 1991; Myers, 1982, 1987; Myers and Ford, 1986), with additional taxonomic contributions by Zimmermann and Zimmermann (1988), La Marca (1992, 1994), and Kaplan (1997). Approximately two-thirds of the species of dendrobatids are assigned to a “basal” grade of brown, nontoxic frogs (including Aromobates, Colostethus, Mannophryne, and Nephelobates), whereas the remaining third is hypothesized to form a clade of putatively aposematic frogs (including Allobates, Ameerega, Dendrobates, Epipedobates, Minyobates, Oophaga, Phobobates, Phyllobates, and Ranitomeya). Compelling evidence for that split is lacking, however, as some of the putatively aposematic taxa have been shown experimentally to be unable to sequester significant amounts of alkaloids (e.g., Daly, 1998), alkaloid profiles for most dendrobatids remain unexamined, and several of the species assigned to the “basal” grade are no less brightly colored than several of the species assigned to the aposematic clade (e.g., Colostethus abditaurantius and C. imbricolus). Furthermore, recent molecular studies (e.g., Clough and Summers, 2000; Vences et al., 2000, 2003a; Santos et al., 2003) have found several putatively aposematic taxa to be more closely related to species of Colostethus than to other toxic species.

Compelling evidence for the monophyly of most genera, especially the “basal” taxa, is also lacking. The nonmonophyly of Colostethus has been recognized for decades (Lynch, 1982), and the naming of Ameerega, Aromobates, Epipedobates, Mannophryne, and Nephelobates has merely exacerbated the problem (Kaiser et al., 1994; Coloma, 1995; Meinhardt and Parmelee, 1996; Grant et al., 1997; Grant and Castro-Herrera, 1998). Colostethus is also the most diverse genus of dendrobatids, with 138 named species recognized currently. Generally, Colostethus is regarded as a group of convenience for all dendrobatids that cannot be referred to one of the other genera (e.g., Grant and Rodríguez, 2001). Detailed investigations of several new species of Colostethus have led to the discovery of novel morphological characters that help elucidate phylogeny (Coloma, 1995; Grant et al., 1997; Grant and Castro-Herrera, 1998; Grant and Rodríguez, 2001; Myers and Donnelly, 2001; Caldwell et al., 2002a), and molecular studies are accumulating data rapidly (e.g., Santos et al., 2003; Vences et al., 2003a), but to date progress has been limited. Molecular evidence for the monophyly of Mannophryne and Nephelobates was presented by La Marca et al. (2002) and Vences et al. (2003a), but the relationship of those genera to other dendrobatids is unclear. Aromobates has been hypothesized to be the monotypic sister group of all other dendrobatids (Myers et al., 1991), but synapomorphies shared with Mannophryne and Nephelobates, also from the northern Andes of Venezuela, cast doubt on that claim; no molecular evidence has been presented for this taxon.

Among the “aposematic” taxa, only Phyllobates is strongly corroborated (Myers et al., 1978; Myers, 1987; Clough and Summers, 2000; Vences et al., 2000; Widmer et al., 2000). No synapomorphy is known for Ameerega or Epipedobates, and they are likely para- or polyphyletic with respect to each other and/or Allobates, Colostethus, Cryptophyllobates, and Phobobates. Schulte (1989) and Myers et al. (1991) rejected Allobates and Phobobates on the basis of errors in the analysis of behavior, lack of evidence, unaccounted character conflict, incorrect character coding, and creation of paraphyly in Epipedobates (as also found by Clough and Summers, 2000; Vences et al., 2000, 2003; Santos et al., 2003), but many authors continue to recognize those genera. Conflicting results have been obtained for some genera. Phobobates was found to be monophyletic by Vences et al. (2000), but paraphyletic by Clough and Summers (2000). Minyobates may or may not be nested within Dendrobates (Silverstone, 1975a; Myers, 1982, 1987; Jungfer et al., 1996a, 2000; Clough and Summers, 2000). Likewise, although neither study recognized Minyobates, it was found to be monophyletic by Santos et al. (2003) but polyphyletic by Vences et al. (2003a). Cryptophyllobates is the most recently named genus, but it contains only two species and its relationship to other dendrobatids is unclear.

Although the recent studies have demonstrated unambiguously the inadequacies of the current state of dendrobatid systematics, they have generated many more questions than decisive answers. To a certain extent, this means that this is an exciting time in dendrobatid systematics. New light is being shed on old problems, which is causing scientists to reconsider their prior beliefs (e.g., regarding the single origin of aposematism; Santos et al., 2003; Vences et al., 2003a). However, much of the current confusion is due to unreconciled conflict among datasets analyzed in isolation (e.g., regarding the monophyly of Minyobates), limited taxon sampling, and failure to include prior evidence (e.g., morphology) in the new analyses. This is not surprising, as most studies to date have been designed to address particular questions in evolutionary biology rather than to resolve dendrobatid phylogeny per se (e.g., Santos et al., 2003). The two kinds of problems are inextricably linked, and more thorough phylogenetic studies may have important consequences for the proposed evolutionary scenarios, but their empirical and analytical requirements differ.

The present study was designed to test current hypotheses of dendrobatid diversification as severely as possible by combining new and prior genotypic and phenotypic evidence in a total evidence analysis. We included as many species of dendrobatids as possible through our own fieldwork, colleagues' ongoing fieldwork, and existing natural history collections. In light of the many outstanding problems in species taxonomy, we included numerous undescribed species and samples of taxonomically problematic species from multiple localities. The primary aim of this paper is to address dendrobatid systematics, and to that end we used the optimal phylogenetic hypothesis to construct a monophyletic taxonomy. Finally, we analyzed the evolution of several character systems given the new hypothesis of relationships.

History of Dendrobatid Systematics

Scientific knowledge of dendrobatid frogs began in 1797 when the first species was named by Cuvier as Rana tinctoria (see Savage et al., in press). Over the next two centuries the number of available species-group names associated with the family swelled to 304, of which 247 are currently recognized and included in Dendrobatidae (fig. 1; for data see appendix 1). New species continue to be described at a rapid rate, especially in the taxonomically challenging genus Colostethus.

Figure 1

Accumulation of dendrobatid species, 1797–2005.

Only named species currently recognized and included in Dendrobatidae are counted. For data see appendix 1.


In this section we review the development of knowledge of the systematics of dendrobatids as background for the present study. Rather than present a strict chronology, this review is divided into three parts. The first part looks at the early history, ending in 1926 when Noble provided the modern content of the group. The second and third parts begin in 1926 with a monophyletic Dendrobatidae and continue to the present, examining the relationships among dendrobatids and between Dendrobatidae and other frogs, respectively. Ford (1993) and Grant et al. (1997) summarized many aspects of the early history and the relationships of Dendrobatidae to other groups, but we also cover some details here.

This review does not treat every paper published on dendrobatids. First, we include only systematics papers (and only the most relevant of these; i.e., species descriptions and synonymies are not detailed), and second, we include only papers published for a scientific audience. Because of the elaborate behaviors, brilliant coloration, diurnal activity, and occurrence of skin toxins in some species, large ecological, ethological, biochemical, and hobbyist literatures exist, and reviewing them all lies beyond the scope and purpose of the present paper. Also, we include only authorship and date of publication of scientific names where relevant; authorship and date of species-, genus-, and family-group names (including new taxa proposed below) are included in appendices 1{ label needed for app[@id='app2'] }3, respectively. We do not address nomenclatural problems. Grant et al. (2006; see also Grant, 2004) and Savage et al. (in press) have pending petitions to the Commission on Zoological Nomenclature regarding the use of the species-group name panamensis and family-group name Dendrobatidae, respectively, and we direct the reader to those papers (especially the latter) for nomenclatural discussion. Throughout, “dendrobatid frogs” and “dendrobatids” refer to species contained in the modern Dendrobatidae, and formal taxonomic names are used as by the author in question.

The most thorough study of amphibian relationships to appear in recent years is that of Frost et al. (2006). Owing to the disproportionate importance of that paper in designing the present study (e.g., in sampling outgroup species), we address it in a separate section (Phylogenetic Placement of Dendrobatidae and Outgroup Sampling).

Part I: 1797–1926, Early History

Although scientific study of dendrobatids had begun roughly 40 years earlier (Cuvier, 1797), little progress was achieved until Duméril and Bibron's (1841) publication. They delimited major groups of frogs based on the occurrence of teeth (vomerine [“palate”] and jaw) and the tongue, but they also employed characters from the tympanum and middle ear, parotoid glands, number of digits, webbing, hand and foot tubercles, vertebrae, and vocal sac to distinguish and group species. Additionally, they employed the relative length of the first finger as a character to arrange the three recognized species of Dendrobates (Duméril and Bibron, 1841: 651). All dendrobatids were placed in Phaneroglossa, but they were allocated to different families based on the presence and absence of maxillary teeth. Duméril and Bibron named Phyllobates bicolor as a new genus and species, and they considered it to be the last hylaeform genus, grouped with either Crossodactylus and Elosia (p. 637) or Hylodes and Phyllomedusa (p. 502). Dendrobates was the first bufoniform genus, grouped with Hylaedactylus ( =  Hyladactylus, currently a junior synonym of the microhylid Kaloula; p. 645). Although the dendrobatid genera were placed in different families, Duméril and Bibron (p. 638; translated freely from the French) actually saw them as being much closer to each other than many subsequent workers would appreciate:

This genus [Phyllobates], by the whole of its structure, makes obvious the passage of the last Hylaeformes to the first species [those of Dendrobates] of the following family, that of Bufoniformes, in which there are no longer teeth on the whole of the upper jaw and which almost always lack them on the palate.

That is, in the transitional, “grade-thinking” of the time (as opposed to the “clade-thinking” of the present), Dendrobates and Phyllobates were adjacent genera.

Fitzinger (1843: 32; see also Fitzinger, 1860) also recognized the resemblance of dendrobatid species. He grouped Dendrobates and Phyllobates in his family Phyllobatae, but he included Crossodactylus and Scinacodes ( =  Hylodes) as well.

Günther (1858) placed all dendrobatids in Opisthoglossa Platydactyla, but Phyllobates was in Hylodidae with Crossodactylus, Hylodes, and Platymantis (now in Ceratobatrachidae; Frost et al., 2006), while Hylaplesia ( =  Hysaplesia  =  Dendrobates) was in its own family, Hylaplesidae.

Cope (1865) named the family Dendrobatidae for Dendrobates and placed it in Bufoniformia. The remaining dendrobatids were placed in Arcifera in the heterogeneous family Cystignathidae. As discussed in detail by Grant et al. (1997: 30), within two years, Cope (1867; see also Cope, 1871) had begun to see the problems with separating dendrobatids solely on the basis of teeth, but he still refused to group them together. All dendrobatids were placed in Raniformia, but Colostethidae (containing Colostethus) was in Ranoid Raniformia, whereas Dendrobatidae (containing Dendrobates) was in Bufonoid Raniformia (he did not address Phyllobates). In his description of Prostherapis, Cope (1868: 137) argued that, although Prostherapis was closest in general appearance to Phyllobates, it was most closely related to Colostethus, and he placed both in his Colostethidae. He also stated that Limnocharis (now a synonym of Crossodactylus) was most closely related to Phyllobates. Subsequently, Cope (1875) restricted Raniformia to the ranoids and applied the name Firmisternia to the bufonoid taxa. This arrangement was based on novel characters of the pectoral girdle and the number of lobes of the liver, as well as the traditional ones dating to Duméril and Bibron (1841).

Boulenger (1882) simplified Cope's scheme somewhat, grouping all dendrobatids in Firmisternia, but he placed Hyloxalus (misspelled Hylixalus), Prostherapis, Phyllodromus, and Colostethus in Ranidae, Dendrobates and Mantella in the separate family Dendrobatidae, and Phyllobates in Cystignathidae. Gadow (1901) divided Ranidae into three subfamilies (Ceratobatrachinae, Dendrobatinae, and Raninae), with the toothed dendrobatids (including Phyllobates) in Raninae, and Dendrobates, Mantella, and Cardioglossa in Dendrobatinae. Gadow was uncomfortable with this arrangement, however, noting (1901: 272):

This mere loss of teeth, and the geographical distribution suggest that these frogs do not form a natural group, but have been developed independently from other Ranidae, the Neotropical Dendrobates from some likewise Neotropical genus like Prostherapis, the Malagasy Mantella from an African form like Megalixalus.

Boulenger (1910) eliminated Dendrobatinae altogether and placed all dendrobatids in Ranidae. However, although he did not formally recant, it seems that he was not entirely convinced that Dendrobatidae was not a valid group, given that Ruthven (1915: 3) acknowledged Boulenger “for assistance in diagnosing the form” Geobatrachus walkeri as a new species and genus of Dendrobatidae, and further specified that “the form falls under Boulenger's definition of the family Dendrobatidae” (Ruthven, 1915: 1). Given that Ruthven only collected the specimens in 1913, his interactions with Boulenger must have occurred after the publication of Les Batraciens in 1910.

Nicholls (1916) did away with Arcifera and Firmisternia and proposed instead to divide Phaneroglossa into four groups based on the structure of the vertebral column, particularly the centra, the groups being descriptively named Opisthocoela (sacral vertebra biconvex, free from coccyx; presacral vertebrae convex anteriorly and concave posteriorly [ = opisthocoelous]); Anomocoela (sacral vertebra ankylosed to coccyx or articulating with single condyle; presacral vertebrae concave anteriorly and convex posteriorly [ = procoelous] or rarely opisthocoelous); Procoela (sacral vertebra free and articulating with double condyle; presacral vertebrae procoelous); and Diplasiocoela (sacral vertebra biconvex; eighth presacral vertebra biconcave, other seven presacrals procoelous). Insofar as he believed the diplasiocoelous condition to occur in all firmisternal taxa, this new arrangement did not affect the placement of dendrobatids.

In a series of four papers, G. K. Noble synthesized published information with his own research on the development and structure of vertebrae, pectoral girdles, thigh musculature, and external morphology to provide the framework for the modern understanding of Dendrobatidae. First, Barbour and Noble (1920) carried out a major taxonomic revision. They followed Peracca (1904: 17) in referring Phyllodromus to Prostherapis, but they went on to include both Prostherapis and Colostethus (the latter based largely on a letter from Boulenger to Barbour) as junior synonyms of Phyllobates. Next, Noble (1922) argued against the close relationship of Dendrobates and Mantella and explicitly endorsed Boulenger's (1910) elimination of Dendrobatidae (p. 8), disputed Nicholls's (1916) claim that all firmisternal species are diplasiocoelous (describing a numberof dendrobatid species as procoelous and transferring them to Procoela [pp. 14–15]), and gathered together Brachycephalus, Atelopus, Rhinoderma, Sminthillus (now a synonym of the brachycephalid Eleutherodactylus), Geobatrachus, Oreophrynella, Phyllobates, Hyloxalus, Chilixalus (now a synonym of Rana), and Dendrobates in Brachycephalidae (pp. 68–69). Noble (1923) subsequently diagnosed Hyloxalus from Phyllobates by the presence of webbing (contra Savage, 1968, who attributed the definition of Hyloxalus as toothed dendrobatids with webbed toes to Dunn, 1931). Finally, on the basis of the occurrence of “leathery scutes on the upper surface of each digit tip”, Noble (1926: 7) united Phyllobates, Hyloxalus, and the toothless Dendrobates in a single, exclusive group, the first time such an arrangement had been proposed (Grant et al., 1997).

Noble (1926) was not only the first to unite the dendrobatids into an exclusive group, but he also provided the hypothesis of family-level phylogeny that has guided thinking ever since by proposing that (p. 9)

Crossodactylus gave rise to Hyloxalus by merely a fusion of the coracoid cartilages. … Hyloxalus gave rise to Phyllobates by a reduction in its digital webs. The latter genus evolved and is evolving directly into Dendrobates by a loss of its maxillary teeth.

That is, although most of the theoretical views Noble held are no longer embraced, such as the notion of group or stock evolution and nonmonophyletic yet natural groups (see below and Grant et al., 1997: 31, footnote 18), the scheme of the webbed, more aquatic species being primitive to the unwebbed, more terrestrial species, and these being primitive to the terrestrial, toothless species has yet to be seriously questioned—or tested.

Part II: 1926–present, Relationships among Dendrobatids

Having grouped dendrobatids together wholly on the basis of anatomical characters, Noble (1927: 103) noted that his conclusion “receives an eloquent support from life history data” as well. He pointed out that males of species of Dendrobates and Phyllobates transport tadpoles to pools, and, further, that “[n]o other Salientia have breeding habits exactly like Dendrobates and Phyllobates” (p. 104).

Noble (1931: 507) formally recognized the group of Phyllobates, Hyloxalus, and Dendrobates as Dendrobatinae, a subfamily of the procoelan Brachycephalidae, and he reiterated that the group evolved from Crossodactylus.

The same year, Dunn (1931) named Phyllobates flotator, a new species with a swollen third finger in males and an umbelliform oral disc, reduced rows of keratodonts, and scattered median papillae in tadpoles. Dunn (1924) had previously observed the same third finger morphology in P. nubicola, also from Panama, and he postulated that these two species formed a group within Phyllobates. In error, Dunn (1924) had attributed these characteristics to P. talamancae, and he later stated (Dunn, 1931) that in his 1924 paper he had mistakenly referred specimens of his new P. flotator to P. talamancae. (However, his [Dunn, 1924: 7] description that “[t]he throat of the male is black” indicates that the specimens mistakenly identified as P. talamancae were P. nubicola, not P. flotator; but see also Savage, 1968.)

Dunn (1931: 389) explicitly followed Noble's (1926) evolutionary scenario but further partitioned Phyllobates into groups, stating:

The Phyllobates from Panama, Costa Rica, and Nicaragua that I have seen fall into three groups; typical Phyllobates, without specialized tadpoles, or modified male third finger (these apparently stem from Hyloxalus, which has webbed toes); Phyllobates which have specialized tadpoles and modified third finger (flotator and nubicola); and Phyllobates which have markings black and yellow instead of black and white, and ventral light markings. (These are close to Dendrobates.)

Dunn (1933: 69) reviewed Hyloxalus and modified it slightly to include species with both webbed and fringed toes. He concluded that six species were attributable to Hyloxalus thus diagnosed, including Hyloxalus fuliginosus, H. bocagei, H. chocoensis, Hylixalus collaris, H. granuliventris (now a synonym of Phyllobates palmatus), and H. panamansis (name subsequently emended to H. panamensis by Dunn, 1940; see Grant, 2004; Grant et al., 2006). Dunn (1933) excluded H. huigrae (now a junior synonym of the brachycephalid “Eleutherodactylusdiastema) and H. beebei—the latter exclusion being the only practical consequence of Dunn's (1933) redelimitation of Hyloxalus. Dunn did not apply his new diagnosis consistently over subsequent years, however; on occasion he returned to Noble's (1923) diagnosis, that is, without reference to fringes (e.g., Dunn, 1941: 89, 1944: 519), but he also applied his own diagnosis of having both webbing and fringes (e.g., Dunn, 1957: 77 [as Prostherapis, see below]). Dunn (1933) noted that males of his new species H. panamensis possessed a swollen third finger, which he had previously observed in Phyllobates nubicola and P. flotator and had used to group them phylogenetically, but he did not attribute any phylogenetic significance to the present observation.

In his discussion of the relationships of Dendrobates auratus, Dunn (1941: 88–89) recognized a group of species with rounded light markings, formed by D. auratus, a species from “the western part of Colombia … [in which] the light color is red or yellow” (presumably D. histrionicus), D. pumilio, and D. speciosus. He also recognized a second group of “typical Dendrobates” with “dorso-lateral light lines like Phyllobates … [but] lacking maxillary teeth” for “tinctorius, trivittatus, etc.”, as well as “lugubris, minutus, and shrevei.” In total, Dunn (1941) now recognized 18 species of Dendrobates, 26 Phyllobates, and 8 Hyloxalus.

Prostherapis remained in the synonymy of Phyllobates, where it had been placed by Barbour and Noble (1920), for over 35 years. The sole exception was Breder (1946: 405) who reported Prostherapis inguinalis from Panama without commenting on the status of the genus. It was Test (1956: 6), acting on the advice of Dunn, who resurrected the genus as a senior synonym of Hyloxalus. A more detailed account of this synonymy was published after Dunn's death (Dunn, 1957: 77), where Dunn clarified that “the presence of webs and fringes on the toes distinguishes Prostherapis from Phyllobates which hasn't got them.”

Bhaduri (1953) studied the urinogenital systems of diverse amphibians, including Dendrobates auratus, D. tinctorius, and Colostethus flotator (as Phyllobates nubicola flotator). He noted several differences among these species, such as the greater posterior extension of the kidneys in Dendrobates than in Phyllobates (p. 56), but he nonetheless concluded that “[t]he structural similarities of the urinogenital organs which I have observed in these two genera lend further support to Noble's view [that Dendrobates and Phyllobates are closely related]” (p. 72).

Rivero (1961) provided accounts for Venezuelan species. In his description of Prostherapis shrevei, he postulated that it was “perhaps a race” of Prostherapis bocagei, but he concluded that the two were distinct, but presumably closely related, species. Rivero (1961) suggested that Phyllobates brunneus and Phyllobates marchesianus might prove to be conspecific, but he did not propose phylogenetic relationships for the other species.

Dunn's arrangement was followed until 1966, by which time Cochran (1966) had become skeptical of the usefulness of toe webbing in diagnosing these groups of frogs. This change was foreshadowed by Cochran and Goin's (1964) description of a new webbed dendrobatid with teeth as Phyllobates mertensi. Cochran (1966) accepted the recognition of Phyllobates and Dendrobates on the basis of maxillary teeth, but she (p. 61; see also p. 64) argued against the further division of toothed species because “[t]he variation in degree of webbing of the species [of Prostherapis] is so great … that no valid reliance can be placed on it to justify such a separation on that characteristic.” Cochran and Goin (1970) employed this taxonomy, even though it had already become outdated by the time their monograph was published.

Although Cochran (1966) treated only the Colombian species, she proposed a number of novel groups. These included a group for D. trivittatus and an as yet undescribed species (D. ingeri), and a second group for D. hahneli and D. lugubris. A third group was further divided into subgroups for D. opisthomelas and D. minutus ventrimaculatus, and for the subspecies of D. tinctorius: D. t. histrionicus, D. t. wittei, D. t. chocoensis, and D. t. confluens. Among Colombian species of Phyllobates, Cochran (1966) recognized a group for P. bicolor, P. mertensi, P. boulengeri, and P. femoralis, with the latter two species more closely related. Another group included P. subpunctatus, P. vergeli, P. chocoensis, and another as yet unnamed species (presumably P. thorntoni, named by Cochran and Goin, 1970). Curiously, a soon-to-be-named subspecies of P. subpunctatus (P. s. walesi) was placed in a different group with P. palmatus. Finally, a group containing P. brunneus, P. pratti, P. latinasus, and P. inguinalis was also proposed.

Savage (1968) ushered in the modern era of dendrobatid research. Although his study focused on the Central American taxa, it was highly influential and arguably the most important paper since Noble's (1926) in establishing a framework for much of the dendrobatid systematics research of the following decades. In addition to addressing a number of species-level taxonomic problems in Central America, Savage divided the Central American species into three groups, and to each of these groups he assigned the oldest available name. He also referred species outside of Central America to each genus, as far as he could, though subsequent authors would have to provide complete assignments. New characters Savage employed to diagnose his three groups included pigmentation of the flesh, size of digital discs, and in larvae the oral disc morphology, rows of keratodonts, and position of the anus.

Savage (1968: 746–747) resurrected Colostethus for his Group I, which included five Central American species and “most species called Phyllobates in South America”. Savage (1968: 765) clarified that Dendrobates lugubris was a toothed species and that recent workers had mistakenly applied that name to Dendrobates truncatus. Consequently, he assigned Phyllobates to his Group II, composed of P. lugubris in Central America, and P. bicolor and P. aurotaenia “among others” in South America. Dendrobates was assigned to his remaining Group III, still composed of toothless dendrobatids, as it always had been.

In the late 1960s, two graduate students undertook studies of the systematics of Dendrobatidae. Stephen R. Edwards wrote his Ph.D. dissertation (Edwards, 1974a) on Colostethus (sensu Savage, 1968, with minor modification). He studied 63 species in his dissertation, including many undescribed species, but only two small papers on dendrobatids were published as a result of this work (Edwards, 1971, 1974b); the bulk of Edwards's dissertation research—including descriptions for the unnamed species in his dissertation and the quantitative phenetic analysis—were never published (which prompted the naming of Colostethus exasperatus; see Duellman and Lynch, 1988) and will therefore not be discussed here (but see discussion below of Rivero, 1990 “1988” and Rivero and Serna, 1989 “1988”). In the first of his papers, Edwards (1971) referred 43 nominal species to Colostethus and described two more species as new; he did not discuss the relationships among the species. In his second publication, Edwards (1974b) named a new species and clarified the identities of another three. More importantly, he also arranged the nominal species into seven groups. Although Edwards (1974b: 1) was explicit that these groups “do not reflect evolutionary or taxonomic units” and that their sole purpose was to facilitate comparisons (e.g., C. vertebralis, shown below in bold, was listed in each appropriate group), this was the first arrangement ever provided for most of these species. The groups were as follows:

  1. C. elachyhistus, C. fraterdanieli, C. kingsburyi, C. subpunctatus, C. variabilis.

  2. C. alagoanus, C. brunneus, C. capixaba, C. carioca, C marchesianus.

  3. C. collaris, C. dunni, C. herminae, C. meridensis, C. riveroi, C. trinitatus [ =  trinitatis].

  4. C. beebei, C. chocoensis, C. fuliginosus, C. granuliventris, C. mandelorum, C. mertensi, C. palmatus, C. shrevei, C. talamancae, C. vergeli.

  5. C. intermedius, C. latinasus.

  6. C. nubicola, C. pratti.

  7. C. alboguttatus, C. bromelicola, C. infraguttatus, C. olfersioides, C. pratti, C. ranoides, C. vertebralis.

  8. C. anthracinus, C. infraguttatus, C. lehmanni, C. ramosi, C. taeniatus, C. vertebralis, C. whymperi.

Because Edwards's dissertation was a quantitative phenetic analysis, he focused largely on meristic data and reported few novel characters. His most lasting contribution in terms of character delimitation was to focus on and demarcate explicitly the different pale lateral stripes found in most species.

Philip A. Silverstone carried out his Ph.D. research on the systematics of Dendrobates (Silverstone, 1970). He published two small papers (Silverstone, 1971, 1975b) on dendrobatid systematics, but most of Silverstone's findings were published in two comprehensive monographs; the first (Silverstone, 1975a) summarized his dissertation on Dendrobates and included accounts for 16 species; the second (Silverstone, 1976) reported his research on Phyllobates and included 20 species.

Silverstone (1975a: 3) did not put much credence in the generic taxonomy he employed (which was largely that of Savage, 1968). He noted that there were species with morphology intermediate between the genera, and that “any rigidly applied definition of more than one genus for dendrobatid frogs could result in unnatural ( =  polyphyletic) groups.” But rather than place all dendrobatids into a single genus, Silverstone (1975a: 3) continued “to recognize the three currently accepted genera as categories of convenience, that is, as taxonomic units convenient to study, but not necessarily natural.” Although he thought the three genera may grade into each other, Silverstone (1975a: 4) implicitly followed Noble's (1926) evolutionary scenario, stating that he was “concerned more with the relationship of Phyllobates to Dendrobates than with that of Phyllobates to Colostethus.”

The generic diagnoses Silverstone used were very similar to Savage's (1968), although he did incorporate new characters (occurrence of the palatine, omosternum, vertebral fusion; he also used fusion and sculpturing of the cranium to diagnose species groups). In terms of content, there were two major differences. First, Phyllobates sensu Savage was, explicitly at least, a group of only three, very similar species, whereas Phyllobates sensu Silverstone included 20 species, most of which had been implicitly referred to Colostethus by Savage. Second, Silverstone went against all previous workers by transferring two toothless species from Dendrobates to Phyllobates. Although all specimens of P. trivittatus and most of P. pictus lacked teeth, Silverstone (1975a) was overwhelmed by evidence from chromosomes and finger morphology that indicated these species should be placed in Phyllobates. Thus, dendrobatid systematics was finally rid of the a priori weighting applied to teeth that had hindered progress since Duméril and Bibron (1841).

In his two monographs, Silverstone (1975a, 1976) proposed numerous species groups, many of which he thought were natural. Within Dendrobates, he proposed the histrionicus group for D. histrionicus and D. leucomelas. Significantly, Silverstone (1975a: 25) clarified that D. histrionicus was not a subspecies of “the large, striped, Guianan species to which D. tinctorius is restricted”, but he remained ambivalent with regard to the putative subspecies of D. histrionicus; he did not separate them formally, but he did attribute diagnostic color patterns to several of them. His reasons for treating all the color patterns as a single species were that they all “lack an omosternum and have the same breeding call” (Silverstone, 1975a: 23). Based on larval morphology, Silverstone (1975a: 23) surmised that “the histrionicus group is more closely related to the pumilio group than to the other two groups of Dendrobates.”

Silverstone's minutus group contained six species: D. altobueyensis, D. fulguritus, D. minutus, and D. opisthomelas from Central America and northwestern South America and D. quinquevittatus and D. steyermarki from the Amazon basin (the former from lowlands, the latter from 1,200 m on the summit of the tepui Cerro Yapacana). Within this group, Silverstone (1975a: 29) hypothesized a close relationship between D. fulguritus and D. minutus on the basis of size and dorsal striping; his decision to treat them as distinct species was due to his having collected them in sympatry. He also conjectured that D. minutus and D. opisthomelas were closely related, as tadpoles of these species were the only ones in the genus with an indented oral disc and dextral anus; Silverstone was not completely convinced of the identity of the tadpoles he assigned to D. altobueyensis and D. fulguritus, but they also had an indented oral disc and dextral anus. Tadpoles of D. quinquevittatus and D. steyermarki were unknown to Silverstone, and he assigned those species to the minutus group on the basis of other characters. He also hypothesized that D. steyermarki was “more closely related to [the western Andean D. opisthomelas] than to any other species of Dendrobates” (Silverstone, 1975a: 36).

Silverstone (1975a) proposed the pumilio group for D. granuliferus, D. pumilio, and D. speciosus. Silverstone (1975a: 38) argued that D. granuliferus and D. pumilio were very closely related, perhaps even conspecific, and that they were “probably geographically and genetically continuous before the onset of orogeny and aridity in Costa Rica.” This would leave D. speciosus as their sister group. As mentioned above, Silverstone hypothesized that the pumilio and histrionicus groups were sister groups.

The tinctorius group included D. auratus, D. azureus, D. galactonotus, D. tinctorius, and D. truncatus. Within this group, Silverstone (1975a) proposed that D. auratus was most closely related to D. truncatus. He also hypothesized that D. azureus had “arisen by isolation of a population of D. tinctorius in forest islands surrounded by unsuitable habitat” (Silverstone, 1975a: 44).

The 20 species of Phyllobates Silverstone (1976) recognized were arranged into four groups, but the relationships among these four groups were not addressed. The bicolor group was the same as Phyllobates sensu Savage (1968) with the addition of two more species. That is, he placed P. aurotaenia, P. bicolor, and P. lugubris in a single group (as had Savage) together with an as yet unnamed taxon (later named Dendrobates silverstonei; Silverstone doubted the inclusion of this species in this group but placed it there due to its superficial resemblance with P. bicolor) and P. vittatus (which Savage considered to be conspecific with P. lugubris). Silverstone did not further resolve the relationships of this group.

The femoralis group included P. anthonyi, P. boulengeri, P. espinosai, P. femoralis, P. tricolor, and P. zaparo. Within this group, Silverstone (1976) proposed the following relationships: (P. tricolor (P. femoralis P. zaparo) (P. anthonyi P. boulengeri P. espinosai)).

The pictus group contained P. bolivianus, P. ingeri, P. parvulus, P. petersi, P. pictus, P. pulchripectus, and P. smaragdinus. Silverstone (1976) was doubtful that this group was monophyletic, but he did think parts of it were. He grouped P. pictus and P. parvulus together based on the shared presence of a calf spot. He also grouped P. petersi and P. pulchripectus on the basis of similar color patterns, and united them with P. bolivianus (although he was more ambivalent about the latter's relationship). Silverstone did not place P. smaragdinus, and he did not propose a scheme of relationships among these groups.

Silverstone (1976) was more certain about the naturalness of the trivittatus group, which contained only the similarly colored P. bassleri and P. trivittatus. Silverstone did not publish further studies on dendrobatid frogs, as he discontinued working in herpetology to pursue a scientific career in botany.

In 1976, Charles W. Myers and John W. Daly began publishing on the systematics implications of their work initiated a decade earlier (Daly and Myers, 1967). They added three new sources of evidence: alkaloid profiles, vocalizations, and behavior. Modern research in dendrobatid alkaloids was initiated by Märki and Witkop (1963), and the accumulated data appeared to have clear systematics implications. Similarly, audiospectrographic analysis of vocalizations had been carried out for several groups of frogs (e.g., Bogert, 1960; Martin, 1972), but not for dendrobatids. Numerous workers had published observations on dendrobatid parental care and other behaviors (Wyman, 1859a, 1859b [reported as Hylodes lineatus; Dendrobates trivittatus fide Boulenger, 1888], Ruthven and Gaige, 1915; Senfft, 1936; Dunn, 1944; Test, 1954; Stebbins and Hendrickson, 1959; Duellman, 1966; Goodman, 1971; Crump, 1972; Bunnell, 1973; Silverstone, 1973, 1975a, 1976; Dole and Durant, 1974), and to these were added the extensive field and laboratory observations of Myers and Daly, who analyzed the phylogenetic implications of these advances.

Based on these and traditional data, Myers and Daly (1976b) named three new species and redescribed D. histrionicus. They also added support to Silverstone's (1975a) pumilio group, and they proposed a group consisting of D. histrionicus, D. lehmanni, and D. occultator (they did not mention D. leucomelas, which Silverstone had grouped with D. histrionicus). That same year, Myers and Daly (1976a) named D. abditus and added it and D. viridis to Silverstone's (1975a) minutus group.

Myers et al. (1978) proposed a restricted application of Phyllobates as an explicitly monophyletic genus (the first in the family). They argued that Phyllobates sensu Silverstone (1976) had been diagnosed on the basis of symplesiomorphy, whereas the occurrence of batrachotoxins was a synapomorphy for a group containing P. aurotaenia, P. bicolor, P. lugubris, P. terribilis, and P. vittatus, and thus resembling Phyllobates sensu Savage (1968). In order to avoid coining new names without evidence of monophyly, Myers et al. (1978) referred the rest of Phyllobates sensu Silverstone (1976) to Dendrobates, pending a comprehensive phylogenetic analysis.

Rivero (1978 “1976”) named three species of Colostethus and proposed that C. haydeeae and C. orostoma were closest relatives (later dubbed the haydeeae group by Rivero, 1980 “1978”: 99). This conjecture was based largely on the supposed occurrence of four anterior and five posterior rows of keratodonts in larvae, although Rivero did note the possibility that the larvae were not of these species. Rivero (1978 “1976”) speculated that C. leopardalis was most closely related to C. alboguttatus, C. collaris, and C. meridensis and concluded that “in spite of the presence of a collar in C. leopardalis and its absence in C. alboguttatus, these two species are more closely related to each other than either is to C. collaris [which has a collar]” (p. 334; translated from the Spanish).

Rivero (1979) suggested that the presence of a dark chest collar delimited a monophyletic group of species confined to the Venezuelan Cordillera de la Costa. Rivero (1979) mentioned the occurrence of similar dark spotting on each side of the chest in several species from southern Colombia to northern Peru, and he (Rivero, 1979: 172) proposed that the collared species were derived from the species with chest spotting. Curiously, Rivero (1984 “1982”) later included C. mandelorum, a species that lacks a dark collar, in this group, and, following Rivero (1979: 173), went on to hypothesize that the “ancestral stock of C. trinitatis … gave origin to the other collared forms of Venezuela and C. mandelorum” (p. 12). The inclusion of this uncollared species in this group was based on the species's “affinity with collared species, its limited altitudinal distribution, and the absence currently of any uncollared species similar to it” (Rivero, 1984 “1982”: 12).

Myers and Daly (1979) further characterized the trivittatus group based on vocalizations, and they added to it D. silverstonei. The following year, Myers and Daly (1980) named a new species (D. bombetes), resurrected D. reticulatus, and assigned both to the minutus group. (They also included an unnamed species, finally described 20 years later as D. claudiae by Jungfer et al., 2000.) Furthermore, they hypothesized that D. abditus, D. bombetes, and D. opisthomelas, all from the western Andes of Colombia and Ecuador, formed a monophyletic group delimited by a “median gap that interrupts the papillate fringe on the posterior (lower) edge of the oral disc” (Myers and Daly, 1980: 20).

Based on finger length and color pattern, Rivero (1980 “1978”) proposed that C. inflexus was part of the haydeeae group (sensu Rivero 1978 “1976”), and that their closest relative was C. alboguttatus. Colostethus inflexus was later placed in the synonymy of C. alboguttatus by Rivero (1984 “1982”), but he did not address the phylogenetic implications of this change. Although he did not retract his previous claim that C. haydeeae and C. orostoma had a larval keratodont row formula of 4/5, Rivero (1980 “1978”) did seriously question its veracity, given that no other Colostethus was known to possess this morphology. La Marca (1985) subsequently identified Rivero's C. haydeeae tadpole as Hyla platydactyla.

Myers (1982) resurrected and redescribed D. maculatus but clarified that he was “unable at this time to demonstrate a close relationship with any other known dendrobatid” (p. 2). Myers (1982: 2) also resurrected D. fantasticus from synonymy with D. quinquevittatus and placed D. vanzolinii, D. fantasticus, D. quinquevittatus, and D. reticulatus in a monophyletic quinquevittatus group delimited by “distinctively reticulate limbs”. Myers (1982) speculated that D. captivus and D. mysteriosus were sister species, but he was unable to present any synapomorphies to corroborate this hypothesis.

In 1982, Lynch published two papers on Colombian dendrobatids. Lynch (1982) named C. edwardsi and C. ruizi and hypothesized that they formed a distinct group within Dendrobatidae, based on the occurrence of an “anal sheath” and putatively derived absence of a tarsal fold or tubercle (also known in dendrobatid literature as “tarsal keel”). He refrained from naming this group formally to avoid encumbering future research; he also observed that no synapomorphies were known for Colostethus and declared that the genus was paraphyletic (although he did not present evidence to that effect).

Lynch and Ruiz-Carranza (1982) described the new genus and species Atopophrynus syntomopus as a dendrobatid. They reported a number of features unknown in other dendrobatids, but they were unable to elucidate the relationships of this taxon with respect to other dendrobatids. They (Lynch and Ruiz-Carranza, 1982: 561) explicitly rejected the absence of teeth as a synapomorphy “because it postulates loss of an attribute.”

Rivero (1984) clarified that C. dunni did not have a throat collar (contra Edwards, 1974a, 1974b) and provided a name, C. oblitteratus, for the MCZ material Edwards had seen.

Myers et al. (1984) combined what had been the pumilio and histrionicus groups into a new, expressly monophyletic histrionicus group delimited by the synapomorphic occurrence of a “chirp call”. This group contained D. arboreus, D. granuliferus, D. histrionicus, D. lehmanni, D. occultator, D. pumilio, D. speciosus, and an unnamed species.

Maxson and Myers (1985) employed microcomplement fixation to compare the serum albumin of several dendrobatids. They concluded that recognition of Phyllobates as a separate group was warranted, and that the “[s]peciation events leading to the living species of true dart-poison frogs (Phyllobates) appear to have occurred within the last five million years” (Maxson and Myers, 1985: 50). They also found that the species of Dendrobates they studied were much more divergent than the species of Phyllobates, and that this was “consistent with accumulating evidence that Dendrobates is a polyphyletic assemblage” (Maxson and Myers, 1985: 50). They suggested that at least four major lineages were represented, and that initial divergence dated back some 60 million years.

Péfaur (1985) described two new species of Colostethus from Venezuela, but he did not discuss their phylogenetic relationships. La Marca (1985: 4) claimed that his new species C. molinarii was “a member of the C. alboguttatus group, a monophyletic assemblage” comprised additionally of C. alboguttatus, C. dunni, C. haydeeae, C. leopardalis, C. mayorgai, C. meridensis, and C. orostoma. However, La Marca (1985) did not offer any evidence in support of this conjecture.

Dixon and Rivero-Blanco (1985) named Colostethus guatopoensis (placed in the synonymy of Colostethus oblitterata by Rivero, 1990 “1988”) and grouped it with C. riveroi on the basis of the shared absence of the outer metatarsal tubercle. This synapomorphy was disputed by La Marca (1996 “1994”), who reported the occurrence of the outer metatarsal tubercle in both species (and considered both species to be valid).

In a series of privately published but nomenclaturally valid (according to ICZN, 1999) papers, Bauer (1986, 1988, 1994) named several genera and speculated on their relationships. Bauer's proposals were based on reinterpretations of Silverstone (1975a, 1976) and Myers and colleagues (mainly Myers et al., 1978, 1984; Myers and Burrowes, 1987) augmented with limited observations of a few species in captivity. Bauer's publications were overlooked by all workers except Wells (1994), and as a result the literature is now quite confusing; for that reason we break from chronological order to summarize Bauer's contributions. Bauer (1986) named Ameerega (type specie: Hyla trivittata) for the species of Phyllobates sensu Silverstone (1976) that were not placed in Phyllobates sensu Myers et al. (1978). Bauer (1988) named Ranitomeya (type species: Dendrobates reticulatus) for Dendrobates captivus, D. fantasticus, D. imitator, D. mysteriosus, D. quinquevittatus, D. reticulatus, and D. vanzolinii. Bauer (1988) attributed the name to “Bauer, 1985”, and it was also employed by Bauer (1986); however, those prior uses do not constitute nomenclatural actions because (1) the 1985 use was in a publication that did not specify authorship (Anonymous, 1985) and (2) the 1986 use did not specify a type species. Only Bauer's 1988 use was sufficient to make Ranitomeya an available name. In that paper, Bauer also named Pseudendrobates, but that is an objective synonym of Phobobates Zimmermann and Zimmermann, 1988 (see below) because it was published later and specified the same type species (Dendrobates silverstonei). Bauer (1994: 1) stated that “Phobobates should be considered a synonym”, but of what he did not say, and he did not provide evidence to substantiate his view. Bauer (1994) proposed the name Oophaga (type species: Dendrobates pumilio) for the histrionicus group of Myers et al. (1984), namely, Dendrobates arboreus, D. granuliferus, D. histrionicus, D. lehmanni, D. occultator, D. pumilio, D. speciosus, and D. sylvaticus. Although Oophaga was never placed in the synonymy of Dendrobates, it was also never used again. Finally, Bauer (1994) named Paruwrobates as a monotypic genus to accommodate D. andinus; Bauer did not address the placement of D. erythromos, although Myers and Burrowes (1987) had grouped them together (and Bauer claimed to be basing his new taxonomy on their paper). In that paper, Bauer also resurrected Prostherapis, but he did not list the content of the genus and the evidence he cited for distinguishing Prostherapis inguinalis from Colostethus latinasus was his erroneous claim that they differ in the occurrence of swelling in the third finger in adult males (see Grant, 2004).

Bauer (1986, 1988, 1994) was the only recent worker to recognize subfamilies within Dendrobatidae. In the most recent proposal (Bauer, 1994), he recognized Dendrobatinae for Dendrobates, Oophaga, Ranitomeya, and Minyobates; Phyllobatinae for Phyllobates and Ameerega; and Colostethinae for Aromobates, Colostethus, and Epipedobates. (Note that Bauer's use of Epipedobates was restricted to Silverstone's femoralis group, and he applied Ameerega to the bulk of Phyllobates sensu Silverstone.) Bauer was apparently unaware of Mannophryne La Marca, 1992. Subfamily diagnoses employed differences in chromosome number, coloration, occurrence of maxillary teeth, skin toxins, webbing, length of first finger, muscle coloration, clutch size, breeding biology, and tadpole specialization. He believed Dendrobatinae and Phyllobatinae to be monophyletic, but thought that Colostethinae was paraphyletic (though he did not say with respect to what); he did not otherwise propose relationships among the subfamilies.

Meanwhile, Myers and Ford (1986) examined Lynch and Ruiz-Carranza's (1982) assertion that Atopophrynus was a dendrobatid. They could find no support for Lynch and Ruiz-Carranza's claim, given that specimens they examined showed major differences from dendrobatids in external morphology, jaw musculature, thigh musculature, skull, finger structure, and hyoid structure, and shared no particular synapomorphy. Consequently, they removed the genus from Dendrobatidae and placed it in Leptodactylidae (transferred with other eleutherodactylines to Brachycephalidae by Frost et al., 2006).

Myers (1987) proposed a major taxonomic rearrangement aimed to better reflect hypotheses of monophyly, whereby “[d]endrobatids that produce lipophilic alkaloids are a monophyletic group that is now partitioned among four genera” (p. 304). Epipedobates (type species: Prostherapis tricolor) was named to accommodate most species of Phyllobates sensu Silverstone (1976) minus the species Myers et al. (1978) had placed in their restricted Phyllobates. Although Myers's intention was the same as Bauer's (discussed above), his designation of a different type species means that the two names may be applied to different groups. Dendrobates was redefined as a monophyletic group delimited by a suite of synapomorphies from larval, adult, behavioral, and alkaloid characters. Dendrobates included the quinquevittatus group of Myers (1982), which had been part of the minutus group (Silverstone, 1975a; Myers and Daly, 1976a, 1980; Myers, 1982). The remainder of the minutus group was transferred to Minyobates, which retained the plesiomorphic states not found in Dendrobates. Dendrobates and Phyllobates were claimed to be sister groups based on “the loss of cephalic amplexus (cephalic embrace sometimes retained in an aggressive context), loss of the primitive oblique lateral line, and first appearance of 3,5-disubstituted indolizidine alkaloids” (Myers, 1987: 305).

Myers and Burrowes (1987) named Epipedobates andinus and postulated that its nearest relative was E. erythromos based on “a few similarities of the color patterns” and “an overall morphological similarity” (Myers and Burrowes, 1987: 16). They followed Vigle and Miyata (1980) in tentatively placing these species in Silverstone's (1976) pictus group. Given their placement in this group, indirect evidence for the close relationship of E. andinus and E. erythromos not cited by Myers and Burrowes is given by their occurrence on the Pacific slopes in contrast to the cis-Andean distribution of the remainder of the pictus group. Myers and Burrowes (1987) also transferred Phyllobates azureiventris to Epipedobates, also in the pictus group.

Zimmermann and Zimmermann (1988) performed a phenetic analysis of 62 characteristics (mostly behavioral, but also including vocalizations and larval morphology) for 32 species. Their analysis resulted in nine groups of decreasing similarity:

  • Colostethus group: C. inguinalis, C. collaris, C. trinitatis, C. palmatus

  • Epipedobates pictus group: E. pulchripectus, E. pictus, E. parvulus

  • Epipedobates tricolor group: E. anthonyi, E. boulengeri, E. espinosai, E. tricolor

  • Epipedobates silverstonei group: E. bassleri, E. silvestonei, E. trivittatus

  • Epipedobates femoralis group: E. femoralis

  • Phyllobates terribilis group: P. lugubris, P. terribilis, P. vittatus

  • Dendrobates leucomelas group: D. auratus, D. azureus, D. leucomelas, D. tinctorius, D. truncatus

  • Dendrobates quinquevittatus group: D. fantasticus, D. imitator, D. quinquevittatus, D. reticulatus, D. variabilis

  • Dendrobates histrionicus group: D. granuliferus, D. histrionicus, D. lehmanni, D. pumilio, D. speciosus

Furthermore, Zimmermann and Zimmermann (1988) proposed Phobobates for their silverstonei group (viz., Dendrobates bassleri, D. silverstonei, and Hyla trivittata) and Allobates for the monotypic femoralis group. However, Schulte (1989: 41) and Myers et al. (1991: 18) rejected those genera, citing errors in the analysis of behavior, lack of evidence, unaccounted character conflict, incorrect character coding, and creation of paraphyly.

In 1989, the Colostethus collaris group, delimited by “a dark band present on the posterior part of the throat and anterior part of the chest in all members”, was proposed by La Marca (1989: 175) for C. collaris, C. oblitteratus (as C. guatopoensis), C. herminae, C. neblina, C. olmonae, C. riveroi, C. trinitatis, and C. yustizi.

Over 60 years after the only previous specimen had been collected, Schulte (1990) rediscovered Dendrobates mysteriosus from Amazonian Peru. Despite some similarities, Schulte (1990: 66) determined that it was necessary to exclude D. mysteriosus from the quinquevittatus group (sensu Silverstone, 1975a, presumably), and he further stipulated that it was not closely related to D. captivus as proposed by Myers (1982). Rather, Schulte (1990: 67) believed D. mysteriosus to be most closely related to D. histrionicus from the lowlands of Pacific Ecuador and Colombia. He based this on shared size, absence of omosternum, occurrence of round spots on a dark background, reproductive behavior, an elevated number of small ova, and a similar fundamental frequency of the call, although no audiospectrographic analysis was performed. None of these characters is unique to the histrionicus group, and several other reported character states conflict with this relationship (e.g., larval mouth parts).

Rivero (1990 “1988”) selectively extracted data from Edwards's unpublished dissertation (1974a) and arranged the species of Colostethus into eight groups, which were soon expanded to nine by Rivero and Serna (1989 “1988”). Numerous species were not placed in any group because of apparent character conflict and other concerns. Although these groups were putatively based on derived characters and were hypothesized to be monophyletic, “characteristics shared by the majority of members” and geographic distribution were attributed evidential significance (Rivero, 1990 “1988”: 4). The content of the groups (as modified by Rivero and Serna, 1989 “1988” and augmented by Rivero and Granados-Díaz, 1990 “1989”; Rivero, 1991a ,1991b; Rivero and Almendáriz, 1991; Rivero and Serna, 1991, 2000 “1995”; La Marca, 1998 “1996”) was as follows:

  • Group I (vertebralis group): C. elachyhistus, C. exasperatus, C. idiomelus, C. infraguttatus, C. mittermeieri, C. peculiaris, C. shuar, C. sylvaticus, C. vertebralis

  • Group II (brunneus group): C. brunneus, C. intermedius [ =  C. kingsburyi fide Coloma, 1995], C. kingsburyi, C. marchesianus, C. olfersioides, C. peruvianus, C. talamancae, C. trilineatus

  • Group III (alagoanus group): C. alagoanus, C. capixaba, C. carioca

  • Group IV (inguinalis group): C. agilis, C. alacris, C. brachistriatus [as C. brachystriatus], C. cacerensis [ =  inguinalis fide Grant, 2004], C. dysprosium, C. erasmios, C. fallax, C. fraterdanieli, C. inguinalis, C. latinasus, C. mertensi, C. nubicola, C. paradoxus [ =  Epipedobates tricolor fide Coloma, 1995], C. pratti

  • Group V (edwardsi group): C. edwardsi, C. ruizi

  • Group VI (fuliginosus group sensu stricto; i.e., sensu Rivero and Serna, 1989 “1988”): C. abditaurantius, C. betancuri, C. chocoensis, C. excisus, C. faciopunctulatus, C. fuliginosus, C. furviventris, C. maculosus [ =  C. bocagei fide Coloma, 1995], C. nexipus, C. palmatus, C. pseudopalmatus, C. ramirezi (?), C. shrevei, C. thorntoni, C. vergeli

  • Group VII (trinitatis group): C. collaris, C. neblina, C. oblitteratus, C. olmonae, C. riveroi, C. trinitatis

  • Group VIII (alboguttatus group): C. alboguttatus, C. duranti, C. haydeeae, C. mayorgai, C. molinarii, C. orostoma, C. saltuensis, C. serranus

  • Group IX (subpunctatus group): C. anthracinus, C. borjai, C. cevallosi, C. citreicola [ =  C. nexipus fide Coloma, 1995], C. degranvillei, C. festae, C. jacobuspetersi, C. lehmanni, C. marmoreoventris, C. mystax, C. parcus [ =  C. exasperatus fide Coloma, 1995], C. pinguis, C. poecilonotus, C. pumilus, C. ramirezi (?), C. ramosi, C. ranoides, C. sauli, C. subpunctatus, C. taeniatus [ =  C. pulchellus fide Coloma, 1995], C. tergogranularis [ =  C. pulchellus fide Coloma, 1995], C. torrenticola [ =  C. jacobuspetersi fide Coloma, 1995], C. whymperi, C. yaguara

Among these groups, Rivero (1990 “1988”: 26) hypothesized that the brunneus group formed (or was close to) the “ancestral stock” from which the other Colostethus were derived. On the same page, he also hypothesized that the brunneus group gave rise to the inguinalis group (see also Rivero, 1991a: 23). He postulated that the fuliginosus group (sensu lato; fuliginosus + subpunctatus groups of Rivero and Serna, 1989 “1988”) was derived from the inguinalis group, and that the members of the fuliginosus group that lack toe webbing “could be close to the ancestral stock that gave rise to [the vertebralis group].” The edwardsi group was conjectured to have arisen from the fuliginosus group (sensu lato), and the alboguttatus group was believed to have arisen from the same ancestral stock as the edwardsi group. However, Rivero (1990 “1988”) also speculated that the trinitatis group (which was identical to La Marca's, 1989, collaris group) may have given rise to the alboguttatus group (which differed only slightly from La Marca's, 1985, alboguttatus group), citing putative intermediate forms as evidence. Rivero (1990 “1988”) was more ambivalent with regard to the relationships of the trinitatis group than he had been previously (Rivero, 1979). He now concluded that the trinitatis group may have arisen from the vertebralis group (as he had argued in 1979), or that both the trinitatis and vertebralis groups may have arisen from the fuliginosus group (sensu lato). Besides Rivero and his colleagues, few authors have recognized these groups (see Coloma, 1995).

Caldwell and Myers (1990) further elucidated the systematics of the Dendrobates quinquevittatus group, which had been revised previously by Myers (1982). In the process, they proposed that D. quinquevittatus sensu stricto was sister to D. castaneoticus, united by the synapomorphic absence of the inner metacarpal tubercle, as well as a number of character states of more ambiguous polarity. As a working hypothesis, they further proposed that this group was sister to a clade united by the synapomorphy of pale limb reticulation (i.e., D. fantasticus, D. quinquevittatus, D. reticulatus, D. vanzolinii), but they were unable to propose any synapomorphies to support this arrangement.

Myers et al. (1991) named a new genus and species, Aromobates nocturnus. They argued that this was the sister of all other dendrobatids on the basis of (1) nocturnal and (2) aquatic behavior, (3) large size, and (4) presence of m. adductor mandibulae externus superficialis in many specimens. They also proposed an informal redefinition of Colostethus based on the occurrence of the swollen third finger in adult males; they were explicit that they were not proposing formal nomenclatural changes. Their Colostethus sensu stricto corresponded with Rivero's (1990 “1988”) and Rivero and Serna's (1989 “1988”) inguinalis group with the addition of Phyllobates flotator and Colostethus imbricolus. The remaining species of Colostethus sensu lato were assigned to Hyloxalus (within which was included Phyllodromus), although no synapomorphies or diagnostic characters (besides the lack of the swollen third finger) were proposed. Almost immediately, Myers (1991; see also Myers and Donnelly, 1997: 25) retreated from this arrangement, given that the swollen third finger also occurs in some species of Epipedobates. Myers et al. (1991) provided a cladogram summarizing their views on the relationships of the dendrobatid frogs, reproduced here as figure 2.

Figure 2

Hypothesized phylogeny of dendrobatids, redrawn from Myers et al. (1991: 29, fig. 20).

All evidence is shown on the cladogram. In this scenario, Aromobates nocturnus is postulated to be the sister species of all other dendrobatids. All of the unquestioned synapomorphies listed for Dendrobatidae apply only to A. nocturnus and are unknown in any other dendrobatid.


In comparing Aromobates nocturnus to other dendrobatids, Myers et al. (1991) speculated that it may be most closely related to the collared species of Venezuelan Colostethus. They listed 10 species (1 undescribed) as definitely possessing a collar and 2 more as possibly having one. They did not define a group for these species, and their list of collared species differed from La Marca's (1989) collaris group ( =  trinitatis group of Rivero, 1990 “1988” and Rivero and Serna, 1989 “1988”) by including species of La Marca's alboguttatus group.

Also in 1991, Myers named Colostethus lacrimosus and, based on several similarities (but no clear synapomorphies), speculated that it may be closely related to C. chocoensis. He also suggested that they, in turn, were related to C. fuliginosus.

In a series of papers in the 1990s, La Marca proposed a number of novel relationships and taxonomic changes. In 1992 he formally named the collaris group as Mannophryne and later (La Marca, 1995) presented a hypothesis of relationships based on five characters from morphology and behavior, shown here as figure 3.

Figure 3

Hypothesized phylogeny of Mannophryne, redrawn from La Marca (1995: 70, fig. 11).

Synapomorphies are: A1, narrow collar, uniformly colored; B1, tadpoles with small papillae (presumably B1 on the cladogram is B0 from the text on p. 53); B2, tadpoles with large papillae (B2 is undefined in the text on p. 53; presumably it refers to B1); C1, uniformly colored dorsum; A2, wide collar without conspicuous pale markings; D1, posteroventral dark band present; A3, wide collar with pale flecks or spots; E1, bright throat coloration reduced, melanophores on anterior part of throat; A4, wide collar with large pale dots.


Although this study purported to be a quantitative cladistic analysis, few characters were used and some characters discussed by the author were ignored, not all characters were scored based on observations (i.e., some states were merely assumed), and monophyly and character polarity were assumed (i.e., no outgroup species were included). In addition to the species originally placed in Mannophryne, the genus currently includes M. caquetio, M. cordilleriana, M. larandina, and M. lamarcai (Mijares-Urrutia and Arends R., 1999).

In discussing the systematics of Colostethus mandelorum (about which he only concluded that the species is not closely related to either Mannophryne or the C. alboguttatus group), La Marca (1993) considered Aromobates nocturnus to be most closely related to the C. alboguttatus group of La Marca (1985). Regardless, instead of transferring the alboguttatus group into Aromobates, La Marca (1994) named it Nephelobates. The group was delimited by the occurrence of elongate teeth (also reported for Aromobates; see Myers et al., 1991; La Marca, 1993) and a dermal covering of the cloaca (also reported for the edwardsi group; Lynch, 1982), and it included N. alboguttatus, N. haydeeae, N. leopardalis, N. mayorgai, N. meridensis, N. molinarii, and N. orostoma; Mijares-Urrutia and La Marca (1997) subsequently included N. duranti and N. serranus. La Marca (1994) did not include C. saltuensis, which had been included in Rivero's alboguttatus group (Rivero, 1990 “1988”), but he did not state his reasons for its exclusion. No explicit hypothesis of relationships has been proposed for the species of Nephelobates, but Mijares-Urrutia and La Marca (1997) reported several larval character-states of unclear polarity, as well as the occurrence of “reduced nasal bones” (p. 134) as a synapomorphy for the genus.

Kaiser et al. (1994) described Colostethus chalcopis from Martinique in the French Antilles. Although they were skeptical of the monophyly of Mannophryne, they conjectured that C. chalcopis could be the sister species to that assemblage on the basis of the shared occurrence of a dark throat collar.

Although Coloma (1995) did not intend to provide a phylogenetic hypothesis for Colostethus, the taxonomic changes he made had numerous phylogenetic implications. For example, some of the species he considered to be synonymous had been placed in different and presumably distantly related groups by Rivero (e.g., Rivero and Almendáriz, 1991, placed C. nexipus in the fuliginosus group, whereas its junior synonym C. citreicola was placed in the subpunctatus group), which called into question the phylogenetic validity (or even taxonomic utility) of those groups. Coloma (1995: 58) also summarized the recognized species groups of Colostethus, arguing that “most of the character states given by Rivero [1990 “1988”] and Rivero and Serna [1989 “1988”] seem to be plesiomorphic at the level used.” Although he concluded that “the phylogenetic relationships within ‘Colostethus’ (sensu lato) constitute an enormous polytomy” (Coloma, 1995: 60), Coloma tentatively supported the following relationships:

  • Some species of Colostethus may be more closely related to some species of Epipedobates (based on the shared occurrence of a swollen third finger in adult males) than to other species of Colostethus.

  • Species in Aromobates, Mannophryne, and the vertebralis and fuliginosus groups may be basal within Colostethus.

  • The edwardsi group is monophyletic.

  • A novel group composed of Aromobates nocturnus, Colostethus awa, C. bocagei, C. nexipus, and C. riveroi may be monophyletic on the basis of shared (albeit facultative) nocturnal behavior. Myers et al.'s (1991) claim that nocturnal activity is plesiomorphic was not addressed.

Grant et al. (1997) reviewed the distribution of the median lingual process in dendrobatids and other frogs. The occurrence of the median lingual process in a putative sister group (see Interfamilial Relationships, below) led them to interpret it tentatively as symplesiomorphic and, consequently, they did not use it to delimit a group within Dendrobatidae.

Kaplan (1997) followed Silverstone (1975a) in studying the distribution of the palatine (neopalatine of Trueb, 1993), and he used these data to further resolve Myers et al.'s (1991) hypothesis of relationships (and he explicitly incorporated Mannophryne and Nephelobates). He concluded that the absence of the palatine delimits a clade composed of part of Hyloxalus sensu Myers et al. (1991), Colostethus sensu stricto, and the aposematic dendrobatids. Kaplan (1997) presented a cladogram, shown here as figure 4. The separate treatment of Epipedobates was due to the presence of a swollen third finger in some species of that genus (Myers, 1991).

Figure 4

Hypothesized phylogeny of dendrobatids, redrawn from Kaplan (1997: 373, fig. 3).

Numbered synapomorphies are: (1) tympanum posterodorsally tilted under anterior edge of massive superficial slip of m. depressor mandibulae, (2) mercaptanlike defensive odor, (3) diurnal activity, (4) riparian–terrestrial habitat preference, (5) smaller size (<50 mm SVL), (6) m. adductor mandibulae externus superficialis absent (“s” pattern), (7) neopalatines absent, (8) finger three of males swollen, and (9) lipophilic alkaloids present.


La Marca (1998 “1996”) reviewed the species of Guayanan Colostethus and assigned C. ayarzaguenai, C. guanayensis, C. murisipanensis, C. parimae, C. parkerae, C. praderioi, C. roraima, C. sanmartini, C. shrevei, and C. tepuyensis to the fuliginosus group sensu lato (i.e., sensu Rivero, 1990 “1988”). He did not note the occurrence of the median lingual process, although it is present in several of these species.

Grant and Castro (1998) proposed the Colostethus ramosi group based on the occurrence of a patch of black, apparently glandular tissue on the ventral and medial surfaces of the distal extreme of the upper arm, just proximal to the elbow (referred to by Grant and Castro as the black arm band). This group presently includes C. cevallosi, C. exasperatus, C. fascianiger, C. lehmanni, C. ramosi, and C. saltuarius (Grant and Ardila-Robayo, 2002), but this character-state also occurs in C. anthracinus and an undescribed species from the slopes of the Magdalena valley, Colombia (T. Grant, personal obs.).

Schulte's (1999) book on Peruvian Dendrobates and Epipedobates included a number of novel phylogenetic arrangements, many of which involved non-Peruvian species as well. Lötters and Vences (2000) strongly criticized many of Schulte's (1999) taxonomic conclusions, and below we exclude the nomina nuda and taxa they placed in synonymy. Schulte (1999) proposed eight groups of Dendrobates and six groups of Epipedobates, and he provided branching diagrams depicting the relationships of each (Schulte, 1999: 24–25, 160–161). The groups he proposed are as follows:


  • Group 1 (amazonicus): D. amazonicus, D. duellmani, D. fantasticus, D. variabilis

  • Group 2 (quinquevittatus): D. quinquevittatus, D. castaneoticus, D. flavovittatus

  • Group 3 (imitator): D. imitator

  • Group 4 (vanzolinii): D. biolat, D. lamasi, D. vanzolinii

  • Group 5 (ventrimaculatus): D. ventrimaculatus

  • Group 6: D. reticulatus, D. rubrocephalus, D. sirensis, D. steyermarki, D. (M.) virolinensis [sic]

  • Group 7: D. captivus

  • Group 8 (histrionicus): D. histrionicus, D. lehmanni, D. mysteriosus


  • Group 1 (giant types [“Riesenarten”]): E. bassleri, E. planipaleae, E. silverstonei, E. trivittatus

  • Group 2 (petersi/pictus): petersi subgroup: E. cainarachi, E. labialis, E. macero, E. petersi, E. pongoensis, E. smaragdinus, E. zaparo; pictus subgroup: E. bolivianus, E. hahneli, E. pictus, E. rubriventris

  • Group 3 (azureiventris): E. azureiventris, Phyllobates [i.e., Phyllobates sensu Myers et al., 1978]

  • Group 4 (femoralis): E. femoralis, E. ingeri, E. labialis, E. myersi, E. zaparo

  • Group 5 (parvulus): E. espinosai, E. parvulus

  • Group 6 (tricolor): E. anthonyi, E. espinosai, E. parvulus, E. subpunctatus, E. tricolor

Rather than detail exhaustively the relationships Schulte (1999) proposed, we limit ourselves to pointing out a few of his more heterodox hypotheses. Without comment he transferred Prostherapis subpunctatus from Colostethus (where it had been placed by Edwards, 1971) to Epipedobates as sister species to E. anthonyi and E. tricolor. Also without comment, he referred Dendrobates steyermarki and Minyobates virolinensis—both of which had been in Minyobates (Myers, 1987; Ruiz-Carranza and Ramírez-Pinilla, 1992)—to Dendrobates, but he did not discuss the relationships of the remaining species of Minyobates. Further, according to his own diagrams he rendered Epipedobates paraphyletic by grouping E. azureiventris with species of Phyllobates. Schulte redefined the histrionicus group to include D. mysteriosus, but he excluded most of the species Myers et al. (1984)—and even Silverstone (1975a) and Myers and Daly (1976b)—had referred to that group, and he once again placed D. leucomelas in that group (as had Silverstone, 1975a). Relationships among most groups were not specified, but some groups (e.g., Groups 1 and 7) were paraphyletic in Schulte's own diagrams, and some species (e.g., D. labialis and D. zaparo; E. parvulus and E. espinosai) were included in multiple groups, with their relationships to each other being different in each group. No new character systems were added in this study, and, although Schulte provided limited group diagnoses and details on natural history, behavior, coloration and color patterns, and external morphology, no explicit synapomorphies were provided for any of his groups.

Grant (1998) named Colostethus lynchi and argued that it was part of the C. edwardsi group on the basis of the occurrence of a cloacal tube (he did not address the occurrence of this character in Nephelobates). More specifically, he argued that C. lynchi was the sister species to the group of C. edwardsi + C. ruizi.

The first attempt to address phylogenetic relationships among dendrobatids with DNA sequence data was published by Summers et al. (1997), although that paper only included the distantly related Dendrobates pumilio, Dendrobates claudiae (as Minyobates sp.), and Phyllobates lugubris (plus C. talamancae, used as the root). Since 1999, nearly a dozen phylogenetic studies of differing scales, scopes, and datasets have appeared (Summers et al., 1999b; Clough and Summers, 2000; Vences et al., 2000, 2003a; Widmer et al., 2000; Symula et al., 2001, 2003; La Marca et al., 2002; Santos et al., 2003). The cladograms that resulted from those studies are redrawn in figures 5Figure 6Figure 7Figure 8Figure 9Figure 10Figure 11Figure 1213. Interpretation of these studies is complicated by their use of different methods, nonoverlapping taxon samples, and heterogeneous datasets, but their findings can be summarized as follows:

Figure 5

Hypothesized phylogeny of Dendrobates, redrawn from Summers et al. (1999b: 261, fig. 1), based on parsimony analysis of cytochrome oxidase I, cytochrome b, and 16S DNA sequences aligned with Clustal W (Thompson et al., 1994) (parameters not specified) and modified manually, excluding ambiguously aligned regions.

Numbers are bootstrap frequencies (unlabeled nodes present in fewer than 50% of replicates).


Figure 6

Hypothesized phylogeny of dendrobatids, redrawn from Clough and Summers (2000: 342, fig. 1), based on parsimony analysis of 12S, 16S, and cytochrome b DNA sequences aligned with Clustal W (Thompson et al., 1994) (parameters not specified) and modified by eye and excluding ambiguously aligned regions.

Numbers are bootstrap frequencies (unlabeled nodes present in fewer than 50% of replicates).


Figure 7

Hypothesized phylogeny of dendrobatids, redrawn from Vences et al. (2000: 37, fig. 1), based on neighbor-joining analysis of 16S DNA sequences aligned manually and excluding highly variable regions.

Numbers are bootstrap frequencies (unlabeled nodes present in fewer than 50% of replicates).


Figure 8

Hypothesized phylogeny of dendrobatids, redrawn from Widmer et al. (2000: 561, fig. 2), based on parsimony analysis of cytochrome b sequences aligned with Clustal W (Thompson et al., 1994) (parameters not specified).

Numbers are parsimony/maximum likelihood/neighbor-joining bootstrap frequencies.


Figure 9

Cladogram summarizing species level phylogeny of Dendrobates (reduced from 30 terminals) proposed by Symula et al. (2001: 2419, fig. 3), based on maximum likelihood analysis (under the GTR + Γ + I model) of cytochrome b, cytochrome oxidase I, 12S, and 16S DNA sequences aligned with ClustalX (Thompson et al., 1997) (parameters not specified) and “a few regions of ambiguous alignment … removed from the analysis”.

Numbers are parsimony bootstrap frequencies.


Figure 10

Hypothesized phylogeny of dendrobatids, redrawn from La Marca et al. (2002: 239, fig. 4), based on maximum likelihood analysis (model not specified) of 16S DNA sequences (alignment method not stated), excluding hypervariable regions.

Numbers are maximum likelihood/parsimony/neighbor-joining bootstrap frequencies.


Figure 11

Hypothesized phylogeny of dendrobatids, redrawn from Santos et al. (2003: 12794, fig. 1), based on unweighted parsimony analysis of the mitochondrial transcription unit H1 (ca. 2,400 bp), aligned with ClustalX (Thompson et al., 1997) “under various parameters … and finally adjusted by eye to produce a parsimonious alignment” whereby “informative sites were minimized” (Santos et al., 2003: 12793).

Parsimony bootstrap frequencies shown above branches, frequency of clades among trees sampled in Bayesian analysis shown below branches. Stars indicate clades found in ≥95% of the replicates or sampled trees.


Figure 12

Hypothesized phylogeny of Dendrobates, redrawn from Symula et al. (2003: 459, fig. 3), based on maximum likelihood (under the GTR + Γ model) analysis of cytochrome b and cytochrome oxidase I DNA sequences aligned with ClustalX (Thompson et al., 1997) (parameters not specified).

Maximum likelihood branch lengths shown above branches, parsimony bootstrap frequencies shown below branches (frequencies >75% shown).


Figure 13

Phylogeny of dendrobatids redrawn from Vences et al. (2003a: 219, fig. 3), based on maximum likelihood analysis (under the GTR + I + Γ model) of 368 bp of 16S rDNA aligned “using the clustal option of Sequence Navigator (Applied Biosystems)” (parameters not specified) and excluding “all regions of the gene fragments that could not be clearly aligned among all taxa” (p. 217).

Numbers above branches are neighbor-joining bootstrap frequencies (frequencies >50% shown); numbers below branches are frequency of clades among trees sampled in Bayesian analysis (frequencies >70% shown).

  • Colostethus: Found to be either para- or polyphyletic by all authors who tested its monophyly.

  • Epipedobates: Found to be monophyletic by Clough and Summers (2000) (with femoralis placed outside in Allobates) but polyphyletic by Vences et al. (2000, 2003a; see also Santos et al., 2003).

  • Phobobates: Found to be monophyletic by Vences et al. (2000) but paraphyletic by Clough and Summers (2000), Santos et al. (2003), and Vences et al. (2003a).

  • Allobates: This small genus fell out in a clade with species of Colostethus in Vences et al. (2000, 2003a) and Santos (2003). (Jungfer and Böhme, 2004 added the enigmatic Dendrobates rufulus to Allobates, but that species has not been included in any analysis.)

  • Phyllobates: Without exception, this genus was found to be monophyletic. The optimal topology found by Widmer et al. (2000) was ((vittatus lugubris) (aurotaenia (bicolor terribilis))) (outgroup taxa were Epipedobates azureiventris and Dendrobates sylvaticus, and the tree was rooted on E. azureiventris). In their more inclusive study, Vences et al. (2003a) found P. aurotaenia to be the sister of the remainder, and P. bicolor to be sister to the Central American species, giving the topology (aurotaenia (terribilis (bicolor (lugubris vittatus)))).

  • Minyobates: Both Clough and Summers (2000) and Vences et al. (2000) found Minyobates to be nested within Dendrobates. Because each analysis used only one species of Minyobates, they did not test the monophyly of Minyobates itself. Vences et al. (2003a) included M. steyermarki (type species), M. minutus, and M. fulguritus and found Minyobates to be paraphyletic with respect to all other Dendrobates. Santos et al. (2003) included M. minutus and M. fulguritus and found them to be the monophyletic sister to the D. quinquevittatus group (i.e., they recovered a monophyletic minutus group sensu Silverstone, 1975b).

  • The Dendrobates histrionicus group is monophyletic in all studies that test its monophyly.

  • The Dendrobates quinquevittatus group is potentially monophyletic. Although the tree presented by Clough and Summers (2000: 524) indicates that Minyobates minutus is the sister species of a monophyletic D. quinquevittatus group, there is in fact no evidence to support this assertion, given that these nodes collapse in the strict consensus. Symula et al. (2003) found Dendrobates leucomelas to be sister to part of the D. quinquevittatus group, with a D. quinquevittatus + D. castaneoticus clade in a basal trichotomy (they rooted the network with D. histrionicus, so it is unknown from their results if D. quinquevittatus + D. castaneoticus or D. histrionicus is more closely related to the D. leucomelas + other D. quinquevittatus group clade).

  • Nephelobates and Mannophryne were both found to be monophyletic by La Marca et al. (2002) and Vences et al. (2003a).

Lötters et al. (2000) erected the new genus Cryptophyllobates for Phyllobates azureiventris (which was placed in Epipedobates by Myers and Burrowes, 1987). The justification for this monotypic genus is somewhat convoluted. On pp. 235–236, the authors stated that “from the genetic point of view, it is apparent that azureiventris is more closely related to Epipedobates than to Phyllobates”, but that “the species is not a member of Epipedobates, from which it differs by at least one apomorphy.” However, they also asserted that “[i]t shares more—but not all—characters with Phyllobates from which it appears genetically well separated.” Similarly, although Vences et al. (2000) found this species to be the sister of Colostethus bocagei, Lötters et al. (2000) stated that they “negate that both species are representatives of the same genus for C. bocagei is cryptically coloured, lacking dorsal stripes at all, and possesses webbed feet.” Insofar as this change did not fix the nonmonophyly of Epipedobates, the creation of this monotypic genus did little to improve matters. Recently, Caldwell (2005) described a new species and referred it to this genus based on the sister-species relationship between the new species and azureiventris recovered in an independent phylogenetic study, despite noting that these species “were nested in a clade of Ecuadorian and Peruvian Colostethus.”

Morales (2002 “2000”) combined Rivero's Groups II (brunneus) and III (alagoanus) into a newly defined trilineatus group (which excluded C. kingsburyi and C. peruvianus) on the basis of an analysis of 12 characters. However, in addition to problems of character individuation (e.g., characters 6, “línea lateral oblicua”, and 10, “lista oblicua anteroinguinal”, are logically dependent; see Grant and Rodríguez, 2001), the monophyly of the group was assumed (the cladogram was rooted on an unspecified “Hylodes” and no other dendrobatids were included), and the putatively derived states of all 12 characters are well known to occur in many dendrobatids.

In the most recent contribution to dendrobatid phylogenetics, Graham et al. (2004) added 12S, tRNAval, and 16S mtDNA sequences from a specimen collected near the type locality of Epipedobates tricolor and analyzed them with the data from Santos et al. (2003). Graham et al. reported that the E. tricolor sample from southern Ecuador was more closely related to Colostethus machalilla than to true E. tricolor, and, as such, they resurrected E. anthonyi from synonymy with E. tricolor. However, the Bremer support value1 reported for the critical node is 0, meaning that this relationship is not recovered in other, equally parsimonious solutions.

Part III: 1926–present, Relationships between Dendrobatidae and Other Frogs

Noble (1931) summarized his research on the evolutionary relationships of anurans. He considered the three genera of dendrobatids, which he had grouped together in his earlier paper (Noble, 1926), to be a subfamily of Brachycephalidae. The other subfamilies were Rhinodermatinae (Geobatrachus, Sminthillus, and Rhinoderma) and Brachycephalinae (Atelopus, Brachycephalus, Dendrophryniscus, and Oreophrynella). Noble (1931: 505; see Grant et al., 1997: 31, footnote 18) maintained his curious view that independently derived groups may constitute natural assemblages:

Each subfamily has arisen from a different stock of bufonids, but as all the ancestral stocks were bufonids residing in the same general region, Brachycephalidae may be considered natural, even though a composite, family.

Particularly, Noble (1931; see also Noble, 1926) reiterated that Dendrobatinae evolved from the elosiine bufonid genus Crossodactylus. Brachycephalidae was included in the suborder Procoela, which also contained Bufonidae and Hylidae, as well as the extinct Palaeobatrachidae.

Noble was aware that his placement of Brachycephalidae in Procoela instead of Diplasiocoela could be viewed as problematic. He (Noble, 1931: 514) pointed out that the frogs he referred to Diplasiocoela “differ strikingly from most other Salientia except Brachycephalidae”, but he reasoned that “[t]he latter are purely neotropical, and as the genera of Brachycephalidae are well defined, they should not be confused with the Diplasiocoela.” He also observed that both Dendrobatinae and the African ranid Petropedetinae (nested well within Diplasiocoela) had “apparently identical” (Noble, 1931: 520) dermal scutes on the upper surface of each digit, but he explained away this similarity as adding “one more to the many cases of parallel evolution in the Salientia”.

Although Noble's general scheme was widely accepted as the standard for decades (e.g., Dunlap, 1960), it attracted extensive criticism almost immediately. For example, Trewavas (1933: 517) concluded that the hyolaryngeal apparatus provided “little support for the inclusion of Dendrobates in the family [Brachycephalidae]” and recommended that the relationships of the family be reconsidered. Davis (1935: 91) criticized Noble's belief that independently derived taxa could be grouped naturally, and he raised each of Noble's (1931) brachycephalid subfamilies (i.e., Brachycephalinae, Dendrobatinae, and Rhinodermatinae) to family rank. Laurent (1942: 18; translated, italics as in original) concluded that the similarities in the initial phases of parental care of larvae in dendrobatids (tadpoles are transported on the male's back) and rhinodermatids (tadpoles are transported in the male's mouth) “constituted a weighty argument in favor of the common ancestry of the Rhinodermatinae and the Dendrobatinae”, and he included both in Dendrobatidae. Orton (1957; see also Orton, 1953) was highly critical of Noble's system because it conflicted with larval morphology; but, beyond her rejection of suborder rank within Salientia, dendrobatids were unaffected. Likewise, Reig (1958) incorporated evidence from a variety of previous studies (e.g., Trewavas, 1933; Davis, 1935; Walker, 1938; Taylor, 1951; Griffiths, 1954) and his own fossil work to provide an extensively modified higher taxonomy, but the placement of Dendrobatidae was unaffected (i.e., Reig's neobatrachian “Superfamily A” [now Hyloidea,  =  Bufonoidea auctorum] was identical to Noble's Procoela with the exclusion of Palaeobatrachidae).

Griffiths (1959, 1963) provided the first major challenge to Noble's placement of Dendrobatidae. Griffiths (1959) reviewed Noble's (1922, 1926, 1931) evidence that dendrobatids were part of Procoela and related to Crossodactylus, and, arguing that (1) “vertebral pattern has not the exact taxonomic validity vested in it by Noble” (p. 481); (2) path of insertion of the m. semitendinosus is ranoid in Hyloxalus; (3) “Noble's claim that Phyllobates has an arciferal stage cannot be held” (p. 482); (4) the bursa angularis oris is found only in firmisternal genera; (5) dermal scutes (which he claimed to be “glandulo-muscular organs”) on the digits occur in petropedetid ranids (as well as Crossodactylus); and (6) the breeding habits of dendrobatids “are found in no other Salientia except in the arthroleptid ranids” (p. 483), proposed “that the Dendrobatinae be redefined as a Neotropical subfamily of the Ranidae” (p. 483). Subsequent reviews either explicitly endorsed (e.g., Hecht, 1963: 31) or did not address (e.g., Tihen, 1965; Inger, 1967; Kluge and Farris, 1969) Griffiths's hypothesis of the relationships of dendrobatids.

However, Lynch (1971: 164; see also Lynch, 1973) supported Noble's hypothesis, arguing that elosiines (including Crossodactylus) and dendrobatids “agree in cranial morphology, vertebral columns, the T-shaped terminal phalanges, the dermal glandular pads on top of the digital pads, and in the presence, in at least some species of each group, of toxic skin secretions” (see fig. 14).

Figure 14

Hypothesized phylogeny of anurans, redrawn from Lynch (1973).


Lynch (1971: 164) also asserted that Crossodactylus and dendrobatids exhibit the ranoid pattern of thigh musculature, which “mitigates the importance of one of the criteria used by Griffiths (1963) to associate the dendrobatids as a Neotropical subfamily the Ranidae.” Interestingly, Lynch (1971: 210–211) also indicated that there was “considerable similarity in myology and osteology” between the Neotropical leptodactyloid Elosiinae and Dendrobatidae and the African ranid subfamily Petropedetinae. Further, although he cautioned that his examination of the African taxa was not exhaustive, he stated that “[t]he similarities are quite striking and probably reflect a community of ancestry rather than parallelism.”

Lynch's (1971, 1973) resurrected version of Noble's (1926) hypothesis stood for 15 years. For example, although Savage (1973) adopted Starrett's (1973) scheme of higher level relationships and did not discuss dendrobatid phylogeny per se, he followed Lynch (1971) in considering Dendrobatidae to be a South American, tropical, leptodactyloid derivative. Bogart (1973: 348) conjectured that “Dendrobatidae may be derived chromosomally from a 26-chromosome ancestor, such as the leptodactylid Elosia” (although he did not examine any African ranoid species for comparison). Duellman (1975) included Dendrobatidae in Bufonoidea (though not explicitly with Crossodactylus). Ardila-Robayo (1979) evaluated 68 characters and found two equally parsimonious topologies, both of which showed Dendrobatidae (“Phyllobatinae”; see Dubois, 1982, and Holthius and Dubois, 1983, for discussion of nomenclature) to be related to elosiines. Like Duellman (1975), Laurent (“1979” 1980) and Dubois (1984) did not address dendrobatid relationships specifically, but they included Dendrobatidae in Bufonoidea (except that the latter replaced Bufonoidea with the senior synonym Hyloidea).

Both the ranoid and hyloid hypotheses have suffered from mistaken observations. Against Griffiths (1959), Kaplan (1997) confirmed Noble's (1926) claim that the pectoral girdle of Colostethus subpunctatus overlaps in ontogeny (which had been denied by Griffiths), and Silverstone (1975a) and Grant et al. (1997) showed that Griffiths's claims regarding dendrobatid thigh musculature were also false. Against Lynch (1971), the thigh musculature in hylodines conforms with Noble's (1922) “bufonid” pattern, not the dendrobatid pattern (Grant et al., 1997: 31; see also Dunlap, 1960), and no species of hylodine tested by Myers and Daly was found to possess lipophilic alkaloids (Grant et al., 1997).

Fifteen years after Lynch (1971) resurrected the hyloid hypothesis, Duellman and Trueb (1986) resurrected the ranoid one. Based on a cladistic analysis of 16 characters, they placed Dendrobatidae in a ranoid polytomy, unrelated to South American leptodactylids. Myers and Ford (1986) did not address the phylogenetic position of dendrobatids, but they listed a number of diagnostic character-states for Dendrobatidae, including (1) the posterodorsal portion of the tympanum concealed beneath the massive superficial slip of the m. depresssor mandibulae, (2) the alary processs of the premaxilla tilted anterolaterally, (3) occurrence of a retroarticular process on the mandible, (4) absence of m. adductor mandibulae externus, (5) single anterior process on hyale, (6) the occurrence of digital scutes, and (7) the m. semitendinosus tendon of insertion piercing the tendon of the m. gracilis.

Shortly thereafter, Ford (1989) completed her doctoral dissertation on the phylogenetic position of Dendrobatidae, based on 124 osteological characters, and found that the most parsimonious solution placed Dendrobatidae as the sister taxon of the Old World ranoid family Arthroleptidae. That dissertation remains unpublished, but it was summarized by Ford and Cannatella (1993; see also Ford, 1993). They reiterated the dendrobatid synapomorphies given by Myers and Ford (1986) and cited Ford's dissertation as finding that “dendrobatids were nested within Ranoidea, close to arthroleptid and petropedetine ranoids” (p. 113; see fig. 15), but they did not list any synapomorphies in support of that hypothesis.

Figure 15

Hypothesized phylogeny of neobatrachians, redrawn from Ford and Cannatella (1993).


The phylogenetic position of Dendrobatidae alternated between the ranoid and hyloid hypotheses through the 1990s. Bogart (1991: 251–252) compared karyotypes, average measurements, and idiograms of several species of petropedetids and hylodines with dendrobatids and concluded that “Hylodes and other hylodine leptodactylids have the more similar karyotypes to the dendrobatid frogs.” Blommers-Schlösser's (1993) redefined Ranoidea excluded Dendrobatidae, but she still considered Dendrobatidae to be part of the more inclusive “firmisternal frogs” group, which is equivalent to Ranoidea sensu lato. However, Blommers-Schlösser (1993) also proposed the novel hypothesis that Dendrobatidae is most closely related to microhylids, brevicipitids, and hemisotids in her Microhyloidea group. Hillis et al.'s (1993) combined analysis of the morphological data from Duellman and Trueb (1986) and their own 28S rDNA sequence data indicated that the hyloid hypothesis was more parsimonious than the ranoid hypothesis. Hedges and Maxson's (1993) neighbor-joining analysis of 12S mitchondrial DNA (mtDNA) sequences also placed dendrobatids among hyloids, as did Hay et al.'s (1995) and Ruvinsky and Maxson's (1996) neighbor-joining analyses 12S and 16S mtDNA data. Haas (1995) described an additional dendrobatid synapomorphy (viz., proximal ends of Certatobranchialia II and III free, lacking synchondritic attachment). He failed to find evidence of a ranoid relationship, but discovered a number of character-states shared with hyloid taxa; however, these characters are of uncertain polarity, and no hylodine was included to rigorously test Noble's hypothesis. Grant et al. (1997) discovered that a median lingual process occurs in many Old World ranoid genera (including those thought to be most closely related to dendrobatids) and several species of dendrobatids but failed to detect it in any nonranoid frog. Burton (1998a) reported a synapomorphy in the musculature of the hand (absence of caput profundum arising from carpals) in Dendrobatidae, Hylodes, and Megaelosia, but not the putative ranoid relatives (but note that this state also occurs in part or all of Ascaphidae, Bombinatoridae, Discoglossidae, Heleophrynidae, Hemisotidae, Pipidae, and Sooglosidae).

Additional support for the ranoid hypothesis has not been proposed, as most studies this decade have found dendrobatids to be nested among hyloids, if not directly related to hylodines. Vences et al.'s (2000) analysis of 12S and 16S mtDNA sequence data showed Dendrobatidae to group with hyloids, not ranoids, as did Emerson et al.'s (2000) analysis of 12S, tRNAval, and 16S mtDNA data (although the latter authors also found Dendrobatidae to be polyphyletic, broken up by Bufo valliceps and Atelopus chiriquiensis). Haas's (2001) study of the mandibular arch musculature of larval and postmetamorphic amphibians included Phyllobates bicolor, which was found to possess the neobatrachian (plus pelobatid) synapomorphy (viz., presence of functionally differentiated m. levator mandibulae lateralis) and lack the three ranoid synapomorphies, hence leaving it in a “hyloid” polytomy. In an explicit cladistic analysis, Haas (2003) assembled a dataset composed of mostly larval characters (but including most traditionally important characters from adult morphology and behavior) and found Dendrobatidae to be sister to his two hylodine species (fig. 16). Vences et al. (2003a) also included two species of hylodines in their analysis of 12S and 16S mtDNA sequences, but they found dendrobatids to be sister to Telmatobius simonsi. Darst and Cannatella (2004) analyzed 12S, 16S, and tRNAval mtDNA sequences and found dendrobatids to be nested within Hylidae (parsimony result shown in fig. 17) or sister to a group consisting of ceratophryines, hemiphractines, and telmatobiines (maximum likelihood; topology not shown). Faivovich et al. (2005) were primarily interested in the relationships among hylids, but their outgroup sample was extensive; their analysis of nuclear and mitochondrial DNA placed dendrobatids as the sister group to the Hylodinae (topology not shown).

Figure 16

Hypothesized phylogeny of anurans, redrawn from Haas (2003), based on parsimony analysis of larval and adult morphology.


Figure 17

Hypothesized phylogeny of anurans, redrawn from Darst and Cannatella (2004: 465, fig. 1), based on parsimony analysis of mitochondrial transcription unit H1 (ca. 2,400 bp), aligned with ClustalX (Thompson et al., 1997) “under a variety of gap penalty weightings”, adjusted manually “to minimize informative sites under the parsimony criterion”, using secondary structure models to help align ambiguous regions, and excluding regions “for which homology of the sites could not be inferred” (Darst and Cannatella, 2004: 463).

Numbers are bootstrap frequencies.


Summary of Historical Review

The picture that emerges from the review of the history of dendrobatid systematics is one of considerable conflict and confusion. There is near universal support for the monophyly of the family, which has not been seriously challenged since it was first proposed by Noble 90 years ago, but the phylogenetic position of Dendrobatidae has alternated between two predominant hypotheses: (1) deeply embedded among ranoids as the sister to petropedetids or arthroleptids or (2) deeply embedded among hyloids as the sister to hylodines. Recent studies based on DNA sequences (mostly mtDNA) have favored the hyloid hypothesis, but there is extensive conflict in the details of both hypotheses. Within Dendrobatidae, the once uncontroversial monophyly of the aposematic taxa has been rejected by mtDNA studies, and there is little agreement on the monophyly and relationships among most genera. The monophyly of Phyllobates has been universally supported, although the relationships among its five species have not. To date, no study has combined DNA sequences with evidence from morphological, behavioral, and biochemical (alkaloid) sources, and all explicit phylogenetic analyses have included a limited sample of the diversity of dendrobatids.

Phylogenetic Placement of Dendrobatids and Outgroup Sampling

Theoretical Background

Although the present study was not designed primarily to test the relationships between dendrobatids and other anurans, that question is key to selecting an adequate sample of outgroup taxa to rigorously test the relationships (including monophyly) and transformation series among dendrobatids. That is, the position taken in this study is that all nondendrobatids constitute “the outgroup,” and outgroup taxa are sampled for the purpose of testing hypothesized patristic and cladistic relationships. Ideally, one would code all nondendrobatids for all included characters; however, given the practical impossibility of that ideal, prior knowledge of phylogeny and character variation must be used to inform sampling of those taxa most likely to falsify ingroup hypotheses (including ingroup monophyly), the scope and scale of outgroup sampling being limited primarily by practical limitations of time and resources (e.g., specimen and tissue availability, laboratory resources, computer power and time). The possibility always exists that expansion of the outgroup sample may lead to improved phylogenetic explanations—a consideration that points the way to increased testing in future research cycles.

Although this approach to outgroup sampling incorporates prior knowledge, it does so in an expressly non-Bayesian way. The effect of prior knowledge in Bayesian approaches is to constrain hypothesis preference toward prior beliefs about ingroup evolution. Here, prior knowledge is used heuristically to maximize the probability of falsifying prior beliefs about ingroup evolution (for discussion of heurism in phylogenetic inference see Grant and Kluge, 2003). That this “probability” is not frequentist, logical, or personal and is not formally quantifiable does not deny its relevance. The goal is to test phylogenetic hypotheses as severely as possible, and prior knowledge is key to that undertaking.

Empirical Background

As summarized above, the phylogenetic placement of Dendrobatidae is among the most controversial problems in anuran systematics. In part, this is because the two cladistic hypotheses that have emerged as the leading contenders are so radically contradictory, effectively placing dendrobatids at opposite extremes of the neobatrachian clade: dendrobatids are placed as sister to hylodine hyloids from South America or are allied to petropedetid or arthroleptid ranoids from Africa. Minimally, evaluation of these hypotheses would require a phylogenetic analysis of Neobatrachia, which was beyond the scope of the present study.

Nevertheless, in a recently completed study Frost et al. (2006) sampled 532 terminals for the mitochondrial H-strand transcription unit 1 (H1; composed of 12S ribosomal, tRNAval, and 16S ribosomal sequences), histone H3 (H3), tyrosinase, rhodopsin, seventh in absentia (SIA), 28S large ribosomal subunit, and Haas's (2003) morphological transformation series in a phylogenetic analysis of living amphibians. That study included approximately 9% of each of the “major” amphibian clades (caecilians, salamanders, and frogs), including eight species (and genera) of dendrobatids and all putative sister groups. Insofar as that study is the most extensive analysis of amphibian phylogeny undertaken to date, we used those results to inform outgroup sampling for the current study.

The Frost et al. (2006) study resulted in four trees of 126,929 steps, the relevant portion of which is shown in figure 18.

Figure 18

Phylogenetic placement of Dendrobatidae according to Frost et al. (2006).

The Frost et al. study sampled 532 terminals (direct optimization parsimony analysis under equal costs for all transformations; see text for details of dataset), 51 of which are included here to show the placement of Dendrobatidae with respect to its closest relatives. All of the terminals shown were included in the present study, including three representatives of Hylidae and two “other bufonids”. Those targeted for additional genotypic and phenotypic evidence are marked in the figure with an asterisk (see text for details). Numbers are Bremer support values.


Relevant to the present study, Frost et al. (2006) corroborated the monophyly of Dendrobatidae. Furthermore, dendrobatids were not found to be closely related to petropedetids, arthroleptids, or any other ranoid and were instead nested deeply among hyloid taxa. Specifically, Dendrobatidae was found to be sister to Thoropa, those taxa were sister to Bufonidae, and that inclusive clade was sister to Cycloramphidae (including Crossodactylus, Hylodes, and Megaelosia as the sister clade of the remaining cycloramphids). Alternative hypotheses of the placement of Dendrobatidae (e.g., placed in a clade with Crossodactylus, Hylodes, and Megaelosia, as favored by Noble, 1926; Lynch, 1971; Haas, 2003; Faivovich, 2005) were tested explicitly by inputting constraint topologies for diagnosis and swapping, but they all required additional transformations (breaking up the Thoropa + Dendrobatidae clade required at least 39 extra steps).

Although detailed knowledge of the placement of Thoropa did not exist prior to that analysis, its placement as the sister of Dendrobatidae is unconventional, to say the least. No morphological synapomorphies have been proposed to unite these taxa, and it was expected that Thoropa would be nested among cycloramphids. Nevertheless, insofar as this is the most parsimonious solution found in the most complete study of amphibian relationships carried out to date, the Frost et al. (2006) hypothesis provides the starting point for further testing. Also, the immediately relevant nodes of the Frost et al. (2006) tree are all well supported (Bremer support for Dendrobatidae + Thoropa  =  39, Dendrobatidae + Thoropa + Bufonidae  =  30); considering that Thoropa was only scored for the mtDNA and H3 loci (i.e., over 1,500 bp of nuDNA were missing), the Bremer value for the Thoropa + Dendrobatidae clade is remarkably high. Furthermore, the general placement of Dendrobatidae is reminiscent of (but not identical to) Noble's (1922) Brachycephalidae, which included the dendrobatids, Brachycephalus, Atelopus, Rhinoderma, Sminthillus (now a synonym of Eleutherodactylus), Geobatrachus, and Oreophrynella (the latter two genera not sampled by Frost et al.). According to Frost et al. (2006), Brachycephalus and Eleutherodactylus are part of the distantly related Brachycephalidae (not shown in fig. 18), but Atelopus and Rhinoderma are placed in the same general neighborhood as Dendrobatidae. As such, the results of Frost et al. (2006) provide both an objectively defensible and subjectively “reasonable” basis for outgroup sampling, and we therefore sampled outgroup taxa from among these closely related groups.

Outgroup Sampling

In light of Frost et al.'s (2006) findings, it is clear that dendrobatids are not closely related to the Old World ranoids and are instead nested among New World hyloids. Despite the relatively high support for the relevant nodes, the actual sister-group relationship of Dendrobatidae remains controversial, and the present study aimed to further test this topology by including relevant morphological characters, additional molecular data, and additional taxa. Especially relevant is the large amount of missing data for Thoropa and the relatively low Bremer support (BS) for the monophyly of Cycloramphidae (BS  =  9) and several of the cycloramphid nodes (BS as low as 4). With that in mind, we targeted the following 46 athesphatanuran outgroup taxa: Allophryne ruthveni, Alsodes gargola, Atelognathus patagonicus, Atelopus spurrelli, Atelopus zeteki, Batrachyla leptopus, Centrolene geckoideum, Centrolene prosoblepon, Ceratophrys cranwelli, Chacophrys pierottii, Cochranella bejaranoi, Crossodactylus schmidti, Cycloramphus boraceiencis, Dendrophryniscus minutus, Edalorhina perezi, Eupsophus calcaratus, Hyalinobatrachium fleischmanni, Hypsiboas boans, Hyla cinerea, Osteocephalus taurinus, Hylodes phyllodes, Hylorina sylvatica, Lepidobatrachus laevis, Leptodactylus fuscus, Leptodactylus discodactylus, Leptodactylus hylaedactyla, Leptodactylus lineatus, Leptodactylus ocellatus, Limnomedusa macroglossa, Megaelosia goeldii, Melanophryniscus klappenbachi, Odontophrynus achalensis, Odontophrynus americanus, Paratelmatobius sp., Physalaemus gracilis, Pleurodema brachyops, Proceratophrys avelinoi, Pseudopaludicola falcipes, Rhaebo guttatus, Rhaebo haematiticus, Rhinoderma darwinii, Scythrophrys sawayae, Telmatobius jahuira, Telmatobius marmoratus, Telmatobius sp., and Thoropa miliaris. Hypsiboas boans was designated as the root for analyses.

All but one of these species were the same ones used by Frost et al. (2006), the exception being Atelopus spurrelli, which we included because (1) sequences proved difficult to generate for the Atelopus zeteki tissue, so adding an additional species was necessary to ensure full coverage of molecular data, and (2) adequate whole specimens of this Chocoan endemic were available at AMNH to allow morphological study.

We included all molecular data from the Frost et al. (2006) analysis for these terminals. To these we added phenotyic characters and sequences for cytochrome oxidase c subunit I, cytochrome b, recombination activating gene 1, and several fragments that were missing from Frost et al. (2006) for 12 of those terminals (marked with an asterisk in fig. 18): Atelopus spurrelli, Atelopus zeteki, Crossodactylus schmidti, Cycloramphus boraceiencis, Dendrophryniscus minutus, Eupsophus calcaratus, Hylodes phyllodes, Megaelosia goeldii, Melanophryniscus klappenbachi, Rhinoderma darwinii, Telmatobius jahuira, Thoropa miliaris. These terminals were targeted for increased sampling because of their phylogenetic proximity to Dendrobatidae, phenotypic affinities, and the availability of whole specimens and other data (e.g., behavior, alkaloid profiles) to score phenotypic characters.

As discussed in greater detail below, with a single exception character-states were coded for each ingroup species and were not extrapolated from other species (e.g., we did not assume that all species of Colostethus lack lipophilic alkaloids and instead only coded species that have been examined for alkaloids); however, we relaxed that requirement to incorporate additional information for outgroup taxa. Specifically, for Crossodactylus alkaloid data were derived from Crossodactylus sp. from Teresopolis, Río de Janeiro, Brazil (Flier et al., 1980; Grant et al., 1997; J. W. Daly, in litt. 09/15/00), chromosome number was assumed to be the same as in the other five species that have been karyotyped (Aguiar et al., 2004), and all other data were coded from Crossodactylus schmidti. For Cycloramphus, most data were taken from Cycloramphus boraceiensis, but osteological data were taken from Cycloramphus fuliginosus. For Eupsophus, DNA sequences and larval data were taken from Eupsophus calcaratus, whereas all other data were taken for Eupsophus roseus (for which material was available at AMNH); see Nuñez et al. (1999) for discussion of the identity of these two species. For Hylodes, most data were obtained from Hylodes phyllodes, but osteology was coded from Hylodes nasus. Finally, for Melanophryniscus, DNA sequences were taken from Melanophryniscus klappenbachi, whereas all other data were scored from Melanophryniscus stelzneri (which is better known and adequately represented at AMNH). Chromosome data were not available for Megalosia goeldii, and there is variation in chromosome number within the genus (Rosa et al., 2003). Insofar as there is no clear empirical evidence to ally Megalosia goeldii with any of the three species for which chromosome data have been reported, we coded Megalosia goeldii as polymorphic. The osteological data reported for Thoropa miliaris were taken from Thoropa lutzi. We assumed that Telmatobius jahuira has the same chromosome number as reported for all other species in the genus (Kuramoto, 1990). All other outgroup data were taken from single species.

Materials and Methods

Conventions and Abbreviations

One of the goals of this study is to propose a monophyletic taxonomy that represents the phylogeny of dendrobatids. The inadequacy of the current taxonomy is widely recognized, and although the general scheme remains that of Myers (1987) and Myers et al. (1991), the recent application of Bauer's overlooked generic names (e.g., Ameerega), the recognition (as well as continued rejection) of Zimmermann and Zimmermann names (e.g., Allobates), the rejection (as well as continued recognition) of Minyobates, and the proposal of a new name (Cryptophyllobates) all indicate that dendrobatid taxonomy is currently in flux with no universally accepted standard around which to structure discussion of dendrobatid diversity. To avoid confusion due to disagreements between the current taxonomies and our proposal for a monophyletic taxonomy, we use binominals only in the introductory sections, above, and after proposing the new taxonomy. Elsewhere (e.g., in the character descriptions) we refer to species using only their trivial names (e.g., fraterdanieli). Currently, 247 nominal species of dendrobatids are recognized, very few of which have the same trivial names. Where giving only the trivial name would engender confusion, we include the author, e.g., sylvaticus Barbour and Noble versus sylvaticus Funkhouser. All species-group names and their original, approximate current, and proposed placements are listed in appendix 1. We also include several species that are undescribed or of unclear identity. For simplicity, we refer to these species by their localities as informal names within quotes (e.g., “Magdalena” in reference to an undescribed species from the Magdalena valley in Colombia, “Curuá-Una” for an undescribed species from the Rio Curuá-Una in Brazil) or as they have been reported in the literature (e.g., Nephelobates sp.). We also assigned unique sample identification numbers to all tissues used in this study, and these numbers are reported as unique terminal identifiers.

Commands used in computer programs are italicized. Sequences incorporated from GenBank are listed in appendix 4. Data for tissues (including GenBank numbers for DNA sequences) and specimens examined are listed in appendices 5 and 6, respectively, referenced with the permanent collection number for the voucher specimen or, if that is unavailable, the tissue collection number, as follows: AMCC (Ambrose Monell Cryo Collection, American Museum of Natural History, New York, USA), AMNH (American Museum of Natural History, New York, USA), ARA (Andrés Acosta field series; specimens at MUJ), BB (Boris Blotto field series), BMNH (The Natural History Museum, London, UK), BPN (Brice P. Noonan field series; specimens at UTA), CFBH (Célio F. B. Haddad specimen collection, Brazil), CFBH-T (Célio F. B. Haddad tissue collection, Brazil), CH (Colección Herpetológica, Panamá), CPI (D. Bruce Means field series, to be deposited at USNM), CWM (Charles W. Myers field series), GB (Godfrey Bourne field series), IAvH (Instituto de Investigación de Recursos Biológicos Alexander von Humboldt, Villa de Leyva, Colombia), ICN (Instituto de Ciencias Naturales, Universidad Nacional de Colombia, Bogotá, Colombia), IRSNB-KBIN (Institut Royal des Sciences Naturelles de Belgique/Koninklijk Belgisch Instituut voor Natuurwetenschappen, Brussels, Belgium), IZUA (Instituto de Zoología, Universidad Austral de Chile, Valdivia, Chile), JAC (Jonathan A. Campbell field series), JDL (John D. Lynch field sereis), JF (Julián Faivovich field series), KRL (Karen R. Lips field series), KU (University of Kansas Natural History Museum, Lawrence, USA), LACM (Natural History Museum of Los Angeles County, Los Angeles, USA), LR (Lily Rodriguez field series), LSUMZ (Louisiana State University Museum of Natural Science, Baton Rouge, USA), MACN (Museo Argentino de Ciencias Naturales Bernardino Rivadavia, Buenos Aires, Argentina), MAD (Maureen A. Donnelly field series), MAR (Marco Antonio Rada; specimens at MUJ), MB (Marcus Breece captive collection), MJH (Martin J. Henzl field series), MLPA (Museo de la Plata, Buenos Aires, Argentina), MHNUC (Museo de Historia Natural Universidad del Cauca, Popyán, Colombia), MPEG (Museu Paraense Emilio Goeldi, Belém, Brazil), MRT (Miguel Rodrigues tissue collection), MUJ (Museo de Historia Natural, Universidad Javeriana, Bogotá, Colombia), MVZ (Museum of Vertebrate Zoology, University of California at Berkeley, USA), MZUSP (Museu de Zoologia da Universidade de São Paulo, Brazil), NK (Museo de Historia Natural Noel Kempff Mercado, Santa Cruz, Bolivia), OMNH (Sam Noble Oklahoma Museum of Natural History, The University of Oklahoman, USA), PK (Philippe Kok field series; specimens at IRSNB-KBIN), RDS (Rafael de Sá tissue collection), RG (Ron Gagliardo, Atlanta Botanical Garden), ROM (Royal Ontario Museum, Toronto, Canada), RWM (Roy W. McDiarmid field series), SIUC (Southern Illinois University at Carbondale, USA), UMFS (University of Michigan Museum of Zoology, Ann Arbor, Michigan, USA, field series), UMMZ (University of Michigan Museum of Zoology, Ann Arbor, Michigan, USA), USNM (National Museum of Natural History, Smithsonian Institution, Washington DC, USA), USNM-FS (National Museum of Natural History, Smithsonian Institution, Washington DC, USA, field series), UTA (University of Texas at Arlington, USA), UVC (Universidad del Valle, Cali, Colombia), WES (Walter E. Schargel field series), ZSM (Zoologisches Museum, München, Germany). Phenotypic data are given in appendix 7; formatted files for all data may be downloaded from

General Analytical Approach: Theoretical Considerations

Choice of Phylogenetic Method

The general goal of phylogenetic systematics is to explain the diversity of life by discovering the evolutionary relationships among species, where inferred transformations from one character-state to another provide the means to choose among competing explanations. That is, phylogenetic hypotheses are composite explanations consisting of both hypotheses of homology (transformation series; Hennig, 1966; see Grant and Kluge, 2004) and hypotheses of monophyly (topology). Farris (1967) expressed this succinctly by analyzing the concept of evolutionary relationship into its component parts of patristic relationship and cladistic relationship.

Operationally, phylogenetic analysis begins by decomposing the observed diversity of living things into its minimal historical units: character-states (sensu Grant and Kluge, 2004) and species (sensu Kluge, 1990; see also Grant, 2002). Although character-states are the evidential basis that underlies phylogenetic inference, they are effectively “bundled” into individual organisms, populations, and species, which constrains the ways in which they evolve and how they may be explained (e.g., females and males evolve as parts of the same lineage; defensible phylogenetic explanations are therefore not permitted to place them in separate clades). Likewise, species, which are the historical entities related through phylogeny (Hennig, 1966), may be decomposed into independently heritable (and independently variable) parts, that is, character-states. This ontological transitivity of taxic and character evolution is the foundation of phylogenetic inference (cf. Farris, 1967).

Once these minimal units have been individuated, all possible historical relationships between character-states and species are defined by pure logic (Siddall and Kluge, 1997; Wheeler, 1998). Phylogenetic analysis proceeds by mapping hypothetical character-state relationships to hypothetical species relationships and evaluating the competing composite hypotheses in terms of the number of character-state transformations they entail.

All phylogenetic methods aim to minimize character transformations. Unweighted (equally weighted) parsimony analysis minimizes hypothesized transformations globally, whereas assumptions (expressed as differential probabilities or costs) about the evolution or importance (e.g., reliability) of different classes of transformations employed in maximum likelihood, Bayesian analysis, and weighted parsimony methods lead to the minimization of certain classes of transformations at the expense of others. Operational considerations aside (e.g., tree-space searching capabilities), disagreements between the results of unweighted parsimony analysis and the other methods are due to the increased patristic distance required to accommodate the additional assumptions.

Kluge and Grant (2006) reviewed the justifications for parsimonious phylogenetic inference previously considered sufficient, namely, conviction (Hennig, 1966), descriptive efficiency (Farris, 1979), minimization of ad hoc hypotheses of homoplasy (Farris, 1983), and statistical, model-based inference (maximum likelihood, Sober, 1988). Finding significant inconsistencies in all of those justifications, Kluge and Grant (2006) proposed a novel justification for parsimony. Drawing on recent advances in the understanding of phylogenetics as a strictly ideographic, historical science and parsimonious inference generally in the philosophy of science literature (e.g., Barnes 2000; Baker, 2003), they argued that by minimizing globally the transformation events postulated to explain the character-states of terminal taxa, equally weighted parsimony analysis maximizes explanatory power. As such, in the present study we analyzed the total, equally weighted evidence under the parsimony criterion (for additional discussion of character weighting and total evidence, see Grant and Kluge, 2003). Given the size and complexity of this dataset, a further advantage of parsimony algorithms (whether weighted or unweighted) is that thorough analysis could be achieved in reasonable times given available hard- and software.

Sources of Evidence

The empirical evidence of phylogenetic systematics consists of transformation series (i.e., the ideographic character concept of Grant and Kluge, 2004). Traditionally, transformation series were derived exclusively from such sources as comparative morphology, molecular biology, and behavior, but as technological advances have made DNA sequencing simpler and less costly, phylogenetic studies have come to rely increasingly on the genotypic evidence of DNA sequences to test phylogenetic hypotheses. The present study exemplifies this trend. Nevertheless, both kinds of data provide evidence of phylogeny, and each has its own suite of strengths and weaknesses.

An important strength of phenotypic data is that the complexity of observed variation allows the historical identity of each transformation series to be tested independently (Grant and Kluge, 2004). By carrying out progressively more detailed structural and developmental studies, researchers are able to refine their hypotheses about the homology of phenotypic variants. However, the phenotype is determined by both the directly heritable components of the genotype and the nonheritable effects of the environment. In contrast, an obvious strength of DNA sequence evidence is that, because DNA is the physical material of genetic inheritance, the potentially confounding effects of environmental factors are avoided altogether.

Nevertheless, DNA sequence character-states are maximally reduced to physico-chemically defined classes of nucleotides, of which there are only four (cytosine, guanine, adenine, and thymine). Whereas this simplicity is advantageous in many kinds of genetics studies, it poses a serious problem for phylogenetics, because no structural or developmental complexity to distinguishes nucleotides that share a common evolutionary history (i.e., those that are homologous, their physico-chemical identity owing to the same transformation event) from those that evolved independently (i.e., those that are homoplastic, their physico-chemical identity owing to independent transformation events). For example, in terms of object properties, all adenines are physico-chemically identical, regardless of whether or not they arose through the same or different transformation events. Moreover, DNA sequences evolve through complete substitutions of one nucleotide for another (meaning that there are no intermediate states from which to infer historical identity) or complete insertions and deletions (meaning that any given nucleotide could be homologous with any other nucleotide). Phylogenetic analysis of DNA sequences must therefore contend with the problem of discovering both transformations between nucleotides and the insertion and deletion of nucleotides. To visualize homologous nucleotides, multiple sequence alignments codify insertions and deletions (indels) as gaps, that is, placeholders that shift portions of the sequence to align homologous nucleotides into column vectors.

Nucleotide Homology and the Treatment of Indels

The method of inferring indels and nucleotide homology (i.e., alignment) and the subsequent treatment of indels in evaluating phylogenetic explanations are of critical importance in empirical studies because, as is now widely appreciated, a given dataset aligned according to different criteria or under different indel treatments may result in strong support for contradictory solutions (e.g., McClure et al., 1994; Wheeler, 1995). Many workers infer indels in order to align nucleotides but then either treat them as nucleotides of unknown identity by converting gaps to missing data, or they eliminate gap-containing column vectors altogether, either because they believe them to be unreliable or because the implementation of a method of phylogenetic analysis does not allow them (Swofford et al., 1996). Others argue that indels provide valid evidence of phylogeny but suggest that sequence alignment (homology assessment) and tree evaluation are logically independent and must be performed separately (e.g., Simmons and Ochoterena, 2000; Simmons, 2004).

The position we take here is that indels are evidentially equivalent to any other classes of transformation events and, as such, are an indispensable component of the explanation of the DNA sequence diversity. Furthermore, because nucleotides lack the structural and/or developmental complexity necessary to test their homology separately, hypotheses of nucleotide homology can only be evaluated in reference to a topology (Grant and Kluge, 2004; see also Wheeler, 1994; Phillips et al., 2000; Frost et al., 2001). In recognition of these considerations, we assessed nucleotide homology dynamically by optimizing observed sequences directly onto topologies (Sankoff, 1975; Sankoff et al., 1976) and heuristically evaluating competing hypotheses by searching tree space (Wheeler, 1996). This is achieved using Direct Optimization techniques (Wheeler, 1996, 2003a, 2003b, 2003c), as implemented in the computer program POY (Wheeler et al., 1996–2003).

In this approach, determination of nucleotide homology is treated as an optimization problem in which the preferred scheme of nucleotide homologies for a given topology is that which requires the fewest transformation events when optimized onto that topology, that is, that which minimizes patristic distance, thus providing the most parsimonious explanation of the observed diversity. Determining the optimal alignment for a given topology is NP-complete2 (Wang and Jiang, 1994). For even a miniscule number of sequences, the number of possible alignments is staggering (Slowinski, 1998), making exact solutions impossible for any contemporary dataset, and heuristic algorithms are required to render this problem tractable. Likewise, finding the optimal topology for a given alignment is also an NP-complete problem (Garey and Johnson, 1977; Garey et al., 1977).

Phylogenetic analysis under Direct Optimization therefore consists of two nested NP-complete problems. POY searches simultaneously for the optimal homology/topology combination, and search strategies must take into consideration the severity and effectiveness of the heuristic shortcuts applied at both levels. In any heuristic analysis, a balance is sought whereby the heuristic shortcuts will speed up analysis enough to permit a sufficiently large and diverse sampling of topologies and homologies to discover the global optimum during final refinement, but not so severe that the sample is so sparse or misdirected that the global optimum is not within reach during final refinement. Ideally, indicators of search adequacy (e.g., multiple independent minimum-length hits, stable consensus; see Goloboff, 1999; Goloboff and Farris, 2001) should be employed to judge the adequacy of analysis, as is now reasonable in analysis of large datasets using prealigned data (e.g., in TNT; Goloboff et al., 2003). However, current hard- and software limitations make those indicators unreachable in reasonable amounts of time for the present dataset analyzed under Direct Optimization, and the adequacy of our analysis may only be judged intuitively in light of the computational effort and strategic use of multiple algorithms designed for large datasets (see below for details).

Total Evidence

The majority of phylogenetic studies, even those legitimately considered “total evidence” (Kluge, 1989), examine either higher level or lower level problems. The former are designed to address relationships between putative clades of multiple species (usually discussed as species groups, genera, families, etc.) by targeting exemplars from each of those units and sampling relatively invariable character systems. The latter are designed to address species limits and relationships among closely related species (and often phylogeographic questions also), and character sampling focuses on more variable systems.

The nestedness of phylogenetic problems both enables and weakens this divide-and-conquer approach. Assuming the monophyly of a group, the relationships within that clade have no bearing on the relationships of that clade to other clades. And assuming the sister-group relationships between the ingroup and outgroup taxa, the relationships within the ingroup clade are independent of the relationships between that clade and more distant relatives. Nevertheless, although this is a defensible (provided the assumed monophyly and placement of the ingroup are supported by evidence) and presently necessary strategy, these assumptions may ultimately be found to be false, and their elimination allows hypotheses at both levels to be more severely tested and may lead to more globally parsimonious explanations. Furthermore, the ripple effects that cladistically distant optimizations may have throughout the topology are unpredictable, so that the inclusion of distant terminals that are not immediately relevant to a problem may affect local topology. The ultimate goal of total evidence is to analyze all evidence from all sources and all terminals at all levels simultaneously.

There are many obstacles, computational and otherwise, that prevent this ideal from being achieved in the foreseeable future. A potential criticism of this study is that we did not include all of the data from Frost et al. (2006). We restricted the outgroup sample to Athesphatanura to prevent the already computationally challenging problem of analyzing 414 terminals from becoming even more intractable. However, once hard- and software improvements permit thorough analysis of larger datasets, studies that extend outgroup sampling beyond Athesphatanura will offer a test of our results. Insofar as this study was designed to test both the species limits of problematic taxa and the relationships among dendrobatid clades, and to further test the relationships between dendrobatids and other anurans by combining the data from Frost et al. (2006) with new data from key outgroup taxa, it is intended to be a step toward the total evidence ideal. That this study aimed to simultaneously address problems of such different hierarchic levels had important consequences in taxon sampling, character sampling, and the analytical strategy that was undertaken.

General Analytical Approach: Implementation

Taxon Sampling

Outgroup taxa and the rationale for their selection are discussed above (Phylogenetic Placement of Dendrobatids and Outgroup Sampling). Selection of ingroup terminals was governed by three considerations: (1) relevance to testing prior phylogenetic claims, (2) availability of tissues (or sequences on GenBank), and (3) availability of specimens for morphological study. In light of the many problems in species-level taxonomy, we also sought to sample as many localities as possible for problematic species.

To facilitate taxonomic changes, every effort was made to include type species of all dendrobatid genera. Both genotypic and phenotypic data were included for type species of as many genera as possible, including (genus name in parentheses): azureiventris (Cryptophyllobates), bicolor (Phyllobates), femoralis (Allobates), inguinalis (Prostherapis), nocturnus (Aromobates), pulchellus (Phyllodromus), pumilio (Oophaga), reticulatus (Ranitomeya), silverstonei (Phobobates), steyermarki (Minyobates), tinctorius (Dendrobates), tricolor (Epipedobates), and trivittatus (Ameerega). We did not include the type species alboguttatus (Nephelobates), fuliginosus (Hyloxalus), latinasus (Colostethus), or yustizi (Mannophryne), because adequate data were not available to allow their inclusion in the present study. Nevertheless, we included numerous, presumably closely related representatives of these genera and made taxonomic changes accordingly.

Phenotypic Character Sampling

We anticipate that a criticism of the present study will be that we were too catholic in the inclusion of phenotypic characters. It is common for morphological systematists to seek characters that are conservative at their level of interest, under the assumption that they are more informative or reliable indicators of relationship, either explicitly (e.g., Kluge, 1993) or, much more commonly, implicitly. As a result, much of the systematics literature—especially the precladistic literature—consists of special pleading for the validity (or not) of characters as “higher-level”, “family-level”, “genus-level”, “species-level” or some other rank-specific indicator. Some characters (e.g., presence or absence of teeth, pectoral girdle architecture, skull morphology), it has been argued, are “good” genus- or family-level characters, others (e.g., external morphology, soft-anatomy) are “good” only at the level of species, and still others are unreliable and should be excluded in their entirety. We disagree.

For evolution to occur, all character transformation events must take place at (or below) the species level, and it is only subsequent cladogenetic events that effectively push character transformations back in history and bring them to delimit clades; there can be no natural law regulating variation of characters among clades. The historical debate over the phylogenetic relevance of anuran teeth illustrates the futility of that approach to systematics: maxillary teeth are absent in all species of Bufonidae—which would make this a conservative, phylogenetically informative character at the family level—but vary intraspecifically in some species of dendrobatids—making this a completely uninformative character according to that view. Given the conceptual definition of characters as transformation series (Grant and Kluge, 2004), all characters have the same evidential status in terms of their ability to test phylogenetic hypotheses. Arguments over the rank-specific relevance or reliability of characters depend on the reification of ranks and lead ultimately to the ad hoc dismissal or overlooking of evidence, and such procedures should be eliminated from systematics.

In addition to the novel characters and character-states discovered in the course of this study, our goal was to include all characters that have figured in debates on the monophyly, placement, and internal relationships of Dendrobatidae. However, because this study aims primarily to test relationships within Dendrobatidae and not the position of Dendrobatidae among other frogs, the sample of characters is strongly biased to reflect variation among dendrobatid terminals.

Numerous characters date to the 19th century (mainly Duméril and Bibron, Cope, Boulenger), and we do not always cite the original sources for these traditional characters. However, we do cite more recent papers that have addressed them in the context of dendrobatid systematics, and we cite original sources for all more recent characters. All phenotypic character-states for caeruleodactylus, humilis, and nidicola were coded from the literature (Lima and Caldwell, 2001; Caldwell et al., 2002a; La Marca et al., 2002; Caldwell and Lima, 2003). Other sources for phenotypic data are cited in the relevant sections below. For the purposes of discussion, phenotypic transformation series are classified broadly as morphological, larval, behavioral, and biochemical, the latter referring to alkaloid profiles. Specific problems or concerns regarding particular characters or character systems are discussed below independently. In anticipation of the expansion of the present dataset, we list states and show illustrations for taxa not included in the present analysis.

Comparative anatomical study aimed to delimit transformation series and not to describe dendrobatid (or outgroup) anatomy per se. We have illustrated either photographically or in line drawings those character-states we believe may cause confusion, and character names and descriptions were intended only to be sufficiently precise to allow hypotheses of homology to be tested. With the exception of characters related to the median lingual process, we coded anatomical characters only from gross dissection under a dissecting microscope. This is a limitation of the present study, as greater insight into character-state identity and homology would undoubtedly be gained from histology (e.g., consider the remarkable insights into pectoral girdle architecture attained by Kaplan, 2004). Osteological character-states were coded from dried or cleared and stained (alcian blue and alizarin red) skeletons. We considered tissue with alizarin red-positive crystals to be calcified and uniformly alizarin red-positive tissue to be ossified.

We applied either Lugol's solution or alcian blue to facilitate coding of muscle characters. All muscles are bound by fibrous connective tissue, so the distinction between tendinous and fleshy origins and insertions is one of degree: Tendinous insertions and origins have a confluence of muscle fibers on a distinct segment of fibrous connective tissue, whereas those that are fleshy appear to insert or originate directly on the adjacent structure. We consider a slip to be a distinct bundle of fasciculi isolated from adjacent fasciculi of the same muscle by epimysium.

We examined the histology of the tongues of several species to individuate characters of the median lingual process (Grant et al., 1997). Tissues were embedded in paraffin, sectioned at 6–10 µm, and stained using either hematoxylin and eosin (H&E) or a trichrome stain consisting of Alcian Blue, Periodic Acid and Schiff's reagent (PAS), and H&E. Specifically, we examined histological sections of baeobatrachus, tepuyensis, panamensis, and auratus (the latter two lacking the MLP). For comparison we examined the histology of Arthroleptis variabilis, Mantidactylus femoralis, Phrynobatachus natalensis, P. petropedetoides, Platymantis dorsalis, and Staurois natator, although none of these species were coded for the present study. We also performed detailed dissections of the tongues of atopoglossus, Arthroleptis stenodactylus, Discodeles bufoniformis, and Discodeles opisthodon.

In addition to the phenotypic characters individuated for this study, other sources of variation will undoubtedly yield novel characters. For example, spermatozoa ultrastructure is promising but has been examined in too few species to warrant inclusion in the present study. Garda et al. (2002) examined the spermatozoa of flavopictus, Aguiar et al. (2003) studied femoralis and an undescribed species referred by them to Colostethus (OMNH 37001–37002), and Aguiar et al. (2002) looked at the spermatozoa of hahneli and trivittatus. (For a recent review of spermatozoa in nondendrobatids, see Scheltinga and Jamieson, 2003).

Relevant to the placement of Dendrobatidae, Haas (2003) presented an impressive matrix of detailed morphological evidence scored across the diversity of anurans, much of which was derived from studies of larval anatomy. The evidential value of such data is manifest, but adequate samples were unavailable for most species included in this study. Haas found (see fig. 16, above) that the four included dendrobatids were monophyletic, and that their sister group was Hylodes + Crossodactylus (Megaelosia was not included in that study).

Bhaduri (1953) studied the urinogenital systems of diverse amphibians, including auratus, tinctorius, and flotator (as Phyllobates nubicola flotator). He noted several differences among these species, such as the greater posterior extension of the kidneys in Dendrobates than in Phyllobates (p. 56), but he nonetheless concluded that “[t]he structural similarities of the urinogenital organs which we have observed in these two genera lend further support to Noble's view [that Dendrobates and Phyllobates are closely related]” (p. 72). Although we scored some visceral characters (e.g., pigmentation of the testes, pigmentation of the large intestine), in light of time constraints and the fact that specific characters used by Bhaduri have not been used since and have therefore not played an important role in dendrobatid systematics, we did not study this system in detail.

Burton (1998a) argued that hand musculature supports a relationship between Dendrobatidae and Hylodinae, “as the unusual condition of lacking any fibrous connection to the tendo superficialis or the adjacent aponeurosis is almost restricted to the hylodine genera Hylodes and Megaelosia, and Dendrobatidae” (p. 8). However, the phylogenetic implications of this character are not clear-cut; assuming hylodine monophyly and a sister-taxon relationship with Dendrobatidae, as implied by Burton, the occurrence of this character-state would optimize as either independently evolved in Dendrobatidae and Hylodes + Megaelosia or as a synapomorphy of the inclusive clade with subsequent loss in Crossodactylus. We did not examine hand musculature in this study.

Trewavas (1933) included tinctorius3 in her study of the anuran hyoid and larynx. We examined the osteology of this system but did not examine its musculature. Our experience with other groups suggests that hyoid musculature may be a rich source of characters, and these characters will be evaluated in the near future.

There are also several morphological variants that have been claimed as characters in the literature that we reject in the present study. First, La Marca (1994, 2004) claimed the occurrence of enlarged, fanglike maxillary teeth as a synapomorphy for Nephelobates, and they also have been reported for Megaelosia (e.g., Lynch, 1971) and Aromobates (Myers et al., 1991), among others. Although we agree with La Marca and Myers et al. that dendrobatid tooth morphology varies and that the teeth of Aromobates and Nephelobates seem strikingly elongate and recurved, we were unable to individuate transformations series for several reasons, as follows (see fig. 19).

Figure 19

Examples of variation in dendrobatid maxillary teeth. A, B: lateral (A) and lingual (B) views of pictus (UMMZ 184099).

Note that the teeth do not protrude beyond the edge of the maxilla. C: lateral view of riveroi (AMNH 134144). D: lateral view of subpunctatus (UMMZ 221159). E: lateral view of undulatus (AMNH 159142). F: lateral view of molinarii (UMMZ 176207). G: lateral view of dunni (UMMZ 167131). H: lateral view of nocturnus (AMNH 129940).


(1) No appropriate reference point to assess relative tooth size has been proposed, and without this it is impossible to compare objectively the size of teeth in specimens of different species and varying body sizes and maxilla sizes and shapes (especially the shape and depth of the facial process). (2) Tooth size varies along the maxilla, and it is unclear which teeth should serve as the basis of comparison. (3) Superficial assessment of tooth size in cleared and stained specimens of a number of species suggested that variation is continuous, which must be accounted for when individuating transformations series. (4) All well-developed maxillary teeth (i.e., those that protrude beyond the edge of the maxilla) are recurved, at least in dendrobatids, and comparison of digital images (which eliminates the effect of relative size) shows the curvature of the so-called fanglike teeth of species referred to Nephelobates and Aromobates is no greater than those referred to Colostethus. In light of these considerations, we coded the presence and absence of maxillary teeth, as well as their structure, but not variation in size and shape.

Similarly, Lynch (1982) characterized edwardsi and ruizi as possessing a conspicuously large and elongate cloacal sheath (vent tube, anal sheath, embudo cloacal), and Rivero (1990 “1988”) subsequently referred to these species as the edwardsi group of Colostethus. Later, La Marca (1994) also claimed the presence of a cloacal sheath as a synapomorphy for Nephelobates, although he made no reference to that structure in the edwardsi group. The cloacal sheath has now been included in numbered diagnoses in species descriptions (e.g., Lötters et al., 2003a), and Grant (1998) cited its synapomorphic occurrence as the basis for including Colostethus lynchi in the edwardsi group. More recently, Grant (2004) noted, without further comment, that “examination of extensive material of most species of dendrobatids has caused me to doubt the validity of that character.”

The reason for that doubt is that, as shown in figure 20, only the two species originally placed in the edwardsi group (exemplified here by edwardsi) possess a conspicuously modified vent (cloacal sheath). Variations among other species of dendrobatids (including lynchi) are minor and cannot be distinguished from artifacts of preservation. Specimens that are positioned differently for fixation (whether floated in formalin or laid out in a fixing tray) vary in apparent vent morphology. For example, when a frog specimen is positioned in a fixing tray, the flaccid thigh muscles and loose skin may be pushed posterodorsally, causing the vent and adjacent tissue to “bunch up”, or anteroventrally, causing the vent and adjacent tissue to be drawn downward, both of which alter the apparent prominence, length, and shape of the vent. Desiccation also affects vent prominence. In light of these observations, the cloacal sheath is restricted to the two known species of the edwardsi group. We did not include the cloacal sheath, thus delimited, in the character matrix because we did not include edwardsi or ruizi due to inadequate material.

Figure 20

Posterior view of several species of dendrobatids, showing variation in morphology of the vent. A: edwardsi (ICN 21936). Contrary to the other species depicted, the vent of edwardsi is conspicuously enlarged and elongated relative to that of other anurans. B: molinarii (UMMZ 176222, paratype), a species referred to Nephelobates by La Marca (1994).

The vertical folds vary as an artifact of preservation. C: alboguttatus (AMNH 10503), the type species of Nephelobates. D: trinitatis (AMNH 125796), a species referred to Mannophryne by La Marca (1992). E: petersi (AMNH 42546). F: petersi (AMNH 42506). This species has never been claimed to be part of or closely related to Nephelobates. Note the differing prominence and apparent shape and size of the vent as an artifact of preservation.


Finally, Savage (1968) was followed by Silverstone (1975a) in identifying dark pigmentation of the flesh as a synapomorphy of Dendrobates and Phyllobates. We paid considerable attention to this character, thanks largely to the numerous large series of skinned specimens collected by C. W. Myers and colleagues and deposited (cataloged and uncataloged) at AMNH. The variation we observed is much more complicated than the simple pigmented/unpigmented of Savage and Silverstone. Pigmentation occurs in diffuse, irregular patches and varies continuously in intensity from being entirely lacking to a few black specks or intense dark gray or black. We were unable to delimit transformation series objectively, and therefore excluded pigmentation of the flesh from this study.

Genotypic Character Sampling

In light of the vastly different levels of diversity included in this study (from within localities to among families), we sought to sample genes of differing degrees of variability. We targeted the mitochondrial H-strand transcription unit 1 (H1), which includes 12S ribosomal, tRNAval, and 16S ribosomal sequence, yielding approximately 2,400 bp generated in 5–7 overlapping fragments. We also targeted a 385-bp fragment of cytochrome b and a 658-bp fragment of cytochrome oxidase c subunit I (COI). In addition to those five mitochondrial genes, we targeted the nuclear protein coding genes histone H3 (328 bp), rhodopsin (316 bp), tyrosinase (532 bp), recombination activating gene 1 (RAG1, 435 bp), and seventh in absentia (SIA, 397 bp), and the nuclear 28S ribosomal gene (ca. 700 bp), giving a total of approximately 6,100 bp of nuclear and mitochondrial DNA. Primers used in PCR amplification and cycle sequencing reactions (and their citations) are given in table 1. Included in this study is a novel primer pair (RAG1 TG1F and TG1R) designed to amplify the RAG1 product using the web-based program Primer3 (Rozen and Skaletsky, 2000), available at As discussed below, the amount of sequence data analyzed per terminal varied (fig. 21), ranging from 426 bp (subpunctatus obtained from GenBank) to 6,245 bp (chlorocraspedus 385), with a mean of 3,740 bp per terminal.

Table 1

PCR Primers Used in This Studyai0003-0090-299-1-1-t01.gif

Figure 21

Number of DNA base-pairs analyzed per terminal.

For specific terminal data see appendices 4 and 5.


As noted above, we targeted loci that varied to differing degrees to test hypotheses of relationships at all levels, and we included multiple samples from the same and different localities of the same species in an effort to address problems in alpha taxonomy. We attempted to sequence all loci for at least one sample from every locality, but we did not sequence nuclear loci for all samples. We chose this strategy because early work on this project showed the nuclear loci to be generally less variable and usually identical in all samples from a given locality.

We augmented our own data with sequences from GenBank, listed in appendix 4, in order to include otherwise unsampled ingroup species and additional localities for taxonomically problematic species, that is, the dataset analyzed includes all species on GenBank as well as samples of some species from multiple localities. Nevertheless, we did not include all GenBank data. First, we only included loci for which we also generated data. For example, we did not include Widmer et al.'s (2000) cytochrome b data because their fragment did not overlap with ours. Second, although we included multiple samples to address taxonomic problems, we did not include all samples from population-level studies (e.g., Symula et al., 2003), as such dense intraspecific sampling was not required and would have impeded analysis by unnecessarily expanding the dataset.

Laboratory Protocols

Whole cellular DNA was extracted from frozen and ethanol-preserved tissues (liver or muscle) using the Qiagen DNeasy kit following the manufacturer's guidelines. PCR amplification was carried out in 25-µl reactions using puRe Taq Ready-To-Go Beads (Amersham Biosciences). The standard PCR program consisted of an initial denaturing step of 3 min at 94°C, 35–40 cycles of 1 min at 94°C, 1 min at 45–62°C, and 1–1.25 min at 72°C, followed by a final extension step of 6 min at 72°C. PCR-amplified products were cleaned and desalted using either the ARRAYIT kit (TeleChem International) on a Beckman Coulter Biomek 2000 robot or AMPure (Agencourt Biosciences Corporation). Cycle-sequencing using BigDye Terminators v. 3.0 (Applied Biosystems) was run in 8-µl reactions, and products were cleaned and desalted by standard isopropanol-ethanol precipitation or using cleanSEQ (Agencourt Biosciences Corporation). Sequencing was done on either an ABI 3700 or ABI 3730XL automated DNA sequencer. Contigs (sets of overlapping sequences) were assembled and edited using Sequencher (Gene Codes).

Molecular Sequence Formatting

To enable integration of incomplete sequence fragments (particularly those from GenBank; see Taxon Sampling Strategy and Character Sampling Strategy, above), accelerate cladogram diagnosis, and reduce memory requirements under Iterative Pass Optimization, we broke complete sequences into contiguous fragments. (This also improves the performance of POY's implementation of the parsimony ratchet; see Heuristic Tree Searching, below.) We did so sparingly, however, as these breaks constrain homology assessment by prohibiting nucleotide comparisons across fragments, that is, it is assumed that no nucleotides from fragment X are homologous with any nucleotides from fragment Y. As the number of breaks increases, so too does the risk of overly constraining the analysis and failing to discover the globally optimal solution.

We therefore inserted as few breaks as were necessary to maximize the amount of sequence data included, minimize the introduction of nucleotides of unknown identity into the sequences (see Character Sampling Strategy, above), and attain maximum length fragments of around 500 bases (see table 2). Breaks were placed exclusively in highly conserved regions (many of which correspond to commonly used PCR primers), as recovery of such highly invariable regions is generally alignment-method independent (unpublished data) and therefore does not prevent discovery of global optima. These highly conserved regions were identified via preliminary ClustalX (Thompson et al., 1997) alignments under default parameters and examination using BioEdit (Hall, 1999). Except for their usefulness in placing fragments derived from different PCR primers and detecting errors, these preliminary alignments were used solely for the purpose of identifying conserved regions; they did not otherwise inform or constrain our phylogenetic analysis. Once appropriate conserved regions were identified, fragments were separated by inserting pound signs (#) at break points. Thus, the multiple fragments of the mitochondrial H1 unit remain in the same file and order, for example.

Table 2

Summary of DNA Sequence Dataai0003-0090-299-1-1-t02.gif

Total Evidence Analysis

We did not discriminate between classes of evidence in the phylogenetic analyses. To allow the molecular data to have bearing on problems of species taxonomy, we treated every specimen sequenced as a separate terminal, that is, we did not fuse putatively conspecific specimens into a single polymorphic terminal, which would prevent the molecular data from addressing alpha taxonomic problems and require that all decisions on species identity be made prior to phylogenetic analysis. Loci not sequenced for particular terminals—either because the primers failed or because other syntopic conspecifics were sequenced instead—were treated as missing for those terminals.

There are three possible methods of incorporating phenotypic evidence for specimens judged to be conspecific but coded separately for genotypic data.

  1. Phenotypic characters may be coded for each specimen separately. The shortcomings of this method are numerous: (a) This approach excludes background knowledge that informs but is not explicitly encoded in the character matrix, such as mating behavior and ontogeny. This could result in males, females, and juveniles being grouped in separate clades. (b) Similarly, tissue samples usually are not available for specimens representing all relevant semaphoronts. As such, many semaphoront-specific characters would be excluded from analysis or have to be coded without being matched with the molecular evidence. (c) For this approach to be applied consistently, evidence obtained from specimens in other studies would also have to be rejected, such as alkaloid profiles and vocalizations and other behaviors, or also scored separately for each individual. Strict application of this approach is clearly infeasible and would result in the exclusion of extensive evidence.

  2. The phenotypic data for the species as a whole can be duplicated for each molecular terminal.

  3. The phenotypic data for the species as a whole can be entered for a single molecular terminal, with those characters treated as missing for other (putatively) conspecific terminals.

The latter two options offer more defensible approaches. The second method has the advantage of minimizing ambiguous optimizations due to missing entries, which may be crucial in examining the evolution of some of the most interesting phenotypic characters (e.g., behaviors). The third approach appears to have the advantage of maximizing the severity of the molecular test of species identity, that is, terminals judged conspecific on phenotypic grounds could not be held together on those grounds alone in phylogenetic analysis. Although we see some validity to this argument, given the relative sizes of the phenotypic and genotypic partitions (ca. 170 characters vs. ca. 6,100 unaligned base pairs), we see no a priori reason to expect morphology to overwhelm the DNA sequence data. Moreover, the identical entries that would potentially hold those specimens together in the face of molecular evidence are, in fact, apomorphies for the species, and total evidence analysis demands that they be considered as such. That is, the goal in total evidence analysis is not to test the results of one data partition against another, but to allow all evidence to interact simultaneously to identify the hypothesis that best explains all the evidence.

As such, we opted to duplicate the morphological entries coded for the species, that is, each conspecific terminal was given identical entries in the phenotypic matrix. Phenotypic characters not expressed in the sequenced semaphoront (e.g., testis color in female specimens) were scored and species-level phenotypic polymorphisms were coded as ambiguities. A caveat is that we did not associate GenBank sequences with phenotypic data unless we lacked our own genotypic data for the taxon (e.g., sauli), and then only for one sample if more than one was on GenBank (e.g., we associated the phenotypic entries for kingsburyi with AY364549 only).

With a single exception, the reported ingroup characters were scored for single species (see Outgroup Sampling, above, for coding of outgroup species). The exception was alagoanus, for which the morphological characters were scored from specimens of olfersioides. The two species were named on the basis of few differences between small samples, and study of external morphology of larger samples from a greater number of localities suggests they may be conspecific (V.K. Verdade, personal commun.), although additional data from color in life, vocalizations, and DNA sequences have yet to be examined. DNA sequences were generated for alagoanus, but whole specimens were not available for examination. Tissue samples were unavailable for olfersioides (which has not been observed in Rio de Janeiro since approximately 1980 and may be extinct; C.F.B. Haddad, personal obs.), but numerous museum specimens were available for morphological study. As such, we combined the data from these two species under the asumption that they represent a unique clade; we refer to the resulting terminal as alagoanus because most of the evidence was scored from that species.

Simultaneous phylogenetic analysis was performed using the program POY (Wheeler et al., 1996–2003) version 3.0.11a and the MPI version 3.0.12a-1109195780.71. All POY runs were parallelized across 95 processors of the AMNH 256-processor Pentium 4 Xeon 2.8 GHz cluster or 16–32 processors of the 560-processor mixed 512 mHz and 1 GHz cluster. Results were visualized using Winclada (Nixon, 1999–2002), and we verified POY results and analyzed implied alignments using NONA (Goloboff, 1999) spawned from Winclada. Although character weighting was used heuristically in tree searching (see below), evidence was weighted equally to assess tree optimality.

Heuristic Character Optimization

Numerous algorithms of varying exhaustiveness have been proposed to optimize unaligned DNA sequences on a given topology. Our search strategy employed three Direct Optimization algorithms; in order of increasing exhaustiveness and execution time, these were Fixed-States Optimization (Wheeler, 1999), Optimization Alignment (Wheeler, 1996), and Iterative Pass Optimization (Wheeler, 2003b).

Although Fixed-States Optimization was proposed as a novel means of conceptualizing DNA sequence homology (Wheeler, 1999), we employed it here simply as a heuristic shortcut. Because Fixed-States is so much faster than the Optimization Alignment algorithm, it allowed more thorough sampling of the universe of trees for subsequent refinement under more exhaustive optimization algorithms. Our general strategy was therefore to examine a large pool of initial candidate trees quickly under Fixed-States and submit those trees as starting points for further analysis under Optimization Alignment. Because the potential exists for the globally optimal tree (or trees that would lead to the global optimum when swapped under a more exhaustive optimization algorithm) to be rejected from the pool of candidates under the heuristic algorithm, we also generated a smaller pool of candidate trees under Optimization Alignment. The resulting optimal and near-optimal candidate trees were then submitted to final evaluation and refinement under Iterative Pass optimization using iterativelowmem to reduce memory requirements. (For details on tree-searching algorithms, see Heuristic Tree Searching, below.)

We did not employ the exact command during most searches, although we did use it in the final stages of analysis to allow accurate matrix-based length verification (Frost et al., 2001). To verify lengths reported in POY, we output the implied alignment (Wheeler, 2003a) and binary version of the optimal topology in Hennig86 format with phastwincladfile and opened the resulting file in Winclada (Nixon, 1999–2002). Because each topology may imply a different optimal alignment, when multiple optimal topologies were obtained we examined them separately by inputting each as a separate file using topofile. Examination of the implied alignments, whether formatted as Hennig files or as standard alignments (impliedalignment), grants another opportunity to detect errors in formatting or sequencing.

Heuristic Tree Searching

Efficient search strategies for large datasets are, to a certain degree, dataset dependent (Goloboff, 1999), and, as discussed above, common indicators of sufficiency are impractical given current technological limitations. Therefore, rather than apply a simple, predefined search strategy (e.g., 100 random addition Wagner builds + TBR branch swapping), we employed a variety of tree searching algorithms, spending more time on those that proved most efficient. Optimal trees from different searches were pooled for tree-fusing and TBR swapping, all of which was followed by refinement under Iterative Pass Optimization (Wheeler, 2003b). The search strategy is summarized in table 3.

Table 3

Summary of Tree Searching Methods Combined in Overall Search Strategyai0003-0090-299-1-1-t03.gif

Random addition sequence Wagner builds (RAS) were performed holding one or three trees. We conducted searches without slop or checkslop, both of which increase the pool of trees examined by swapping suboptimal trees found during the search; although these steps can be highly effective, initial trials showed they were too time consumptive for the present dataset.

The parsimony ratchet (Nixon, 1999) was proposed for analysis of fixed matrices. Given that under dynamic homology there are no prespecified column vectors to be reweighted, the original approach had to be modified. In the current version of POY, the ratchet is programmed to reweight randomly selected DNA fragments. The present dataset was broken into 31 fragments (see table 2), so ratchetpercent 15 randomly reweighted five fragments, regardless of their length or relative position. We reweighted 15–35% of the fragments and applied weights of 2–35 times.

As a complementary approach, we also performed quick searches (few random addition sequence Wagner builds + SPR) under indel-transversion-transition costs of 3-1-1, 1-3-1, and 1-1-3 and included the resulting topologies in the pool of trees submitted to fusing and refinement under equal weights, following the general procedure of d'Haese (2003). Reweighting in this method is not done stochastically and therefore differs from both Nixon's (1999) original version and POY's implementation of the ratchet and technically is not a simulated annealing or Metropolis-Hastings-type strategy like the others; however, because it weights sets of transformations drawn from throughout the entire dataset, it is likely to capture different patterns in the data and may actually be a closer approximation to the original ratchet than POY's implementation. Both approaches are effective methods to escape local optima.

We also performed constrained searches by calculating the strict consensus of trees within an arbitrary number of steps of the present optimum, saving the topology as a treefile, constructing the group inclusion matrix (Farris, 1973) in the program Jack2Hen, and then employing constraint in the subsequent searches. To calculate the consensus we included trees within 100–150 steps of the current optimum, the goal being to collapse enough nodes for swapping to be effective, but few enough nodes for significant speedups in RAS + swapping to find optimal arrangements within the polytomous groups (see Goloboff, 1999: 420). This is effectively a manual approximation of Goloboff's (1999) consensus-based sectorial search procedure, the main difference being that we collapsed nodes based only on tree length and not relative fit difference (Goloboff, 1999; Goloboff and Farris, 2001).

Using constraint files generated in the same way, we also input the current optimum as a starting point for fusing and/or ratcheting. This strategy avoids spending time on RAS builds of the unconstrained parts of the tree (which tend to be highly suboptimal) and seeks to escape local optima in the same way as unconstrained ratcheting, discussed above; however, there is a trade-off in that the arrangements may be less diverse but are likely to be, on average, closer to optimum, than those examined through RAS.

As a further manual approximation of sectorial searches, we analyzed subsets of taxa separately by defining reduced datasets with terminals files that listed only the targeted terminals. Rigorous searches (at least 100 RAS + TBR for each of the reduced datasets) of these reduced datasets were then performed, and the results were then used to specify starting topologies for additional searching of the complete dataset.

Static matrices may be thoroughly analyzed in a fraction of the time required to perform an equivalent analysis under dynamic homology. We therefore output implied alignments of current optima from POY and ran multiple replicates of RAS + 50 rounds of the parsimony ratchet using Winclada and NONA. Improvements were not always attained through this procedure, but when they were, we then input the optimal cladogram(s) from the static search as a starting point for further analysis in POY.

As a final attempt to discover more parsimonious solutions, we also rearranged branches of current optima manually. As a general search strategy this would obviously be highly problematic, if for no other reason than that it would bias analyses. However, we performed this step primarily to ensure that the “received wisdom” and other arrangements were evaluated explicitly in the analysis. The procedure was to open the current optimum in Winclada, target taxa whose placement was strongly incongruent with current taxonomy, and move them to their expected positions (or in polytomies, depending on the precision of the expectations). The resulting topologies were saved as treefiles that were read into POY as starting topologies for diagnosis and refinement (e.g., tree fusing). In this way we ensured that the more heterodox aspects of our results were not due to simply failing to evaluate the more orthodox alternatives during the automated searches.

Heuristic Data Exploration

To estimate support (sensu Grant and Kluge, 2003), we calculated Bremer (decay) values for all nodes present in the strict consensus of equally parsimonious solutions (Bremer, 1994). To accomplish this we output the implied alignment and optimal trees in Hennig86 format using phastwincladfile, converted it to NEXUS format in Winclada, and then generated a NEXUS inverse-constraints batch file in PRAP (Müller, 2004), which was analyzed in PAUP* 4.0b10 (Swofford, 1998–2002). Bremer analysis consisted of 1 RAS + 5 iterations of the parsimony ratchet (reweighting 25% of characters by 2) for each clade. More thorough analysis involving more rigorous tree searches of the unaligned data would undoubtedly lower the estimates; as is always the case with heuristic analysis, the Bremer values reported are an upper bound.

Species Identification

One of the goals of this study was to address problems in determining species identity. Through the course of the study species were identified on phenotypic grounds. Here we examine the bearing of evidence from DNA sequences and phylogenetic analysis on alpha taxonomic problems (see also Total Evidence Analysis, above).

As a means of identifying species limits, the results of phylogenetic analysis should be interpreted in light of several caveats: (1) Phylogenetic analysis presupposes that the genealogical relationships among the entities analyzed are phylogenetic (Davis and Nixon, 1992); as such, it will impose a hierarchy even on entities that are related tokogenetically, for example. In such cases, the branching structure would be an analytical artifact and finding that a species is or is not monophyletic would be irrelevant. (2) Species are historical individuals, and, as such, all parts of a given species need not form a clade (Skinner, 2004; see also Frost and Kluge, 1994). Incongruence between the history of different parts and the whole may be due to any number of natural phenomena, such as lineage sorting and partial/temporary introgression, none of which denies the historical individuality of the species. (3) Given a cladogram alone, there is no objective basis for identifying species limits, that is, there is no way to discriminate intra- from interspecific hierarchic structure without additional information. For example, dividing a pectinate cladogram of N terminals into 1 species, N − 1 species, and N species are all cladistically valid delimitations. As such, phylogenetic structure can only disconfirm hypotheses of species identity (but consider points 1 and 2); finding that the parts of a putative species form a clade does not deny that the clade may be composed of multiple species.

In spite of the above caveats, phylogenetic analysis is a valid (if fallible) species discovery operation (Frost et al., 1998). Conceptually, species are minimal historical individuals, meaning that species boundaries occur at the point where the properties of contemporary individuals dissolve (Kluge, 1990; Grant, 2002). Historical individuality may therefore be apprehended both from “below” by discovering the constituent parts that interact and “above” by individuating entities that are historically distinct. Incongruence between the results of complementary discovery operations (those directed from above and below, in this case) indicates heuristically that further study is warranted (Grant, 2002). Ultimately, species individuation requires diagnostic characters, and phylogenetic analysis facilitates their discovery.

To address the limits of problematic species, we considered (1) cladogram topology (cladistic distance), (2) branch lengths (patristic distance), and (3) uncorrected pairwise distance4 (uncorrected p, or number of base mismatches divided by total sequence length; no length variation was observed for this locus) of cytochrome b sequences within and between localities and/or closely related species. We focused on that sequence because (1) it is sufficiently variable and (2) it is almost completely represented in our dataset.

Our primary reason for including pairwise distances in this analysis is that they provide a rapid heuristic for species identification without conducting a complete phylogenetic analysis, in the same way that artificial dichotomous keys are efficient identification tools (Grant, 2002). We do not advocate using pairwise distances to delimit species. First, there is no justification for setting some arbitrary distance (e.g., 5%)—phenetic or otherwise—as “sufficient” for granting species status. Given variation in evolutionary rates and sampling density, it is expected that intraspecific variation may be greater in some species than interspecific variation among others. Indeed, the inability to distinguish between rate variation and artifacts due to taxon sampling (including extinction) casts doubt on all studies that base conclusions on degree of divergence or distance. What matters is the total evidence (including other loci, morphology, behavior, etc.) for the historical reality of the putative species and clades, for which character-state transformations must be identified to diagnose minimal historical individuals, not degree of similarity (pair- or otherwise). Second, as two-taxon statements, pairwise distances do not distinguish between symplesiomorphy and synapomorphy and therefore fail to explain the observed variation. Third, pairwise distance only discriminates among samples, that is, it is a relational concept and therefore cannot diagnose any particular entity (see Frost, 2000). Nevertheless, because they do not require extensive sampling or detailed analysis (phylogenetic or otherwise), pairwise comparisons are extremely fast and simple and therefore highly heuristic, and as such they are a useful starting point in examining species identity.

Phenotypic Characters

0. Dorsal Skin Texture (fig. 22): smooth  =  0; posteriorly granular  =  1; strongly granular  =  2; spiculate  = 3. Nonadditive.

Figure 22

Character 0, dorsal skin texture.

A: State 0, smooth (galactonotus, AMNH live exhibit). B: State 1, posteriorly tubercular (fraterdanieli, MHNUC 364). C: State 2, granular (macero, AMNH 129473). D: State 3, spiculate (Dendrophryniscus minutus, AMNH 93856).


Living and well-preserved anuran skin always has some texture, so even “smooth” skin may appear shagreen or faintly granular under high magnification. In state 0 all dorsal surfaces lack distinct tubercles or granules (e.g., histrionicus, abditaurantius). In state 1, granules or tubercles are scattered irregularly over the dorsal surfaces, being more distinct and prominent posteriorly, especially in the sacral region and on the thigh and/or shank, and absent or weaker and sparser anteriorly (e.g., boulengeri, fraterdanieli). These granules or tubercles are often distinctly elevated and conical. State 2 consists of rounded or flattened granules distributed densely and evenly (e.g., granuliferus, parvulus). Spiculate skin (state 3) is restricted to outgroup species; the skin of Dendropryniscus minutus is conspicuously spiculate, but in others (e.g., Atelopus spurrelli) the distinctly spiculate skin is only evident under magnification. Although state 1 is intermediate in the amount of granulation, the individual granules or tubercles of states 1 and 2 are qualitatively different, and there is no developmental evidence to suggest that transformations between states 0 and 2 pass through state 1.

Dorsal skin texture has generally been used descriptively in species-level taxonomic studies (e.g., Myers et al., 1995, in distinguishing between pumilio and granuliferus; Silverstone, 1976, in distinguishing between femoralis and boulengeri). Jungfer (1989) reviewed the “red-backed granulated” Amazonian dendrobatids but did not explicitly delimit them as a group.

Unlike some other anurans (e.g., centrolenids), skin texture of dendrobatids varies only minimally (or not at all) in relation to season and/or reproductive activity, with variation involving only the protuberance of granules or tubercles and not the occurrence of different states. Nevertheless, care must be exercised in coding this character (and others involving dermal structures) because it is prone to alteration due to preservation. Inadequately fixed or preserved specimens tend to lose granularity or even slough the epidermis. Even well-preserved specimens fixed according to the standard procedure of laying the specimen in a fixing tray prior to immersion in formalin are often less granular than they were in life. Conversely, granularity may be exaggerated in desiccated specimens. As noted by Myers and Daly (1979: 5, footnote 1), the best means of preserving skin texture (as well as other dermal characters such as hand and foot tubercles) is to float them completely in formalin immediately. (However, we do not mean to endorse this method generally, as it complicates examination of the vast majority of morphological characters.)

Heyer (1983: 322) provides electronmicrographs showing the skin texture for Cycloramphus boraceiensis. We coded pulcherrimus according to Duellman's (2004) description.

1. Paired Dorsal Digital Scutes: absent  =  0; present  =  1.

All species of dendrobatids have distinctive paired dermal scutes atop digital discs, although they may be inconspicuous on first and last digits and are generally most strongly expressed on the third finger and fourth toe (i.e., on discs that are most expanded). Noble (1926: 7) cited this character as evidence uniting dendrobatids in a single, exclusive group, and since then it has been used consistently to diagnose dendrobatids. Noble and Jaeckle (1928) examined the histology of the digital discs and illustrated (but did not discuss) the digital scutes of a specimen they identified as Phyllobates latinasus (actual species unknown but probably not latinasus; see Grant, 2004, for discussion of latinasus taxonomy) and Hylodes nasus (as Elosia bufonia). Noble (1931) noted the occurrence of the digital scutes in his Elosiinae (p. 504) and dendrobatids (p. 507). Although he did not explicitly state that it was homologous in the two groups, that was implied by his hypothesis that dendrobatids arose from “Crossodactylus or a form closely allied to it”. He further observed (p. 520) that both dendrobatids and the African ranid Petropedetinae had “apparently identical” dermal scutes on the upper surface of each digit (he did not comment on the shared occurrence of this state in his Elosiinae), but he explained away this similarity as adding “one more to the many cases of parallel evolution in the Salientia”. Liem and Hosmer (1973: 473) also noted that the myobatrachid genus Taudactylus has “expanded digital discs with a median longitudinal groove dorsally”. Lynch (1979) illustrated the discs of groups known to possess digital scutes or scutelike structures. Lynch (1979: 7) clarified that the scutes are “flaplike structures”, which distinguishes them from superficially similar digits of some Eleutherodactylus that exhibit only a median groove. La Marca (1995: fig. 9) provided scanning electron micrographs of the digital scutes of collaris, herminae, oblitterata, neblina, olmonae, riveroi, trinitatis, yustizi, and an undescribed species. Griffiths (1959: 482) claimed that the scutes are “really glandulo-muscular organs and probably function to facilitate adhesion to foliage etc.”, but no evidence has been presented in support of his thesis and their functional significance remains unknown.

2. Supernumerary Tubercles on Hand: absent  =  0; present  =  1.

Most dendrobatids possess a large, subcircular palmar tubercle and an elliptical thenar tubercle. Many nondendrobatids also possess distinct supernumerary tubercles scattered over the fleshy part of the palm (e.g., Lynch and Duellman, 1997). As part of their polymorphism, some dendrobatids exhibit a tiny tubercle-like thickening on the outer edge (not the fleshy part) of the palm, but we do not consider this to be homologous with the supernumerary tubercles of other taxa.

3. Distal Tubercle on Finger 4 (fig. 23): absent  =  0; present  =  1.

Figure 23

Character 3, distal subarticular tubercle of finger IV.

A: State 0, absent (degranvillei, AMNH 90876). B: State 1, present (pictus, AMNH 79209).


Most dendrobatids possess both proximal and distal subarticular tubercles on finger IV (state 1). Grant and Rodríguez (2001) noted that in some species the distal tubercle on finger IV is absent (state 0) and that, although this is often associated with reduction in the length of finger IV (Character 4), some species that lack this tubercle show no reduction in finger length (e.g., melanolaemus, pumilio), which demonstrates the transformational independence of the two characters. This independence is further reinforced by examination of outgroup taxa, as Thoropa miliaris possesses a long finger IV and lacks the distal subarticular tubercle.

4. Finger IV Length (fig. 24): surpassing distal subarticular tubercle of finger III  =  0; reaching distal half of distal subarticular tubercle of finger III  =  1; not reaching distal subarticular tubercle of finger III  =  2. Additive.

Figure 24

Character 4, length of finger IV.

A: State 0, surpassing distal subarticular tubrcle of finger III (histrionicus, AMNH 88259). B: State 1, reaching distal half of distal subarticular tubercle of finger III (tricolor, USNM 286082). Note also the strong preaxial swelling of finger III. C: State 2, not reaching distal subarticular tubercle of finger III (insperatus, KU 149684). Note also the absence of the distal subarticular tubercle of finger IV.


The length of finger IV is assessed by pressing it against finger III to determine if it extends well beyond the distal subarticular tubercle (state 0), reaches the distal half of, but does not surpass, the distal subarticular tubercle (state 1), or does not reach the distal subarticular tubercle (state 2). In the latter state, finger IV extends to a point approximately midway between the proximal and distal subarticular tubercles Although it is possible that we have conflated transformations involving the length of finger III, the fact that finger II reaches the distal half of the distal subarticular tubercle in all species supports indirectly the hypothesis that variation is due exclusively to transformations of finger IV, that is, if the observed variation is due to changes in the length of finger III, then the same change would also have had to affect the relative length of finger II. And it is further supported by the loss of the distal subarticular tubercle in species with a relatively short finger IV (see Character 3, above). Given the constancy of the length of finger II, this character is equivalent to the traditional taxonomic coding of finger IV versus finger II (i.e., when both are pressed against finger III, in state 0 IV is longer than II, in state 1 fingers IV and II are equal, and in state 2 IV is shorter than II).

5. Relative Lengths of Fingers I and II: I << II (II 1.2 or more times longer than I)  =  0; I < II  =  1; I  =  II  =  2; I > II  =  3. Additive.

Traditionally, the relative lengths of fingers I and II have been assessed by pressing these two fingers together at the point midway between the two digits. However, this is highly dependent on the investigator's judgment of the midway point between the two digits, that is, bringing finger I further toward finger II (or vice versa) can affect coding of this character. Kaplan (1997) measured the length of each finger from the same point at the base of the palmar tubercle to the tip of each finger, which is more precise and less prone to error, and we employed his method here. Kaplan's method also assumes that there are no carpal changes that affect the distance from the palm to finger tips differentially (no such variation was detected). Any method of measuring finger length requires that the fingers be straight; when well-preserved hands were unavailable, digits were straightened for measurement. In state 0 finger II is at least 20% longer than finger I; in state 1 finger II is less than 15% longer than finger I; in state 2 the fingers are subequal in length; in state 3 finger I is unambiguously longer than finger II.

Although developmental data are unavailable, gross morphology suggests that state transformations are due to variation in the length of finger I and not the length of finger II, that is, the length of finger II relative to finger III was not observed to vary, as it reaches the midlevel of the distal subarticular tubercle in all taxa. However, it is possible that two characters have been conflated, that is, one involving variation in the length of finger I, the other variation in the length of finger II. We did not attempt to relate the differences in relative lengths with the underlying osteology, which could also reveal that multiple characters have been conflated.

6. Digital Discs: absent  =  0; present  =  1.

The differentiation of the digital terminations into expanded discs with adhesive pads has long been used to infer anuran relationships (e.g., Cope, 1867). Numerous authors (e.g., Noble and Jaeckle, 1928; Green, 1979; Emerson and Diehl, 1980; Rivero et al., “1987” 1989; Ba-Omar et al., 2000) have examined the structure (and function) of the disc apparatus in a diversity of frogs and have found them to differ in only minor structural details (e.g., number of epidermal cell layers). We are unaware of any anuran that possesses finger discs but lacks toe discs (or vice versa), or that possesses discs on some but not all digits (although degree of expansion certainly varies among digits; see below). We have therefore treated the origin (and loss) of digital discs as a single transformation series. All dendrobatids possess digital discs, but they are absent in several of the sampled outgroup taxa.

7–10. Expansion of Finger Discs (Fig. 25)

In the dendrobatid literature, expansion of finger discs is generally treated as one or at most two characters. Duellman and Simmons's (1988) standard diagnosis coded only the disc of finger III, and Dendrobates species descriptions often report the expansion of discs I and II–IV separately. However, there is no logical dependency between the discs of different digits, and the distribution of states in the matrix shows that the expansion of each digital disc is transformationally independent, and, as such, they are defensibly coded separately for analysis. A trend is that the disc of finger I is often (but not always) less expanded than those of the other fingers, but this does not violate the transformational independence of these characters.

We detected four discrete states in finger (and three in toe) disc expansion, shown schematically in figure 25. All dendrobatids have digital discs, so some degree of expansion is always detectable, although it may be extremely slight. This is exemplified by elachyhistus and pumilio, in which the disc of finger I appears unexpanded or at most weakly expanded (state 0). States 1 and 2 are found in most dendrobatids; state 3 is found in those species with greatly expanded discs (e.g., tinctorius). State 3 was only observed among fingers II–IV.

Figure 25

Characters 7–10 and 31–35, schematic illustration of the four states observed in the expansion of digital discs.

The digital shaft is indicated by the inner crosshatched area and the outer edges of the disc are indicated by the heavier outer line.


Polder (1973: 17) and Silverstone (1975a) claimed that some species of dendrobatids are sexually dimorphic in the expansion of the digital discs, with males possessing larger discs than females. Neither author provided quantitative data, however, and when Myers and Daly (1976b: 203) tested the claim quantitatively in histrionicus they found it to be unsupported. Although we detected (and coded) polymorphism in disc expansion in some species (including histrionicus), we concur with Myers and Daly (1976b) that it does not reflect differences between sexes. For example, in leucomelas the finger discs of male AMNH 137309 are larger than those of female AMNH 137310, but no more expanded than those of female AMNH 46051. We did not test the hypothesis that discs of males are statistically (i.e., on average) larger than those of females (Silverstone, 1975a: 8) because that question is unrelated to the problem of homologizing character-states and inferring transformation events (Grant and Kluge, 2003, 2004).

7. Finger Disc I: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

8. Finger Disc II: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2; greatly expanded  =  3. Additive.

9. Finger Disc III: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2; greatly expanded  =  3. Additive.

10. Finger Disc IV: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2; greatly expanded  =  3. Additive.

11–18. Finger Fringes (Fig. 26)

The occurrence and extent of lateral keels and/or fringes on fingers and toes has been cited in most alpha taxonomic studies of dendrobatids for the past several decades at least (e.g., Edwards, 1971). Duellman and Simmons (1988: 116) noted that “the development of fringes on the fingers is variable, so standard comparison is made with the second finger.” Nevertheless, as discussed above in reference to expansion of digital discs, because the fringes on each edge of each finger vary independently we coded them as separate characters.

Although lateral dermal expansions of the digits are commonly described in the dendrobatid literature, explicit delimitations of character-states are generally lacking, which has led to considerable confusion. They are generally referred to as either keels or fringes. As noted by Lynch and Duellman (1997: 33) for species of Eleutherodactylus, “there is a continuum from keels to fringes, and in some cases the distinction is arbitrary.” Although such arbitrariness is relatively harmless in descriptive taxonomic studies, the cumulative effect of arbitrary delimitations can be disastrous in phylogenetic analyses. Coloma (1995: 6–7) noted that fringes may be absent, poorly developed, or well developed. He further clarified that “[w]hen it was difficult to distinguish between a real fringe and a preservation artifact, I describe the dermal modification as a ‘keel’ ”, which, although explicit, actually engenders greater confusion because keels are generally considered to be real dermal modifications (e.g., Lynch and Duellman, 1997).

The strength of keeling (extent of dermal thickening) varies extensively, leading La Marca (1996 “1994”: 6) to differentiate between keels and fringes as “very low” and “conspicuous but not folding around the toes”, respectively. However, although we agree that these descriptors encompass the observed variation, and despite numerous dissections, we were unable to individuate character-states objectively. Any attempt to subdivide state 0 into multiple character-states must overcome two difficulties: (1) apparently continuous variation (as suggested by external examination and gross dissections) and (2) the fact that these dermal expansions are highly prone to postmortem modification, either due to desiccation (as indicated by Coloma, 1995) or simply as an artifact of preservation (and variation in preservation techniques). It is likely that the greater precision attained through histological study could overcome both of these problems, but that was beyond the scope of the present study. For the purpose of phylogenetic analysis, we individuated only two character-states: fringes absent (state 0) and fringes present (state 1).

In state 0, the extent of lateral dermal expansion varies from absent (i.e., the side of the digit is smoothly rounded and there is no detectable dermal thickening along the lateral margin) to conspicuously keeled. In state 1, the skin that extends from the dorsal surface extends ventrad and appears to fold over the side of the digit, which we refer to as a fringe (see fig. 26). In ventral (palmar) view the folding over can be seen to create a deep longitudinal crease or groove. We have not detected evidence that the folding over varies as an artifact of preservation, providing a basis to distinguish this state objectively. This state is approximately equivalent to La Marca's (1996 “1994”: 6) “flaps”, which he diagnosed as “folding around the toes”. The strength of fringes varies from a weak flap (e.g., toes of degranvillei) to a strong flap that wraps around much of the ventral surface of the digit (the latter condition found only on toes; see below), but we were unable to delimit distinct states.

Figure 26

Characters 11–18, finger fringes.

In Megaelosia goeldii (AMNH 103949) fringes are present on pre- and postaxial edges of all fingers.


Webbing between the fingers does not occur in any dendrobatid we examined. Donoso-Barros (1965 “1964”: 486) described “rudimentary web between 2nd and 3rd fingers” in riveroi, but finger webbing was not reported by La Marca (1996 “1994”) and is absent in the specimens we examined. Similarly, Coloma (1995) described and illustrated webbing between the fingers in an undescribed species (as Colostethus chocoensis; see Grant et al., 1997: 24, footnote 13), and Grant et al. (1997: 25) mentioned the possible occurrence of webbing on the hands of atopoglossus. However, closer examination of the same specimens of “Colostethus chocoensis” and atopoglossus revealed that the apparent webbing is due to flattening of the loose skin of the hand, as considered by Grant et al. (1997). Lynch (1971: 30) reported similar mistaken reports among “leptodactyloids”.

11. Finger Fringe: I preaxial: absent  =  0; present  =  1.

12. Finger Fringe: I postaxial: absent  =  0; present  =  1.

13. Finger Fringe: II preaxial: absent  =  0; present  =  1.

14. Finger Fringe: II postaxial: absent  =  0; present  =  1.

15. Finger Fringe: III preaxial: absent  =  0; present  =  1.

16. Finger Fringe: III postaxial: absent  =  0; present  =  1.

17. Finger Fringe: IV preaxial: absent  =  0; present  =  1.

18. Finger Fringe: IV postaxial: absent  =  0; present  =  1.

19. Metacarpal Ridge (fig. 27): absent  =  0; weak  =  1.

Figure 27

Character 19, metacarpal ridge.

State 1, present (abditaurantius, ICN 9853).


The metacarpal ridge or fold is a dermal thickening running from the postaxial edge of the base of finger IV along the outer edge of the palm toward the palmar tubercle. In most species an edge is formed where the relatively flattened palm meets the rounded side of the hand, but we did not consider this to be a metacarpal ridge unless dermal thickening could be detected, either by gross inspection or by making a transverse incision. Although there is some variation among species in the degree of expression of the metacarpal ridge, it was minor and we were unable to delimit discrete states. As with other dermal characters, the metacarpal ridge may be exaggerated or lost as an artifact of preservation.

20–21. Finger III Swelling

Reproductively active males of numerous dendrobatids present swollen third fingers, a condition that is unknown in nondendrobatids. The “swelling” is due to the occurrence of extensive glandular tissue, the large granules often being evident in gross dissection or even through the skin. In light of the important role this character has played in recent discussions of dendrobatid systematics (e.g., Myers et al., 1991; but see Myers, 1991), we review its usage here.

Although a number of species possessing a distinctly enlarged third finger in males had been described previously (e.g., trilineatus, the holotype of which is a male), the first worker to describe and illustrate the swollen third finger was Dunn (1924: 7–8) for nubicola. Descriptions of “digital dilatations” or “enlargements” in the earlier literature referred to the expanded digital disc apparatus (e.g., Cope, 1867: 130, 1887: 55). When Dunn (1931) named flotator, he grouped it with nubicola based in part on the shared occurrence of the swollen third finger in males. Dunn (1933) noted that males of panamensis possess a swollen third finger, but he did not attribute any phylogenetic significance to the observation.

Over the 50 years following Dunn's first report of the swollen third finger in males, the state of the third finger was mentioned sporadically in diagnoses and descriptions (e.g., among papers that deal with species with swollen third fingers in males, it was mentioned by Dunn, 1931, 1957; Funkhouser, 1956; Savage, 1968; Cochran and Goin, 1970: 60 [only for their Phyllobates inguinalis, as “flanges” on the third finger of males]; Edwards, 1971; and Silverstone, 1971, 1976; but not by Cochran and Goin, 1964; Cochran, 1966; or Silverstone, 1975b), but was not illustrated again until 1974, when Edwards provided a schematic representation in his unpublished (but widely distributed; see Myers et al., 1991: 30, footnote 14) dissertation (Edwards, 1974a).

The character has been mentioned fairly consistently since 1974, but miscoding is common, probably due in part at least to the inadequacy of Dunn's (1924) and Edwards's (1974a) illustrations, both of which depicted (1) roughly equal expansion on both sides (preaxial and postaxial) of the digit and (2) distally exaggerated swelling, neither of which is found in all (or even most) of the species with swollen third fingers. Similarly, although accurate for a few species, Duellman and Simmons's (1988: 117) description that “the basal segment of the third finger is distinctly swollen in males” does not apply to most of the species with clearly swollen third fingers in males (and none of the species they addressed in their paper). The expansion of the third finger can be much more subtle than Dunn's (1924) and Edwards's (1974a) illustrations suggest, and significant variation occurs in the extent and location of the swelling.

Silverstone (1976: 33) noted in his account of tricolor that not all adult males of a given sample may express the swollen third finger, a finding that was corroborated more generally by Myers et al. (1991), who speculated that expression is likely under hormonal control. This and additional difficulties related to the coding of this character were discussed by Myers et al. (1991, 1998, see especially fig. 4), Myers (1991), Myers and Donnelly (1997), and Grant and Rodríguez (2001).

20. Finger III Swelling in Adult Males: absent  =  0; present  =  1.

This character was scored for awa from Coloma (1995) because no adult males were included in the series we examined.

21. Morphology of Swollen Third Finger in Males (fig. 28): pre- and postaxial swelling  =  0; weak preaxial swelling  =  1; strong preaxial swelling  =  2; swelling extending from wrist, mainly preaxial on digit  =  3. Nonadditive.

Figure 28

Character 21, morphology of swollen third finger in males.

A: State 0, pre- and postaxial swelling (mertensi, ICN 43698). B: State 1, weak preaxial swelling (insperatus, KU 149676). C: State 2, strong preaxial swelling (nubicola, AMNH 114574). D: State 3, swelling extending from wrist, mainly preaxial on digit (baeobatrachus, AMNH 140650).


22. Carpal Pad (fig. 29): absent  =  0; present  =  1.

Figure 29

Character 22, male supracarpal pad (undulatus, AMNH 159134).


Myers and Donnelly (2001) discovered the carpal pad in undulatus. It consists of a conspicuous nonglandular thickening and heavy melanosis of the skin above the wrist of males. We did not find this character to be present in any other species, but we include it here in anticipation of future discoveries.

23. Male Excrescences on Thumb: absent  =  0; present  =  1.

Although nuptial excrescences are common among outgroup taxa, most dendrobatids lack nuptial excrescences (state 0), the sole exception being oblitterata, which was reported as possessing nuptial excrescences (state 1) by La Marca (1995: 66).

We coded Telmatobius jahuira for this character following Lavilla and Ergueta (1995).

24. Morphology of Male Excrescences on Thumb: large, cornified spines  =  0; small, uncornified spines  =  1; nonspinous asperities  =  2. Additive.

Lavilla and Ergueta (1995: 49, translated freely from the Spanish) described the nuptial excrescences of Telmatobius jahuira as “few cornified spines separated by large, unkeratinized spaces”.

25. Female Excrescences on Thumb: absent  =  0; present (large, cornified spines)  =  1.

See Noble (1931: 122, 126) for illustrations and comments on the large, cornified spines on the thumb of females of species of Crossodactylus.

26. Thenar Tubercle (fig. 30): absent or small, inconspicuous swelling  =  0; large, conspicuous, well-defined tubercle  =  1.

Figure 30

Character 26, thenar tubercle.

A, B: State 0, absent or small, inconspicuous swelling (pumilio, AMNH 102262). In this specimen, the thenar tubercle appears absent in both palmar aspect and profile. C, D: Another specimen of the same species (pumilio, AMNH 102263). In this specimen, the thenar tubercle is inconspicuous but clearly seen in profile (also scored as state 0). E, F: State 1, large, conspicuous, protuberant tubercle (nubicola, AMNH 114574).


Most dendrobatids have a conspicuous, protuberant, elliptical thenar (inner metacarpal) tubercle (state 1). Silverstone (1975a) noted that the thenar tubercle is inconspicuous or absent in leucomelas (state 0). Likewise, Caldwell and Myers (1990) illustrated and discussed the absence of the thenar tubercle in castaneoticus and quinquevittatus, which they interpreted as a synapomorphy uniting these two species in an exclusive clade (they did not make comparisons with leucomelas). Other species also exhibit the same morphology (e.g., pumilio).

Caldwell and Myers (1990: 16) noted some variation in the expression of the thenar tubercle in quinquevittatus; in some specimens it is altogether undetectable, whereas in others “possible vestiges of it” were detected as “possibly represented by slight epidermal thickening”. Our observations concur with theirs. Given the propensity for such subtle dermal features to be lost as an artifact of preservation (due to skin sloughing, desiccation, or inadequate fixation, among other causes), we combined the apparent complete absence and inconspicuous epidermal thickening as state 0. Expression of the thenar tubercle is not dependent on overall body size; leucomelas is quite large, and the thenar tubercles of the small species nubicola and stepheni (roughly the same snout-vent length as pumilio) are large and well defined.

27. Black Arm Gland in Adult Males: absent  =  0; present  =  1.

This character was identified, discussed, and illustrated photographically by Grant and Castro-Herrera (1998; see also Grant and Ardila-Robayo, 2002) and used to delimit the ramosi group. It remains unclear if this patch of black, thickened tissue on the ventral and medial surfaces of the distal extreme of the upper arm and often extending onto the inner surface of the lower arm is glandular, but its absence in females and juveniles and exaggeration in sexually active males suggests it is involved in amplexus and probably under hormonal control. In addition to the species listed by Grant and Ardila-Robayo (2002), this character is also present in anthracinus and the undescribed species referred to herein as “Ibagué”.

28. Tarsal Keel: absent  =  0; present  =  1.

The tarsal keel is a dermal structure that extends obliquely along the plantar surface of the tarsus. Regardless of its point of origin (see Character 29), it always terminates medially, not on the margin of the tarsus (see Character 30). Silverstone (1975a, 1976) used variation in this structure to diagnose species groups in Dendrobates and Phyllobates, and Lynch (1982) cited the loss of the tarsal keel in edwardsi and ruizi to delimit the edwardsi group of Colostethus.

Silverstone (1975a: 8) treated the “tarsal fold” and “tarsal tubercle (at the proximal end of the tarsal fold)” as separate characters. He considered the tarsal fold to be present in all Dendrobates and the tarsal tubercle to be both present and absent in Dendrobates. However, the tarsal fold and tarsal tubercle form a single structure, the tubercle simply being an increased thickening of the proximal portion of the keel. This is especially clear in many nonaposematic dendrobatids (which were not the focus of Silverstone's work) in which the proximal end of the keel is conspicuously enlarged and may be described as tubercle-like, but is sharply curved to run across the tarsus and does not conform to the rounded structures usually referred to as tubercles.

29. Morphology of Tarsal Keel (fig. 31): straight or very weakly curved, extending proximolaterad from preaxial edge of inner metatarsal tubercle  =  0; tubercle-like (i.e., enlarged) and strongly curved at proximal end, extending from metatarsal tubercle  =  1; short, tubercle-like, curved or directed transversely across tarsus, not extending from metatarsal tubercle  =  2; weak, short dermal thickening, not extending from metatarsal tubercle  =  3. Additive.

Figure 31

Character 29, morphology of tarsal keel.

A: State 0, straight or very weakly curved, extending proximolaterad from preaxial edge of inner metatarsal tubercle (imbricolus, AMNH 102082). B: State 1, tuberclelike and strongly curved at proximal end, extending from metatarsal tubercle (degranvillei, AMNH 90876). C: State 2, short, tuberclelike, curved or directed transversely across tarsus, not extending from metatarsal tubercle (“Neblina species”, AMNH 118657). D: State 3, weak, short dermal thickening, not extending from metatarsal tubercle (pumilio, AMNH 102261). The hind limb is rotated to view the inconspicuous tarsal keel in profile.


30. Tarsal Fringe (fig. 32): absent  =  0; present  =  1.

Figure 32

Character 30, tarsal fringe.

State 1, present (Megaelosia goeldii, AMNH 103950).


The tarsal fringe consists of a conspicuous dermal flap that runs along the entire length of the preaxial edge of the tarsus; it is continuous with the fringe on toe I. The tarsal fringe differs from the tarsal keel (characters 28, 29) in that the latter extends proximolaterad across the tarsus to terminate at roughly the middle of the tarsus on the plantar surface, whereas the former never crosses the tarsus and extends along its entire length.

31–35. Expansion of Toe Discs

Like finger discs, dendrobatid literature generally treats the degree of expansion of toe discs as a single character. However, as discussed above under finger discs, toe discs vary independently of one another and are defensibly treated as separate characters. Toe discs exhibit three of the four character-states found in fingers; the greatest expansion found in finger discs (finger disc state 3) does not occur in toe discs. (Character-states are figured schematically in fig. 25, above.)

31. Toe Disc I: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

32. Toe Disc II: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

33. Toe Disc III: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

34. Toe Disc IV: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

35. Toe Disc V: unexpanded  =  0; weakly expanded  =  1; moderately expanded  =  2. Additive.

36–45. Toe Webbing

Webbing has been used consistently in dendrobatid systematics since Noble (1923, 1926) diagnosed Phyllobates from Hyloxalus on the basis of reduced webbing. Although webbing can be argued to form a single, integrated functional unit (as can the entire organism), functional independence is at most secondary to historical independence in phylogenetic inference (Grant and Kluge, 2004), and there is ample evidence that the extent of webbing along each edge of each digit varies independently. Coding follows the nomenclature proposed by Savage and Heyer (1967) and subsequently modified by Myers and Duellman (1982), which quantifies webbing in terms of the number of free phalanges, assessed in relation to subarticular tubercles (e.g., in Character 40, state 6, the two distal phalanges are free of webbing). We consider toe fringes (defined as for fingers, above) to be homologous with webs. We do not consider lateral fringes that meet between the toes to constitute a web unless it is expanded relative to lateral fringes, that is, if the continuous lateral fringes are broader at the base than along the sides of the digits, we construe this as being a web.

Among the sampled outgroup taxa, McDiarmid (1971: 33) noted that the interdigital webbing of Atelopus and Dendrophryniscus “is not a membrane, as defined by Peters (1964) but rather a thickened integumentary connection between digits, similar to the webbing encountered in many of the more terrestrial anurans, such as toads of the genus Bufo.” This suggests that the interdigital webbing of these species may not be homologous with that of other anurans. Nevertheless, although the distinction is clear in Dendrophryniscus minutus, it is less so in the sampled species of Atelopus, and we have therefore treated webbing as a single transformation series and allowed character congruence to be the ultimate arbiter (although much more extensive sampling of relevant taxa will be required to fully resolve the question).

36. Webbing: Toe I Preaxial: absent  =  0; fringe  =  1.

37. Webbing: Toe I Postaxial: absent  =  0; fringe  =  1; 2  =  2; 1.5  =  3; 1  =  4; 0  =  5. Additive.

Coloma (1995: 51) reported basal webbing (I2–3.5II) for talamancae and toachi, but there is no trace of webbing in the specimens we examined in this study.

38. Webbing: Toe II Preaxial: absent  =  0; 2.5  =  1; 2  =  2; 1  =  3; 0  =  4. Additive.

Coloma (1995: 51) reported basal webbing (I2–3.5II) for talamancae and toachi, but there is no trace of webbing in the specimens we examined.

39. Webbing: Toe II Postaxial: absent  =  0; 2 (without fringe)  =  1; 2 (with fringe)  =  2; 1.5  =  3; 1  =  4; 0  =  5. Additive.

40. Webbing: Toe III Preaxial: absent  =  0; fringe  =  1; 3.5 (without fringe)  =  2; 3.5 (with fringe)  =  3; 3  =  4; 2.5  =  5; 2  =  6; 1.5  =  7; 1  =  8. Additive.

Coloma (1995: 51) reported more extensive webbing (equivalent to state 4) for talamancae than we observed (state 2).

41. Webbing: Toe III Postaxial: absent  =  0; 3 without fringe  =  1; 3 with fringe  =  2; 2.5  =  3; 2  =  4; 1.5  =  5; 1  =  6. Additive.

42. Webbing: Toe IV Preaxial: absent  =  0; 4 without fringe  =  1; 4 with fringe  =  2; 3.5  =  3; 3  =  4; 2.5  =  5; 2  =  6; 1  =  7. Additive.

43. Webbing: Toe IV Postaxial: absent  =  0; fringe  =  1; 4  =  2; 3.5  =  3; 3  =  4; 2.5  =  5; 2  =  6; 1  =  7. Additive.

Coloma (1995: 51) reported basal webbing (IV4.5–3V) in talamancae, but there is no trace of webbing in the specimens we examined.

44. Webbing: Toe V Preaxial: absent  =  0; fringe  =  1; 2.5 (with fringe)  =  2; 2  =  3; 1.5  =  4; 1  =  5. Additive.

45. Webbing: Toe V Postaxial: absent  =  0; fringe  =  1.

This character was coded for insulatus, pulcherrimus, and sylvaticus Barbour and Noble from Duellman (2004), who reported it as absent in them all.

46. Metatarsal Fold (fig. 33): absent  =  0; weak  =  1; strong  =  2. Additive.

Figure 33

Character 46, metatarsal fold.

A: State 1, weak (“Neblina species”, AMNH 118657). B: State 2, strong (degranvillei, AMNH 90876).


The metatarsal fold is a dermal thickening running from the postaxial edge of the base of toe V (often coextensive with the fringe, if present) along the outer edge of the sole toward the outer metatarsal tubercle. In most species an edge is formed where the relatively flattened sole meets the rounded side of the foot, but we did not consider this to be a metatarsal ridge or fold unless actual dermal thickening could be detected, either by gross inspection or by dissection. A weak metatarsal fold (state 1) is a ridge; strong dermal folds (state 2) are often folded over or angled relative to the surface of the sole.

47. Cloacal Tubercles: absent  =  0; present  =  1.

Grant et al. (1997) identified and figured this pair of tubercles adjacent to the cloaca near the base of the thighs. They also discussed difficulties in scoring this character due to postmortem artifacts.

48–66. External Coloration

Much of the diversity of dendrobatids involves variation in external color and color pattern. Among species referred to Colostethus, for example, variation in the pattern of lateral stripes and ventral color serves as one of the main tools for diagnosis. However, color and color pattern are perhaps the most confounding–and therefore undersampled in this study–sources of variation. Several aposematic dendrobatids (e.g., pumilio, histrionicus, and tinctorius) are renowned for their astonishing intra- and interpopulational variation in color and color pattern, and the difficulties posed by this immense and often continuous and overlapping variation can be immediately appreciated by glancing at a few pages of Myers et al.'s (1976b) account of histrionicus. Practically speaking, the three main difficulties are (1) detection of objective boundaries between different characters and character-states, (2) requirement of many states per character, and (3) distinguishing between color and color pattern. We made every effort to incorporate as much of the variation as possible, but much of it was overwhelming. Also, for ease of coding (especially preserved specimens) we focused more on color pattern than color, but by doing so we undoubtedly conflated characters and character-states. For example, we scored both auratus and reticulatus as having the thighs pale with dark spots, even though the thighs are different colors. Future studies will undoubtedly advance considerably beyond the current project by scoring more of this diversity of color and color pattern.

48. Iridescent Orange or Golden Spot at Dorsal Limb Insertions: absent  =  0; present  =  1.

Note that the photo of quinquevittatus in Caldwell and Myers (1990: 11) shows that this character is not redundant with or nonindependent of the thigh coloration characters.

49. Pale Paracloacal Mark (fig. 34): absent  =  0; present  =  1.

Figure 34

Character 49, pale paracloacal mark.

State 1, present (degranvillei, AMNH 90880).


This is a pale, elongate mark at the base of the thigh. The shape of the spot varies from a straight vertical line to a sickle extending as a pale longitudinal stripe along the posterior surface of the thigh. The paracloacal mark originates adjacent to the vent at the base of the thigh, not in the groin or on the top the thigh (as does the pale mark in femoralis, for example; see character 50).

50. Thigh Dorsal Color Pattern (fig. 35): pale with dark spots (forming reticulum when spots are close together)  =  0; solid dark  =  1; dark with pale spots/bands  =  2; solid pale  =  3; brown with dark brown bands/blotches  =  4; dark with pale longitudinal stripe  =  5. Nonadditive.

Figure 35

Character 50, dorsal thigh color pattern.

A: State 0, pale with dark spots (quinquivittatus, AMNH 124069). B: State 1, solid dark (petersi, AMNH 111000). Note that the pale spot is confined to the inguinal regions and does not extend onto the dorsal surface of the thigh. C: State 2, dark with pale spots/bands (aurotaenia, AMNH live exhibit). D: State 3, solid pale (terribilis, AMNH live exhibit). E: State 4, brown with dark brown bands/blotches (inguinalis, LACM 42409). F: State 5, dark with pale longitudinal stripe (flavopictus, AMNH 88642).


51. Discrete Pale Proximoventral Calf Spot (fig. 36): absent  =  0; present  =  1.

Figure 36

Character 51, discrete pale proximoventral calf spot.

State 1, present (imbricolus, AMNH 102082).


Silverstone (1975a, 1975b) used the absence (state 0) and presence (state 1) of a discrete, pale spot on the proximal portion of the concealed surface of the shank to diagnose species and species groups. In life it is a bright flash mark. A number of species (e.g., fraterdanieli) have bright flash coloration on the concealed surface of the shank, but it does not form a discrete spot and we therefore follow Silverstone in scoring this character as absent for those species.

52–57. Pale Lateral Stripes

Edwards (1974a) used the combinations of pale lateral stripes (or lines) to diagnose species of Colostethus, identifying dorsolateral, oblique lateral, and ventrolateral stripes. Previous workers (e.g., Savage, 1968) had drawn attention to these characteristics as well, but Edwards standardized the distinction between the three stripes and has been followed by most authors. Nevertheless, caution must be employed when consulting the literature, as terminology varies. For example, what is referred to here as the oblique lateral stripe was referred to as a dorsolateral stripe by Edwards (1971) and Haddad and Martins (1994), and consistently as the inguinal stripe by La Marca (e.g., 1985, 1996 “1994”, 1998 “1996”; see also Myers and Donnelly, 2001). Duellman and Simmons (1988) discussed these characters as “pale longitudinal stripes”, and Coloma (1995) followed their usage. Duellman (2004) distinguished between the oblique lateral and dorsolateral stripes in his Summary of Taxonomic Characters but used them interchangeably in the text (e.g., idiomelus and sylvaticus are diagnosed as lacking oblique lateral stripes and possessing dorsolateral stripes, but the converse is true for both species; e.g., see Duellman's figs. 5F and 6F).

Edwards (1974a) was concerned only with the mostly cryptically colored dendrobatids then referred to Colostethus and not the more conspicuously colored species referred to Dendrobates and Phyllobates. The broader sample of the present study showed that there are (at least) two distinct “dorsolateral” stripes, which we have designated A (Character 52) and B (Character 53), the latter also having been confused previously with the oblique lateral stripe.

52. Dorsolateral Stripe A (Does Not Drop to Thigh; fig. 37): absent  =  0; present in juveniles only (i.e., lost ontogenetically)  =  1; anterior, narrow, faint  =  2; complete  =  3. Nonadditive.

Figure 37

Character 52, dorsolateral stripe A.

A, B: State 1, present in juveniles (A), absent in adults (B) (terribilis, A: captive-raised specimen; B: AMNH live exhibit). C: State 2, anterior, narrow, faint (atopoglossus, holotype UVC 12068). D: State 3, complete (aurotaenia, AMNH live exhibit).


This dorsolateral stripe runs posteriad from the eyelid toward the tip of the urostyle. It does not cross the flank toward the groin (oblique lateral stripe), nor does it drop to the top of the thigh (dorsolateral stripe B). Myers et al. (1978) reported that in bicolor and terribilis the dorsolateral stripe is present in juveniles and lost ontogenetically (state 1). This “loss” is peculiar, however, as it is due to the hypertrophy of the bright dorsolateral stripes that expand ontogenetically to cover the entire dorsum, thus creating a uniformly colored, stripeless color pattern.5 In state 2, the dorsolateral stripe is short, narrow, and inconspicuous (often more conspicuous in juveniles than adults), running from the posterior edge of the eye to a point just past the insertion of the arm. When present, the dorsolateral stripe of most species is complete, reaching or surpassing the level of the sacrum, and persists in adults (state 3).

The ontogenetic loss of the dorsolateral stripe is suggestive of additivity (i.e., absent ↔ present in juveniles only ↔ present throughout ontogeny); however, given the peculiarity of this particular “loss” through expansion, the additivity absent ↔ present throughout ontogeny ↔ present in juveniles only may be more appropriate. Regardless, it is unclear where state 2 would fit into this series, as there is no evidence that the dorsolateral stripe extends posteriorly through development, nor that state 2 is the result of reduction from a complete dorsolateral stripe. We therefore did not specify the additivity of this transformation series.

53. Dorsolateral Stripe B (Drops to Top of Thigh, Not Groin; fig. 38): absent  =  0; present  =  1.

Figure 38

Character 53, dorsolateral stripe B.

State 1, present (femoralis, AMNH 140646).


This dorsolateral stripe extends posteriad from the eyelid along the dorsolateral edge of the body and turns abruptly ventrad at a position immediately anterior to the thigh. This stripe was considered to be dorsolateral by Silverstone (1975a) and Caldwell and Myers (1990) for quinquevittatus, but oblique lateral (“lateral”) by Silverstone (1976) for femoralis. The confusion is understandable, as its path is intermediate between these two characters. Unlike the oblique lateral stripe, it does not run diagonally along the flanks but remains dorsal until almost the level of the thigh, but unlike dorsolateral stripe A, it drops toward the thigh posteriorly.

54. Ventrolateral Stripe (fig. 39): absent  =  0; wavy series of elongate spots  =  1; straight  =  2. Nonadditive.

Figure 39

Character 54, ventrolateral stripe.

A: State 1, wavy series of elongate, interconnected spots (espinosai, AMNH 104875). In this specimen the spotting forms a fairly contiguous wavy stripe, but it is common for the elongate spots to be separated, forming a broken stripe. B: State 2, straight (talamancae, AMNH 69829, photo by R. Zweifel). Note also that the pale dorsolateral stripe does not drop toward the thigh posteriorly (Character 52, state 3).


The ventrolateral stripe (VLS) runs along the ventral edge of the flank between the belly and the usually dark coloration of the flank. It may be present as a wavy series of elongate, often interconnected spots (state 1) or as a straight line (state 2). The ventrolateral stripe can be difficult to detect in preserved specimens, even those in which the ventrolateral stripe was prominent in life, because of the degradation of iridophores, especially in taxa with fairly pale ventral surfaces. In some of these cases the ventrolateral stripe can be detected as a lack of melanophores. However, the iridophores break down fairly quickly in preservative, often revealing a deeper layer of underlying melanophores invisible in living or freshly preserved specimens.

Coloma (1995: 47–48) reported that some specimens of pulchellus have “an interrupted white ventrolateral line”, but we did not observe this in the specimens examined. Caldwell and Lima (2003) reported the ventrolateral stripe as absent and described the holotype of nidicola as having “irregular white blotches, not forming a stripe”. However, a wavy VLS is evident in the photograph shown in their figure 3B (gravid female). Among the trivittatus specimens examined, the ventrolateral stripe is present in all specimens from Suriname but absent in all but one of the specimens from Peru (AMNH 43204, in which it is a series of small elongate spots on the left and a single, large elongate spot on the right).

55. Oblique Lateral Stripe: absent  =  0; present  =  1.

The pale oblique lateral stripe extends from the groin diagonally across the flanks toward the eye.

56. Oblique Lateral Stripe Length (fig. 40): partial  =  0; complete  =  1.

Figure 40

Character 56, oblique lateral stripe length.

A: State 0, partial (panamensis, AMNH 69836, photo by R. Zweifel). B: State 1, complete (fraterdanieli, MHNUC 364).


Edwards (1974b) distinguished oblique lateral stripes (OLS) that extend from the groin part-way to the eye (partial, state 0) or all the way to the eye (complete, state 1). There is some individual variation in the anterior extension of the partial OLS, but it usually terminates prior to and does not extend past the level of the insertion of the arm. There is no evidence that the stripe develops from one end to the other, which is why we did not combine length with presence/absence as an additive multistate character (i.e., absent ↔ partial ↔ complete).

Edwards (1974b: 10) described the OLS of sauli as incomplete, which is supported by both his painting (p. 6) and the color plate of the same specimen in Coloma (1995: plate 1A). However, Coloma (1995) explicitly compared sauli only to those species having a complete oblique lateral stripe, and we have also observed it to be complete. We therefore scored this character as polymorphic.

57. Oblique Lateral Stripe Structure (fig. 41): solid  =  0; series of spots  =  1; diffuse  =  2. Nonadditive.

Figure 41

Character 57, oblique lateral stripe structure.

A: State 0, solid (pulchripectus, AMNH 137290). B: State 1, series of spots (mertensi, ICN 43698). C: State 2, diffuse (trilineatus, AMNH 171974).


The oblique lateral stripe (OLS) of most species consists of a solid line of pale pigmentation (e.g., nubicola; state 0). Lynch and Ruiz-Carranza (1985) identified state 1 (series of well-defined spots) in agilis, and Myers et al. (1991: 2, 3, figs. 1, 3) illustrated it photographically for nocturnus. Grant and Rodríguez (2001) discussed variation in this character and described and illustrated photographically state 2. As shown in Grant and Rodríguez (2001: 9, fig. 6), state 2 may also include spots, but they are smaller, less distinct, and arranged irregularly (not in a line).

58. Gular–Chest Markings (fig. 42): absent  =  0; present  =  1.

Figure 42

Character 58, markings on gular–chest region, state 1 (present).

A: Diffuse, white-spotted blotches (awa, AMNH 111542). B: Discrete, small dark spots (vertebralis, USNM 28232). Despite their differing shapes and patterns, we treated the occurrence of these markings as a single character-state.


A number of species from the Andes of southern Colombia, Ecuador, and Peru possess highly variable dark spots or blotches on the posterolateral portion of the gular–chest region. Myers et al. (1991) compared these markings with the collars of several Venezuelan species and considered the possibility that they may be homologous. We code them as different transformations series here, the difference being that the gular–chest markings are always separated medially and do not form a continuous transverse band.

Coloma (1995: 10) reported several variants in the shape and pattern of the gular–chest markings. Much of this variation is intraspecific, and Coloma reported ontogenetic changes. Consequently, until this variation is better understood, we treated all of these variants as homologous and subsumed their occurrence within a single character-state. Although Coloma (1995: 10) discussed them in the same context, the markings on the mental region and the pair of spots on the posterior chest do not occur in the same region, and we did not treat them as part of this transformation series.

Coloma (1995) reported the presence of diffuse bandlike markings in bocagei, but it was absent from all the specimens we examined.

59. Dark Dermal Collar (fig. 43): absent  =  0; present  =  1.

Figure 43

Character 59, dermal collar, state 1 (present) in trinitatis.

A: Male (UMMZ 167474). B: Female (UMMZ 167471). In this species, the dermal collar of males is diffuse and broad, but is clearly distinguished from the fainter gray stippling of the adjacent surfaces.


The dermal collar (“chest markings” of La Marca, 1995) is a continuous transverse band of dark pigmentation that extends across the posterior throat, anterior to the arms. Although La Marca (1996 “1994”) reported sexual variation in its occurrence, we observed it to be present in adults of both sexes of all species that possess the dermal collar (although we did observe polymorphism in males of neblina), so we did not code males and females as separate semaphoronts.

Rivero (1978 “1976”: 330; translated from the Spanish) noted that “[i]n almost all specimens [of leopardalis] a faint dark collar may be detected, never as clear and well defined as in C. collaris, and generally confined to the sides of the throat.” Similarly, Myers et al. (1991) noted the occurrence of faint collarlike pigmentation on the throat of many specimens of nocturnus. Closer examination and dissection revealed that these dark collars are not caused by melanophores in the skin, as they are in other collared species (e.g., collaris), but instead by melanophores in the epimysium of the m. interhyoideus and connective tissue in the hyomandibular sinus (i.e., anterior to the pectoral apparatus) that are visible through the semitranslucent skin (fig. 44). The density of melanophores varies among individuals, with males having greater density (and therefore a more prominent collar) than females. Dense subdermal pigmentation may also occur in species with dark dermal pigmentation (e.g., galactonotus; see fig. 44), and individuals with dermal collars may (or may not) also present extensive subdermal pigmentation. In leopardalis (e.g., UMMZ 17170) the subdermal pigmentation is not as concentrated but still accounts for the faint collar reported by Rivero. Some degree of melanosis of the collar region is widespread among dendrobatids. However, as observed in pigmentation of the flesh generally, variation is continuous from a few melanophores scattered across the throat to a solid subdermal collar. As discussed above, we suspect there are valid transformation series here, but we were unable to delimit them objectively for the present study.

Figure 44

Extensive subdermal melanosis of the collar region.

A, B: nocturnus (AMNH 130008). C, D: galactonotus (AMNH 128233). Note also the irregular (clumped) stippling or faint, diffuse spotting in A (character 61, state 6; see below).


60. Dark Lower Labial Stripe (fig. 45): absent  =  0; present  =  1.

Figure 45

Character 60, dark lower lip line.

State 1, present (fraterdanieli, MHNUC 364).


In fraterdanieli Grant and Castro-Herrera (1998) indicated the occurrence of a distinctive dark (black or brown) line along the lower lip and contrasting with the pale adjacent coloration.

61–64. Male and Female Throat and Abdominal Coloration

The color and color pattern of the throat and abdominal regions of adult males and females provide some of the most useful characters for discriminating among species of dendrobatids. Sexual dimorphism is common, especially in throat color and color pattern, but most states occur in both sexes.

In a series of field experiments, Narins et al. (2003) showed that both vocalization and pulsation of the dark vocal sac of femoralis are necessary to elicit aggressive responses, and they have experimented with this system to reveal the extent of temporal and spatial integration of these stimuli required for an aggressive response (Narins et al., 2005). Given the extent of species- and sex-specific variation in throat coloration in many dendrobatids, it is likely that integration of visual and vocal cues will be found to be widespread. To date Narins et al.'s investigations have only examined aggressive male–male interactions in this single species, but application of their procedures more generally promises to reveal fascinating insights into the role of throat coloration in aggresive and other interactions between males, females, and juveniles across multiple species.

A few points apply to the coding of all the following characters. First, as discussed above, we emphasized color pattern over color. Second, spotting, marbling, and reticulation form a continuous gradient that, though unambiguous in the extremes, we were unable to delimit objectively. We therefore treated these as a single character-state, although we undoubtedly overlooked additional transformations by doing so. Third, it may be difficult to discriminate between pale spotting/reticulation/marbling on a dark background versus dark spotting/reticulation/marbling on a pale background, as the distinction has to do with adjacent coloration and the relative concentration of pale and dark pigmentation. Many species are unambiguously one or the other, but others were either ambiguous or exhibited both states and were therefore coded as polymorphic. Fourth, we also treated irregular stippling (i.e., clumped stippling) and the occurrence of diffuse dark spotting as a single state because we were unable to discriminate two states objectively. Finally, the collar and gular–chest markings (characters 58, 59) are independent of the region referred to as the throat. For our purposes, throat refers to the region of the central gular region, that is, the area of the vocal sac in males.

61. Male Throat Color (figs. 44A, 46): pale, free or almost free of melanophores  =  0; dark due to absence of iridophores  =  1; evenly stippled  =  2; pale with discrete dark spotting/reticulation/marbling  =  3; solid dark  =  4; dark with discrete pale spotting/reticulation/marbling  =  5; irregular (clumped) stippling or faint, diffuse spotting  =  6. Nonadditive.

Figure 46

Character 61, male throat color.

A: State 0, pale, free or almost free of melanophores (“Neblina species”, AMNH 118689). B: State 1, dark due to absence of iridophores (abditaurantius, ICN 9853). This character-state is inconspicuous in preserved specimens but obvious in living or recently prepared specimens. C: State 2, evenly stippled gray (infraguttatus, AMNH 104846). Note that the gular–chest markings (character 58) of infraguttatus do not interfere with the even stippling of the throat. D: State 3, pale with dark spots (“nubicola-spC”, MHNUC 321). E: State 4, solid dark (inguinalis, LACM 42329). F: State 5, dark with discrete pale spotting/reticulation/marbling (tricolor, USNM 286082).


In state 0 (pale, free or almost free of melanophores) the throat appears immaculate, but closer inspection may reveal sparse, inconspicuous melanophores. State 1 (dark due to absence of iridophores) is conspicuous in life but may easily be overlooked in preserved specimens. See Grant and Castro-Herrera (1998) for this character-state in life. The spotting/reticulation/marbling of the vocal sac is often irregular. State 6 (dark with pale medial stripe) is restricted to only boulengeri and espinosai. In both species the medial “stripe” varies from one or more elongate spots to a solid stripe. Also, the adjacent dark surfaces sometimes include scattered pale spots. States 0–5 are shown in figure 46; state 6 is shown in figure 44A.

62. Female Throat Color (fig. 47): pale, free or almost free of melanophores  =  0; irregular (clumped) stippling or faint, diffuse spotting  =  1; solid dark  =  2; dark with discrete pale spotting/reticulation/marbling  =  3; pale with discrete dark spotting/reticulation/marbling  =  4; dark with pale medial longitudinal stripe  =  5. Nonadditive.

Figure 47

Character 62, female throat color.

A: State 0, pale, free or almost free of melanophores (undulatus, AMNH 159128). B: State 1, irregular (clumped) stippling or faint, diffuse spotting (nocturnus, AMNH 130018). C: State 2, solid dark (hahneli, AMNH 96190). D: State 3, dark with discrete pale spotting/reticulation/marbling (imbricolus, AMNH 102083). E: State 4, pale with discrete dark spotting/reticulation/marbling (fraterdanieli, AMNH 148021). F: State 5, dark with pale medial longitudinal stripe (boulengeri, USNM 145281).


63. Male Abdomen Color (fig. 48): pale, free or almost free of melanophores  =  0; pale with discrete dark spotting/reticulation/marbling  =  1; evenly stippled  =  2; dark with discrete pale spotting/reticulation/marbling  =  3; irregular (clumped) stippling or faint, diffuse spotting  =  4; solid dark  =  5. Nonadditive.

Figure 48

Character 63, male abdomen color.

A: State 0, pale, free or almost free of melanophores (“Neblina species”, AMNH 118689). B: State 1, pale with discrete dark spotting/reticulation/marbling (quinquevittatus, AMNH 124069). C: State 2, evenly stippled (talamancae, AMNH 113893). D: State 3, dark with discrete pale spotting/reticulation/marbling (infraguttatus, AMNH 104846). E: State 4, irregular (clumped) stippling or faint, diffuse spotting (nocturnus, AMNH 130012). F: State 5, solid dark (inguinalis, LACM 42329).


64. Female Abdomen Color (fig. 49): pale, free or almost free of melanophores  =  0; pale with discrete dark spotting/reticulation/marbling  =  1; solid dark  =  2; dark with discrete pale spotting/reticulation/marbling  =  3; irregular (clumped) stippling or faint, diffuse spotting  =  4; evenly stippled  =  5. Nonadditive.

Figure 49

Character 64, female abdomen color.

A: State 0, pale, free or almost free of melanophores (undulatus, AMNH 159128). B: State 1, pale with dark spotting/reticulation/marbling (fraterdanieli, AMNH 39360). C: State 2, solid dark (silverstonei, AMNH 91845). D: State 3, dark with discrete pale spotting/reticulation/marbling (infraguttatus, AMNH 104849). E: State 4, irregular (clumped) stippling or faint, diffuse spotting (nocturnus, AMNH 130018). F: State 5, evenly stippled (riveroi, AMNH 134141).


Coloma (1995: 54) described vertebralis as having “dark stippling on abdomen in females, darker in males”; however, none of the females in the series AMNH 17458, 17604–08, 140977–141011 possesses any stippling on the abdomen (but all males do). Although we coded riveroi as having the abdominal region evenly stippled, in life it is posteriorly orange (Donoso-Barros, 1965 “1964”).

65. Iris Coloration (fig. 50): lacking metallic pigmentation and pupil ring  =  0; possessing metallic pigmentation and pupil ring  =  1.

Figure 50

Character 65, iris coloration.

A: State 0, lacking metallic pigmentation and pupil ring (bicolor, AMNH live exhibit). B: State 1, with metallic pigmentation and pupil ring (subpunctatus, uncataloged MUJ specimen from Colombia: Bogotá, D.C., campus of Universidad Nacional de Colombia).


Silverstone (1975a: 8) noted that in life the iris of the species he included in Dendrobates is “black (or rarely dark brown) and is never reticulated”. Similarly Silverstone (1976: 3) stated that in the species he included in Phyllobates “the iris is black or brown (rarely bronze) and never reticulated”. We diagnose this character somewhat more precisely, but we believe our intentions are the same.

The iris coloration of most dendrobatids includes metallic pigmentation (bronze, copper, gold, silver) producing a metallic iris with a black reticulated pattern or a black iris with metallic flecks. Additionally, a distinct metallic ring around the pupil invariably occurs in irises with metallic pigmentation. A number of the aposematic dendrobatids lack all metallic pigmentation in the iris, giving rise to the solid black or brown iris mentioned by Silverstone.

This character can only be coded from living specimens. We dissected the eyes of preserved specimens of several species but failed to detect differences between pigmented and unpigmented irises. We therefore relied on explicit field notes, personal observations, and high-quality photographs. In addition to personal observations and unpublished field notes and photographs, this character was scored from the following published accounts: arboreus (Myers et al., 1984); aurotaenia (Silverstone, 1976; Lötters et al., 1997a); azureiventris (Kneller and Henle, 1985; Lötters et al., 2000); awa (Coloma, 1995); baeobatrachus (Lescure and Marty, 2000); bicolor (Myers et al., 1978; Lötters et al., 1997a); bocagei (Coloma, 1995); boulengeri (Silverstone, 1976); caeruleodactylus (Lima and Caldwell, 2001); claudiae (Jungfer et al., 2000); delatorreae (Coloma, 1995); degranvillei (Boistel and de Massary, 1999; Lescure and Marty, 2000); elachyhistus (Coloma, 1995; Duellman, 2004); flotator (Savage, 2002), Eupsophus roseus ([for E. calcaratus] Nuñez et al., 1999); granuliferus (Myers et al., 1995; Savage, 2002); herminae (La Marca, 1996 “1994”); Hylodes phyllodes (Heyer et al., 1990); ideomelus (Duellman, 2004); imitator (Symula et al., 2001); infraguttatus (Coloma, 1995); insperatus (Coloma, 1995); insulatus (Duellman, 2004); kingsburyi (Coloma, 1995); lehmanni Myers and Daly (Myers and Daly, 1976b); lugubris (Silverstone, 1976; Savage, 2002); macero (Rodríguez and Myers, 1993; Myers et al., 1998); machalilla (Coloma, 1995); molinarii (La Marca, 1985); nexipus (Frost, 1986; Hoff et al., 1999); nidicola (Caldwell and Lima, 2003); nocturnus (Myers et al., 1991); nubicola (Savage, 2002), parvulus (Silverstone, 1976); pictus (Köhler, 2000); petersi (Rodríguez and Myers, 1993; Myers et al., 1998); pulchellus (Coloma, 1995); pulchripectus (Silverstone, 1975a); pulcherrimus (Duellman, 2004); pumilio (Myers et al., 1995; Savage, 2002); quinquevittatus (Caldwell and Myers, 1990); reticulatus (Myers and Daly, 1983); rubriventris (cover of Herpetofauna 19(110); see also Lötters et al., 1997b); sauli (Coloma, 1995); silverstonei (Myers and Daly, 1979); speciosus (Jungfer, 1985); sylvaticus Barbour and Noble (Duellman, 2004); sylvaticus Funkhouser (Myers and Daly, 1976b [as histrionicus]; Lötters et al., 1999); talamancae (Coloma, 1995); terribilis (Myers et al., 1978); toachi (Coloma, 1995); trinitatis (Wells, 1980c; La Marca, 1996 “1994”); trivittatus (Myers and Daly, 1979); undulatus (Myers and Donnelly, 2001); vanzolinii (Myers, 1982); ventrimaculatus (Lötters, 1988 [as quinquevittatus]; Lescure and Bechter, 1982 [as quinquevittatus]); vertebralis (Coloma, 1995); vicentei (Jungfer et al., 1996b); vittatus (Silverstone, 1976; Savage, 2002); zaparo (Silverstone, 1976).

66. Large Intestine Color (fig. 51): unpigmented  =  0; pigmented anteriorly  =  1; pigmented extensively  =  2. Additive.

Figure 51

Character 66, large intestine color.

A: State 0, unpigmented (“Neblina species”, AMNH 118679). Note that the distended tissue is translucent. B: State 1, anteriorly pigmented (pratti, SIUC 07654). C: State 2, extensively pigmented (beebei, ROM 39631).


The large intestine of most species is unpigmented (state 0), being either white or (when distended) translucent. In some species, heavy melanosis forms a solid black coloration extending posteriad from the front of the large intestine. In state 1 the melanosis is confined to the anterior quarter of the large intestine; in state 2 it extends beyond the midlevel of the large intestine. The ontogeny of this character invariably progresses from state 0 through state 1 to state 2, which we interpret as evidence for the additivity of this transformation series.

67. Adult Testis (Mesorchium) Color (fig. 52): unpigmented  =  0; pigmented medially only  =  1; entirely pigmented  =  2. Additive.

Figure 52

Character 67, adult testis (mesorchium) color.

State 2, entirely pigmented testes (claudiae, AMNH 124257) in ventral view.


Testis color is scored from adults only. In all dendrobatids we have examined, testis pigmentation increases ontogenetically, with the mesorchia of juveniles being invariably entirely unpigmented white (state 0) and melanosis beginning medially (state 1) and eventually covering the testis entirely (state 2), forming either a dark reticulum or a solid dark color. Ontogenetic series show this character to develop from state 0 to state 1 to state 2, which we interpret as evidence of additivity.

Polymorphism among adults is rare. Grant (2004) found that of 40 specimens of panamensis scored, the left testes of two were unpigmented whereas the right testes were pigmented brown. Grant (2004) also documented unusual variation in the testis pigmentation of inguinalis. Testes of all adults had some degree of dark pigmentation, but it varied from being confined to medial and anterior surfaces to engulfing the entire testis; this variation was not correlated with adult size, extent of dark ventral pigmentation, or maturity. It is likely that such variation is hormonally controlled and related to sexual activity, but no evidence exists to support this conjecture. Among the included outgroup taxa, Lötters (1996) reported that, although most species of Atelopus possess permanently unpigmented testes, in some species the testes become pigmented with the onset of the breeding season.

68. Color of Mature Oocytes (fig. 53): unpigmented (uniformly white or creamy yellow)  =  0; pigmented (animal pole brown or black)  =  1.

Figure 53

Character 68, mature ova color.

A: State 0, white or yellowish (Atelopus spurrelli, AMNH 50983). B: State 1, pigmented (brown) (“Neblina species”, AMNH 118679).


The entire oocyte may be white or creamy yellow (state 0) or differential localization of pigment granules in the animal cortex may give rise to a dark (brown or black) animal hemisphere and white or creamy yellow vegetal hemisphere (state 1).

Duellman and Trueb (1986) explained egg pigmentation as an adaptation to exposure to sunlight, and they listed a number of anuran groups in support of that hypothesis. However, it is unclear if that adaptive explanation holds among dendrobatids, given that many species with pigmented eggs lay clutches that are not exposed to sunlight. For example, Myers and Daly (1979) found “a clutch of 30 eggs … on a curled dry leaf that was completely concealed by another leaf of the cut-over forest floor”, yet that species has pigmented eggs. It has also been conjectured (e.g., Duellman and Trueb, 1986) that this melanosis either raises egg temperature by better absorbing ambient heat or provides protection from exposure to ultraviolet radiation. Missing from previous discussions is an evaluation of the polarity of the transformations. Until this is evaluated through phylogenetic analysis, it is impossible to know if a particular instance of pigmentation (or lack of pigmentation) is apomorphic (and therefore a candidate for an explanation of adaptation) or symplesiomorphic.

Duellman and Trueb (1986: 535) reported that Rhinoderma darwinii possesses unpigmented ova, but specimens examined had pigmented ova.

69. M. Semitendinosus Insertion (figs. 54, 55): “bufonid type” (ventrad)  =  0; “ranid type” (dorsad)  =  1.

Figure 54

Character 69, m. semitendinosus insertion.

Photograph (left) and outline drawing (right) of ventral view of distal thigh of Thoropa miliaris (AMNH 17044), showing state 0, ventrad “bufonid” path of insertion. Arrow indicates the m. semitendinosus tendon of insertion.


Figure 55

Character 70, m. semitendinosus binding tendon.

State 1, present (aurotaenia, AMNH 161109), photograph (left) and outline drawing (right) showing view of the concealed surface of the knee. The mm. gracilis complex is deflected ventrally to reveal the dorsad “ranid” path of the m. semitendinosus and the secondary binding tendon that straps it to the outer edge of the mm. gracilis complex.


Noble's (1922) seminal work brought thigh musculature to the forefront of studies of anuran relationships, and since then the path of insertion of the distal tendon of the m. semitendinosus has played an important role in discussions of dendrobatid relationships (reviewed by Grant et al., 1997). Noble identified two predominant morphologies: (1) the putatively primitive “bufonid type” in which the tendon of the m. semitendinosus inserts ventrad to the tendon of insertion of the mm. gracilis complex, and (2) the putatively derived “ranid type” in which it inserts dorsad to the mm. gracilis.

Noble also reported a number of “intermediate” morphologies, including that of dendrobatids. However, the m. semitendinosus of dendrobatids clearly inserts dorsal to the mm. gracilis, the apparent intermediacy owing to a secondary binding tendon (see Character 70, below). Similarly, Noble interpreted intermediate morphologies as providing evidence for the “inward migration” of the tendon of insertion of the m. semitendinosus from the presumptively primitive “bufonid” position to the derived “ranid” position. However, he relied on phylogenetic evidence to establish character additivity, not ontogenetic (or other) evidence. The groups Noble considered most primitive had “bufonid type” insertion, those he thought were most derived had “ranid type”, and variants were treated as intermediate between the two. Such reasoning is obviously fallacious, as it conflates the premises of analysis with the conclusions. We are unaware of developmental evidence of inward migration of the m. semitendinosus tendon of insertion from the “bufonid” to the “ranid” position.

70. M. Semitendinosus Binding Tendon (fig. 55): absent  =  0; present  =  1.

As first described and illustrated by Noble (1922: 41 and plate XV, fig. 6)6, the dendrobatid thigh has a well-defined binding tendon7 that straps the m. semitendinosus distal tendon to the dorsal edge of the inner surface of the mm. gracilis complex (state 1; fig. 55). In some large species, such as palmatus and nocturnus, this binding tendon is robust and conspicuous, giving the impression that the distal tendon of the m. semitendinosus actually pierces or penetrates the distal mm. gracilis tendon (e.g., Dunlap, 1960: 66). However, even in these species the m. semitendinosus tendon does not pass through the tendinous tissue, but rather between the binding tendon and the gracilis muscle (therefore differing from myobatrachids; Noble, 1922; Parker, 1940). This tendinous tissue also forms a secondary tendon that inserts on the inner (posterior) surface of the proximal head of the tibiofibula. Another secondary tendon is often present, arising near the ventral edge of the inner surface of the m. gracilis and leading to the thick sheet of connective tissue that wraps around the knee. Distal to the binding tendon, the m. semitendinosus tendon expands to insert along the long axis of the ventral surface of the tibiofibula.

71. M. Levator Mandibulae Externus Division: undivided (“s”)  =  0; divided (“s + e”)  =  1.

In her dissertation, Starrett (19688) found that the jaw adduction musculature of dendrobatids includes a single muscle originating from the zygomatic ramus of the squamosal that lies deep (internal, medial) to the mandibular ramus of the trigeminal nerve (V3), which she interpreted as the presence of the m. adductor posterior mandibulae subexternus and absence of the m. adductor mandibulae externus superficialis, or condition “s” in her system (state 0). Silverstone (1975a) found this in all 41 species he examined, and other workers (Myers et al., 1978, 1991; Myers and Daly, 1979; Myers and Ford, 1986; Grant et al., 1997; Grant, 1998) have found this in almost all dendrobatids (see below). Our observations conform with the accounts of previous workers, with the exception that fibers generally originate on the anterior edge of the annulus tympanicus as well. In addition to the “s” morphology, among other frogs Starrett (1968) reported the presence of both muscles, or “s + e” in her system (state 1), as well as the absence of the m. adductor posterior mandibulae subexternus and presence of the m. adductor mandibulae externus superficialis, “e” (not observed in the present study).

Haas (2001) reevaluated the homology of these muscles based on detailed developmental studies across the diversity of amphibians. He concluded that the “subexternus” and “externus” muscles are actually different portions of the same muscle (slips, in our terminology; see Materials and Methods, above)—the m. levator mandibulae externus profundus and m. levator mandibulae externus superficialis, respectively. As such, Starrett's (1968) three conditions (“s”, “e”, and “s + e”) involve two distinct transformation series: (1) the m. levator externus is constant in both “s” and “e” and the transformation series is formed by changes in the position of V3, deep (internal, medial) in the “s” and superficial (external, lateral) in the “e”; (2) “s + e” is formed by the division of the m. levator mandibulae externus into profundus and superficialis slips, with V3 lying between them. Because we did not observe the “e” condition in any of the species coded for morphology in the present study, we scored only the latter transformation series.

The sole published exception to the undivided m. levator mandibulae externus (“s”; state 0) morphology in dendrobatids is nocturnus, in which Myers et al. (1991) found the m. levator externus of some specimens (or one side) to form two distinct slips (“s + e”; state 1). Myers et al. (1991) interpreted this observation as possibly indicative of both the ranoid origin of dendrobatids and the primitiveness of nocturnus within the dendrobatid clade. The only other exception we observed was a specimen of vanzolinii (AMNH 108332) in which V3 pierces the m. levator mandibulae externus. However, we did not code the superficial and medial fibers as different slips (i.e., “s + e”) because the fibers are tightly bound both dorsal and ventral to the nerve and are not segregated by connective tissue septa (as they are in nocturnus, for example) and therefore do not form distinct slips (for similar individual variation, see Lynch, 1986). That said, only a single specimen was available for dissection, and the symmetry of this morphology suggests that further study may reveal a phylogenetically relevant change in the path of V3.

Additional intraspecific variation was observed in Hylodes phyllodes. Of the 11 frogs in the series AMNH 103885–95, in 2 (AMNH 103888, 103890) V3 runs medial to a distinct m. levator mandibulae externus superficialis (“s + e”; state 1) on both sides of the head, whereas the remaining 9 specimens all lack that slip (“s”; state 0). As in nocturnus, and in contrast to vanzolinii, the lateral fibers form a distinct slip. Indeed, in H. phyllodes all of the lateral fibers appear to originate on the rim of the annulus tympanicus, whereas the medial slip originates from the squamosal. Although this latter consideration is suggestive of nonhomology of the m. levator mandibulae externus superficialis in these taxa, we coded them as the same state and subjected that hypothesis to the test of character congruence.

72–75. M. Depressor Mandibulae

Starrett (1968) identified three distinct slips of the m. depressor mandibulae of dendrobatids: a massive, superficial slip originating from the dorsal fascia overlying the scapula and m. levator posterior longus, a deeper slip originating from the otic ramus of the squamosal, and an additional slip of fibers originating on the tympanic annulus. The combined morphology was codified as DFSQdAT. Lynch (1993: 37) refined the delimitation of this condition as “one in which some number of superficial fibers of the squamosal portion of the m. depressor mandibulae extend medial to the crest of the otic ramus of the squamosal and overlie the fibers of the m. levator posterior longus.” Silverstone (1975a) confirmed that all 41 species of dendrobatids he examined have this morphology, and this was further confirmed in additional species by Myers and coworkers (e.g., Myers et al., 1978, 1991; Myers and Daly, 1979; Myers and Ford, 1986).

Lynch (1993) rejected the anatomical findings of Starrett (1968), at least as they applied to Eleutherodactylus. Of greatest relevance to dendrobatids is his finding (pp. 37–38) that the superficial “DF” fibers actually are “bound tightly to deeper fibers … that originate on the lateral face of the otic ramus of the squamosal”. That is, the fibers from the two origins are not segregated by connective tissue septa and therefore do not constitute distinct slips. Our dissections confirm Lynch's observations in dendrobatids and the sampled outgroup taxa as well, leading us to follow him in discarding Starrett's terminology.

Nevertheless, regardless of whether the depressor muscle is divided into distinct slips or not, the variation in fiber origins constitutes a valid transformation series. In all specimens examined, some fibers originate from the otic ramus of the squamosal. In all dendrobatids examined, the vast majority of fibers originates form the dorsal fasciae. Lynch (1993) referred to the portion of the m. depressor mandibulae that originates medial to the crest of the squamosal on the m. temporalis as a “dorsal flap”, and we follow his terminology here (Character 73). We scored as Character 74 the origin of fibers posterior to the crest of the squamosal. The occurrence of fibers originating on the posterior edge of the annulus tympanicus is coded in Character 75.

Manzano et al. (2003) presented an extensive survey of the m. depressor mandibulae in anurans. Among dendrobatids, they examined auratus, pictus, and subpunctatus. Our observations and coding conform generally with theirs, with the following exceptions: (1) Manzano et al. did not recognize the dorsal flap as a separate character. (2) Manzano et al. reported that the superficial “slip” of auratus is divided into anterior and posterior “slips”, whereas that of pictus and subpunctatus consists of a single, wide, fan-shaped muscle. We examined 20 uncataloged AMNH skinned carcasses of auratus from Isla Tobago, Panama, constituting 40 depressor muscles. In that series, variation is continuous between an uninterrupted fan-shaped muscle, the occurrence of a slight division across the thoracic sinus, and well-defined separate branches, with conspicuous bilateral asymmetry in some specimens. Given the continuous variation, we were unable to delimit states objectively. Moreover, the degree of individual variation suggests that the differences are likely nongenetic, although we cannot offer any direct evidence to that effect. (3) Manzano et al. reported fibers originating from the annulus tympanicus in Rhinoderma darwinii. However, although we observed fibers to extend toward the annulus, in the specimens we examined (AMNH 37849, 58082), the fibers invariably run along the cartilage and ultimately attach to the squamosal.

72. M. Depressor Mandibulae Dorsal Flap: absent  =  0; present  =  1.

73. M. Depressor Mandibulae Origin Posterior to Squamosal: absent  =  0; present  =  1.

74. M. Depressor Mandibulae Origin on Annulus Tympanicus: no fibers originating from annulus tympanicus  =  0; some fibers originating from annulus tympanicus  =  1.

Among dendrobatids, the fibers that attach to the annulus tympanicus are generally deep and easily overlooked, but careful dissection showed them to be present in all dendrobatids examined.

75. Tympanum and m. Depressor mandibulae relation: tympanum superficial to m. depressor mandibulae  =  0; tympanum concealed superficially by m. depressor mandibulae  =  1.

Myers and Daly (1979: 8) pointed out that in dendrobatids “the large superficial slip of the depressor mandibulae muscle tends to slightly overlap the tympanic ring and, in any case, holds the skin away from the rear part of the tympanum, thus accounting for the fact that the tympanum is only partially indicated externally” (see also Myers and Ford, 1986; Myers et al., 1991). Daly et al. (1996) further discussed this character and compared conditions found in Mantella.

76. Vocal Sac Structure: absent  =  0; median, subgular  =  1; paired lateral  =  2. Nonadditive.

This character was coded following Liu (1935). Although we coded the vocal sac for the sampled species Megaelosia goeldii, in which males lack vocal sacs and slits and presumably do not call (Giaretta et al., 1993), other species of Megaelosia possess paired lateral vocal sacs (e.g., M. lutzae; Izecksohn and Gouvêa, 1987 “1985”). Lynch (1971) reported the state for Eupsophus roseus (coded for E. calcaratus).

77–78. M. Intermandibularis Supplementary Element (Fig. 56)

Tyler (1971) described variation in superficial throat musculature of hylids and other anurans, reporting the differentiation of the m. intermandibularis to form a supplementary element in two species of Colostethus [as Calostethus], two species of Dendrobates, and two species of Phyllobates. He did not list the species of dendrobatids he examined or describe the dendrobatid condition in any detail. La Marca (1995: 45) noted the occurrence of “supplementary elements of the anterolateral type attached to the ventral surface of the m. submentalis”. Of relevance to the current study, Burton (1998b) also reviewed the occurrence and variation in supplementary elements of numerous Neotropical hyloid groups.

The treatment of the supplementary element in phylogenetic analysis is somewhat problematic, as the homology of the elements in different taxa is debatable. Following Tyler's (1971) terminology, dendrobatids possess an anterolateral element: Fibers originate on the lingual surface of the anterior portion of the mandible and run anteriomediad to insert on the ventral surface of the m. submentalis, with the more posterior fibers underlying (superficial to) the deeper fibers of the m. intermandibularis. Tyler also identified apical and posterolateral elements in other groups of anurans, and these conditions have been largely supported by subsequent workers. Tyler (1971) effectively treated each of the morphologies as nonhomologous (i.e., the differences in morphologies was treated as evidence that each of the supplementary elements was independently derived), but other workers have treated them as a homologous entity with subsequent variation (e.g., Burton, 1998b; Mendelson et al., 2000).

The shared origin of the supplementary element on the lingual surface of the mandible superficial to the deeper primary sheet of the m. intermandibularis and the fact that the different morphologies never co-occur are sufficient evidence to treat the supplementary elements of different anurans as a homologous structure. We therefore submitted the hypothesis of supplementary slip homology to the simultaneous test of character congruence by coding its occurrence as one character and the variation in the element as a second character.

77. M. Intermandibularis Supplementary Element Occurrence: absent  =  0; present  =  1.

78. M. Intermandibularis Supplementary Element Orientation: 0  =  anterolateral; 1  =  anteromedial.

Among the sampled species that possess the supplementary element, we observed two patterns. In the first (state 0), the fibers radiate anterolaterally from a sagittal raphe. In the second, the fibers extend anteromedially from the mandible. Burton (1998b) reported both of these morphologies for a variety of Neotropical “leptodactylids”. However, our coding deviates from Burton's (1998b) in that he treated species with anterolateral supplementary slips (e.g., Hylodes spp.) and without supplementary slips but having all fibers directed mediad or anterolaterad (e.g., Thoropa miliaris) as “variants of the same general pattern” (pp. 67–68). Burton based this decision on “the fact that two of these variants may occur within the same genus (e.g., Cycloramphus), or within the same species (C[audiverbera]. caudiverbera)” (p. 67). Co-occurrence of this nature does not constitute evidence of character-state identity (if it did, polymorphism would be conceptually impossible), and we scored Hylodes phyllodes and Thoropa miliaris differently.

Manzano and Lavilla (1995) reported an apical supplementary element in Rhinoderma darwinii; according to the terminology employed herein, it is anterolateral (state 0).

79–86. Median Lingual Process (Figs. 57, 58)

Grant et al. (1997) discovered the median lingual process (MLP; fig. 57) in dendrobatids and documented its occurrence and variation throughout Anura (for additional dendrobatid records, see Myers and Donnelly, 1997; Grant and Rodríguez, 2001). Of greatest interest was the finding that an apparently homologous modification of the tongue occurs in dendrobatids and several Old World ranoids (including the putative sister group postulated by Griffiths, 1959) but is entirely lacking among all hyloid taxa. The functional significance of the MLP remains unknown. Variation in the MLP has been illustrated extensively by Grant et al. (1997) and Myers and Donnelly (1997); here we provide illustrations for novel characters.

Figure 57

Anterior view of the open mouth of the dendrobatid praderioi (CPI 10203) showing the short, tapered median lingual process (MLP).


As a first effort to understand the distribution and diversity of the MLP, Grant et al. (1997) allocated the observed variation to four “types”. For phylogenetic analysis it was necessary to decompose those types into their component transformation series. Given the relevance of this anatomical structure to the placement of Dendrobatidae, we examined the histology of the tongues of eight species to gain a better understanding of its structure. Additional data were gathered through gross dissection. Although several of the characters we observed do not vary among the MLP-possessing taxa sampled in the present study, they vary independently in the broader context of the evolution of the MLP in anurans, and we therefore score all of these characters here.

To discover differences between the type C processes of Old World and New World taxa, we examined the histology of two species of the dendrobatids tepuyensis and baeobatrachus, and we compared them to Phrynobatrachus natalensis and Phrynobatrachus petropedetoides. These two species of Phrynobatrachus have retractile type C processes, which we hoped would maximize the morphological differences between them and the nonretractile type C processes of the dendrobatids. To gain insight into the mechanism of retraction and protrusion, one of the specimens of Phrynobatrachus petropedetoides had the MLP fully protruded, whereas the other one had it retracted below the surface of the tongue. We also examined the type A processes of Arthroleptis variabilis, Mantidactylus femoralis, Platymantis dorsalis, and Staurois natator (the latter two species were included only for comparative purposes to delimit transformation series more rigorously and were not coded for phylogenetic analysis). Due to lack of specimens, we did not examine any type B or D processes.

The MLP of all examined taxa (i.e., types A and C, retractile and nonretractile) is formed through the same modification of the basal portion of the m. genioglossus, supporting the hypothesis that they are homologous structures. As seen in sagittal and transverse section of baeobatrachus (fig. 58) the m. genioglossus basalis is extended dorsally to protrude above the lingual surface as the median lingual process. In all taxa, muscle fibers are replaced distally by loose, presumably collagenous connective tissue, with elastic fibers forming the walls of the process. Although we did not stain specifically for nervous tissue, no major nerves were detected within the MLP. Additional histological findings are discussed below under the relevant characters.

Figure 58

Longitudinal histological section of the tongue of baeobatrachus (AMNH 140672) showing that the median lingual process (MLP) is an extension of the m. genioglossus.

Note lack of muscle fibers toward the tip of the MLP.


79. MLP Occurrence: absent  =  0; present  =  1.

To count the origin of the MLP as a single event (and not as multiple origins of each of the nested characters), we scored the occurrence of the MLP as a separate character. That is, species that lack the MLP were scored as state 0 for this character and missing (“–“) for the remaining MLP characters.

80. MLP Shape: short, bumplike  =  0; elongate  =  1.

We considered the MLP to be short and bumplike if it its length (height) was no greater than its width at the base and elongate if its length was greater than its width at the base.

81. MLP Tip: blunt  =  0; tapering to point  =  1.

82. MLP Texture: smooth  =  0; rugose  =  1.

Grant et al. (1997) found most MLPs to be smooth relative to the lingual surface (state 0) but that in some species the MLP is rugose, textured like the adjacent surfaces of the tongue.

83. MLP Orientation when Protruded: upright  =  0; posteriorly reclined  =  1.

When protruded, Grant et al. (1997) reported upright MLPs pointing straight dorsad (state 0) and posteriorly reclined MLPs (state 1).

84. MLP Retractility (fig. 59): nonretractile  =  0; retractile  =  1.

Figure 59

Character 84, median lingual process (MLP) retractility, state 1 (retractile).

A: Transverse section of a protruded MLP in Phrynobatrachus natalensis (AMNH 129714). B: Transverse section of a retracted MLP in Phrynobatrachus petropedetoides (AMNH 129626).


Following the reasoning of Grant et al. (1997), retractility was inferred from the position of the MLP in preserved specimens. We were unable to detect any histological differences between retractile and nonretractile processes. However, the fact that even very small series of some species show the MLP in various stages of retraction whereas very large samples of others do not include a single retracted lingual process suggests that this is not merely an artifact of preservation.

Comparison of retracted and protruded processes provides some clues as to the mechanism involved in retractility. As seen in figure 59A of the protruded process of Phrynobatrachus natalensis in transverse view, the connective tissue that extends to the tip of the MLP is very loose, with large spaces between the fibers and fibroblasts. In contrast, in a specimen of Phrynobatrachus petropedetoides with the MLP completely retracted below the surface of the tongue (fig. 59B), the loose connective tissue is much denser with no spaces between the fibers and fibroblasts, reminiscent of a squeezed sponge. This is characteristic of hydrostatic organs such as the feet of mollusks and suggests that protrusion and retraction of the MLP is achieved by the displacement of some sort of fluid.

85. MLP-Associated Pit: absent  =  0; present  =  1.

Grant et al. (1997) noted the absence (state 0) and presence (state 1) of a pit immediately posterior to the MLP into which fits the posteriorly reclined MLP of some species. Although all of the species with posteriorly reclined MLPs sampled in the present study also have an associated pit, the observations of Grant et al. (1997) establish the transformational independence of the two characters.

86. MLP Epithelium (fig. 60): glandular  =  0; nonglandular  =  1.

Figure 60

Character 86, median lingual process (MLP) epithelium.

A: State 0, glandular (Phrynobatrachus petropedetoides, AMNH 129593). The surface of the MLP is pitted with invaginations of the epithelium that form alveolar glands. B: State 1, nonglandular (Phrynobatrachus natalensis, AMNH 129732). The alveolar glands are absent, and the surface of the MLP consists of unmodified stratified epithelium.


In state 0 the surface of the MLP is pitted with invaginations of the epithelium that form alveolar glands, as occur over the rest of the surface of the tongue. In state 1, these glandular invaginations do not occur, and the MLP is covered in unmodified, stratified epithelium.

87–98. Larvae

In addition to specimens examined, larval data were taken from Ruthven and Gaige (1915), Fernández (1926), Funkhouser (1956), Donoso-Barros (1965 “1964”), Savage (1968), Duellman and Lynch (1969), Hoogmoed (1969), Edwards (1971, 1974b), Lynch (1971), McDiarmid (1971), Silverstone (1975a; 1976), Lescure (1976), Duellman (1978, 2004), Myers and Daly (1979), Cei (1980), Lescure and Bechter (1982), Heyer (1983), La Marca (1985), Lavilla (1987), Formas (1989), Caldwell and Myers (1990), Donnelly et al. (1990), Heyer et al. (1990), Mijares-Urrutia (1991), Myers et al. (1991), van Wijngaarden and Bolaños (1992), Rodríguez and Myers (1993), Haddad and Martins (1994), Juncá et al. (1994), Coloma (1995), Ibáñez and Smith (1995), La Marca (1996 “1994”), Mijares-Urrutia and La Marca (1997), Kaplan (1997), Lötters et al. (1997b), Grant and Castro-Herrera (1998), Faivovich (1998), Lindquist and Hetherington (1998), Lötters et al. (2000), Caldwell et al. (2002a), Caldwell and Lima (2003), Nuin (2003), and Castillo-Trenn (2004).

87. Larval Caudal Coloration: vertically striped  =  0; scattered melanophores clumped to form irregular blotches  =  1; evenly pigmented  =  2. Additive.

Caldwell et al. (2002a) figured the larvae of caeruleodactylus and marchesianus, the tails of which possess conspicuous, dark, broad, vertical stripes (state 0). The larval tails of the majority of species possess variable amounts of irregular, scattered melanophores clumped to form diffuse blotches, ranging from inconspicuous reticulation to large blotches (state 1). There is extensive ontogenetic and individual variation in the amount and intensity of this diffuse spotting, as documented for kingsburyi by Castillo-Trenn (2004), which prevented dividing the variation observed within this character-state into additional states. In some species, the larval tails are evenly pigmented brown, gray, or black (state 2).

88. Larval Oral Disc (fig. 61): “normal”  =  0; umbelliform  =  1; absent  =  2; suctorial  =  3. Nonadditive.

Figure 61

Character 88, larval oral disc.

A, B: Ventral (A) and lateral (B) views of State 0, “normal” (“Neblina species”, AMNH 118673). C, D: Ventral (C) and lateral (D) views of State 1, umbelliform disc (nubicola, AMNH 94849). Note also the submarginal papillae scattered over the surface of the oral disc (character 91).


As described by Haas (2003: 54), “[t]he oral disk is formed by the upper and lower lips, i.e., flat, more or less expansive flaps of skin set off from the mouth and jaws and commonly bearing labial ridges with keratodonts.” What is here referred to as the “normal” larval oral disc (state 0) consists of a thick, fleshy upper lip that is fully attached medially and lacks marginal papillae and a lower lip that is entirely free but relatively narrow and bears marginal papillae. The umbelliform (funnel-shaped) oral disc (state 1; fig. 61) is greatly enlarged relative to state 0. The upper lip is free and the marginal papillae extend around the entire circumference of the disc. Among dendrobatids, state 1 is known only in flotator, nubicola, and two unnamed species not included in this study. Dendrobatids that lack the oral disc (state 2) are endotrophic. The suctorial oral disc (state 3) is confined to the outgroup.

89. Lateral Indentation of Larval Oral Disc: absent (not emarginate)  =  0; present (emarginate)  =  1.

90. Marginal Papillae of Larval Oral Disc: short  =  0; enlarged  =  1; greatly enlarged  =  2. Additive.

The marginal papillae of most dendrobatids (e.g., boulengeri) are numerous (>50 in late stages) and relatively small (state 0).9 The marginal papillae of some species (e.g., pumilio) are fewer (<30 in late stages) and uniformly larger (state 1). For illustrations exemplifying this state, see Silverstone (1975a) and Haddad and Martins (1994). The remarkable larvae of caeruleodactylus and marchesianus possess only 12–18 (in late stages) greatly and irregularly enlarged marginal papillae (state 2). For illustrations of this state, see Caldwell et al. (2002a).

91. Submarginal Papillae of Larval Oral Disc (fig. 61): absent  =  0; present  =  1.

Among dendrobatids, submarginal papillae (see Altig and McDiarmid, 1999a) are known to occur only in larvae with umbelliform oral discs (e.g., nubicola). However, nondendrobatids that lack umbelliform discs also possess submarginal papillae (e.g., Duellmanohyla uranochroa; see Altig and McDiarmid, 1999b), demonstrating the transformational independence of these two characters.

92. Median Gap in Marginal Papillae of Lower Labium: absent  =  0; present  =  1.

Among dendrobatids, the median gap in the marginal papillae of the lower labium was illustrated and discussed by Myers and Daly (1980; see also 1987) and was claimed by them to be a synapomorphy for abditus, bombetes, and opisthomelas; Ruiz-Carranza and Ramírez-Pinilla (1992) added virolinensis to the group.

93. Anterior Larval keratodont Rows: 0  =  0; 1  =  1; 2  =  2. Additive.

Haddad and Martins (1994) reported the absence of anterior keratodont rows in hahneli, and we confirmed their observations in tadpoles of early stages. However, tadpoles of later stages possess a single anterior keratodont row, and we therefore coded this species as state 1. The unusual shape of the mouth, illustrated by Haddad and Martins, is retained until metamorphosis.

94. Posterior Larval keratodont Rows: 0  =  0; 1  =  1; 2  =  2; 3  =  3. Additive.

Haddad and Martins (1994) reported the absence of posterior keratodont rows in hahneli, and we confirmed their observations in small (e.g., stages 25 or 26) tadpoles. However, tadpoles of later stages possess two posterior keratodont rows, and we therefore coded this species as state 2.

95. Larval Jaw Sheath: absent  =  0; lower jaw only, not keratinized  =  1; entire, keratinized  =  2. Additive.

96. Larval Vent Tube Position: dextral  =  0; median  =  1.

Myers (1987) claimed the position of the larval vent tube among the differences between Minyobates and Dendrobates, with the former possessing the putatively primitive dextral position (state 0) and the latter being medial (state 1). Myers and Daly (1979) reported ontogenetic variation from dextral to median in silverstonei, and Donnelly et al. (1990) later reported that the vent tube of aurotaenia, lugubris, terribilis, and vittatus migrates from medial at Gosner (1960) stages 24 and 25 to dextral by stage 37 (or earlier). Caldwell and Myers (1990: 8) reported the frequency of dextral, sinistral, and median vent tubes in castaneoticus and summarized variation in this character and noted that “the rare sinistral condition [is] so far known only in the variation of the Dendrobates quinquevittatus complex.” Similar intraspecific variation has been reported in other anurans as well (e.g., Altig and McDiarmid, 1999a). Because ontogenetic series were unavailable for most dendrobatids, we coded the position of the larval vent tube as observed in the most developed larvae examined. As such, for example, we coded the vent tube of aurotaenia, lugubris, terribilis, and vittatus as dextral.

97. Spiracle: absent  =  0; present  =  1.

Among the coded dendrobatids, the unusual condition of the absence of the spiracle has been reported for only degranvillei (Lescure, 1984) and nidicola (Caldwell and Lima, 2003), both of which possess endotrophic larvae.

98. Lateral Line Stitches: absent  =  0; present  =  1.

The lateral line system is composed of receptor organs (mechanoreceptive neuromasts and electroreceptive ampullary organs) and the nerves that innervate them, both of which develop from the lateral line placodes (Schlosser, 2002a). Insofar as is known, the lateral line system is entirely lacking only in direct developing anurans, but the accessory organs (stitches) derived from the primary neuromasts fail to develop in some species of multiple families of anurans (Schlosser, 2002b). That is, the apparent absence of the lateral line system in these taxa is due to the absence of the stitches. The occurrence of stitches varies among dendrobatids, and we coded their absence (state 0) and presence (state 1). Only transverse stitches have been reported for anurans (Schlosser, 2002b), and dendrobatids are no exception.

Stitches have been reported (as the lateral line system), described, and/or illustrated by several authors (e.g., Mijares-Urrutia, 1991; La Marca, 1996 “1994”; Myers and Donnelly, 1997, 2001; Castillo-Trenn, 2004), but they are frequently overlooked, and their absence is usually not reported (e.g., Duellman, 2004). Although stitches are large and conspicuous in some species, they are barely detectable in others, owing to their small size and the pigmentation of the surrounding areas, so it is not safe to assume that failure to mention the lateral line stitches signifies their absence. As such, when coding character-states from the literature, we scored the lateral line stitches as absent when the authors were explicit or provided adequate illustrations or unusually thorough descriptions that (we assume) would have noted stitches had they been visible.

In addition to the presence and absence of stitches, we observed variation in the system of rami they form. However, as also observed by Castillo-Trenn (2004) in kingsburyi and R. W. McDiarmid (personal commun.) in other anurans, we found that the pattern of rami varies extensively within species, and too little is known about ramus ontogeny and other variation to allow transformation series to be delimited at present.

99–116. Behavior

The behavior of a number of dendrobatids has been documented and, beginning with Noble (1927), interpreted phylogenetically by several authors (e.g., Myers and Daly, 1976b; Weygoldt, 1987; Zimmermann and Zimmermann, 1988; Summers et al., 1999b). However, although behavior is unquestionably a valid source of evidence of phylogenetic relationships, its interpretation as phylogenetic characters requires special attention because particular behaviors are context dependent. Certain aspects of the behavioral repertory of a given species may be stereotyped and consistent across populations, but behavioral differences among populations have been documented (e.g., Myers and Daly, 1976b), and variation among individuals and over time (especially under different conditions) in the same individual are well known (especially in vocalizations; e.g., Juncá, 1998; Grant and Rodríguez, 2001), and the (external and/or internal) causes of this variation are, for the most part, a complete mystery. Even highly stereotyped responses may be context dependent, which casts some degree of doubt on the significance of observations made on captive specimens and the validity of comparisons to wild individuals. In light of these potential problems, we proposed hypotheses of homology of behaviors judiciously.

An example of overinterpretation of behavioral observations is Zimmermann and Zimmermann (1988), who performed a phenetic analysis of 62 variables (including vocalizations and larval morphology) for 32 species. We disregarded some of their “characters” because they are invariable in the ingroup and most of the outgroup (e.g., pulsation of flanks and/or throat; upright posture; female follows male; inflate body), defined too subjectively or arbitrarily to allow comparison (e.g., oviposition near a stream; male defends large territory, male defends small territory), demonstrably nonindependent (e.g., larvae carnivorous and/or herbivorous and larvae mostly herbivorous, microphagous), or are otherwise problematic. For example, their character “larvae carried singularly or in group” is problematic because, although differences almost certainly exist among species (bombetes has only been observed to carry up to three tadpoles [T. Grant, personal obs.; A. Suárez-Mayorga, personal commun.], whereas fraterdanieli carries up to at least 12 [T. Grant, personal obs.], and palmatus nurse frogs transport up to 31 tadpoles [Lüddecke, 2000 “1999”]), the number of dorsal tadpoles observed is highly variable (e.g., Myers and Daly, 1979: 326, reported a male silverstonei found carrying a single larva and two others carrying nine larvae; see also Lüddecke, 2000 “1999”: 315, table 3) due, potentially at least, to differences in clutch size, egg survivorship, rate of development, and the fact that a single load of tadpoles may be deposited all at once or one or a few at a time (Ruthven and Gaige, 1915). Clearly there are legitimate transformation series hidden in these observations, but more information is needed before characters can be delimited.

There is extensive missing data for behavioral characters, which necessarily limits the impact of these characters on the present analysis. However, one of our motivations for coding it nonetheless is that standardized codification facilitates future work. One of the most difficult aspects of individuating and scoring transformation series for phylogenetic analysis is that the behaviors are often complex and the ways they may be described by different observers may vary greatly. By delimiting and scoring these characters, we hope to draw attention to them for use in future comparative behavioral studies. Especially problematic are absences; for the present purposes, we coded conspicuous behaviors not reported in detailed studies as absent, but it is possible that they were simply not noticed. A similar problem is that even detailed notes may fail to mention expected observations, such as diurnal activity in dendrobatids. We did not make assumptions regarding the latter characters and only scored them from personal observation or explicit statements.

In addition to personal observations and unpublished field notes, behavioral data (not including vocalizations) were taken from the following published sources: Dunn (1933, 1941, 1944), Eaton (1941), Trapido (1953), Test (1954, 1956), Funkhouser (1956), Stebbins and Hendrickson (1959), Sexton (1960), Savage (1968, 2002), Duellman and Lynch (1969, 1988), Hoogmoed (1969), Mudrack (1969), Myers (1969, 1982, 1987), Edwards (1971), Lynch (1971), Crump (1972), Silverstone (1973, 1975a, 1975b,1976), Polder (1974), Durant and Dole (1975), Lescure (1975, 1976, 1991), Lüddecke (1976, 2000 “1999”), Wells (1978, 1980a, 1980b, 1980c), Myers et al. (1978, 1984), Myers and Daly (1976a, 1979, 1980, 1983), Cei (1980), Limerick (1980), Weygoldt (1980, 1987), Vigle and Miyata (1980), Zimmermann and Zimmermann (1981, 1984, 1985, 1988), Kneller (1982), Hardy (1983), Heyer (1983), Dixon and Rivero-Blanco (1985), Jungfer (1985, 1989), Frost (1986), Formas (1989), Summers (1989, 1990, 1992, 1999, 2000), Caldwell and Myers (1990), Praderio and Robinson (1990); Aichinger (1991), Morales (1992), van Wijngaarden and Bolaños (1992), Brust (1993), Duellman and Wild (1993), Giaretta et al. (1993), Rodríguez and Myers (1993), Juncá et al. (1994), Kaiser and Altig (1994), Coloma (1995), Cummins and Swan (1995), Jungfer et al. (1996a), La Marca (1996 “1994”, 1998 “1996”), Caldwell (1997), Fandiño et al. (1997), Grant et al. (1997), Caldwell and de Araújo (1998, 2004, 2005), Juncá (1998), Grant and Castro-Herrera (1998), Morales and Velazco (1998), Boistel and de Massary (1999), Caldwell and de Oliveira (1999), Haddad and Giaretta (1999), Hoff et al. (1999), Summers et al. (1999b), Köhler (2000), Kok (2000), Lescure and Marty (2000), Lötters et al. (2000), Bourne et al. (2001), Downie et al. (2001), Hödl and Amézquita (2001), Lima et al. (2001, 2002), Myers and Donnelly (2001), Summers and Symula, (2001), Caldwell and Lima (2003), Giaretta and Facure (2003, 2004), Lima and Keller (2003), Grant (2004), Lehtinen et al. (2004), Toledo et al. (2004), and Summers and McKeon (2004).

99. Advertisement Calls: buzz  =  0; chirp  =  1; trill  =  2; retarded trill  =  3. Nonadditive.

Male advertisement calls played an important role in the systematics studies of Myers and Daly (e.g., 1976b). For example, the histrionicus group of Myers et al. (1984) is delimited, in part, by a synapomorphic “chirp” call. Data are available for numerous species (for partial review, see Lötters et al., 2003b), but their use in systematics has been predicated on their identification as a buzz (Myers and Daly, 1976b: 225), chirp (Myers and Daly, 1976b: 226), trill (Myers et al., 1978: 325), retarded trill (Myers and Daly, 1979: 18), or retarded chirp (Myers and Burrowes, 1987: 16), and the diversity of dendrobatid calls extends far beyond these few types. Although Lötters et al. (2003b) aimed to expand and standardize the definitions of these calls, they were aware that known calls of most species of dendrobatids do not correspond to any of these types, and additional characterizations such as peeps, cricketlike chirps, croaks, whistled trills, or harsh peep train (e.g., Rodríguez and Myers, 1993; Grant and Castro-Herrera, 1998; Bourne et al., 2001) have been employed, although none of these is defined precisely.

It is clear that these call types are composites of temporal and spectral transformation series that should be decomposed into independent characters for phylogenetic analysis. Unfortunately, the necessary analysis of advertisement call variation was outside the scope of the present study, and we scored advertisement calls according to the published scheme in order to test prior hypotheses (e.g., the buzz call as a synapomorphy of the histrionicus group). This is highly suboptimal, mainly because (1) many species were scored as unknown simply because their calls did not fit within the current classification and not because data were unavailable, and (2) extensive information on spectral and temporal modulation could not be incorporated. We hope this may be rectified in future studies. Species were coded according to Lötters et al. (2003b).

100. Male Courtship: Stereotyped Strut: absent  =  0; present  =  1.

Dole and Durant (1974), Wells (1980a), and Lüddecke (2000 “1999”) reported the occurrence of this behavior (state 1) in collaris, panamensis (as inguinalis; see Grant, 2004), and palmatus, respectively. Lüddecke (2000 “1999”: 309, see also p. 210 for illustration) described it as “a stereotyped rigid-looking strut [the male] performs in the silent intervals between advertisement calls”.

101. Male Courtship: Jumping Up and Down: Absent  =  0; present  =  1.

Wells (1980c: 195) described this character as follows:

When a female or brown male moved near a calling black male, the usual response of the black male was to jump up and down on his calling perch. … Often the male would run for a few centimeters and jump so that his front feet rose 1–2 mm off the ground. Similar behavior has been reported in a closely related species (C. collaris), although males of that species apparently leap higher off the substrate than do male C. trinitatis.

102. Female Courtship: Crouching: absent  =  0; present  =  1.

According to Lüddecke (2000 “1999”: 309), in this behavior the female crouches in front of, but does not slide underneath, the male.

103. Female Courtship: Sliding under Male: absent  =  0; present  =  1.

Lüddecke (2000 “1999”: 309) reported for palmatus that the female crouches and then “slides completely under the male” as one of the final stages of courtship.

104. Timing of Sperm Deposition: after oviposition  =  0; prior to oviposition  =  1.

In 1980 both Limerick (1980) and Weygoldt (1980) reported that sperm deposition in pumilio appears to occur prior to oviposition (state 1). This unusual occurrence has since been reported for additional species by several authors (Jungfer, 1985; Weygoldt, 1987; Jungfer et al., 1996a, 2000; Lötters et al., 2000). Jungfer et al. (1996a) claimed this as a synapomorphy of Dendrobates and rationale for placing Minyobates in its synonymy, as done subsequently by Jungfer et al. (2000). Nevertheless, several species of Dendrobates sensu Jungfer et al. have been explicitly reported to have post-oviposition fertilization (e.g., histrionicus fide Zimmermann, 1990: 69; arboreus fide Myers et al., 1984: 15), which suggests that the phylogenetic interpretation of this character is not as straightforward as Jungfer et al. implied.

105. Reproductive Amplexus (fig. 62): absent  =  0; axillary  =  1; cephalic  =  2. Nonadditive.

Figure 62

Character 105, reproductive amplexus.

State 2, cephalic amplexus (anthonyi, AMNH live exhibit) shown in anterior (A) and lateral (B) aspects.


Myers et al. (1978: 324–325) first described and illustrated cephalic amplexus in tricolor. Although reproductive amplexus is absent in numerous dendrobatids (a variety of pseudo-amplectant positions–including cephalic grasping–may be employed in aggressive and/or courtship behavior), cephalic amplexus was cited by numerous authors (e.g., Duellman and Trueb, 1986; Myers and Ford, 1986; Myers et al., 1991) as a dendrobatid synapomorphy, with the absence in numerous dendrobatids explained as a derived loss. A similar amplectant position was reported for Hypsiboas faber by Martins and Haddad (1988), but the sampled outgroup species exhibit axillary amplexus.

106. Cloacal Apposition: absent  =  0; present  =  1.

Crump (1972: 197) first reported the occurrence of this character-state in granuliferus, in which the male and female face opposite directions and bring their cloacae into contact.

107. Egg Deposition Site: aquatic  =  0; terrestrial: leaf litter, soil, under stones  =  1; terrestrial: phytotelmata  =  2. Additive.

This character is coded additively to reflect the increasing or decreasing degree of association with ground-level standing or flowing water.

108. Egg Clutch Attendance: none  =  0; male  =  1; female  =  2; both  =  3. Nonadditive.

See discussion of sex of nurse frog (Character 110, below) for the rationale behind treating biparental care as a separate state instead of polymorphism.

109. Dorsal Tadpole Transport (fig. 63): absent  =  0; present  =  1.

Figure 63

Character 109, dorsal larval transport.

State 1, present (fraterdanieli, specimens at UVC). This male nurse frog was transporting 12 tadpoles.


Noble (1927: 103) noted that his grouping of dendrobatids on morphological grounds “receives an eloquent support from life history data” as well, pointing out that males of species of Dendrobates and Phyllobates transport tadpoles on their back to pools (state 1), and, further, that “[n]o other Salientia have breeding habits exactly like Dendrobates and Phyllobates” (p. 104). For the present purposes, we refer only to dorsal transport by genetic parents (see below), and we follow Ruthven and Gaige (1915: 3) in referring to the parent that performs larval transport as the nurse frog.

Among outgroup taxa, males of Rhinoderma darwinii transport larvae, which Laurent (1942: 18) claimed as evidence of close relationship to dendrobatids. However, male Rhinoderma transport young in their hypertrophied vocal sacs (see Noble, 1931: 71 for illustration), whereas dendrobatids transport tadpoles on their backs. Several other anurans transport their young on their backs (e.g., Hemiphractus, Stefania, Gastrotheca), but they do so beginning with the egg clutch, whereas in dendrobatids transport is exclusively post-hatching. Among Neotropical anurans, the only species reported to have terrestrial (nontransported) eggs and dorsally transported tadpoles is Cycloramphus stejnegeri (Heyer and Crombie, 1979). Tadpole transport is not known for the sampled species of Cycloramphus (C. boraceiensis), but Giaretta and Facure (2003) reported male egg attendance (also present in C. dubius and C. juimiria; C.F.B. Haddad, personal obs.), which leaves open the possibility of tadpole transport. Outside of the Neotropics, apparently identical parental care occurs in the dicroglossids Limnonectes finchi and L. palavanensis (Inger and Voris, 1988) and the sooglossid Sooglossus sechellensis (Lehtinen and Nussbaum, 2003). In the hemisotid Hemisus marmoratus, maternal tadpole transport may occur as the female digs a subterranean tunnel to a water body (Lehtinen and Nussbaum, 2003).

Dorsal tadpole transport by parents is here coded as a single transformation series, but even the extremely limited evidence available suggests this is much more complex and probably involves multiple characters. Stebbins and Hendrickson (1959: 509) reported in subpunctatus that “[t]he tadpoles are anchored to the back of the frog by a sticky mucus.” Myers and Daly (1980: 19) further noted that

[i]n some dendrobatids, this attachment is accomplished solely by mere surface adhesion between the mucus and the tadpoles' flattened or slightly concave bellies, and the larvae are easily moved about and dislodged. … In other dendrobatids … the mucus attachment seems almost gluelike and the tadpoles are very resistant to being dislodged.

To this we add only that it is common for tadpoles to wriggle around freely on the nurse frog's back without being prodded (especially if few tadpoles are being transported by a large frog, such as bicolor), giving the impression that they adjust themselves to the nurse frog's movements.

Ruiz-Carranza and Ramírez-Pinilla (1992) studied the histology of the contact surfaces of nurse frogs and transported tadpoles in virolinensis and found numerous modifications in both the dorsal integument of the nurse frog and the ventral integument of the larvae. Lüddecke (2000 “1999”) found experimentally that recently hatched larvae of palmatus did not mount a rubber model moistened with water, mounted but immediately abandoned a rubber model treated with either male or female skin secretions, and would only mount and settle on a live frog, with no sexual discrimination. In a less controlled experiment with hatching anthonyi T. Grant found that gently touching the jelly capsule with a finger was sufficient to stimulate hatching, immediate mounting, settling, and attachment (i.e., the tadpoles remained attached to the finger submerged in water for >1 min until they were forcibly dislodged); however, the male nurse frog had already removed most of the tadpoles from the clutch, which may have “primed” the remaining embryos for hatching and transport. As coded in Character 110, the sex of the nurse frog varies among species, and little is known about the biology of this kind of sex role reversal. Much more research is required to understand the evolution of dorsal tadpole transport in dendrobatids.

110. Sex of Nurse Frog: male  =  0; female  =  1; both  =  2. Nonadditive.

Among species that transport larvae, the role of the nurse frog is typically assumed by one sex (Wells, 1978, 1980a, 1980b, 1980c). However, in some species, both sexes have been observed carrying tadpoles. Myers and Daly (1983) found experimentally that in anthonyi (as tricolor) the father was normally responsible for tadpole transport and would actively prevent the mother from approaching the developing clutch, but that removal of the male shortly after breeding led to female brood care and larval transport. They suggested that parental care is competitive, that is, the sexes compete to care for the offspring. This is at least consistent with Aichinger's (1991) observation of 38 male nurse frogs and only a single female nurse frog. Caldwell and de Araújo (2005) reported that in brunneus and femoralis males usually transport larvae but females will occasionally perform this role, and Silverstone (1976: 38) reported nurse frogs of both sexes for petersi. It is unknown how widespread this behavior is (i.e., if both sexes are usually potential carriers, even though one sex predominantly assumes this role, as in tricolor), but it is not universal. It is unknown how widespread this behavior is (i.e., if both sexes are usually potential carriers, even though one sex predominantly assumes this role, as in tricolor), but it is not universal. Host Lüddecke (in litt., 08/31/00) found experimentally that palmatus does not exhibit this behavior; in his experiments, Lüddecke found that mothers ate their eggs when the fathers were removed. As noted for Character 108, Lüddecke (2000 “1999”) also found that tadpoles mounted males or females indiscriminately, which suggests that a potential for female transport may be primitive.

Given the paucity of experimental data, it is unclear if all cases of both sexes transporting tadpoles are the result of the same mechanism and/or a transformation event. For the time being, we coded each species based on available information. We have therefore scored species as having males (state 0), females (state 1), or both sexes (state 2) assume the role of nurse frog. This character individuation will undoubtedly require modification as more data are obtained on this behavior.

We coded biparental transport as a separate state rather than an ambiguous polymorphism because the behavioral modifications required to achieve biparental care do not apply to male or female care alone, that is, it involves more than just the co-occurrence of states 1 and 2. Also, we did not specify any particular additivity for this transformation series, as there is no evidence that the shift between sexes requires a cooperative (or competitive) intermediate biparental stage (although this could be indicated by phylogenetic analysis).

Savage (2002) reported male nurse frogs in talamancae, and Summers and McKeon (2004: 62, fig. 3) scored femoralis, hahneli (as “hahnei”), talamancae, and trilineatus (as “trilieatus”) as having exclusively male parental care. However, nurse frogs of both sexes have been reported for femoralis (Silverstone, 1976; Lescure, 1976; for a recent report see Caldwell and de Araújo, 2005), hahneli (as pictus; Aichinger, 1991) and trilineatus (Aichinger, 1991), and exclusively female nurse frogs have been reported for talamancae (Grant, 2004 and references therein). Insofar as Savage and Summers and McKeon did not dispute those reports or provide specimen documentation for confirmation, we dismiss their reports as erroneous.

Adding to the behavioral complexity and the difficulty in coding this character, Weygoldt (1980) found in pumilio that mothers transport their larvae to water, whereas males may perform nonparental infanticide by transporting unelated tadpoles and not depositing them in water, thus achieving sexual interference. As noted above, we did not score nonparental transport as part of this transformation series.

111. Larval Habitat: ground level pool or slow-flowing stream or other body of water  =  0; phytotelmata  =  1; nidicolous  =  2. Nonadditive.

Note that there is a logical dependency between larval habitat and dorsal tadpole transport (Character 108) in that nidicolous larvae are, by definition (Altig and Johnston, 1989; McDiarmid and Altig, 1999), not transported. Nevertheless, the two characters are not coextensive and are clearly independent: Lack of transport may also be associated with either state, and nurse frogs may transport larvae to either a ground level body of water or phytotelm.

Although “phytotelm” often refers to chambers above ground (e.g., bromeliads), technically the term applies to any chambers in a plant. Moreover, whether on or above the ground, these phytotelmata are expected to be biologically equivalent (e.g., both microhabitats offer limited space, nutrients, and other resources, and have a potentially high risk of predation), and we therefore did not discriminate between ground-level and higher phytotelmata. For example, we followed Caldwell and de Araújo (1998; 2004) in scoring castaneoticus as a phytotelm breeder because it uses Brazil nut husks.

112. Larval Diet: detritivorous  =  0; predaceous  =  1; oophagous  =  2; endotrophic  =  3. Nonadditive.

The vast majority of anurans have detritivorous tadpoles (state 0). We assumed that larvae found in ground level pools or streams or other large bodies of water (i.e., state 0 of Character 111) are detritivorous; unless diet is actually known, larvae of other habitats were coded as unknown (“?”) for this character. Numerous species of dendrobatids are aggressive predators that consume con- and heterospecific tadpoles and arthropod larvae as an important component of their diet (Caldwell and de Araújo, 1998; state 1). Several species consume sibling oocytes (oophagous, state 2), either exclusively (histrionicus group; Limerick, 1980) or as part of a predaceous diet (state 1; e.g., vanzolinii; Caldwell and de Oliveira, 1999). We coded the latter taxa as polymorphic (see also Character 113, provisioning of oocytes for larval oophagy).

Four species with endotrophic larvae (state 3) have been described: chalcopis (not included in this study), degranvillei, nidicola, and stepheni (for review and description of nidicola see Caldwell and Lima, 2003). Some amount of larval growth prior to deposition (e.g., during transport; Wells, 1980b) is probably widespread, but complete endotrophy is much more limited and tends to be correlated with a variably reduced morphology. Nevertheless, the unmodified larva of chalcopis (Kaiser and Altig, 1994) demonstrates the transformational independence of endotrophy and the various morphological reductions (see also Altig and Johnston, 1989). Likewise, the occurrence of endotrophy is independent of tadpole habitat: degranvillei is transported (Lescure and Marty, 2000; tadpole transport was also predicted for chalcopis by Juncá et al., 1994), whereas the remaining endotrophic tadpoles are nidicolous.

113. Provisioning of Oocytes for Larval Oophagy: biparental  =  0; female only  =  1.

Caldwell and de Oliveira (1999) reported provisioning of eggs for consumption by sibling tadpoles in vanzolinii, as did Bourne et al. (2001) in beebei. In these species, egg provisioning is stimulated by male courtship behavior and is therefore biparental (state 0), and larval diets include a variety of foods (for additional records, see Lehtinen et al., 2004). In other oophagous species (e.g., histrionicus) tadpole care is undertaken exclusively by the female. An alternative way to delimit state 1 is as obligate oophagy, as it appears that tadpoles of these species feed only on eggs (obligate oophagy demonstrated experimentally for pumilio by Brust, 1993; Pramuk and Hiler, 1999), whereas the others are predaceous (Caldwell and de Araújo, 1998).

Zimmermann and Zimmermann (1988) reported biparental provision of oocytes in ventrimaculatus (as quinquevittatus) in captivity, but Summers et al. (1999b) reported exclusively male care in Peruvian ventrimaculatus. Caldwell and Myers (1990) hypothesized that ventrimaculatus is a complex of cryptic species, which is supported by this behavioral variation.

114. Adult Association with Water: aquatic  =  0; riparian (<3 m from water)  =  1; independent of water (up to ca. 30 m from water)  =  2. Additive.

Myers et al. (1991) characterized nocturnus as aquatic, which they contrasted with species such as panamensis (as inguinalis; see Grant, 2004) and latinasus, which are riparian and independent of streams, respectively. Postmetamorphic frogs of any species may be found in or near water, and environmental variation must be taken into account (i.e., during dry seasons or at drier localities frogs that are otherwise found well into the forest will congregate near sources of water), but the degree of commitment to or dependency on an aquatic environment segregates dendrobatids into at least three groups. Among dendrobatids, nocturnus appears to be the only aquatic species, that is, individuals are generally found immersed in water (state 0). A much greater number of dendrobatids are riparian (state 1). These species are almost entirely confined to the areas immediately adjacent to streams, where they establish and defend streamside territories (e.g., Wells, 1980a, 1980c). When disturbed these species seek refuge in water and not in leaf litter or debris beside the stream. The third group of species is effectively independent of water (state 3). As noted by Funkhouser (1956: 78) for espinosai, these species “scurry under debris for safety; they do not take to water even when it is close by”, and territorial and courtship behaviors occur well away from ground water. Despite their relative independence from water, the density of these frogs may be greater nearer to streams, even in extremely wet environments such as the Colombian Chocó (T. Grant, personal obs.) where general moisture requirements are unlikely to be a limiting factor. This is probably due to reproductive factors: Many of these species are known to transport larvae from terrestrial nests to streams or ground-level pools, and it is predictable that selection would favor preference for sites closer to more permanent, larger bodies of water.

A potential fourth character-state is arboreality. For example, Myers et al. (1984) named arboreus in recognition of that species' arboreal habitat preference, whereas other species (e.g., fraterdanieli) are active exclusively on the ground and only climb into vegetation (never more than 1 m) to sleep. However, between these two extremes lie variations that defy simple codification. For example, bombetes is a leaf-litter frog that climbs up to 30 m above ground to deposit larvae in bromeliads (T. Grant, personal obs.; A. Suárez-Mayorga, personal commun.). Similarly, histrionicus forages in leaf litter on the ground but calls from perches in vegetation above ground (Silverstone, 1973; Myers and Daly, 1976b). Clearly there are evolutionary transformation events embedded in these behavioral variations, but the extent to which variation is obligate or facultative is unclear, and we have chosen to group putatively arboreal and terrestrial species as state 2. Assuming the additivity of this transformation series, the transformation from state 1 to state 2 applies to all of these species (as coded), and we have failed to recognize the additional transformation(s) from state 2a (terrestrial) to state 2b (arboreal).

115. Diel Activity: nocturnal  =  0; diurnal  =  1.

Myers et al. (1991) cited the transformation from nocturnal to diurnal activity as evidence for the monophyly of all dendrobatids minus nocturnus. As has been noted by several authors (e.g., Myers et al., 1991; Coloma, 1995; Duellman, 2004), some other species (e.g., riveroi, bocagei, nexipus) exhibit crepuscular or limited nocturnal activity, at least facultatively (e.g., on brightly moonlit nights). Although the conditions that surround this behavior are unclear, we coded these species as polymorphic.

116. Toe Trembling: absent  =  0; present  =  1.

We have observed several species to exhibit toe trembling or toe tapping, whereby usually the fourth toe (sometimes also the third) trembles or twitches rapidly up and down. Little is known about this behavior. Most observations derive from captive individuals, and there is no known function. It does not appear to be involved in intraspecific visual communication, as individuals do not alter their behavior notably when an individual begins toe trembling, and toe trembling may be observed in individuals that are isolated or in groups. Toe trembling is not continuous and only occurs in active frogs. However, although quantitative data are lacking, onset and/or vigor of toe trembling in dendrobatids does not seem to correlate with any particular stimulus and does not obviously originate as an epiphenomenon (Hödl and Amézquita, 2001). Toe trembling may (or may not) occur while foraging and during inter- and intraspecific interactions with individuals of the same or opposite sex. Hartmann et al. (2005) reported toe (and finger) trembling for the hylid Hyspiboas albomarginatus, which they interpreted to be visual signaling.

117. Hyale Anterior Process: absent  =  0; present  =  1.

All dendrobatids examined possess a single anterior process on each hyale (state 1), and it is both present and absent (state 0) in the sampled outgroup species. Myers and Ford (1986) cited the occurrence of a second anterior process on the hyalia of Atopophrynus syntomopus as evidence that it is not a dendrobatid; we did not sample this taxon in the present study and therefore did not test their hypothesis.

118. Shape of Terminal Phalanges: T-shaped  =  0; knobbed  =  1.

Following Lynch's (1971) terminology, the species sampled in this study possess T-shaped and knobbed phalanges.

119–128. Epicoracoids

Pectoral girdle architecture has been key in all discussions of dendrobatid relationships since Boulenger (1882). Character-states have generally been delimited in terms of the overlap or fusion of the epicoracoids and/or the presence or absence of epicoracoid horns (for historical usages, see Kaplan, 2004), with the epicoracoids of dendrobatids characterized as entirely fused and nonoverlapping and lacking epicoracoid horns, as in “firmisternal” taxa.10 However, this is clearly an oversimplification (e.g., Noble, 1926; Kaplan, 1994, 1995, 2000, 2001, 2004). Recently, Kaplan (2004) divided girdle architecture variants into separate transformation series relating to degree of fusion (freedom) and overlap (nonoverlap), and he proposed explicit character-states, which we employ here.

Of most relevance to the problem of dendrobatid phylogeny, Noble (1926) claimed that the entirely fused epicoracoids of subpunctatus overlap during ontogeny, a finding that was challenged by Griffiths (1959), Lynch (1971a), and Ford (1989), but ultimately vindicated by Kaplan (1995). However, Kaplan (1995) interpreted differences between the overlap in subpunctatus and “arciferal” taxa (e.g., Bufo) as evidence that the overlap is nonhomologous and therefore not evidence of common ancestry (contra Noble, 1926).

To date, the only dendrobatid in which overlap has been detected is subpunctatus. Kaplan (1995: 302) also examined abditaurantius (adult), palmatus (adult), and virolinensis (Gosner, 1960, stages 42–43) and “did not find any evidence of overlap”, and Griffiths (1959) reported that overlap is absent in trinitatis (not trivittatus, as reported by Kaplan, 1995).

Although we did not perform the detailed histology necesary to score these characters precisely, epicoracoid morphology has played an important role in all previous discussions of dendrobatid phylogeny and we believe it would be inappropriate to exclude it altogether. Therefore, although we are cognizant of the potential errors that may be incorporated into the analysis, we coded degree of fusion and overlap in adults (or near adults) as precisely as possible through examination of cleared and stained whole specimens. Although Kaplan (1995: 301) stated that in subpunctatus “the girdle halves overlap in adults except for a small area of ventral fusion”, this was not visible in cleared and stained specimens, and so for consistency we coded this species as lacking overlap. Insofar as we did not detect overlap in any other dendrobatid, and Kaplan (1995) argued that overlap in subpunctatus is nonhomologous with the overlap of “arciferal” taxa, coding the occurrence of overlap in this taxon would result in an autapomorphy and therefore would not affect the results of the present analysis.

119. Epicoracoid Fusion: fused from anterior tips to posterior tips  =  0; fused from anterior tips of epicoracoids to level midway between the posterior levels of the procoracoids and the anterior ends of the coracoids, free posteriorly  =  1; fused from anterior tips to a level slightly posterior to medial ends of clavicles, free posteriorly  =  2. Additive.

States 0, 1, and 2 correspond to states E, C, and A, respectively, of Kaplan (2004: 94). State 1 is intermediate in the degree of fusion, which we considered to be evidence for the hypothesis of 0 ↔ 1 ↔ 2 additivity for this transformation series.

120. Epicoracoid Overlap: nonoverlapping  =  0; overlapping from level slightly posterior to level of procoracoids to anterior level of sternum  =  1; overlapping from level between posterior level of procoracoids and anterior ends of coracoids to posterior level of coracoids  =  2; overlapping from level slightly posterior to medial ends of clavicle to level slightly posterior to anterior level of sternum  =  3. Nonadditive.

States 0, 1, 2, and 3 correspond to states B, E, C, and A, respectively, of Kaplan (2004: 94). Because variation in overlap involves complex changes in epicoracoid structure, we were unable to find evidence to select one hypothesis of additivity over another; we therefore treated this character nonadditively.

121. Angle of Clavicles (fig. 64): directed laterally  =  0; directed posteriorly  =  1; directed anteriorly  =  2. Nonadditive.

Figure 64

Character 121, angle of clavicles.

A: State 0, directed laterad (steyermarki, AMNH 118572). B: State 1, directed anteriad, approximately parallel to the posterior margin of the coracoid (opisthomelas, AMNH 102582). C: State 2, directed anteriad (Eupsophus roseus, KU 207501).


In most dendrobatids each clavicle runs laterad, perpendicular to the sagittal plane (state 0). In some species, the clavicles are directed posteriad, running approximately parallel to the posterior margin of the coracoid (state 1). Clavicles directed anteriad (state 2) are confined to certain species in the outgroup. The intermediacy of the laterally directed clavicles is suggestive of additivity; however, pectoral girdle ontogeny does not proceed in this way, and we therefore scored this transformation series nonadditively.

122. Acromion Process: cartilaginous, distinct  =  0; fully calcified (or ossified), continuous with clavicle and scapula  =  1.

The acromion processes of some taxa are cartilaginous (state 0) in mature specimens, whereas in others they are extensively calcified or ossified (state 1). We did not distinguish between extensive calcification and ossification. As with other osteological characters that vary ontogenetically (characters 127, 134–138), adult females are most extensively ossified.

123. Prezonal Element (Omosternum): absent  =  0; present  =  1.

124. Prezonal Element (Omosternum) Anterior Expansion: not expanded distally, tapering to tip  =  0; weakly expanded, to 2.5× style at base of cartilage or equivalent  =  1; extensively expanded distally, 3.5× style or greater  =  2. Additive.

125. Prezonal Element (Omosternum) Shape of Anterior Terminus: rounded or irregularly shaped  =  0; distinctly bifid  =  1.

126. Prezonal Element (Omosternum) Shape of Posterior Terminus: simple  =  0; notched, forming two struts  =  1; continuous with epicoracoid cartilage  =  2. Nonadditive.

127. Prezonal Element (Omosternun) Ossification: entirely cartilaginous  =  0; medially ossified (cartilaginous base and tip)  =  1; basally ossified (cartilaginous tip)  =  2; entirely ossified  =  3. Additive.

128. Suprascapula Anterior Projection: cartilaginous  =  0; heavily calcified  =  1.

129. Sternum Shape: simple (rounded, irregular)  =  0; medially divided  =  1.

The posterior termination of the sternum is either simple (rounded or irregularly shaped; state 0) or distinctly divided medially, forming either two prongs or two broad, rounded lobes. We also observed independent variation in the lateral expansion of the sternum. For example, even though the sternum of both species is medially divided, in panamensis (UMMZ 167459) it is broadly expanded, whereas in juanii (ICN 5097) the sternum is tapered. However, we also observed confounding intermediate and other variation and were unable to individuate states objectively for the current study.

130. Zygomatic Ramus of Squamosal (fig. 65): elongate, slender, pointed  =  0; very long and slender  =  1; robust, truncate, and elongate  =  2; shorter and less robust but still well defined  =  3; well defined, moderate length, abruptly directed ventrad  =  4; inconspicuous, poorly differentiated  =  5; very small, inconspicuous, hooklike  =  6; miniscule bump  =  7; robust, elongate, in broad contact with the maxilla  =  8. Nonadditive.

Figure 65

Character 130, zygomatic ramus of squamosal.

A: State 0 (Eupsophus roseus, KU 207501). B: State 1 (Cycloramphus fuliginosus, KU 92789). C: State 2 (nocturnus, AMNH 130041). D: State 2 shown in a dissected whole specimen (palmatus, AMNH 20436). E: State 3 (trinitatis, AMNH 118389). F: State 4 (trivittatus, AMNH 118428). G: State 5 (espinosai, AMNH 118417). H: State 6 (bocagei, UMMZ 182465). I: State 7 (Melanophryniscus stelzneri, AMNH 77710). J: State 8 (Megaelosia goeldii, AMNH 103952; squamosal colored red).


The zygomatic ramus of the squamosal varies extensively and forms a series of complex morphological transformations. In state 0, the zygomatic ramus is elongate (approximately half the length of the ascending ramus and extending well anterior past the tympanic ring), slender, gently curved, and pointed. State 1 is a very long and slender process. State 2 is robust, truncate, and elongate (extending anterior to the tympanic ring, but not as long as state 2). State 3 is shorter and less robust than state 2 but is still a conspicuous shaft that usually extends anterior to the tympanic ring. Like the processes of states 0, 1, and 2, the axis of state 3 is at most only weakly inclined toward the maxilla. The zygomatic ramus of state 4 is also well defined, but it is distinctly and abruptly directed ventrad, its axis pointing almost straight down at the maxilla, that is, a line from the zygomatic ramus would intersect the posterior extreme of maxilla, and it does not extend anterior to the tympanic ring. State 5 is an inconspicuous, poorly differentiated process. The zygomatic ramus of most of the sampled species is a very small, inconspicuous, hooklike process (state 6). McDiarmid (1971) considered the zygomatic ramus to be absent in Melanophryniscus; however, we detected a miniscule bump (state 7) in Melanophryniscus stelzneri, which we considered to be homologous with the zygomatic ramus. (Regardless, we did not observe this state in any other species included in the present analysis, so coding it as “absent” or “a miniscule bump” has no bearing on the outcome of analysis.) In Megaelosia goeldii, the robust zygomatic ramus extends anteroventrad to be in broad contact with the maxilla (state 8).

131. Orientation of Alary Process of Premaxilla: directed anterolaterally  =  0; directed dorsally  =  1; directed posterodorsally  =  2. Additive.

Myers and Ford (1986) claimed the anterolaterally tilted alary process as a synapomorphy of dendrobatids, although several other taxa also share this state (e.g., Lynch, 1971). We treated this transformation series additively (0 ↔ 1 ↔ 2) based on the argument that the rearrangement in skull architecture required to alter the orientation of the alary process would necessitate passing through the intermediate stage.

132. Palatines: absent  =  0; present  =  1.

Variation in the occurrence of the palatine bones among dendrobatids has been documented by numerous authors (e.g., Silverstone, 1975a; Myers and Ford, 1986), and Kaplan (1997) interpreted the character phylogenetically. Trueb (1993) considered the neobatrachian palatine to be nonhomologous with the palatine of other vertebrates, and she is almost certainly correct. Nevertheless, this bone would unquestionably be identified as a palatine if anurans were found to be rooted on a neobatrachian. As such, the validity of Trueb's distinction rests on the phylogenetic position of neobatrachians, that is, it is a conclusion of phylogenetic analysis, not a premise. We therefore follow Haas (2003) in referring to this bone as the palatine.

133. Quadratojugal–Maxilla Relation: overlapping  =  0; separated  =  1.

In dendrobatids, the quadratojugal and maxilla are never in contact or tightly bound but are instead loosely bound by ligamentous tissue. In some species, the two bones overlap (state 0), whereas in others the anterior tip of the quadratojugal does not reach the level of the posterior tip of the maxilla.

134. Nasal–Maxilla Relation (fig. 66): separated  =  0; in contact  =  1.

Figure 66

Character 134, nasal–maxilla relation.

A: State 0, nasal and maxilla broadly separated (silverstonei, AMNH 91847). B: State 1, greater magnification showing nasal and maxilla overlapping or fused (bassleri, AMNH 43402).


The nasal and maxilla may be separate (state 0) or contact each other. We did not distinguish between overlap and fusion because gross examination under a dissecting microscope proved inadequate to determine the status of many specimens and the necessary histological study was infeasible for the present study.

135. Nasal–Sphenethmoid Relation (fig. 67): free, separate  =  0; overlapping or fused  =  1.

Figure 67

Character 135, nasal–sphenethmoid relation.

A: State 0, separate (bassleri, AMNH 43402). B: State 1, overlapping or fused (nocturnus, AMNH 130014). In this case the nasals clearly overlap but are not fused with the sphenethmoid, but in other species the distinction between the bones is not as clear.


In state 0, the nasal and sphenethmoid do not overlap, whereas in state 1 those bones are either overlapping or fused. We did not distinguish between overlapping and fusion as the necessary histological analysis was infeasible for the present study.

136. Frontoparietal Fusion: entirely free  =  0; fused posteriorly  =  1; fused along entire length  =  2. Additive.

Ontogenetic variation in frontoparietal fusion suggests that it proceeds anteriorly. We therefore treated this character additively.

137. Frontoparietal–Otoccipital Relation: free  =  0; fused  =  1.

Among dendrobatids, there is variation in the relation of the frontoparietal and otoccipital (i.e., the fused prootic and exoccipital; Lynch, 1971: 52), being free (state 0) in some taxa and fused (state 1) in others. Lynch (1971) documented variation in this character in numerous outgroup taxa.

138. Exoccipitals: free, separate  =  0; fused sagittally  =  1.

The exoccipital portions of the fused otoccipital bones (see Lynch, 1971: 52) may be separated by cartilage (i.e., chondrocranial ossification may be incomplete; state 0) or may be fused sagittally (state 1). A further potential state is for them to abut but not fuse, but we did not observe this among the specimens examined.

139. Maxillary Teeth: absent  =  0; present  =  1.

Variation in the occurrence of teeth has been used consistently in dendrobatid systematics (see Grant et al., 1997 for discussion). In the more recent literature, Edwards (1971: 147) stated that dendrobatids “can be divided into two groups—those species lacking maxillary teeth (Dendrobates) and those having maxillary teeth (Phyllobates and Colostethus).” Silverstone (1975a) showed that the situation is more complicated due to character conflict and polymorphism. (See also Materials and Methods for uncoded variation in maxillary tooth size and shape.)

140. Maxillary Tooth Structure: pedicellate  =  0; nonpedicellate  =  1.

Most anurans have pedicellate teeth, whereby the tooth is divided into a pedicel and crown (Parsons and Williams, 1962). Parsons and Williams (1962: 377) examined the teeth of bocagei (as Phyllobates bocagii) and palmatus (as Phyllobates granuliventris) and found that “the division is certainly not marked in gross structure and is quite probably lacking.” Myers et al. (1991: 11) further pointed out that there is no “pattern of physical separation of crowns from pedicels (breakage is irregular)”, and that “the loss or significant obfuscation of the usual amphibian pedicellate condition warrants attention as a possible synapomorphy for the Dendrobatidae.” We coded tooth structure from gross examination of cleared and stained specimens only, although histological study is required to address this problem decisively.

141. Vomerine Teeth: absent  =  0; present  =  1.

142. Retroarticular Process of Mandible (fig. 68): absent  =  0; present  =  1.

Figure 68

Length variation in character 142, retroarticular process of the mandible.

A: nocturnus, AMNH 130041. B: riveroi, AMNH 134142. C: vittatus, AMNH 118386. D: lehmanni Myers and Daly, AMNH118442. E: pratti, AMNH118364. F: “Neblina species”, AMNH 118667.


Myers and Ford (1986) noted the occurrence of a retroarticular process on the mandible as a distinguishing characteristic of dendrobatids, and Ford and Cannatella (1993) listed it as one of two unique synapomorphies. Although many dendrobatids are characterized by conspicuously elongate retroarticular processes, Myers et al. (1991: 11) noted that in nocturnus the process is “present, but always short (compared with other dendrobatids) although somewhat variable in length.” As shown in figure 68, there is considerable interspecific variation in the length of the retroarticular process. However, we were unable to delimit states, in part because no clear choice for a standard reference point has been identified.

143. Expansion of Sacral Diapophyses (fig. 69): unexpanded  =  0; moderately expanded  =  1; strongly expanded  =  2. Additive.

Figure 69

Character 143, expansion of sacral diapophyses.

A: State 0, unexpanded (riveroi, AMNH 134143l). B: State 1, moderately expanded (pumilio, AMNH 118514). C: State 2, greatly expanded (Melanophryniscus stelzneri, AMNH 77710).


The shape of the sacral diapophyses has been used since Boulenger (1882). Ford (1989) mistakenly cited Duellman and Trueb (1986) as having placed Dendrobatidae among ranoids based in part on their sharing round-shaped sacral diapophyses (Duellman and Trueb did not include that character in their matrix), but it has, nonetheless, played an important role in anuran systematics.

The state found in dendrobatids has usually been referred to as round or cylindrical (e.g., Duellman and Trueb, 1986), but the sacral diapophyses are invariably elliptical in cross section. For this reason we refer instead to the degree of expansion of the sacral diapophyses. Emerson (1982) quantified expansion by measuring the angle formed by the expansion. Here we code this character as the ratio of the width of the tip of the diapophysis to the width of the base of the diapophysis. Variation is not continuous; all states are separated by gaps. Unexpanded diapophyses are subequal at the base and tip. Moderately expanded diapophyses are 1.5–2.5 times wider at the tip than at the base; greatly expanded diapophyses are at least 2.7 times greater at the tip. Sacral diapophyses often bear irregular flanges that we did not include in the measurement of width.

144–146. Vertebral Fusion

Noble (1922: 15) reported fusion of vertebrae 2 + 3 and 8 + 9 (i.e., 8 + sacrum) in two species of dendrobatids (pumilio [as Dendrobates typographicus] and probably histrionicus or sylvaticus Funkhouser [discussed under the tentative name Dendrobates tinctorius]). Silverstone (1975a: 5) summarized his observations of vertebral fusion in dendrobatids as “absent in the 17 specimens of Colostethus examined, present in only two of the 29 specimens of Phyllobates examined, and present in 28 of the 46 specimens of Dendrobates examined.” He also noted that vertebral fusion varies intraspecifically, and we also found intraspecific variation among equivalent semaphoronts.

144. Vertebra 8 and Sacrum: free  =  0; fused  =  1.

145. Vertebrae 1 and 2: free  =  0; fused  =  1.

Myers et al (1991:11) reported ventral fusion of vertebrae 1 and 2 in nocturnus (AMNH 130047); however, close examination showed that the vertebrae are tighly bound by ligamentous tissue but are not fused.

146. Vertebrae 2 and 3: free  =  0; fused  =  1.

147–172. Alkaloid Profiles

Dendrobatid frogs are known to possess a diverse array of over 450 alkaloids (Daly et al., 1999; J. W. Daly, in litt., 01/25/05). Use of alkaloid profiles as transformation series is complicated, in part, because it appears that “some, if not all … ‘dendrobatid alkaloids’ may have a dietary origin” (Daly et al., 1994a; see also Myers and Daly, 1976b: 194–197; Myers et al., 1995), which means that the occurrence of a given alkaloid may be determined not by the genotype but by availability of the dietary source in the environment (making this a nonheritable characteristic, i.e., not a character). Saporito et al. (2003) identified a species of siphonotid millipede as the likely dietary source of spiropyrrolizidine and Saporito et al. (2004) identified certain species of formicine ants as the natural dietary source of two pumiliotoxins found in pumilio. Dumbacher et al. (2004) identified melyrid beetles as the probable dietary source of batrachotoxins for the New Guinean passerine birds Pitohui and Ifrita and further conjectured that this is the likely source of the alkaloids in Phyllobates as well. There is often considerable variation in the alkaloid profiles of conspecifics from both the same and disjunct populations (e.g., Myers et al., 1995). Captive reared offspring of wild-caught, toxic frogs are nontoxic if fed crickets and fruit flies, but readily accumulate alkaloids if present in the diet (either as a pure supplement to a fruit fly diet or in leaf-litter arthropods; Daly et al., 1992, 1994a, 1994c).

Nevertheless, despite the environmental dependency, there is also clearly a heritable aspect to the alkaloid uptake system. It has been found experimentally that azureiventris, panamensis, and talamancae do not accumulate detectable amounts of alkaloids when ingested from the diet (Daly et al., 1994c; Daly, 1998). Furthermore, among sequestering species there is differential accumulation, as suggested indirectly by the occurrence of different alkaloid profiles among microsympatric species (Daly et al., 1987; Myers et al., 1995) and demonstrated directly by feeding experiments (Daly et al., 1994c, 2003; Garraffo et al., 2001), that is, the uptake systems of different species either (1) are capable of sequestering only a subset of the alkaloids ingested in the diet or (2) vary drastically in the efficacy of accumulation of different classes of alkaloids. Either way, this variation is heritable. Furthermore, Daly et al. (2003) demonstrated selective alkaloid modification by certain dendrobatid species and not others (see Character 171). As with all phenotypic characters, the expression of alkaloid characters is due to the combination of genotype plus environment (for a detailed discussion of the meaning of “genetic” see Sarkar, 1998). Hypotheses of homology can therefore be proposed defensibly, albeit cautiously, for alkaloid profiles.

Given that it is the capability to accumulate a class of toxin that is treated as the character, we coded alkaloid profiles as “any instance” (Campbell and Frost, 1993). That is, we treated the demonstrated occurrence of a given class of alkaloid in one or more populations of a species as evidence that the entire species is capable of accumulating that class of alkaloid (i.e., it is coded as present), even if that class of alkaloid was not detected in all samples. This is not intended as a general endorsement of that method of codifying polymorphism (for theoretical arguments, see Grant and Kluge, 2003, 2004), but rather as a consequence of this particular biological problem. Given current understanding of the alkaloid uptake system of these frogs, it is most likely that the absence of a class of alkaloid in some but not all individuals is due to dietary deficiency and not a character-state transformation. This assumption is testable, and it may be found that (1) this assumption is borne out (i.e., the alkaloid is sequestered when present in the diet), (2) such species are truly polymorphic (i.e., character history and species history do not track each other perfectly, either due to ancestral polymorphism, a character-state transformation event subsequent to the most recent cladogenetic event, or some other phenomenon), or (3) multiple species have been conflated. This is exemplified by lugubris:

Only one of several populations of p[hyllobates] lugubris had barely detectable amounts of batrachotoxins. Some but not all populations had trace levels of other alkaloids. … Alkaloids including a batrachotoxin, were fed to captive-raised P. lugubris and found to be readily accumulated into skin (J. W. Daly, unpublished results). Thus, the frog has a functional accumulating “system” and the lack or near lack of alkaloids in wild-caught frogs must reflect low availability or non-targeting of alkaloid- or batrachotoxin-containing arthropods. (J. W. Daly, in litt. 02/02/00).

It is also possible that a species is capable of accumulating an alkaloid not detected in any population because the dietary source of the precursor is absent at all sampled localities (i.e., failure to detect accumulation in wild-caught specimens does not decisively demonstrate that the species is incapable of sequestration). However, by coding these taxa as lacking the ability to accumulate the toxin we incorporated all available evidence. The hypothesis that a taxon is incapable of accumulating a class of toxin is falsifiable and can be tested both by examining more specimens and populations and through feeding experiments. For example, although no histrionicotoxin could be detected in wild lehmanni Myers and Daly (Myers and Daly, 1976b), Garraffo et al. (2001: 421) report that “[f]eeding experiments indicated that D. lehmanni readily accumulated histrionicotoxin into skin when fed alkaloid-dusted fruit flies.”

Although a dietary source is either known or assumed for dendrobatid alkaloids, the actual arthropod(s) responsible have yet to be discovered for the vast majority of these, that is, most of the alkaloids are unknown elsewhere in nature. Potential sources were reviewed by Daly et al. (1993: 226, 2005). For example, pyrrolizidines are known to occur in the ants Solenopsis xenovenenum, Monomorium spp. from New Zealand, and Megalomyrmex from Venezuela. Pyrrolidines (including 2,5-pyrrolidines, known among amphibians only in dendrobatids) occur in Solenopsis, Monomorium, and Megalomyrmex. Decahydroquinolines were detected in extracts of virgin queens of the thief ant Solenopsis (Diphorhoptrum) azteca from Puerto Rico. 3,5-Disubstituted indolizidines occur in ants of the genera Monomorium and Solenopsis. Coccinellines were first discovered in the ladybug beetles Coccinellidae. Monocyclic piperidines occur in Solenopsis. Spiropyrrolizidine is likely sequestered from a millipede (Saporito et al., 2003), two pumiliotoxins found in pumilio are obtained from formicine ants (Saporito et al., 2004), and scheloribatid mites are the probable source of a third pumiliotoxin in pumilio (Takada et al., 2005). The source of batrachotoxins in dendrobatids remains unknown, although melyrid beetles are likely (Dumbacher et al., 2004).

That the actual dietary source is unknown is an important consideration, given the discovery by Daly et al. (2003) that some species convert dietary pumiliotoxin to allopumiliotoxin via a specific hydroxylation event (see Character 171, below). Whereas prior to this discovery it was assumed that all of the over 450 alkaloids known in these frogs were incorporated without modification into the skin, one must consider the possibility that some portion of this diversity of alkaloids may result from the modification of precursors. Nevertheless, it is unclear how widespread metabolic conversion may be, as the following 12 alkaloid classes have been administered in feeding experiments with no evidence for any metabolism (J. W. Daly, in litt., 01/25/05): batrachotoxin, histrionicotoxins, allopumiliotoxins, decahydroquinolines, 3,5-pyrrolizidines, 3,5-indolizidine, 5,8-indolizidine, 5,6,8-indolizidine, pyrrolidine, piperidine, spiropyrrolizidine, and coccinelline-like tricyclics.

Given the dietary origin of the alkaloids and how little is known about the alkaloid uptake system, we were conservative in delimiting alkaloid characters for phylogenetic analysis. Instead of coding the occurrence of each of the over 450 dendrobatid alkaloids as a separate character, we scored the occurrence of the major and minor classes of alkaloids, following Daly et al. (1987, 1993) and incorporating more recent developments (e.g., Daly et al., 1994c, 2003, 2005; Daly, 1998; Garraffo et al., 1993, 1997, 2001; Daly et al., 1999; Mortari et al., 2004; J. W. Daly, in litt., 01/25/05). We followed Myers (1987) and Myers et al. (1995) in coding 3,5-indolizidines and 5,8-methylindolizidines as distinct characters. In only coding the occurrence of general classes of alkaloids, we consciously overlooked more refined, potentially phylogenetically informative data in an attempt to avoid introducing error due to the nature of alkaloid accumulation in these frogs. Furthermore, in the majority of species, numerous alkaloids of the same class co-occur, which suggests that sequestration acts at the level of the class of alkaloid, not individual alkaloids; that is, it appears that it is the ability to sequester alkaloids with certain chemical properties that evolves, not the ability to sequester a particular alkaloid.

We did not distinguish between major, minor, and trace occurrences of alkaloids (i.e., we treated all as “present”), but, following Daly's recommendation (J. W. Daly, in litt., 02/02/00), we did not consider “trace, trace” occurrences as evidence of presence of an alkaloid, as merely having recently consumed an alkaloid-containing prey item could give this result.11 We also did not discriminate based on uptake efficiency. For example, although uptake of piperidines is poor in most species (e.g., auratus, in which they are trace alkaloids), and uptake of piperidine 241D appears highly efficient in speciosus (in which this is a major or minor alkaloid), we coded piperidines identically (i.e., present). It should be clarified that, despite the fact that the trivial names of the classes of alkaloids are often derived from species that possess it (e.g., pumiliotoxin from pumilio), compounds are assigned to a class based on molecular structure and chemical properties, not taxonomic distribution.

It has been speculated that certain alkaloids could share common precursors, specifically a 2,6-disubstituted(dehydro)piperidine as a precursor in the biosynthesis of histrionicotoxins, gephyrotoxins, indolizidines, and decahydroquinolines (Daly et al., 1987: 1065), and more generally that the monocyclic piperidines are possible precursors for the more complex, piperidine-ring-containing alkaloids and the monocyclic pyrrolidines for the more complex, pyrrolidine-ring-containing dendrobatid alkaloids (Daly et al., 1993: 251). Nevertheless, with the exception of allopumiliotoxin 267A (see Character 171), there is no evidence that they share a common biosynthetic origin, and even if they do, this would pertain to the arthropods, not the frogs' uptake system. Historical independence is demonstrated by the fact that no classes of alkaloids share identical taxonomic distributions.

We coded unambiguously only those species whose alkaloid profiles have been examined; taxa whose profiles have not been examined were coded as unknown (“?”). The discovery that an untested species is embedded within a toxic clade would provide a strong prediction that the species may also sequester alkaloids and would therefore guide chemists in their search for novel, potentially useful toxins. Negative findings often are not explicitly reported in the literature; however, in cases where species have been examined using techniques that would detect a particular compound and the compound was not reported, we coded it as absent (e.g., epibatidine). If there was any doubt as to the tests samples were subjected to, we coded the character as unknown (“?”).

Data were taken from reviews (Daly et al., 1987, 1993, 1999), the primary literature (Tokuyama et al., 1992; Garraffo et al., 1993, 2001; Badio and Daly, 1994; Myers et al., 1995; Daly et al., 1997., 2003; Fitch et al., 2003; Saporito et al., 2003; Mortari et al., 2004; Darst et al., 2005), and an exhaustive summary of published and unpublished alkaloid profiles and corrections to previous accounts provided by John W. Daly (in litt., 01/25/05). To facilitate coding from the literature we list the individual nonsteroidal alkaloids for each class, following the convention of Daly et al. (1987). We did not include unclassified alkaloids, although they may provide relevant information once their structures are elucidated. We did not list unpublished alkaloids in the character descriptions (although we did code their presence in the matrix), and we only listed alkaloids that occur in the species sampled in the present study.

147. Ability to Sequester Alkaloids: absent  =  0; present  =  1.

We coded the general ability to sequester alkaloids separately from the individual classes of alkaloids sequestered in order to count the gain and loss as a single transformation event. We scored species that are incapable of sequestering any alkaloid as state 0 for this character and missing (“–“) for all other lipophilic alkaloid characters; we coded species that are able to sequester any alkaloid as state 1 for this character and present and absent for each of the particular alkaloid classes. That is, although the origin of the ability to sequester alkaloids necessarily entails the ability to sequester some particular class(es) of alkaloid(s) (i.e., there is a logical relation of nested dependency between these characters), the fact that no taxon possesses only a single class of alkaloid would mean that the alternative approach of treating each origin and loss as entirely unrelated events would count the origin of sequestration as multiple events. The biological assumption underlying this coding is that there exists a single genetic basis for the sequestration of all classes of lipophilic alkaloids and that modifications to it account for the differential ability to sequester distinct classes. This assumption is consistent with the limited understanding of the uptake mechanism but has not been subjected to critical test (i.e., no attempt has been made to isolate the genetic basis of sequestration).

Darst et al. (2005) reported the occurrence of alkaloids based on thin layer chromatography, which allowed us to score several additional species. Unfortunately, that method does not discriminate between or lead to the identification of different alkaloids, so this is the only character that can be coded from their results.

148. Batrachotoxins (BTX): absent  =  0; present  =  1.

The steroidal batrachotoxins are known to occur in only five species of frogs (aurotaenia, bicolor, lugubris, terribilis, and vittatus), and their shared occurrence was treated as evidence of the monophyly of those species in a restricted Phyllobates (Myers et al., 1978). Given the extreme toxicity of BTX relative to other dendrobatid alkaloids, the ability of these frogs (and the inability of other dendrobatids) to sequester BTX is likely related to their modified sodium channel that is insensitive to BTX (as demonstrated for aurotaenia and terribilis; Daly et al., 1980).

149. Histrionicotoxins (HTX): absent  =  0; present  =  1.

235A, 237F, 239H, 259A, 261A, 263C, 265E, 283A, 285A, 285B, 285C, 285E, 287A, 287B, 287D, 291a.

Alkaloid 283A′ (found in sylvaticus Funkhouser) is closely related to and was treated as an HTX by Daly et al. (1987), but it was not included by Daly et al. (1993) or Daly et al. (2005).

150. Pumiliotoxin (PTX): absent  =  0; present  =  1.

207B, 209F, 225F, 237A, 251D, 253F, 265D, 265G 267C, 267D, 277B, 281A, 293E, 297B, 305B, 307A, 307B, 307D, 307F 307G, 307H, 309A, 309C, 321A, 323A, 325B, 353A.

151. Allopumiliotoxins (aPTX): absent  =  0; present  =  1.

225E, 237B, 241H, 251I, 253A, 267A, 297A, 305A, 307C, 309D, 321C, 323B, 325A, 339A, 339B, 341A, 341B, 357.

152. Homopumiliotoxins (hPTX): absent  =  0; present  =  1.

223G, 249F, 251L, 256R, 265N, 317, 319A, 319B, 321B.

153. Decahydroquinoline (DHQ): absent  =  0; present  =  1.

193D, 195A, 209A, 209J, 211A, 211K, 219A, 219C, 219D, 221C, 221D, 223F, 223Q, 223S, 231E, 243A, 245E, 249D, 249E, 251A, 253D, 267L, 269AB, 269A, 269B, 271D, 275B.

154. 3,5-Disubstituted Pyrrolizidines: absent  =  0; present  =  1.

167F, 195F, 209K, 223B, 223H, 237G, 251K, 253I, 265H, 265J, 267H.

167F and 209K were formerly classified as the 3,5-disubstituted indolizidines 167B and 209D.

155. 3,5-Disubstituted Indolizidines: absent  =  0; present  =  1.

195B, 211E, 223AB, 223R, 237E, 239AB, 239CD, 239E, 249A, 271F, 275C, 275F.

156. 5,8-Disubstituted Indolizidines: absent  =  0; present  =  1.

181B, 193E, 197C, 203A, 205A, 207A, 209B, 209I, 217B, 219F, 221A, 221K, 223D, 223J, 225D, 231C, 233D, 235B, 237D, 237H, 239A, 239B, 239C, 239D, 239F, 239G, 241C, 241F, 243B, 243C, 243D, 245B, 245C, 245D, 251B, 251U, 253B, 263F, 257C, 259B, 261D, 271A, 273B, 279D, 295A, 295B.

157. Dehydro-5,8-indolizidines: absent  =  0; present  =  1.

245F, 245H.

158. 5,6,8-Indolizidines: absent  =  0; present  =  1.

195G, 207Q, 223A, 231B, 233G, 237L, 249H, 251M, 253H, 259C, 263A, 263D, 265I, 265L, 267J, 273A, 275E, 277C, 277E, 279F, 293C.

159: 4,6-Quinolizidines: absent  = 0; present  =  1.

195C, 237I.

160. 1,4-Quinolizidines: absent  =  0; present  =  1.

207I, 217A, 231A, 233A, 235E', 247D, 257D.

161. Lehmizidines: absent  =  0; present  =  1.


162. Epiquinamide: absent  =  0; present  =  1.


163. 2,5-Pyrrolidine (PYR): present  =  0; absent  =  1.

183B, 197B, 223N, 225C, 225H, 277D, 279G.

164. 2,6-Piperidines (PIP): absent  =  0; present  =  1.

197E, 211I, 211J, 213, 221L, 223K, 225B, 225I, 237J, 239I, 239L, 239O, 241D, 241G, 253J, 255A, 267K, 267C.

165. Gephyrotoxin (GTX): absent  =  0; present  =  1.

287C, 289B.

166. Coccinelline-like Tricyclics: absent  =  0; present  =  1.

191B, 193A, 193C, 201B, 205B, 205E, 207J, 207P, 207R, 219I, 221G, 221M, 235M, 235P.

167. Spiropyrrolizidines: absent  =  0; present  =  1.

Referred to as pyrrolizidine oximes by Daly et al. (1993).

222, 236, 252A, 254.

168. Indolic Alkaloids (Chimonanthine/Calycanthine): absent  =  0; present  =  1.

346B, 346C.

169. Epibatidines: absent  =  0; present  =  1.

208/210, 308/310.

170. Noranabasamine ( = Pyridyl-Piperidines): absent  =  0; present  =  1.

This pyridine alkaloid is known in nature only from aurotaenia, bicolor, and terribilis (Daly et al., 1993, 1999).


171. Pumiliotoxin 7-Hydroxylase: absent  =  0; present  =  1.

Feeding experiments by Daly et al. (2003) demonstrated the existence in several species of dendrobatids of an enantioselective mechanism that converts PTX (+)-251D to the more highly toxic allopumiliotoxin (aPTX) (+)-267A. That is, contrary to other alkaloid characters, which code the ability to sequester a class of alkaloid, this character applies to the occurrence of the 7-hydroxylase, as evidenced by the occurrence of the hydroxylated compound.

Coding this character is somewhat more complicated than coding the other alkaloid characters, because in this case occurrence of aPTX 267A may be due to either (1) the hydroxylation of PTX 251D or (2) the sequestration of aPTX 267A from a dietary source (aPTX is known to occur in some arthropods). This creates the potential for both false negatives and false positives. Direct evidence for the occurrence of 7-hydroxylase may only be obtained through feeding experiments. Further evidence on the distribution of the pumiliotoxin 7-hydroxylase obtained indirectly from the alkaloid profiles of wild-caught specimens (see Daly et al., 2003: 11095, table 1) requires the assumption that all aPTX 267A occurs through metabolism of ingested PTX 251D, which, at least in the case of anthonyi (reported as tricolor; for taxonomy of these species, see Graham et al., 2004), is false (assuming multiple species have not been conflated). Daly et al. (2003) reported wild-caught specimens as possessing trace amounts of aPTX 267A, but feeding experiments revealed that the species is incapable of hydroxylating PTX 251D and the occurrence of aPTX 267A represents a false positive for the presence of 7-hydroxylase. Nevertheless, in the absence of direct evidence from feeding experiments, such as is available for anthonyi, we coded all trace, minor, and major occurrences of aPTX 267A as the presence of the 7-hydroxylase, which allows the results of phylogenetic analysis to serve as a tool for designing future feeding experiments to test hypothesized occurrence of 7-hydroxylase (e.g., finding that a species that possesses aPTX 267A is embedded in a clade of species incapable of 7-hydroxylation would suggest the occurrence may be due to sequestration from a dietary source and not biosynthetic conversion).

Conversely, the absence of 7-hydroxylase can only be assured in the presence of PTX 251D. We coded the failure to detect aPTX 267A as “absent” (state 0) only when PTX 251D was detected. If PTX 251D was not detected (but other PTXs were), we coded this character as unknown (“?”) (e.g., truncatus). If available evidence indicates that a species is incapable of sequestering pumiliotoxins, we coded this character as missing (“–”) (e.g., trivittatus).

Direct evidence for the presence of the pumiliotoxin 7-hydroxylase through feeding experiments was found for auratus, galactonotus, and castaneoticus, and direct evidence for the absence of pumiliotoxin 7-hydroxylase through feeding experiments was found in tricolor and bicolor (Daly et al., 2003). Other species are coded on the basis of wild-caught specimens, with data derived from Daly et al. (1987, 1993, 2003).

172. Tetrodotoxin (TTX): absent  =  0; present  =  1.

Daly et al. (1994b) reported the occurrence of TTX in panamensis (as Colostethus inguinalis; see Grant, 2004). They also examined aqueous extracts of eight additional species referred to Colostethus (the “Colostethus species” reported as being “common, nr Villa María, Caldas, Colombia” is fraterdanieli), and nocturnus, pumilio, and bicolor. Daly et al. (1994b: 283) cautioned that the negative results for pumilio and bicolor were based on methanol extracts,

which would have extracted only minimal amounts of tetrodotoxin. … Thus, very low levels of tetrodotoxin-like compounds … might have escaped detection because of the low efficiency of methonol in extracting such compounds. But levels approaching those reported for C. inguinalis [ =  panamensis] … would have been detected even in methanol extracts.

173. Chromosome Number: 18  =  0; 20  =  1; 22  =  2; 24  =  3; 26  =  4; 28  =  5; 30  =  5. Additive.

Karyological data have been reported for 35 of the dendrobatids included in the present study: panamensis and pumilio (Duellman, 1967), auratus and pumilio (León, 1970), trivittatus (Bogart, 1970, 1973, 1991), trinitatis (Rada de Martínez, 1976), auratus, granuliferus, histrionicus, lugubris, pumilio, and sylvaticus Funkhouser (as histrionicus from NW Ecuador) (Rasotto et al., 1987), conspicuus [as brunneus], femoralis, fraterdanieli, olfersioides, palmatus, pictus, subpunctatus, talamancae, trivittatus, truncatus, vanzolinii [as quinquevittatus], vertebralis, and an undescribed species referred to Colostethus (Bogart, 1991), caeruleodactylus, marchesianus (sensu stricto; see Caldwell et al., 2002b) and two undescribed species referred to Colostethus (Veiga-Menoncello et al., 2003a), nidicola and stepheni (Veiga-Menoncello et al., 2003b), chalcopis, leopardalis, herminae, neblina, olmonae, and trinitatis (Kaiser et al., 2003), flavopictus, femoralis, hahneli, and trivittatus (Aguiar et al., 2002). Thirty of those species are included in the present study.

For outgroup taxa, data were taken from Kuramoto's (1990) review. Data published subsequently were taken from Silva et al. (2001) for Cycloramphus boraceiensis, Rosa et al. (2003) for Megaelosia, Ramos et al. (2002) for Atelopus zeteki, and Aguiar et al. (2004) for Crossodactylus and Hylodes phyllodes.

Coding chromosome variation as transformation series is complicated by imprecision in determining chromosome homology. For the most part, chromosomes are simply arranged according to size and named (numbered) consecutively. That all variation in chromosome morphology is reported in relation to chromosome identity (which is a function of relative chromosome size) is a serious problem. Rarely, more detailed considerations are brought to bear (e.g., see Bogart, 1991, regarding the homology of chromosome 4 in pictus and chromosome 5 in trivittatus), but this is done so infrequently as to be of little use in the present study. A further limitation of available karyological data relates to the variation in techniques and kinds of data reported. For example, nucleolar organizing regions (NORs) are reported for only 11 of the dendrobatids included in this study, and in just those few species at least six NOR states are apparent. Likewise, in light of the confounding variation he observed, Bogart (1991: 245) cautioned that “[i]t is evident that analysis of chromosome arms would be of little value for understanding karyotype evolution in the family Dendrobatidae. It is also evident that dendrobatid chromosomes have undergone extensive restructuring via translocations and inversions.”

There are undoubtedly many additional transformation series in chromosome morphology, but we coded only chromosome number because (1) it is reported in all karyological studies, (2) it is less dependent on individual chromosome identity (see references above), and (3) it has been employed previously in studies of dendrobatid systematics. Nevertheless, inferring transformation series solely from chromosome number necessarily assumes that the same chromosome(s) are gained or lost in each change in total number of chromosomes, which future research will undoubtedly determine to have been an oversimplification.


General Results

Direct optimization parsimony analysis followed by the parsimony ratchet analysis of the implied alignment resulted in 37 equally parsimonious solutions of 46,520 steps. Forty nodes collapse in the strict consensus of these optimal topologies, all of which involve conspecific terminals only. Further swapping those 37 trees using the implied alignment recovered a total of 25,872 trees, with no additional nodes collapsed in the strict consensus. The CI (Kluge and Farris, 1969) and RI (Farris, 1989) for the phenotypic characters12 on the total evidence solutions are 0.14 and 0.76, respectively. We begin by summarizing higher-level relationships, shown in figure 70, and proceed to the relationships among dendrobatids in figures 71Figure 72Figure 73Figure 74Figure 7576. Rather than describe the cladogram and associated support values exhaustively, we emphasize information not depicted on the cladogram, especially the unambiguous transformations that delimit clades and the bearing of the current results on species-level problems. Only unambiguous synapomorphies shared by all most parsimonious trees are reported.

Figure 70

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: outgroup relationships and monophyly and placement of dendrobatids.

Numbers above branches are Bremer support values. Family group names applied as in Frost et al. (2006). Cycloramphinae and Hylodinae were nested within Cycloramphidae in Frost et al.'s study. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 71

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Unidentified species taken from GenBank are labeled as originally published. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 72

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Unidentified species taken from GenBank are labeled as originally published. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 73

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without sample numbers were taken from GenBank. Note that Morales (Morales, 2002 “2000”) identified “PortoWalter2” 356 as gasconi, and “Curuá-Una” 354 as brunneus (see fig. 72). Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 74

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Unidentified species taken from GenBank are labeled as originally published. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 75

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Unidentified species taken from GenBank are labeled as originally published. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Figure 76

Strict consensus of 25,872 most parsimonious trees of 46,520 steps: relationships among dendrobatids.

Numbers above branches are Bremer support values. Numbers following terminal names are unique sample identifiers. Terminals without numbers or with alphanumeric identifiers (GenBank numbers) were not sequenced for the present study or Frost et al. (2006) and were taken from GenBank. Unidentified species taken from GenBank are labeled as originally published. Upper right inset shows entire cladogram and corresponding figure numbers, with present view in black.


Dendrobatid Monophyly and Outgroup Relationships

Dendrobatid monophyly was supported strongly in the present analysis (fig. 70). Unambiguous phenotypic transformations include the loss of supernumerary tubercles on the hand (Character 2, 1 → 0), gain of the tarsal keel (Character 28, 0 → 1), the “ranid” type insertion of the distal tendon of the m. semitendinosus (Character 69, 0 → 1), gain of the m. semitendinosus binding tendon (Character 70, 0 → 1), occurrence of the dorsal flap of the m. depressor mandibulae (Character 72, 0 → 1), relation of the tympanum and m. depressor mandibulae (Character 75, 0 → 1), orientation of the m. intermandibularis supplementary element (Character 78, 0 → 1), maxillary tooth structure (Character 140, 0 → 1), the occurrence of the retroarticular process of the mandible (Character 142, 0 → 1), and the reduction in chromosome number from 26 to 24 (Character 175, 4 → 3). Behavioral synapomorphies include the loss of reproductive amplexus (Character 104, 1 → 0), the gain of dorsal tadpole transport (108, 0 → 1), and the origin of toe trembling (Character 115, 0 → 1).

Our results generally resemble those of Frost et al. (2006) regarding the phylogenetic position of dendrobatids and the relationships among outgroup taxa, but with a few significant exceptions. Of greatest relevance to the problem of dendrobatid relationships, the current study refuted Frost et al.'s (2006) placement of Thoropa and dendrobatids as sister groups and instead placed Thoropa inside Cycloramphidae, with Hylodinae recovered as the sister group of dendrobatids (as first suggested by Noble, 1926). In addition to the genotypic transformations that optimize unambiguously to this node, phenotypic transformations include the origin of digital scutes (Character 1, 0 → 1) and the formation of digital discs (Character 6, 0 → 1), the origin of T-shaped terminal phalanges (Character 118, 1 → 0), and the occurrence of an oblique lateral stripe (Character 55, 0 → 1). Except for the removal of hylodines and insertion of Thoropa, the relationships among cycloramphines are identical to those of Frost et al. (2006). As was found by Frost et al., the next more inclusive clade includes Bufonidae, and then Cycloramphinae.

The greatest difference between Frost et al.'s (2006) results and the present hypothesis involves the placement of leptodactylids. The clades here labeled Leptodactylidae 1 and Leptodactylidae 2 were a monophyletic group in Frost et al.'s study, and that clade was sister to Centrolenidae, together forming the Amnibatrachia. Here, centrolenids are the sister of all included taxa except hylids, Leptodactylidae 1 is sister to all but the centrolenids and hylids, and Leptodactylidae 2 is sister to Bufonidae + Hylodinae + Dendrobatidae, that is, it is separated from Leptodactylidae 1 by ceratophryids and cycloramphines.

Relationships among Dendrobatids

The eastern Colombian species palmatus is sister to a clade that includes all species that possess the median lingual process (MLP) (fig. 71). Eight unambiguous phenotypic transformations unite palmatus with the MLP clade, including the origins of fringes on the preaxial edges of fingers II and III (Characters 13 and 15, 0 → 1).

Within the MLP clade, tepuyensis and the undescribed species “Thomasing” are relatively robust frogs with extensive webbing. Their monophyly is strongly supported (Bremer support  =  41), although only a single phenotypic synapomorphy optimizes unambiguously to this node (expansion of toe disc I, Character 31, 1 → 2). Percent pairwise distances are shown in table 4.

Table 4

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of tepuyensis and “Thomasing”ai0003-0090-299-1-1-t04.gif

The identification of sample 606 as tepuyensis will likely require revision. That species was described by La Marca (1998 “1996”) from Auyántepui, whereas sample 606 was taken over 200 km away on Mt. Ayanganna (ca. 60 km WNW of Kaieteur, Guyana). Given the high degree of endemism of many tepui species, it is doubtful that these samples are conspecific. Nevertheless, we compared the voucher specimen of the tissue sample (ROM 39637, the only specimen of this species collected at this locality) to a series of 33 specimens of tepuyensis from the type locality and failed to detect diagnostic differences. Our prediction is that additional specimens and/or molecular data will reveal that these are different species, but for the present we apply the name tepuyensis to specimens from both localities. “Thomasing” is an undescribed species from Mt. Thomasing, Mazaruni-Potaro, Guyana.

The monophyly of the remaining species of the MLP clade is supported by three unambiguous phenotypic transformations: the gain of the pale paracloacal marks (Character 49, 0 → 1), completion of the oblique lateral stripe (Character 56, 0 → 1), and the origin of an endotrophic larval diet (Character 112, 0 → 3).

Among the species of this clade are several that resemble, superficially at least, degranvillei. Species delimitation is hindered by extensive morphological variation within syntopic series, making this a prime example of the relevance of DNA sequence data in discovering cryptic diversity. Samples 278, 279, and 1336 were all collected in Guyana (details below). Although we did not detect morphological differences, our results indicate that they are not conspecific with degranvillei sensu stricto. The degranvillei data obtained from GenBank were generated by Vences et al. (2003), who stated that their sample of degranvillei was from Saül, French Guiana, which is quite close to the type locality and, therefore, likely to represent degranvillei sensu stricto. We therefore refer to the Guyanan species as “degranvillei.” Cytochrome b sequences for the Vences et al. specimen were not available, but the pairwise distance between “Tafelberg” and the “degranvillei” is 17.5%.

The two samples of praderioi were collected at 1,310 m on Roraima, Guyana. Sample 1336 of the Guyanan “degranvillei” was also collected on Roraima but was taken at 1,075 m. The two remaining Guyanan “degranvillei”samples were taken in the Merume mountains, and “Ayanganna” was collected on Mt. Ayanganna, ca. 50 km WNW of Kaieteur, Guyana.

Despite the morphological similarity and geographic proximity of praderioi and “degranvillei” on Roraima, and only <300 m difference in elevation between the localities, the pairwise distance is 10.1–10.4% (see table 5). The pairwise distance between the three “degranvillei” samples is only 1.6–1.8%, despite the much greater geographic distance.

Table 5

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of “degranvillei” from Guyana, praderioi, and the Undescribed Species “Ayanganna”ai0003-0090-299-1-1-t05.gif

According to the topology alone, it is possible that “Ayanganna” and praderioi may be conspecific. Nevertheless, they differ morphologically (e.g., webbing) and at 8.3% of their cytochrome b sites, leaving little doubt that they are different species.

The sister species beebei and roraima are diminutive, geographically proximate species that both possess the median lingual process and breed in phytotelmata (for breeding behavior in beebei, see Bourne et al., 2001). Pairwise distances are shown for beebei and roraima in table 6. There is no confusion surrounding the identity of beebei, with the exception that the French Guianan species discussed under that name (e.g., Kok, 2000; Lescure and Marty, 2000) is not conspecific with beebei sensu stricto from Guyana (among other differences, the French Guianan species lacks the median lingual process).

Table 6

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of beebei and roraimaai0003-0090-299-1-1-t06.gif

La Marca (1998 “1996”) described roraima based on a single immature specimen from 2,700 m near the peak of Mt. Roraima. Although there are several inconsistencies in La Marca's description and illustrations, and the immaturity of the holotype impedes identification, the material included in the present study was collected at the type locality and agrees with the description sufficiently to conclude that it is roraima. Samples 1337 and 1338 were taken from adults CPI 10216 and CPI 10217. Sample 1339 is from an untagged tadpole collected in a bromeliad, which establishes conclusively adult and larval conspecificity.

The clade composed of baeobatrachus, degranvillei, stepheni, “Tafelberg”, and “Brownsberg” has a Bremer value of 24. Five unambiguous phenotypic changes occur at this node, including the loss of the distal subarticular tubercle on finger IV (Character 3, 1 → 0).

The nomenclatural history of baeobatrachus and stepheni is convoluted. The name “baeobatrachus” first appeared in Edwards's widely distributed but formally unpublished Ph.D. dissertation (Edwards, 1974a). The type locality Edwards intended to designate was Ducke Reserve in Amazonas State, just outside Manaus (Brazil). The two samples included here (514, 515) are from that locality. On the 15th anniversary of the completion of Edwards's dissertation, Martins (1989) described stepheni with the explicit intent of providing a name for Edwards's “baeobatrachus”. The type locality of stepheni is Presidente Figueiredo, also in Amazonas State and approximately 100 km from Manaus. Apparently unaware of this development or the fact that Edwards's “baeobatrachus” was not an available name, and despite having cited a paper that deals with the reproductive biology of stepheni (viz., Juncá et al., 1994), in a popular article Boistel and Massary (1999) presented a color photograph and brief but validating diagnosis under the name Colostethus baeobatrachus. Boistel and Massary did not specify a type locality or voucher specimen, but Kok (2000) provided a complete redescription based on material from Montagne Belvédère in French Guiana and deposited at IRSNB. The four samples included here (14, 42, 43, 44) were taken from that series. Immediately thereafter, Kok (2001) determined that baeobatrachus and stepheni were indistinguishable and placed them in synonymy. Published sonograms of stepheni at Reserva Ducke (Juncá, 1998) and baeobatrachus in French Guiana (Lescure and Marty, 2000) are very similar, the sole potentially relevant difference occurring in the dominant frequencies: in stepheni it is given as 4.6–4.8 kHz and in baeobatrachus 5.12–5.83 kHz. Sample sizes are very small though, and such minor differences are commonly observed within species.

Nevertheless, the ca. 17% pairwise distance between the Reserva Ducke and Montagne Belvédère samples strongly suggests they are not conspecific (see table 7). Furthermore, tadpoles of stepheni are nidicolous with reduced mouth parts and a median vent tube (Juncá et al., 1994; Juncá, 1998), whereas a male nurse frog was collected at Serra do Navio, Amapá, Brazil transporting three tadpoles with fully developed mouth parts and dextral vent tube.13 Assuming that the Montagne Belvédère and Serra do Navio samples are conspecific, there is strong evidence that these baeobatrachus and stepheni are not conspecific, despite the apparent lack of diagnostic characters in adult morphology. That baeobatrachus and stepheni are valid species is further supported by phylogenetic analysis, which places the undescribed species “Brownsberg” (1328), from Guyana, as sister to baeobatrachus to the exclusion of stepheni.

Table 7

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of stepheni, baeobatrachus, and the Undescribed Species “Tafelberg” and “Brownsberg”ai0003-0090-299-1-1-t07.gif

The remaining species, “Tafelberg” (sample 1326) is another undescribed species from Guyana that is closely related to (and potentially conspecific with) the GenBank degranvillei (see above for comments on the identity of this sample). Cytochrome b sequences were unavailable for the GenBank degranvillei sample and morphological comparisons were not made, but the number of unambiguous transformations in 12S and 16S sequences that occur on the terminal branches (13 for “Tafelberg,” 14 for degranvillei) suggests they are not conspecific.

The other clade shown in figure 71 is composed mainly of species currently referred to Aromobates, Mannophryne, and Nephelobates. The monophyly of this clade is strongly supported (Bremer support  =  38) and is delimited by 54 unambiguous genotypic changes, although there are no unambiguous phenotypic transformations at this node. Following the current taxonomy, Aromobates nocturnus and Colostethus saltuensis are nested within Nephelobates. The latter species was included in the alboguttatus group of Rivero (1990 “1988”), but was excluded without comment when La Marca (1992) named his own version of the alboguttatus group formally as Nephelobates. Likewise, the affinities of nocturnus and the species of both Nephelobates and Mannophryne were noted when Aromobates was described (referring to those as yet unnamed genera as the alboguttatus and collaris groups, respectively; Myers et al., 1991), and Kaiser et al. (1994), Meinhardt and Parmelee (1996), and Grant et al. (1997) questioned the monophyly of those genera relative to Aromobates. More specifically, prior to naming Nephelobates, La Marca (1993) considered nocturnus to be most closely related to his alboguttatus group, which is corroborated in this study.

Although Nephelobates is paraphyletic, the monophyly of the equally controversial Mannophryne is solidly corroborated in this analysis. This clade has a Bremer value of 40 and may be diagnosed morphologically by the synapomorphic dermal collar, which optimizes unambiguously to this node (Character 59, 0 → 1). The conclusions, based on morphological criteria, that the collarlike gular–chest markings of several Ecuadorian species (e.g., elachyhistus) are not homologous with the dermal collar of these Venezuelan species and that the diffuse collar of nocturnus is due to nonhomologous subdermal pigmentation (see Characters 58 and 59) are supported by the distant relationships of these taxa in the optimal cladograms.

Among the nominal species included in the cladogram, the herminae samples are not conspecific; at this time we are unable to speculate as to which of these (if either) corresponds to herminae sensu stricto. The cytochrome b sequences for the two samples of nocturnus are identical.

The clade shown in figure 72 is a large, primarily cis-Andean (east of the Andes) group. Ten unambiguous phenotypic transformations delmit this clade, including the origin of swelling of finger III in adult males (Character 20, 0 → 1) and the loss of palatines (Character 132, 1 → 0).

The sister of the remainder of this clade is alagoanus, from the Atlantic forest of Brazil, followed by the undescribed “Neblina species” and undulatus. “Neblina species” was collected at the base of the tepui Neblina, in Venezuela; the cytochrome b sequences of the two specimens are identical. Myers and Donnelly (2001) described undulatus from the Yutajé massif, also in Venezuela. The three samples were all collected in the same vicinity (see table 8 for pairwise distances).

Table 8

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of Syntopic Specimens of undulatusi0003-0090-299-1-1-t08.gif

Among the species included in the present analysis, the only trans-Andean (west of the Andes) species in this clade are the sister species talamancae and the undescribed “Magdalena”. The affinities of talamancae have never been clear (e.g., Rivero, 1990 “1988” was unable to assign it to any of his groups), probably because it differs considerably from the trans-Andean species with which it was compared. However, the placement of these species among these cis-Andean species is strongly supported and highlights the overall resemblance of these species (e.g., for photographs of talamancae and kingsburyi, see Coloma, 1995). Moreover, the discovery of the undescribed “Magdalena” species fills in the gap in the distribution between talamancae and the remaining species.

“Magdalena” and talamancae share the unambiguous transformation from an evenly stippled to solid dark throat in adult males (Character 61, 2 → 4). These two species are allopatric. “Magdalena” is known only from sites on the floor of the middle Magdalena river valley in Colombia, and talamancae is widespread from the Pacific lowlands of South America in Ecuador and Colombia north to Nicaragua. Pairwise distances for cytochrome b sequences of these species are shown in table 9. The talamancae samples are from two localities in Panama (Bocas del Toro: 325, 326; Coclé: 1147) and one in Nicaragua (361, 362, 408).

Table 9

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of talamancae and “Magdalena”ai0003-0090-299-1-1-t09.gif

Although the most parsimonious cladogram recovers monophyletic Panamanian and Nicaraguan samples of talamancae, the genetic distances between the samples are consistent with the hypothesis of continuous gene flow. The greatest pairwise distance is between the samples from Nicaragua and Coclé, with the intermediate sample from Bocas del Toro also intermediate genetically.

Much of the diversity of small, brown, relatively nondescript cis-Andean dendrobatids has been associated with the old (see appendix 1) names brunneus, marchesianus, and trilineatus. Progress in documenting the diversity of Amazonian dendrobatids has been hindered by confusion surrounding these nominal species. Grant and Rodríguez (2001) clarified the identity of the western Amazonian trilineatus, and Caldwell et al. (2002b) redescribed marchesianus based on extensive new material and vocalizations from the type locality (in the vicinity of the Rio Uaupes in Amazonian Brazil) and clarified that all populations referred to that species from elsewhere (e.g., Santa Cecilia, Ecuador) were heterospecific (T. Grant has also examined material of this species from the adjacent region of Colombia). In the same year, Morales (2002 “2000”) provided an account for marchesianus based on examination of a syntype and specimens from other localities, but his redescription is incomplete (e.g., it does not address intraspecific variation or make comparisons with other species) and disagrees in several key points with that of Caldwell et al. (2002b), as well as the Colombian material examined by T. Grant, and the account is therefore rejected. We included DNA sequences for numerous specimens referred to trilineatus by Grant and Rodríguez (2001), as well as material from the same or nearby localities, but we were unable to include sequences for marchesianus sensu stricto.

The remaining taxonomic problem involving an old name is brunneus. Grant and Rodríguez (2001) provided data for topotypic and other material, but they did not attempt to address decisively the problem of brunneus identity. La Marca et al. (2004) improved matters considerably by clarifying that the “brunneus” from northern Venezuela was in fact a new species (which they named Colostethus pittieri) most closely related to humilis. In what appears on the surface to be the most thorough study of the systematics of these frogs, Morales (2002 “2000”) provided an account for brunneus. Like the remainder of the accounts in that paper—including those for the 11 new species named therein—the account of brunneus does not address variation within brunneus or compare that species to others and is therefore highly unsatisfactory. Nevertheless, Morales's account of brunneus is the most recent attempt to clarify its identity, and we therefore apply the name in his sense. We included DNA sequences from several of the specimens examined by Morales and referred by him to several species, including brunneus and his new species conspicuus and gasconi.

Although we apply the name brunneus in the sense of Morales (2002 “2000”), and samples 352 and 1278 were both referred to that species by him, Morales also referred sample 354 of a distantly related species from Curuá-Una (shown in fig. 73) to brunneus. The minimum pairwise distance between that sample and either of the others he referred to brunneus is 16.6%. We therefore exclude that sample from the pairwise comparisons in table 10 and instead include it with the other samples from Curuá-Una (see below). The pairwise cytochrome b distances between brunneus and its sister species from Parque Estadual Guajará-Mirim (“PEG-M2”) are 14.3–15.3%.

Table 10

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of brunneusi0003-0090-299-1-1-t10.gif

Terminals identified as “PEG-M2” represent one of three undescribed species of dendrobatids collected at Parque Estadual Guajará-Mirim, Rondônia, Brazil (see table 11) The pairwise distances between the samples of this species and brunneus are 14.3–15.3%.

Table 11

Uncorrected Pairwise Distances Between Cytochrome b Sequences of “PEG-M2”i0003-0090-299-1-1-t11.gif

The next clade includes juanii, from Villavicencio, Colombia, zaparo, from eastern Ecuador, and the widespread femoralis. The monophyly of zaparo and femoralis is strongly supported (BS  =  147), and they are united by 161 unambiguous transformations.

The occurrence of zaparo and femoralis in this clade of otherwise cryptically colored, nontoxic frogs conflicts strongly with the traditional placement of these species with bright, toxic species such as petersi and pictus (e.g., Silverstone, 1976). Nevertheless, the distant placement of these species found by previous studies (e.g, Santos et al., 2003; Vences et al., 2003a) could not be refuted by the inclusion of phenotypic and additional DNA evidence. Furthermore, both femoralis and zaparo appear to be incapable of accumulating alkaloids (Darst et al., 2005; J. W. Daly in litt., 02/02/00), which suggests that the remarkable resemblance of these nontoxic and toxic species may be due to Batesian mimicry.

The type locality of femoralis is Yurimaguas, Peru, but the taxon is distributed throughout much of the Amazon basin (Silverstone, 1976). Morphologically, specimens referred to femoralis exhibit minor variations in coloration (e.g., thickness of lateral stripes, size and extent of bright thigh flash marks; see Silverstone, 1976). We generated cytochrome b sequences for 17 samples of femoralis collected at the following eight localities, covering much of the nominal species' range (pairwise distances shown in table 12): Porto Walter, Acre, Brazil (397, 398, 1309); Cusco Amazónico, Madre de Dios, Peru (78, 128, 129); Cuyabeno, Sucumbíos, Ecuador (399, 400, 1306); Parque Estadual Guajará-Mirim, Rondônia, Brazil, (393, 394); Reserva Ducke, Amazonas, Brazil (520); Panguana, Huánuco, Peru (nearest to the type locality and therefore tentatively treated as femoralis sensu stricto) (526); Sipaliwini, Suriname (1325); Rio Curuá-Una, Pará, Brazil (395, 396, 1298).

Table 12

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of femoralis and zaparoai0003-0090-299-1-1-t12.gif

The taxonomy of zaparo is less problematic, but we include it in table 12 as a point of reference for femoralis. Vences et al. (2003a) united these two species formally in Allobates. The species Duellman and Mendelson (1995) referred to as zaparo is a distantly related, probably toxic species (details discussed below).

As shown in table 12, the pairwise distances between zaparo and femoralis samples are 12.2–15.3%. Forty-three unambiguous transformations unite the zaparo samples, and 38 unite those of femoralis. Although the cladogram is consistent with the recognition of a single species for material currently referred to femoralis (i.e., the taxon is monophyletic), the extensive patristic and pairwise distances are suggestive of multiple species. Cytochrome b distance is low within localities (0.0–0.8%) and much higher between localities (3.9–14.6%). This is strongly suggestive that a different species occurs at each of these localities (i.e., eight species), which would greatly increase the known diversity of this clade.

Thirty-four unambiguous transformations establish the monophyly of the clade shown in figure 73, with a Bremer value of 10. The caeruleodactylus–”PEG-M3” clade is united by 36 unambiguous transformations. Lima and Caldwell (2001) named caeruleodactylus, and Caldwell et al. (2002a) described its distinctive tadpole. Although we did not include marchesianus in the present analysis, the conspicuous dark vertical stripes on the tail and greatly enlarged marginal papillae provide evidence that it is the sister species of caeruleodactylus (see Caldwell et al., 2002a). Pairwise distances between specimens of caeruleodactylus are shown in table 13. The samples were collected at the type locality.

Table 13

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of Syntopic Specimens of caeruleoactylusi0003-0090-299-1-1-t13.gif

Sample 1277 from Rio Ituxi was referred to gasconi by Morales (2002 “2000”), as was the distantly related sample 356 from Porto Walter (the pairwise distance between cytochrome b sequences of these two specimens is 15.6%). The type locality given for this species is “Jainu on the left side of the Río Juruá, Amazonas, Brazil” (Morales, 2002 “2000”: 30, translated from the Spanish). Although Porto Walter is located on the Rio Juruá, Rio Ituxi is slightly closer, and on that basis we refer these terminals to gasconi. Comparison with topotypes will be required to confirm the identity of these samples. The pairwise cytochrome b distance between this species and caeruleodactylus is 14.0–14.5%, between this species and the undescribed “Curuá-Una” (see below) 13.5–14.2%, and between this species and “PEG-M3” 15.8–16.4%. Pairwise cytochrome b distances within gasconi are given in table 14.

Table 14

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of gasconiai0003-0090-299-1-1-t14.gif

Samples from near Rio Curuá-Una represent an undescribed species (“Curuá-Una”; see table 15). The pairwise distances between the samples of this species and those of “PEG-M3” (see below) are 9.9–10.9%. Morales (2002 “2000”) referred sample 354 to the distantly related brunneus; as mentioned above, the minimum pairwise distance between this specimen and either of the others Morales referred to brunneus is 16.6%.

Table 15

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of an Undescribed Species from Near the Rio Curuá-Una in Brazilai0003-0090-299-1-1-t15.gif

“PEG-M3” is one of three undescribed species of dendrobatids collected at Parque Estadual Guajará-Mirim, Rondônia, Brazil. The pairwise distance between the samples of this species and “Curuá-Una” is 9.9–10.9% (for distances within “PEG-M3” see table 16). In the current analysis, humilis is nested within the samples of “PEG-M3”. However, it is highly unlikely that the populations are conspecific: The sample of humilis was collected at 2,100 m in the Venezuelan Andes (La Marca et al., 2002), whereas “PEG-M3” is from the Amazonian lowlands of western Brazil. Sequence data for humilis are limited to approximately 500 bp of 16S, so additional molecular data is likely to overturn this result. La Marca et al. (2004) considered humilis to be most closely related to their new species pittieri, which we did not test here.

Table 16

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of “PEG-M3”i0003-0090-299-1-1-t16.gif

The clade composed of conspicuus, insperatus, the unidentified Ecuadorian species reported by Santos et al. (2003; no locality given), and an undescribed species from Cuyabeno, Ecuador is well supported (Bremer support  =  56) and united by 70 unambiguous transformations. The samples of the Brazilian species conspicuus were collected at Porto Walter and sample 614 was referred to conspicuus by Morales (2002 “2000”). Bremer support for the monophyly of these specimens is 157, and 157 synapomorphies optimize to it unambiguously. The remaining three species in this clade (branch length  =  62, Bremer support  =  56) are all from Ecuador. Cytochrome b data are not available for insperatus and the unnamed “Colostethus sp.”, but pairwise distances are shown in table 17 for conspicuus and “Cuyabeno”.

Table 17

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of conspicuus and an Undescribed Species from Cuyabeno, Ecuadorai0003-0090-299-1-1-t17.gif

The remaining terminals in figure 73 are allied to trilineatus. In their redescription of trilineatus based on extensive material from Peru, Grant and Rodríguez (2001) noted variation within and between localities that could be representative of greater species diversity. The present study included DNA sequences of putative trilineatus samples from Cusco Amazónico, Madre de Dios, Peru (74 and 112; specimens not examined by Grant and Rodríguez, 2001, but referred explicitly to trilineatus by Morales, 2002 “2000”) and Panguana, Huánuco, Peru (527; Grant and Rodríguez, 2001, referred material from this locality to trilineatus; Morales, 2002 “2000”, referred specimens from this locality to marchesianus, but that species is endemic to the Rio Uaupes of Brazil and adjacent Rio Vaupés of Colombia; Caldwell et al., 2002a, 2002b; T. Grant, personal obs.). The type locality of Yurimaguas is closest to Panguana. Also included here are samples of one of two dendrobatid species collected at Porto Walter, referred to as “PortoWalter2”. As mentioned above, one of these specimens (356) was referred to gasconi by Morales (2002 “2000”). A single sample each is available from São Francisco (516), Reserva Ducke (524), and Rio Ituxi (404), and several samples each from Parque Estadual Guajará-Mirim (359, 616, 1288), Manaus (620, 405, 1318), Cusco Amazónico (74, 112), and Porto Walter (355, 356, 1265, 1269, 1273).

The trilineatus clade has a Bremer support value of 18, with 25 unambiguous transformations at this node. As shown in figure 73 and table 18, the pattern of diversification is suggestive of eight species—one at each locality. The minimum pairwise cytochrome b distances between localities is 5.7% between the trilineatus from Cusco Amazónico and the samples from Porto Walter. Despite the geographic and cladistic proximity of the Reserva Ducke and São Francisco samples, these samples (which are united by 25 unambiguous transformations) are separated by a patristic distance of 244 steps and their cytochrome b sequences are 15.3% different.

Table 18

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of trilineatus and Related Undescribed Speciesai0003-0090-299-1-1-t18.gif

The clade shown in figure 74 is united by 89 unambiguous transformations, including reduction in the length of finger IV (Character 4, 0 → 1), lengthening of finger I (Character 5, 2 → 3), and swelling of finger III in adult males (Character 20, 0 → 1). The next clade, shown at the top of figure 74, includes the nubicola group and Silverstone's (1976) tricolor group + machalilla. This inclusive clade is delimited by 84 unambiguous transformations in DNA sequences.

The nubicola group, represented by flotator, nubicola, and the undescribed species “nubicola-spC” to be named by T. Grant and C. W. Myers (in preparation) is delimited by 46 unambiguous transformations, including the gain of a straight pale ventrolateral stripe (Character 54, 0 → 2), pale male abdomen color (Character 63, 3 → 0), anterior pigmentation of the large intestine (Character 66, 0 → 1), and several synapomorphies relating to the larval oral disc (Character 88, 0 → 1; Character 89, 1 → 0; Character 91, 0 → 1; and Character 94, 3 → 0). This clade includes data downloaded from GenBank that were attributed to pratti from western Colombia by Vences et al. (2003a). However, one of the authors of that study informed us that they did not examine a voucher specimen (S. Lötters, in litt. 2/23/2005), and nubicola and pratti occur in sympatry and are often confused by collectors. The three sampled species resemble each other greatly; the Central American species flotator was considered a synonym of nubicola until 1995 (Ibáñez and Smith, 1995), but these two species are not sisters and differ in 18.4% of their cytochrome b sites (table 19).

Table 19

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of flotator, nubicola, and “nubicola-spC”ai0003-0090-299-1-1-t19.gif

The sister of the nubicola clade includes several taxonomically problematic taxa. Lötters et al. (2003b) noted differences in the vocalizations of boulengeri and concluded that more than one species was probably involved. Cytochrome b sequences are unavailable for the GenBank specimen for comparison, but only eight unambiguous transformations group boulengeri 280 with the other species of this clade. Cytochrome b sequences are also unavailable for GenBank specimen Epipedobates sp. QCAZ16589, but the occurrence of only four unambiguous transformations to distinguish it from espinosai 1139 suggests that it may be conspecific with espinosai. Like Santos et al. (2003), we found that machalilla is nested within this clade of otherwise toxic species. However, the specimens we sequenced fall together, whereas the Santos et al. sequence obtained from GenBank is sister to tricolor. Graham et al. (2004) reported this sample of machalilla to be most closely related to anthonyi instead of tricolor, although that conclusion was not supported in their analysis (the critical node had a Bremer value of 0). In our analysis, only a single unambiguous synapomorphy unites tricolor and machalilla, and the clade has a Bremer value of only 1. There is little unambiguous evidence to group these samples with anthonyi to the exclusion of machalilla samples 73 and 132 (only five transformations), but it is worth noting that samples 73 and 132 are united by 24 unambiguous transformations and differ in only 9.

Grant and Castro-Herrera (1998) noted the extensive within- and among-population variation in fraterdanieli and left open the possibility that this may be a complex of similar species. The samples of fraterdanieli were collected in Colombia near Popayán, Cauca, at approximately 1,800 m in the Cordillera Occidental (1226, 1227, 1228). An additional sample was collected at 2,800 m in the Departamento de Caldas in the Cordillera Central (1230). All localities face the Cauca valley. As seen in table 20, the three Cauca fraterdanieli cytochrome b sequences are identical. The pairwise distance between those samples and the Caldas specimen is 6.5%. Likewise, the three Cauca specimens are united by 42 unambiguous transformations, and the Caldas sample differs by an additional 63 unambiguous transformations. This pattern of diversity is strongly suggestive that these are two different species. The type locality of frateranieli is in the Cordillera Central in Antioquia, at approximately 1,900 m (Silverstone, 1971). Minimally, comparison with samples from lower elevations in the Cordillera Central is required to determine which of these populations (i.e., that from approximately the same elevation but much further south in the Cordillera Occidental or that from the Cordillera Central in roughly the same region but much higher elevation) is conspecific with fraterdanieli sensu stricto or if additional species have been conflated under this name.

Table 20

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of fraterdanieliai0003-0090-299-1-1-t20.gif

The terminals labeled pratti 1144 and and “pratti-like” 1224 are morphologically indistinguishable but are almost certainly not conspecific. Sample 1144 was collected at El Copé, Coclé, central Panama, and 1224 is from Jungurudó, Darién, near the boarder with Colombia. Roberto Ibáñez noted differences in their vocalizations (in litt., 12/20/2003). Further, only female nurse frogs are known to occur in pratti (Grant, 2004), whereas a male nurse frog was collected at Jungurudó. The behavioral sample size at the Darién locality is inadequate to eliminate the possibility that this is simply intraspecific variation, but these behavioral differences are further reinforced by the observation that, although these two samples are united by 104 unambiguous transformations, the patristic distance between them is 163. Finally, the pairwise distance between their cytochrome b sequences is 10.6%. As such, despite the lack of morphological differences between these frogs, there is considerable evidence that they represent different species. Resolution of this problem, though evidentially straightforward, is nomenclaturally complicated. The type locality of pratti is in western Colombia, and the relationship of topotypic pratti to either of these samples has not yet been assessed. The proximity of the Darién species to the type locality suggests it may be pratti sensu stricto, but direct evidence is required. As noted above, the specimen reported as pratti from western Colombia by Vences et al. (2003a) is probably a misidentified specimen of nubicola, but that too requires confirmation.

Grant (2004) removed panamensis from the synonymy of inguinalis on morphological grounds, and, although there are several points of resemblance, imbricolus differs extensively from both species (e.g., ventral coloration, color and definition of flash marks, degree of webbing, sexual dimorphism, and occurrence of tetrodotoxin). Although the identities of inguinalis and imbricolus are not problematic, panamensis is widespread and highly variable. Dunn (1933) and Savage (1968) drew attention to differences between western and eastern samples. Grant (2004) found that variation between localities was no greater than is observed in samples from each locality and therefore concluded that the samples of panamensis constituted a single species. The two panamensis samples included for DNA sequence data are from distant localities: 1150 is from El Copé in central Panama, whereas 1223 is from extreme eastern Panama at Caná, Darién at the eastern extreme of the distribution, near the Colombian border.

Both the cladistic results and the pairwise cytochrome b comparisons (table 21) support Grant's (2004) conclusion that inguinalis is not conspecific with the Panamanian species previously assigned to its synonymy. However, the present results suggest that the two samples of panamensis represent different species. The pairwise distance between the cytochrome b sequences for these two samples is 11.4%. Furthermore, the cladogram shows the western sample to be more closely related to imbricolus, from which its cytochrome b sequence differed by only 3.9%. Denser sampling at intervening localities, as well as additional data (e.g., vocalizations, behavior), are required to address this problem decisively.

Table 21

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of imbricolus, inguinalis, and Two Distant Localities of panamensisai0003-0090-299-1-1-t21.gif

The next large clade includes the majority of the species referred to Phyllobates by Silverstone (1976), Ameerega by Bauer (1986), and Epipedobates by Myers (1987). More specifically, it is equivalent to the combination of Silverstone's (1976) pictus and trivittatus groups, with the addition of species described subsequently. The clade is delimited by 131 unambiguous transformations, including the almost unique gain of conspicuously granular dorsal skin (Character 0, 1 → 2) and ability to sequester lipophilic alkaloids (Character 146, 0 → 1).

Unlike other widespread Amazonian species, such as femoralis (discussed above), and despite the known degree of external color and color pattern variation (Silverstone, 1976) of nominal trivittatus, the pattern and extent of diversity are not suggestive of more than a single species (see table 22). We included samples of trivittatus from seven localities covering (albeit sparsely) most of the known range of the species, as follows (listed approximately from southwest to northeast): Tambopata Reserve, Madre de Dios, Peru (319, 320, 322); Porto Walter, Acre, Brazil (385, 387, 1297, 1305); Panguana, Huánuco, Peru (518); Leticia, Amazonas, Colombia (1350); Balbina, north of Manaus, Amazonas, Brazil (519); south of Manaus, Amazonas, Brazil (627, 628); and Para, Suriname (1329).

Table 22

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of trivittatusai0003-0090-299-1-1-t22.gif

The monophyly of trivittatus is established by 96 unambiguous transformations, and the pairwise cytochrome b distances between these samples and Guyanan pictus are 13.2–14.3% and southeastern Brazilian flavopictus are 10.9–12.5%. Conversely, the variation within trivittatus is low, despite the great distances between localities. Pairwise cytochrome b distances between localities are 0.5–3.4%. Although the higher values are as great as or greater than those between some closely related species (e.g., auratus and truncatus; see below), there are no major gaps (i.e., pairwise distances appear to vary continuously) or geographic trends, and cladistic relationships do not suggest historically isolated populations.

Duellman and Mendelson (1995) referred sample 127 from northern Peru to zaparo, but they also noted that theirs was the first record of that taxon outside the Río Pastaza drainage. The present results demonstrate conclusively that this species is not conspecific with zaparo, despite their morphological resemblance, and we therefore place the name in quotes. Sufficient data (e.g., locality) are unavailable to determine if “zaparo” and Santos et al.'s (2003) parvulus are conspecific.

One of the more unexpected species-level results is the grouping of the GenBank sample of pictus from Bolivia, near the type locality, with pictus 1331 from Guyana. Despite the great geographic distance between these localities, the samples appear to be conspecific.

“PortoWalter1” is another undescribed species from Porto Walter, Brazil. The sister of this species is rubriventris. Although only 12 unambiguous transformations diagnose “PortoWalter1” from rubriventris (three “PortoWalter1” synapomorphies and nine rubriventris autapomorphies) only 566 bp of 16S data were available for rubriventris. Cytochrome b sequences are identical in samples of “PortoWalter1”, except for sample 626, which differs from the others in a single nucleotide.

Like trivittatus and femoralis, hahneli is another widespread Amazonian species. The type locality for hahneli is Yurimaguas, Peru. We included 12 samples of hahneli from four localities, as follows: Cusco Amazónico, Peru (79, 109, 110), Leticia, Colombia (1354); south of Manaus, Brazil (386, 391, 392, 1304), and Porto Walter, Brazil (382, 388, 389, 390). Pairwise distances between the cytochrome b sequences of these samples are given in table 23, and these sequences differ from those of pulchripectus in 10.4–11.7% of their sites. The hahneli samples are united by 51 unambiguous transformations. The sample from Leticia differs from the others in 6.5–7.5% of its cytochrome b sequence—much more than occurs between other samples, despite the greater geographic distance between other samples (e.g., Cusco Amazónico and Manaus). Likewise, the clade containing the remaining hahneli samples is united by 31 unambiguous transformations. This suggests that samples from Leticia and other localities are not conspecific. It is unclear which of these species is conspecific with hahneli sensu stricto. The cytochrome b distances between the remaining hahneli localities are 2.1–3.1%. Although Santos et al. (2003) did not provide specimen data, the topology suggests their specimen (QCAZ13325) is conspecific with the Leticia hahneli.

Table 23

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of hahneliai0003-0090-299-1-1-t23.gif

In figure 75, the Colombian species subpunctatus is sister to a clade diagnosed by 76 unambiguous transformations, including several changes in hand and foot morphology (Characters 13, 15, 36–44), the appearance of posteriorly angled clavicles (Character 121, 0 → 1), gain of palatine bones (Character 132, 0 → 1), and the shift to riparian habitat (Character 113, 2 → 1).

Santos et al. (2003) resurrected maculosus from synonymy with bocagei, where it had been placed by Coloma (1995). Key to that interpretation is the identity of the specimen they identified as bocagei sensu stricto, because that species falls out with sauli both here and in Santos et al.'s (2003) analysis. However, no locality or other data were provided for that specimen, and an alternative possibility is that the remaining samples (including those identified here as bocagei from Cuyabeno) are conspecific with topotypic bocagei and the sister of sauli is an undescribed species. Additional data are required to assess the alternative hypotheses.

Santos et al. (2003) also omitted locality data for the unidentified specimens Colostethus sp. QCAZ16511, Colostethus sp. QCAZ16504, and Colostethus sp. QCAZ16503, which complicates understanding the diversification of these dendrobatids. As noted above, one possibility is that these and related terminals are conspecific. Nevertheless, in light of the patristic distances between these terminals (128 steps between QCAZ16511 and QCAZ16504; 87 between QCAZ16504 and QCAZ16503), our provisional interpretation is that Colostethus sp. QCAZ16511 and Colostethus sp. QCAZ16504 are different, possibly undescribed species, and that Colostethus sp. QCAZ16503 is conspecific with the terminals from Cuyabeno (e.g., patristic distance between QCAZ16503 bocagei 1267  =  8 steps). Although the topology is consistent with Colostethus sp. QCAZ16511 being conspecific with maculosus sensu Santos et al. (2003), 50 and 47 unambiguous transformations in mtDNA subunit H1 occur at these terminal nodes, respectively, suggesting they represent different species.

The clade composed of delatorreae, pulcherrimus, and sylvaticus Barbour and Noble is delimited by 38 unambiguous transformations. These species are all from mid- to high elevations in the Andes of northern Ecuador (delatorreae) and northern Peru (pulcherrimus and sylvaticus Barbour and Noble). Duellman (2004) recently named pulcherrimus and compared it to the similar sylvaticus Barbour and Noble. The two samples of pulcherrimus (118 and 119) are topoparatypes (Cajamarca, Peru), and both samples of sylvaticus Barbour and Noble were collected at 2,820 m in Ayacaba, Peru. The species are closely related, but the pairwise distances between their cytochrome b sequences are 13.0–13.3% (see table 24) and each is diagnosed by approximately 100 unambiguous transformations.

Table 24

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of pulcherrimus and sylvaticus Barbour and Nobleai0003-0090-299-1-1-t24.gif

The sister group to that clade includes nexipus, azureiventris, chlorocraspedus, and an unidentified species sequenced by Santos et al. (2003; no locality data given). The known species form a distinctive group of relatively brightly colored frogs with dorsolateral stripes, the latter being an unambiguous synapomorphy of the clade (Character 52, 0 → 3). In total, the clade is delimited by 51 unambiguous transformations. Lötters et al. (2000) proposed the genus Cryptophyllobates for the putatively aposematic azureiventris. However, Daly (1998:171) reported that in a feeding experiment this species did not accumulate dietary alkaloids. The recently described species chlorocraspedus is as brightly colored as azureiventris, and wild-caught samples lacked detectable levels of alkaloids also (J. W. Daly, in litt., 01/28/05). Although they were not included in the present study, the two recently named species patitae (Lötters et al., 2003a) and eleutherodactylus (Duellman, 2004) are also likely part of this clade. The samples of chlorocraspedus have identical cytochrome b sequences, with the exception of sample 385, which differs in two nucleotides (0.5%). Samples of nexipus were included from two localities at different elevations (Cataratas Ahuashiyacu, 14 km NE of Tarapoto, 730 m: 75, 130, 131; and San Martín, 6 km ESE of Shapaja, 300 m: 123). The specimen from the lower elevation is identical to two of the three specimens from the higher elevation; those specimens differ from one of the 730 m specimens in two nucleotides. Santos et al. (2003) omitted locality data for the sample they identified as nexipus, but 45 unambiguous transformations optimize to the terminal node (all from mtDNA subunit H1) and 42 unambiguous changes delimit the remaining specimens as a clade (all from Peru), suggesting that they may not be conspecific.

The terminals referred to as “Ibagué” are an undescribed species from the slopes of the Magdalena valley in Colombia (see table 25 for pairwise cytochrome b distances). The species possesses the black arm band in adult males and is thus the sole exemplar of the ramosi group included in present study (Grant and Castro, 1998; Grant and Ardila-Robayo, 2002). Other species that possess this structure (and are included in the ramosi group) are anthracinus, cevallosi, fascianiger, exasperatus, lehmanni Silverstone, ramosi, and saltuarius. “Ibagué” is nested in a clade with vertebralis and pulchellus, all of which are small, identically striped, and similarly colored Andean frogs. Forty-two changes optimize unambiguously to the node including vertebralis and 34 unambiguous transformations unite “Ibagué” with pulchellus.

Table 25

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of “Ibagué”ai0003-0090-299-1-1-t25.gif

Rivero (1991a) described idiomelus based on a single specimen (from Venceremos, Peru), but Duellman's (2004) account was based on extensive material, including adults and larvae from several localities; all of the samples sequenced in the present study were referred to idiomelus by Duellman (2004). Three specimens (120–122) are from 2,180 m at Abra Pardo de Miguel, San Martín, and the other two (77 and 126) are from 2,150 m at Pomachochas, Amazonas. The five specimens form a clade, but the Abra Pardo locality is paraphyletic with respect to Pomachochas. Pairwise cytochrome b distances are given in table 26.

Table 26

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of idiomelusai0003-0090-299-1-1-t26.gif

Originally described from Loja, Ecuador, elachyhistus is a widespread, highly variable Andean species. Duellman (2004) recently redescribed elachyhistus from several localities in northern Peru, including those included in the present study. Based on the current results, it is clear two species have been conflated, a southern species from Cajamarca, Peru (105, 106, and 107), which is sister to insulatus, and a northern species from Piura, Peru (108, 114, 115, 116, 117); see table 27 for pairwise cytochrome bnamed-content> distances. Locality data were not given by Santos et al. (2003) for the GenBank elachyhistus included, but the data provided in GenBank report it as being from Ecuador, which suggests that the northern species is elachyhistus and the southern species is undescribed.

Table 27

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of Two Species Currently Referred to elachyhistusai0003-0090-299-1-1-t27.gif

Although toxic species also occur elsewhere in the cladogram (e.g., anthonyi, petersi), the remaining clade, shown in figure 76, consists of exclusively brightly colored and (insofar as is known) toxic species. Evidence for the monophyly of this clade is given by 63 unambiguous changes, including origin of smooth dorsal skin (Character 0, 1 → 0), loss of the oblique lateral stripe (Character 55, 1 → 0), the loss of metallic pigmentation of the iris (Character 65, 1 → 0), use of phytotelmata as larval habitat (Character 110, 0 → 1), and origin of the ability to sequester lipophilic alkaloids (Character 147, 0 → 1).

The clade composed of aurotaenia, bicolor, lugubris, terribilis, and vittatus constitutes Phyllobates sensu Myers et al. (1978), and its monophyly is established by 146 unambiguous transformations, including the lengthening of finger I (Character 5, 2 → 3) and the ability to accumulate batrachotoxin (Character 148, 0 → 1). Species identities are uncontroversial, the sole potential exception being the possibility that terribilis represents the southern extreme of clinal variation of bicolor (Myers et al., 1978; Lötters et al., 1997a). That hypothesis is rejected in the current phylogenetic analysis, which places aurotaenia and terribilis as sister species to the exclusion of bicolor. This result is also consistent with cytochrome alic>b pairwise distances (table 28). The pairwise distance between terribilis and bicolor is 7.0%, whereas the pairwise distance between terribilis and aurotaenia is only 5.7%. The two lugubris samples are from Panama (329) and Nicaragua (366), representing opposite extremes in the species' distribution. These specimens form a clade, and there is no indication in morphology or otherwise that lugubris may refer to more than a single species. Nevertheless, the distance between cytochrome b sequences of the two samples is 6.0%. The terribilis samples are from the type locality in western Colombia (1135) and bred in captivity (1232). The aurotaenia, bicolor, and one of the vittatus samples (839) were bred in captivity; the second vittatus sample is GenBank sequence AF128582.

Table 28

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of aurotaenia, bicolor, lugubris, terribilis, and vittatusai0003-0090-299-1-1-t28.gif

Maxson and Myers (1985) proposed that the South American species bicolor and terribilis were sisters and were, in turn, sister to a clade composed of lugubris, aurotaenia, and vittatus, the latter two being sisters. In addition to a plausible biogeographic argument, bicolor and terribilis were grouped on the basis of the shared ontogenetic “loss” (through hypertrophy; see Character 52, above) of dorsolateral stripes (DLS). Widmer et al. (2000) tested that hypothesis with a 520-bp cytochrome b dataset and concurred that bicolor and terribilis were sister-species. However, they found that aurotaenia was placed in a clade with the other South American species, and that the two Central American species, lugubris and vittatus, were sisters. Our greatly enlarged dataset corroborates Widmer et al.'s (2000) hypothesis of Central and South American monophyly, but we found that bicolor is the sister of aurotaenia + terribilis. Maxson and Myers (1985; see also Myers et al., 1978) hypothesized that the persistent DLS was “a primitive pattern that is retained by the adults of aurotaenia, lugubris, and vittatus.” According to our results, it is equally parsimonious for the persistent DLS to have evolved in the ancestor of Phyllobates, with either two “losses” of the DLS in adults in bicolor and terribilis or a single “loss” of the DLS in adults in the ancestor of the South American clade and reversion in aurotaenia, or for the DLS in juveniles only to be ancestral for Phyllobates and the persistent DLS to have evolved independently in aurotaenia and the ancestor of lugubris and vittatus.

Myers (1987) designated steyermarki as the type species of Minyobates, which he proposed for several species previously included in Silvestone's (1975a) minutus group of diminutive Dendrobates. Further, Myers hypothesized that steyermarki and its relatives were placed outside of a Phyllobates + Dendrobates clade that included the remainder of Silverstone's minutus group. In our results, steyermarki is placed as the sister species of all species traditionally associated with Dendrobates, including a clade that corresponds to the bulk of Silverstone's minutus group. The placement of these species in a clade exclusive of Phyllobates refutes Myers's hypothesis (but see discussion of castaneoticus and quinquevittatus, below), but is identical to the findings of Vences et al. (2003a). These clades are not strongly supported, owing to the lack of evidence for steyermarki, for which only phenotypic characters and 547 bp of 16S (the latter sequenced by Vences et al., 2003a) could be included.

Silverstone (1976) named the distinctive species fulguritus from the Chocó region of western Colombia, and the included sample is from near Bahía Solano. The sister-species claudiae and minutus strongly resemble each other morphologically. Nevertheless, their cytochrome b sequences are 8.1–8.3% dissimilar (see table 29). The monophyly of this group of three species is established by 69 unambiguous transformations, including the occurrence of dorsolateral and oblique lateral stripes (Characters 52 and 55) and the fusion of vertebrae 2 + 3 (Character 146, 0 → 1).

Table 29

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of claudiae, fulguritus, and minutusai0003-0090-299-1-1-t29.gif

The sister group of the fulguritus clade contains most of the Amazonian species of Silverstone's minutus group. Evidence for the monophyly of this group is given by 67 unambiguous transformations, including the expansion of finger discs II–IV (Characters 8–10). Caldwell and Myers (1990) removed ventrimaculatus from the synonymy of quinquevittatus (see below), but they noted that the nominal taxon, which occurs throughout the Amazon region from Peru to French Guiana, probably consists of a complex of similar species. Symula et al. (2003) presented molecular evidence that at least two distantly related species are included in Peruvian “ventrimaculatus”. Assuming the validity of amazonicus and variabilis (but see Lötters and Vences, 2000, and Caldwell and Myers, 1990, respectively) the current results suggest five species of “ventrimaculatus”, one at Rio Ituxi, Brazil (376, 377), a second at Manaus, Brazil (1310), a third at Leticia, Colombia (1349) and French Guiana (GenBank AY263248), a fourth at Pompeya, Ecuador (374), and a fifth at Porto Walter, Brazil (375, 1311). Although the cladogram does not falsify the hypothesis that the samples from Rio Ituxi and Manaus are a single species, 72 unambiguous synapomorphies unite the two Rio Ituxi specimens, 58 autapomorphies optimize unambiguously to the Manaus terminal node, and cytochrome b sequences differ in 8.1% of their sites (table 30). Despite the considerable geographic distance between the Leticia and French Guiana samples, they form a clade, and the patristic difference between them is minimal (14 steps; cytochrome b data are unavailable for the French Guiana sample); however, this is based on analysis of only 560 bp of 16S for the French Guiana specimen. The Leticia and Pompeya localities are closest to the type locality of Sarayacu, Ecuador, suggesting that one of them may be conspecific with ventrimaculatus sensu stricto.

Table 30

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of Nominal ventrimaculatusai0003-0090-299-1-1-t30.gif

The sister clade to the minutus group consists of the remaining species traditionally referred to Dendrobates. Fifty-four unambiguous transformations delimit this node, including the origin of even caudal pigmentation in larvae (Character 87, 1 → 2).

Caldwell and Myers (1990) clarified the identity of quinquevittatus (removing the unrelated ventrimaculatus from its synonymy in the process; see above). They proposed a close relationship between quinquevittatus and the clearly heterospecific castaneoticus. They did not discuss the relationships of galactonotus, but its placement in the tinctorius group by Silverstone (1975a) was uncontroversial. The monophyly of galactonotus, castaneoticus, and quinquevittatus was first proposed by Vences et al. (2003a), although these three taxa were unresolved in their topology. In the present study, 105 unambiguous synapomorphies optimize to this node, and Bremer support is 37, leaving little doubt as to the reality of this clade. Nevertheless, the occurrence of galactonotus in this clade is unexpected, as its morphology shares little with the diminutive castaneoticus and quinquevittatus. Pairwise cytochrome b distances for these species are shown in table 31.

Table 31

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of castaneoticus, galactonotus, and quinquevittatusai0003-0090-299-1-1-t31.gif

The next clade is delimited by 37 unambiguous transformations. The first clade included in this group consists of the histrionicus group of Myers et al. (1984). The evidence for the monophyly of this group is overwhelming, consisting of 132 unambigously optimized synapomorphies. These include several larval modifications (Characters 90, 93, and 94), origin of tadpole transport by female nurse frogs (Character 109, 0 → 1) and larval oophagy (Character 111, 1 → 2), and fusion of the sacrum and vertebra 8 (Character 143, 0 → 1) and vertebrae 2 and 3 (Character 145, 0 → 1).

Myers and Daly (1976b) illustrated and discussed the extensive variation within what they considered to be the single species histrionicus, distributed throughout the Pacific lowlands of western Colombia and northwestern Ecuador. In the same paper, they named lehmanni Myers and Daly, based primarily on differences in vocalizations, coloration and color pattern, and, especially, the absence of histrionicotoxins from skin alkaloid profiles. Nevertheless, E. Zimmermann (1986:135) claimed that histrionicus and lehmanni Myers and Daly crosses produced fertile offspring, and Garraffo et al. (2001) showed experimentally that lehmanni Myers and Daly efficiently sequesters histrionicotoxins administered in the diet. Based on differences in vocalizations and coloration and color pattern, Lötters et al. (1999) resurrected sylvaticus Funkhouser from the synonymy of histrioncus for the southernmost populations in southern Colombia and northern Ecuador.

The histrionicus samples included here were both collected in Chocó department, Colombia, but are from distant localities and involve different color morphs. Sample 336 was taken along Quebrada Vicordó (locality D of Myers and Daly, 1976b; see their Plate 1C for color morph), whereas sample 498 is from Sierra Mecana (approximately 6°15′N, 77°21′W), north of Bahía Solano; the two localities are separated by >100 km. The lehmanni Myers and Daly sample is from the region of the type locality. The sample of sylvaticus Funkhouser is from Ecuador. Cytochrome b sequences were not available for the GenBank specimens shown in the cladogram.

The cladogram is consistent with the validity of these three species. The two samples of histrionicus were recovered as monophyletic; their cytochrome b sequences differ from each other in only a single base (0.3%) and are approximately 5% different from both lehmanni Myers and Daly and sylvaticus Funkhouser (see table 32). Similarly, sylvaticus Funkhouser, which had been in the synonymy of histrionicus until recently, is more closely related to lehmanni Myers and Daly. Although it has never been postulated that these two nominal species may be conspecific to the exclusion of histrionicus, that hypothesis is not ruled out by the current results. Their cytochrome b sequences are only 2.9% dissimilar, which is less than the distance between the closely related species bicolor and terribilis (7.0%) and minutus and claudiae (8.1–8.3%), for example, but is greater than is observed between some specimens of the clearly heterospecific auratus and truncatus (2.3–3.1%; see below). Regardless of their low degree of pairwise dissimilarity, lehmanni Myers and Daly and sylvaticus Funkhouser are still diagnosable on the basis of phenotypic evidence (Myers and Daly, 1976b; Lötters et al., 1999) and therefore are valid species.

Table 32

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of histrionicus, lehmanni Myers and Daly, and sylvaticus Funkhouserai0003-0090-299-1-1-t32.gif

Also included in this clade are a number of small species allied phenetically to pumilio. The systematics of these species has been confounded by the astonishing intra- and interpopulational variation in coloration (e.g., Myers and Daly, 1983). Only pumilio is not represented by singletons in the cladogram, and, as such, the monophyly of those species was not tested. Nevertheless, consideration of patristic and pairwise (table 33) distance supports the historical reality of these species.

Table 33

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of arboreus, pumilio, speciosus, and vicenteiai0003-0090-299-1-1-t33.gif

The sister of the histrionicus group is equivalent to Silverstone's (1975a) tinctorius group, with the exclusion of galactonotus (see above). This clade is individuated by 92 unambiguously optmized synapomorphies. Hoogmoed (1969) described azureus from Vier Gebroeders Mountain in southern Sipaliwini, near the Brazilian border. Its resemblance to tinctorius was noted in the original description, and Silverstone (1975a) considered it to be closely related to and potentially derived from that species. The extensive variation in tinctorius discovered subsequently has only strengthened the suspicion that these two nominal taxa are conspecific. The two samples of azureus were obtained from the region of the type locality in Suriname (1330) and in adjacent Brazil (534). One of the tinctorius samples is also from near the Tafelberg airstrip, Sipaliwini, Suriname (1327), and the other is from Brazil.

The cladogram indicates that tinctorius is paraphyletic with respect to azureus. Furthermore, as shown in the pairwise comparisons (table 34), the two azureus samples are identical and differ from the Brazilian tinctorius sample in only a single nucleotide (0.3%). The pairwise distance between the Brazil and Suriname tinctorius is greater than that between it and azureus. All of this is consistent with the hypothesis that these samples are conspecific.

Table 34

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of azureus and tinctoriusai0003-0090-299-1-1-t34.gif

Despite the considerable variation in color and color pattern in auratus, there are no known problems surrounding the identities of auratus and truncatus (see table 35). Silverstone (1975a) hypothesized that these two species are closely related, and the available evidence corroborates that claim with a total of 84 unambiguously optimized synapomorphies. Samples of auratus are from two localities in Bocas del Toro, Panama (327, 334, 335) and one in Nicaragua (365). One truncatus sample was captive raised (1151); the other two were taken in western Colombia.

Table 35

Percent Uncorrected Pairwise Distances Between Cytochrome b Sequences of auratus and truncatusai0003-0090-299-1-1-t35.gif

Summary of Relationships among Dendrobatids

This study resolved the relationships among most of the included terminals. Results are generally consistent with prior findings, especially the species groups proposed by Silverstone (1975a, 1976). At the level of genera, Allobates (whether restricted to the femoralis group or applied more generally; see below), Ameerega, Dendrobates (including or excluding Minyobates, Oophaga, and Ranitomeya), Epipedobates, Mannophryne, Oophaga, Phyllobates, and Ranitomeya were all found to be monophyletic. Nephelobates was found to be paraphyletic with respect to Aromobates nocturnus and Colostethus saltuensis. As expected, the greatest incongruence between generic grouping and phylogeny involves Colostethus, which was shown to be vastly nonmonophyletic. Nevertheless, the density of taxon sampling allowed coherent clades to be delimited, which will permit a monophyletic taxonomy to be developed below.

In addition to resolving the relationships among species, this study sheds light on the identities of numerous problematic species. Comparison with topotypic material is required to determine which species are new, and in several cases the lack of locality data for the sequences reported by Santos et al. (2003) makes it difficult to assess species identity (especially in relation to bocagei). Nevertheless, consideration of cladistic and patristic distances suggests the 367 dendrobatid terminals included in this analysis represent 156 species. Available evidence clearly delimits two distantly related species conflated under the name elachyhistus, corroborating Duellman's (2004) suspicions. The widespread Amazonian taxon ventrimaculatus is paraphyletic with respect to amazonicus, duellmani, fantasticus, reticulatus, variabilis, and an unidentified species reported by Santos et al. (2003), which, in combination with patterns of patristic distances for all data and pairwise distances for cytochrome b sequences, suggests this taxon includes five species. Although femoralis, fraterdanieli, hahneli, and nexipus were all recovered as monophyletic, pairwise cytochrome b distances and patristic distances for all data suggest they may include multiple species (femoralis: 8 spp.; fraterdanieli: 2 spp.; hahneli: 2 spp.; nexipus: 2 spp.). Grant and Rodríguez (2001) speculated that trilineatus may be a complex of species, which is borne out by the current analysis; we identify provisionally eight species in the trilineatus clade. Similarly, although the diversity of small, cryptically colored Amazonian frogs is greater than the current taxonomy indicates, and the names proposed by Morales (2002 “2000”) are available to associate with several of these species, Morales's taxonomy is difficult to apply because it treated some specimens of distantly related species as conspecific and some conspecific specimens as different taxa. Although we have referred populations to Morales's names, this is provisional and topotypic material must be examined to clarify identities. Specimens from Guyana that are mophologically indistinguishable from degranvillei are not conspecific with a sample from near the type locality in French Guiana. Among controversial species, our results confirm the independence of baeobatrachus and stepheni. Our evidence also suggests that azureus is nested within, and conspecific with, tinctorius.

As a quick heuristic to help identify species, pairwise comparisons of cytochrome b sequences proved useful, but they are not a panacea. Focusing on well-delimited, uncontroversial species, intraspecific cytochrome b sequence distances ranged from 0.0 to 6.0%. The greatest intraspecific distances were between Nicaraguan and Panamanian samples of lugubris (6.0%) and talamancae (5.7%). The localities for these pairs of samples are also separated by large geographic distance, but the cytochrome b sequences of auratus samples from Nicaragua and Panama are identical. Similarly, we expected evidence to indicate that the widespread and phenotypically variable Amazonian species trivittatus is composed of multiple species, as was found in femoralis; however, trivittatus DNA sequences are relatively homogeneous across its distribution, suggesting the existence of a single species. Minimally, these results highlight the pitfalls of generalizing across taxa and suggest caution in interpeting pairwise comparisons alone.

Among closely related species of unproblematic identity, the least interspecific cytochrome b distance is 2.3% and 3.9% in the auratus–truncatus and vicentei–pumilio pairs, respectively. Among putative sister-species pairs, the greatest cytochrome b distance is 18.6% between castaneoticus and galactonotus. Given the degree of morphological divergence between these species, this is unsurprising. However, it is only slightly greater than that observed between the morphologically more similar (but more distantly related) castaneoticus and quinquivittatus (17.4%). As mentioned above, the Central American species flotator and nubicola were considered conspecific until recently (Ibáñez and Smith, 1995), yet they are not each other's closest relatives and their pairwise cytochrome b distance is 18.4%. Whether these differences in pairwise distances between closely related species are due to incomplete taxon sampling (i.e., they are not as closely related as they were presumed to be) or variation in evolutionary rates is unknown.

A Monophyletic Taxonomy

Preliminary Considerations

Evolutionary relationships are the explanatory framework that unifies all areas of biology, and the results of the present study provide a coherent foundation to understand the many fascinating and useful characteristics of dendrobatid frogs. To facilitate understanding and application of the phylogenetic results, we propose a revised taxonomy for dendrobatid frogs that reflects as closely as is presently feasible (see below) current knowledge of phylogeny (figs. 77, 78).

Figure 77

Graphic summary of the proposed higher-level taxonomy of Athesphatanura.


Figure 78

Graphic summary of the proposed taxonomy of Dendrobatoidea.


Our adherence to Linnaean nomenclature and the strictures of the Code (ICZN, 1999) is pragmatic and not intended as a complete endorsement. The imposition of Linnaean ranks is arbitrary and artificial, skewing both thought and analysis as they continue to be treated as identifying objectively equivalent entities, despite pleas to the contrary. If scientific language is to accurately reflect understanding of evolutionary relationships, then it is clear that sooner or later Linnaean nomenclature will have to be abandoned or transformed significantly.

The best known alternative is the PhyloCode (e.g., de Queiroz and Gauthier, 1990), which eliminates ranks. However, the PhyloCode also institutes a number of conventions that would, should they be adopted, surely impede scientific progress, for example, by increasing the frequency with which minor changes in topology would lead to extreme changes in taxonomy (i.e., nomenclatural instability). Consider, for example, that the finding of Darst and Cannatella (2004; see also Faivovich et al., 2005; Frost et al., 2006) that hemiphractines are distant relatives of other hylids makes Ford and Cannatella's (1993) node-based definition of Hylidae apply to all hyloids except brachycephalids (parsimony) or all hyloids (maximum likelihood).

Kluge (2005) recently proposed a novel system to represent phylogeny exactly and eliminate the drawbacks of the Linnaean system without abandoning its strengths (e.g., designation of “types” for bookkeeping purposes, the principle of priority to encourage progress), all or much of which is likely to be implemented (if not endorsed explicitly) simply because it is designed expressly to encourage scientific progress. Indeed, some aspects of his proposal, such as the naming of all clades, may be inevitable by-products of the growth of scientific knowledge, whether the Code is overhauled or not (e.g., by simply shifting ranked names toward the tips, thus pushing the bulk of cladistic structure above the family level where the Code does not apply). Nevertheless, Kluge's proposal has not yet been vetted by the scientific community, and for the immediate need to translate the phylogeny of dendrobatids into a monophyletic taxonomy we continue to apply the existing Code.

Over the past four decades the number of recognized dendrobatid species has exploded from 70 to 247, and there is no indication that discovery of new species in this clade will wane in the foreseeable future. Compared to other vertebrate groups, anuran families are large and cumbersome. Consider, for example, that Frost et al.'s (2006) new taxonomy recognizes only 42 families for approximately 5,000 species of anurans—prior to the Frost et al. update there were only 33 recognized families of anurans. In comparison, current taxonomy recognizes approximately 220 families of birds to accommodate roughly 10,000 species, 500 families of fishes for 28,000 species, and 130 families for roughly the same number of mammals as there are anurans. Indeed, in all of these groups the orders are approximately equivalent to the families in anuran nomenclature (e.g., there are 26 recognized orders of mammals).

This recognition of few families for frogs is not due to an active decision by the herpetological community but rather tradition and the fact that, as exemplified by dendrobatids, much of the diversity of frogs has been discovered so recently and rapidly (over 45% of recognized amphibian species have been named since 1985; D. R. Frost, unpublished data) without any major revamping of the higher-level taxonomy. This is understandable, given that monophyly is more important than the arbitrary ranking of clades, and by that argument there is no need to elevate the rank of the dendrobatid clade. However, the retention of the old family units also results in an underappreciation of diversity and actually obscures patterns of diversification. Insofar as the purpose of naming clades is to facilitate further research, Linnaean ranks, artificial as they are, are a useful means of carving off chunks of diversity for scientific discussion (if this were not the case, then the optimal Linnaean solution to nonmonophyly would always be accretion, with recognized taxa growing ever larger and obscuring more phylogenetic structure as knowledge increases), and in this sense anuran taxonomy is much less refined than in other vertebrate groups. We therefore elevate the rank of the dendrobatid clade to superfamily (Dendrobatoidea) and propose a new arrangement of families, subfamilies, and genera to better reflect the diversity and phylogeny of this clade.

For taxonomic purposes, we have examined specimens of all but a few species of dendrobatids, but available material of many species was not adequate to permit their inclusion in the phylogenetic analysis (genera and species included are listed below in boldface). We therefore refer them to genera provisionally as both an efficient means of summarizing what is known about those species and as explicit phylogenetic hypotheses to be tested in future studies. To permit provisional reference of species that were not included, we fit names to the cladogram somewhat loosely, that is, names refer to demonstrably monophyletic groups, but much of the finer cladistic structure remains unnamed. This was done as a working compromise between two extreme alternatives.

The two alternatives are (1) to maintain the status quo until knowledge is “sufficiently complete” to merit taxonomic revision by allowing all species to be placed with certainty, or (2) to propose a new taxonomy for the species included in this analysis and treat all others as incertae sedis. Alternative (1) is tantamount to a plea for ignorance and is antiscientific. There is no objective basis for determining when any system of scientific knowledge is “sufficiently complete” for any purpose. It is a fundamental characteristic of science that future evidence (or discovery operations) may overturn any prior hypothesis, and rejecting current knowledge simply because it may ultimately be wrong would prevent all progress. Alternative (2) is equally unsatisfactory because it effectively hides the evidence that already exists regarding the relationships of those species. New taxonomies build upon prior ones, and those prior ones had some empirical basis, however limited. Finally, provisional placement facilitates content increasing progressive problem shifts (sensu Lakatos, 1978) by increasing the testability of phylogenetic hypotheses (logically, the more species included in the hypotheses, the greater the potential to falsify it) and, further, by facilitating alpha taxonomy and the discovery of new species. For example, in the current system, a new species of Colostethus should, in principle, be compared to ca. 120 species ranging from Nicaragua to southeastern Brazil and Bolivia. Most taxonomists are regional specialists and lack the resources to undertake such comparisons, which frequently leads to extensive errors by either referring to different species under the same name or naming species that are not diagnosable in a broader context. A taxonomy that reflects current knowledge of phylogeny will point to appropriate comparisons and thereby greatly facilitate species-level work.

As noted above (Materials and Methods), we included genotypic and phenotypic data for 13 type species (genus name in parentheses): azureiventris (Cryptophyllobates), bicolor (Phyllobates), femoralis (Allobates), inguinalis (Prostherapis), nocturnus (Aromobates), pulchellus (Phyllodromus), pumilio (Oophaga), reticulatus (Ranitomeya), silverstonei (Phobobates), steyermarki (Minyobates), tinctorius (Dendrobates), tricolor (Epipedobates), and trivittatus (Ameerega). We did not include the type species alboguttatus (Nephelobates), fuliginosus (Hyloxalus), latinasus (Colostethus), or yustizi (Mannophryne) because adequate data were not available. Nevertheless, we included numerous putatively closely related species and made taxonomic changes accordingly. That is, we treated the sampled species as proxies for the type species in the same way that the sampled species were treated as representative of the complete diversity of dendrobatids. In both cases, further sampling may prove these assumptions to be false, but in the meantime it is better to present a taxonomy derived from a hypothesis of relationships supported by evidence that can form the basis for future testing than to retain the current taxonomy that misrepresents current understanding of phylogeny. In the following accounts, species included in the phylogenetic analysis are in bold.

Unless otherwise noted, only named species are listed in the following accounts. Additional undescribed species of several taxa were included in the phylogenetic analysis, but these are discussed only in the Comment sections. For each named clade we report a summary of the unambiguous transformations (including phenotypic synapomorphies and branch length) and Bremer support. Additionally, generic acounts include a standardized diagnosis designed to allow species to be referred to taxa efficiently following the examination of few, conspicuous, and, insofar as is possible, easily accessible characters, as well as generalities that are taxonomically useful but difficult to individuate as hypotheses of homology. The purpose of these general characterizations is to facilitate rapid identification, and, as such, descriptions are much less precise than in the delimitation and analysis of transformation series. Unambiguous molecular transformations for each named group are given in appendix 8.

The Higher-level Taxonomy of Athesphatanura Frost et al., 2006

As noted above, the higher-level relationships within Athesphatanura differ from the Frost et al. (2006) hypothesis in two major ways: (1) Leptodactylidae is divided into two distantly related groups, neither of which is the sister of Centrolenidae (i.e., Diphyabatrachia is refuted); one is placed as the sister to all Leptodactyliformes except Centrolenidae, and the other is placed inside Hesticobatrachia as the sister to bufonids, dendrobatids, and hylodines. (2) Thoropa (and therefore Thoropidae) move from being the sister of dendrobatids to be nested among cycloramphids, while the hylodines move from being placed with cycloramphids to being the sister of dendrobatoids. Where regulated taxa were concerned, we applied the Code. However, there are different ways in which changes to unregulated taxa can be incorporated, with the opposing strategies focusing on content (i.e., any change in content entails a different phylogenetic hypothesis and, therefore, demands a different name) and nomenclatural stability (i.e., changes in content entail only a reformulation of the existing name and do not necessitate a new name). There are merits to both approaches, and we devised a taxonomy that introduces the fewest new names required to accomodate the new phylogenetic structure and reformulates the remaining groups. Specifically, we propose new names for the additional nodes created by the movement of the leptodactylids, and we reformulate Chthonobatrachia and Hesticobatrachia (which are otherswise unchanged) to include Leiuperidae, and Agastorophrynia to include Hylodidae and exclude Thoropa.

The higher-level taxonomy of Athesphatanura is summarized in table 36. The paraphyly of Hylidae in our analysis owes to having rooted on the hylid species Hypsiboas boans. Insofar as our results do not otherwise differ from Frost et al. (2006) with regard to Hylidae or Leptodactyliformes, we omit those taxa from the following accounts.

Table 36

The Higher-Level Taxonomy of Athesphatanura Frost et al., 2006i0003-0090-299-1-1-t36.gif

Family: Centrolenidae Taylor, 1951

Centrolenidae Taylor, 1951. Type genus: Centrolene Jiménez de la Espada, 1872.

Allophrynidae Goin et al., 1978: 240. Type genus: Allophryne Gaige, 1926.

Immediately more inclusive taxon

Leptodactyliformes Frost et al., 2006.

Sister group

Cruciabatrachia new taxon.

Content (4 Genera)

Allophryne Gaige, 1926; “Centrolene” Jiménez de la Espada, 1872; Cochranella Taylor, 1951; Hyalinobatrachium Ruiz-Carranza and Lynch, 1991.

Characterization, diagnosis, and support

Branch length  =  39. Bremer support  =  15. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Tropical southern Mexico to Bolivia, northeastern Argentina, and southeastern Brazil.


Our optimal hypothesis of centrolenid relationships is identical to that of Frost (2006), and we direct the reader to that paper for comments. Our results differ in the placement of Centrolenidae relative to other hyloid lineages. Frost et al. (2006) found Centrolenidae to be grouped with Leptodactylidae, together forming Diphyabatrachia. We did not find support for Diphyabatrachia in the present analysis, and instead found Centrolenidae to be the sister group of a large, predominantly Neotropical radiation, named below.

Unranked Taxon: Cruciabatrachia New Taxon

Immediately more inclusive taxon

Leptodactyliformes Frost et al., 2006.

Sister group

Centrolenidae Taylor, 1951.


Leptodactylidae Werner, 1896 (1838); Chthonobatrachia Frost et al., 2006.

Characterization, diagnosis, and support

Branch length  =  21. Bremer support  =  14. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Cosmopolitan in temperate and tropical areas (but with most species concentrated in the southern hemisphere) except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


Cruciabatrachia, frogs of the Southern Cross (Crux), referencing the predominantly southern range of this clade.


Following Frost et al. (2006), we include Somuncuria in this group provisionally on the basis of the evidence suggested by Lynch (1978), who placed Somuncuria as the sister taxon of Pleurodema.

Family: Leptodactylidae Werner, 1896 (1838)

Cystignathi Tschudi, 1838. Type genus: Cystignathus Wagler, 1830.

Plectromantidae Mivart, 1869, Proc. Zool. Soc. London, 1869: 291. Type genus: Plectromantis Peters, 1862.

Adenomeridae Hoffmann, 1878, In Bronn (ed.). Type genus: Adenomera Steindachner, 1867.

Leptodactylidae Werner, 1896. Type genus: Leptodactylus Fitzinger, 1826.

Immediately more inclusive taxon

Cruciabatrachia new taxon.

Sister group

Agastorophrynia Frost et al., 2006.

Content (4 Genera)

Hydrolaetare Gallardo, 1963; Leptodactylus Fitzinger, 1826; Paratelmatobius Lutz and Carvalho, 1958; Scythrophrys Lynch, 1971

Characterization, diagnosis, and support

Branch length  =  42. Bremer support  =  14. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Tropical Mexico throughout Central and South America.


Following Frost et al. (2006), we place Hydrolaetare in this group because of its presumed close relationship to Leptodactylus (Heyer, 1970), although we suggest that this proposition merits further study.

Unranked Taxon: Chthonobatrachia Frost Et Al., 2006

Immediately more inclusive taxon

Cruciabatrachia new taxon.

Sister group

Leptodactylidae Werner, 1896 (1838).


Ceratophryidae Tschudi, 1838; Hesticobatrachia Frost et al., 2006.

Characterization, diagnosis, and support

Branch length  =  27. Bremer support  =  14. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Cosmopolitan in temperate and tropical areas except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


Our Chthonobatrachia is equivalent to the group proposed by Frost et al. (2006) with the inclusion of Leiuperidae Bonaparte, 1850, which was previously in the synonymy of Leptodactylidae. See Frost et al. (2006) for further comments.

Family: Ceratophryidae Tschudi, 1838

Ceratophrydes Tschudi, 1838: 26. Type genus: Ceratophrys Wied-Neuwied, 1824.

Immediately more inclusive taxon

Chthonobatrachia Frost et al., 2006.

Sister group

Hesticobatrachia Frost et al., 2006.

Content (3 Subfamilies)

Batrachylinae Gallardo, 1965; Ceratophryninae Tschudi, 1838; Telmatobiinae Fitzinger, 1843.

Characterization, diagnosis, and support

Branch length  =  30. Bremer support  =  19. No phenotypic character-states optimize unambiguously to this node (see appendix 8 for diagostic DNA sequence transformations). See Frost et al. (2006) for discussion of additional synapomorphies.


Andean and tropical lowland South America from Colombia and Venezuela south to extreme southern Argentina and Chile.


Frost et al. (2006) recognized two subfamilies within Ceratophryidae: Telmatobiinae (for Telmatobius) and Ceratophryinae (for the tribes Batrachylini and Ceratophryini). Our results show Ceratophryinae to be paraphyletic with respect to Telmatobiinae. Rather than dissolve all of these groups in the synonymy of Ceratophryidae, below we elevate Batrachylini and Ceratophryini of Frost et al. (2006) to subfamilies.

Subfamily: Batrachylinae Gallardo, 1965

Batrachylinae Gallardo, 1965: 83. Type genus: Batrachylus Bell, 1843.

Immediately more inclusive taxon

Ceratophryidae Tschudi, 1838.

Sister group

Unnamed group composed of Ceratophryinae Tschudi, 1838; Telmatobiinae Fitzinger, 1843.

Content (2 Genera)

Atelognathus Lynch, 1978 and Batrachyla Bell, 1843.

Characterization, diagnosis, and support

Branch length  =  44. Bremer support  =  25. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Central to extreme southern Argentina and Chile (Patagonia).


This taxon is equal to Batrachylini of Frost et al. (2006).

Subfamily: Ceratophryinae Tschudi, 1838

Ceratophrydes Tschudi, 1838: 26. Type genus: Ceratophrys Wied-Neuwied, 1824; Stombinae Gallardo, 1965: 82. Type genus: Stombus Gravenhorst, 1825.

Immediately more inclusive taxon

Ceratophryidae Tschudi, 1838.

Sister group

Telmatobiinae Fitzinger, 1843.

Content (3 Genera)

Ceratophrys Wied-Neuwied, 1824; Chacophrys Reig and Limeses, 1963; Lepidobatrachus Budgett, 1899.

Characterization, diagnosis, and support

Branch length  =  47. Bremer support  =  40. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Andean and tropical lowlands of South America from Colombia and Venezuela south to extreme southern Argentina and Chile.


This taxon is equal to Ceratophryini of Frost et al. (2006).

Subfamily: Telmatobiinae Fitzinger, 1843

Telmatobii Fitzinger, 1843: 31. Type genus: Telmatobius Wiegmann, 1834.

Immediately more inclusive taxon

Ceratophryidae Tschudi, 1838.

Sister group

Ceratophryinae Tschudi, 1838.

Content (1 Genus)

Telmatobius Wiegmann, 1834.

Characterization, diagnosis, and support

Branch length  =  73. Bremer support  =  66. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Andean South America, Ecuador to Chile and Argentina.


This taxon is equal to Telmatobiinae of Frost et al. (2006).

Unranked Taxon: Hesticobatrachia Frost Et Al., 2006

Immediately more inclusive taxon

Chthonobatrachia Frost et al., 2006.

Sister group

Ceratophryidae Tschudi, 1838.


Cycloramphidae Bonaparte, 1850; Calamitophrynia new taxon.

Characterization, diagnosis, and support

Branch length  =  26. Bremer support  =  14. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Cosmopolitan in temperate and tropical areas except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


Our Hesticobatrachia is equivalent to the group proposed by Frost et al. (2006) with the inclusion of Leiuperidae Bonaparte, 1850, which was previously in the synonymy of Leptodactylidae. See Frost et al. (2006) for further comments.

Family: Cycloramphidae Bonaparte, 1850

Cyclorhamphina Bonaparte, 1850. Type genus: Cycloramphus Tschudi, 1838.

Rhinodermina Bonaparte, 1850. Type genus: Rhinoderma Duméril and Bibron, 1841.

Alsodina Mivart, 1869. Type genus: Alsodes Bell, 1843.

Grypiscina Mivart, 1869. Type genus: Grypiscus Cope, 1867 “1866”.

Odontophrynini Lynch, 1969. Type genus: Odontophrynus Reinhardt and Lütken, 1862 “1861”. (Odontophrynini subsequently named more formally by Lynch, 1971: 142.)

Thoropidae Frost et al., 2006. Type genus: Thoropa Cope, 1865.

Immediately more inclusive taxon

Hesticobatrachia Frost et al., 2006.

Sister group

Calamitophrynia new taxon.


Southern tropical and temperate South America.

Content (12 Genera)

Alsodes Bell, 1843; Crossodactylodes Cochran, 1938; Cycloramphus Tschudi, 1838; Eupsophus Fitzinger, 1843; Hylorina Bell, 1843; Limnomedusa Fitzinger, 1843; Macrogenioglottus Carvalho, 1946; Odontophrynus Reinhardt and Lütken, 1862 “1861”; Proceratophrys Miranda-Ribeiro, 1920; Rhinoderma Duméril and Bibron, 1841; Thoropa Cope, 1865; Zachaenus Cope, 1866.

Characterization, diagnosis, and support

Branch length  =  32. Bremer support  =  8. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). See Frost et al. (2006) for discussion of synapomorphies.


Southern tropical and temperate South America.


Cycloramphidae, as defined here, differs from Cycloramphidae sensu Frost et al. (2006) in that it excludes Hylodinae (see below) and includes Thoropa (and therefore Thoropidae). We do not recognize subfamilial divisions within Cycloramphidae. See Frost et al. (2006) for additional discussion.

Unranked Taxon: Calamitophrynia New Taxon

Immediately more inclusive taxon

Hesticobatrachia Frost et al., 2006.

Sister group

Cycloramphidae Bonaparte, 1850.


Leiuperidae Bonaparte, 1850; Agastorophrynia Frost et al., 2006.

Characterization, diagnosis, and support

Branch length  =  27. Bremer support  =  20. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Cosmopolitan in temperate and tropical areas except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


Calamitophrynia, from the Greek calamitas (misfortune), phryne (toad), and suffix ia (having the nature of), referencing the loud calls of so many of the species in this clade.


The placement of Leiuperidae at such cladistic distance from Leptodactylidae was unexpected and contradicts the findings of Frost et al. (2006).

Family: Leiuperidae Bonaparte, 1850

Leiuperina Bonaparte, 1850. Type genus: Leiuperus Duméril and Bibron, 1841.

Paludicolina Mivart, 1869. Type genus: Paludicola Wagler, 1830.

Pseudopaludicolinae Gallardo, 1965. Type genus: Pseudopaludicola Miranda-Ribeiro, 1926.

Immediately more inclusive taxon

Calamitophrynia new taxon.

Sister group

Agastorophrynia Frost et al., 2006.

Content (7 Genera)

Edalorhina Jiménez de la Espada, 1871 “1870”; Engystomops Jiménezde la Espada, 1872; Eupemphix Steindachner, 1863; Physalaemus Fitzinger, 1826; Pleurodema Tschudi, 1838; Pseudopaludicola Miranda-Ribeiro, 1926; Somuncuria Lynch, 1978.

Characterization, diagnosis, and support

Branch length  =  36. Bremer support  =  23. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Mexico throughout Central and South America.


Following Frost et al. (2006), we include Somuncuria in this group provisionally on the basis of the evidence suggested by Lynch (1978), who placed Somuncuria as the sister taxon of Pleurodema.

Unranked Taxon: Agastorophrynia Frost Et Al., 2006

Immediately more inclusive taxon

Hesticobatrachia Frost et al., 2006.

Sister group

Leiuperidae Bonaparte, 1850.


Bufonidae Gray, 1825; Nobleobatia new taxon.

Characterization, diagnosis, and support

Branch length  =  22. Bremer support  =  5. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8). Although the origin of diurnal activity (Character 115) is ambiguous due to incomplete coding of phenotypic characters in the present analysis, it is a likely synapomorphy for Agastorophrynia. Melanophryniscus, Dendrophryniscus, Atelopus, and almost all species of Nobleobatia new taxon are diurnal, whereas leiuperids, cycloramphids, ceratophryids, leptodactylids, centrolenids, and hylids are entirely or predominantly nocturnal.


Cosmopolitan in temperate and tropical areas except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


Agastorophrynia here is equivalent to that of Frost et al. (2006) with the inclusion of their Hylodinae, which is removed from their Cycloramphidae, and the exclusion of Thoropidae, which is found to be a junior synonym of Cycloramphidae.

Family: Bufonidae Gray, 1825

Bufonina Gray, 1825. Type genus: Bufo Laurenti, 1768.

Atelopoda Fitzinger, 1843. Type genus: Atelopus Duméril and Bibron, 1841.

Phryniscidae Günther, 1858. Type genus: Phryniscus Wiegmann, 1834.

Adenomidae Cope, 1861 “1860”. Type genus: Adenomus Cope, 1861.

Dendrophryniscina Jiménezde la Espada, 1871 “1870”. Type genus: Dendrophryniscus Jiménez de la Espada, 1871 “1870”.

Platosphinae Fejérváry, 1917. Type genus: Platosphus d'Isle, 1877 (fossil taxon considered to be in this synonymy because Platosophus  =  Bufo sensu lato).

Bufavidae Fejérváry, 1920. Type genus: Bufavus Portis, 1885 (fossil taxon considered to be in this synonymy because Bufavus  =  Bufo sensu lato).

Tornierobatidae Miranda-Ribeiro, 1926. Type genus: Tornierobates Miranda-Ribeiro, 1926.

Nectophrynidae Laurent, 1942. Type genus: Nectophryne Buchholz and Peters, 1875.

Stephopaedini Dubois, 1987 “1985”. Type genus: Stephopaedes Channing, 1978.

Immediately more inclusive taxon

Agastorophrynia Frost et al., 2006.

Sister group

Nobleobatia new taxon.

Content (47 Genera)

Adenomus Cope, 1861 “1860”; Altiphrynoides Dubois, 1987 “1986”; Amietophrynus Frost et al., 2006; Andinophryne Hoogmoed, 1985; Anaxyrus Tschudi, 1845; Ansonia Stoliczka, 1870; Atelophryniscus McCranie, Wilson, and Williams, 1989; Atelopus Duméril and Bibron, 1841; Bufo Laurenti, 1768; Bufoides Pillai and Yazdani, 1973; Capensibufo Grandison, 1980; Chaunus Wagler, 1828; Churamiti Channing and Stanley, 2002; Cranopsis Cope, 187 “1876”; Crepidophryne Cope, 1889; Dendrophryniscus Jiménez de la Espada, 1871 “1870”; Didynamipus Andersson, 1903; Duttaphrynus Frost et al., 2006; Epidalea Cope, 1865; Frostius Cannatella, 1986; Ingerophrynus Frost et al., 2006; Laurentophryne Tihen, 1960; Leptophryne Fitzinger, 1843; Melanophryniscus Gallardo, 1961; Mertensophryne Tihen, 1960; Metaphryniscus Señaris, Ayarzagüena, and Gorzula, 1994; Nannophryne Günther, 1870; Nectophryne Buchholz and Peters, 1875; “NectophrynoidesNoble, 1926; Nimbaphrynoides Dubois, 1987 “1986”; Oreophrynella Boulenger, 1895; Osornophryne Ruiz-Carranza and Hernández-Camacho, 1976; Parapelophryne Fei, Ye, and Jiang, 2003; Pedostibes Günther, 1876 “1875”; Pelophryne Barbour, 1938; Peltophryne Fitzinger, 1843; Phrynoidis Fitzinger, 1843; Pseudobufo Tschudi, 1838; Pseudepidalea Frost et al., 2006; Rhaebo Cope, 1862; Rhamphophryne Trueb, 1971; Rhinella Fitzinger, 1826; Schismaderma Smith, 1849; Truebella Graybeal and Cannatella, 1995; Vandijkophrynus Frost et al., 2006; Werneria Poche, 1903; “Wolterstorffina” Mertens, 1939.

Characterization, diagnosis, and support

Branch length  =  78. Bremer support  =  37.

Unambiguously optimized phenotypic synapomorphies of this clade are (1) origin of spiculate skin texture (Character 0, 1 → 3), (2) m. depressor mandibulae origin not extending posterior to the squamosal (Character 73, 1 → 0), (3) loss of m. intermandibularis supplementary element (Character 77, 1 → 0), (4) origin of median gap in marginal papillae of lower labium of tadpoles (Character 92, 0 → 1), (5) epicoracoids overlapping from level between posterior level of procoracoids and anterior ends of coracoids to posterior level of coracoids (Character 120, 1 → 2), (6) prezonal element of pectoral girdle (omosternum) absent (Character 123, 1 → 0), (7) maxillary teeth absent (Character 139, 1 → 0), (8) reduction of chromosome number to 22 (Character 173, 4 → 2). See Frost et al. (2006) for discussion of additional synapomorphies.


Cosmopolitan in temperate and tropical areas except for the Australo-Papuan region, Madagascar, Seychelles, and New Zealand.


The internal structure of Bufonidae is identical to that of Frost et al. (2006) for the terminals sampled, as is the placement of Bufonidae relative to other groups, the exception being the content of the sister group, named below. See Frost et al. (2006) for further comments and structure within Bufonidae.

Unranked Taxon: Nobleobatia New Taxon

Immediately more inclusive taxon

Calamitophrynia new taxon.

Sister group

Bufonidae Gray, 1825.


Hylodidae Günther, 1858; Dendrobatoidea Cope, 1865.

Characterization, diagnosis, and support

Branch length  =  53. Bremer support  =  25.

Unambiguously optimized phenotypic synapomorphies of this clade are (1) origin of paired dorsal digital scutes (Character 1, 0 → 1), (2) differentiation of digital discs (Character 6, 0 → 3), (3) reduction of postaxial webbing of toe II (Character 39, 3 → 1/2), (4) reduction of postaxial webbing of toe III (Character 41, 3 → 2), (5) reduction of preaxial webbing of toe IV (Character 42, 3 → 2), (6) reduction of postaxial webbing of toe IV (Character 43, 4 → 1), (7) reduction of pretaxial webbing of toe V (Character 44, 4 → 1), (8) origin of the pale oblique lateral stripe (Character 55, 0 → 1), (9) origin of T-shaped terminal phalanges (Character 118, 1 → 0).


Most of tropical Central and South America and Atlantic forest of Brazil.


Nobleobatia, formed from Noble (a surname) and batia (from the Greek bates, a walker). We name this taxon in honor of G. K. Noble, who, in 1926, was the first to propose the immediate relationship between the genera now referred to Hylodidae and Dendrobatoidea.


The strongly supported placement of dendrobatids and hylodids as sister taxa corroborates one of the most controversial hypotheses of anuran relationships.

Family: Hylodidae GÜnther, 1858

Hylodinae Günther, 1858. Type genus: Hylodes Fitzinger, 1826.

Elosiidae Miranda-Ribeiro, 1923. Type genus: Elosia Tschudi, 1838.

Immediately more inclusive taxon

Nobleobatia new taxon.

Sister group

Dendrobatoidea Cope, 1865.


Southern tropical and temperate South America.

Content (3 Genera)

Crossodactylus Duméril and Bibron, 1841; Hylodes Fitzinger, 1826; Megaelosia Miranda-Ribeiro, 1923.

Characterization, diagnosis, and support

Branch length  =  113. Bremer support  =  74.

Unambiguously optimized phenotypic synapomorphies of this clade are (1) origin of preaxial fringe on finger II (Character 13, 0 → 1), (2) origin of preaxial fringe on finger III (Character 15, 0 → 1), (3) origin of tarsal fringe (Character 30, 0 → 1), (4) origin of preaxial fringe on toe I (Character 36, 0 → 1), (5) origin of fringe on postaxial fringe on toe V (Character 45, 0 → 1), (6) loss of oocyte pigmentation (Character 68, 1 → 0), (7) loss of fibers of m. depressor mandibulae originating from the annulus tympanicus (Character 74, 1 → 0), (8) origin of paired lateral vocal sacs (Character 76, 1 → 2), (9) gain of lateral line stitches (Character 98, 0 → 1).


The Atlantic forest of Brazil.


Frost et al. (2006) found these genera to form a clade nested within Cycloramphidae, which they referred to as Hylodinae.

Superfamily: Dendrobatoidea Cope, 1865

Phyllobatae Fitzinger, 1843. Type genus: Phyllobates Duméril and Bibron, 1841.

Eubaphidae Bonaparte, 1850. Type genus: Eubaphus Bonaparte, 1831.

Hysaplesidae Günther, 1858. Type genus: Hysaplesia Boie In Schlegel, 1826. (Note that this taxon was named as Hylaplesidae, derived from Hylaplesia, an incorrent subsequent spelling of Hysaplesia.)

Dendrobatidae Cope, 1865. Type genus: Dendrobates Wagler, 1830.

Immediately more inclusive taxon

Agastorophrynia Frost et al., 2006.

Sister group

Hylodidae Günther, 1858.

Content (2 Families)

Dendrobatidae Cope, 1865 and Aromobatidae new family.

Characterization, diagnosis, and support

Branch length  =  178. Bremer support  =  159.

Unambiguous phenotypic transformations are (1) (Character 2, 0 → 1), (2) gain of the tarsal keel (Character 28, 0 → 1), (3) origin of the metatarsal fold (Character 46, 0 → 1) (4) the “ranid” type insertion of the distal tendon of insertion of the m. semitendinosus (Character 69, 0 → 1), (5) gain of the m. semitendinosus binding tendon (Character 70, 0 → 1), (6) occurrence of the dorsal flap of the m. depressor mandibulae (Character 72, 0 → 1), (7) m. depressor mandibulae overlapping posterodorsal portion of tympanum (Character 75, 0 → 1), (8) orientation of the m. intermandibularis supplementary element (Character 78, 0 → 1), (9) loss of reproductive amplexus (Character 105, 1 → 0), (10) dorsal tadpole transport (Character 109, 0 → 1), (11) origin of toe trembling (Character 116, 0 → 1), (12) complete fusion of epicoracoids (Character 119, 1/2 → 0), (13) nonoverlapping epicoracoids (Character 120, 1 → 0), (14) medial ossification of prezonal element (omosternum) of pectoral girdle (Character 127, 0 → 1), (15) maxillary teeth nonpedicellate (Character 140, 0 → 1), (16) occurrence of the retroarticular process of the mandible (Character 142, 0 → 1), and (17) reduction in chromosome number from 26 to 24 (Character 173, 4 → 3).

Additional characteristics useful in diagnosing dendrobatoids are the occurrence of dorsal scutes on the digital tip, shared only with the sister clade Hylodidae (among Neotropical frogs).


Dendrobatoid frogs occur throughout large parts of Nicaragua, Costa Rica, Panama, Colombia, Ecuador, Peru, Bolivia, Venezuela, Guyana, Suriname, French Guiana, Brazil, and the Lesser Antilles.


As discussed above, the elevation of the dendrobatid clade to superfamily is proposed to allow more information on the phylogeny and biology of the group to be conveyed in the working taxonomy. To maintain rank equivalency, Dubois (1992) recognized Dendrobatoidae as an epifamily (redundant with Dendrobatidae) within the superfamily Ranoidea. Frost et al. (2006) applied Dendrobatoidea to the clade of dendrobatids + Thoropa (i.e., Dendrobatidae sensu lato + Thoropidae). In the present analysis Thoropa is nested among cycloramphids, and the sister group of dendrobatids is the cycloramphid subfamily Hylodinae (Crossodactylus, Hylodes, and Megaelosia). In recognition of the placement of the hylodine genera outside of Cycloramphidae, we recognize them as a family, Hylodidae Günther, 1858 (see above).

Gross examination reveals fused, nonoverlapping epicoracoid cartilages (i.e., firmisterny in the traditional sense) in dendrobatoids, although histological study has shown this to differ in one species (Noble, 1926; Kaplan, 1995; see also Kaplan, 2004).

The Taxonomy of Dendrobatoidea Cope, 1865

Below we propose a formal taxonomy for dendrobatoid frogs. The structure of this taxonomy is outlined in table 37, and the placement of all species is detailed in appendix 1.

Table 37

The Taxonomy of Dendrobatoidea Cope, 1865i0003-0090-299-1-1-t37.gif

Family: Aromobatidae New Family

Aromobatidae new family. Type genus: Aromobates Myers, Daly, and Paolillo, 1991.

Immediately more inclusive taxon


Sister group


Content (2 Subfamilies)

Anomaloglossinae new subfamily; Aromobatinae new subfamily.

Characterization, diagnosis, and support

Branch length  =  71. Bremer support  =  41. No phenotypic character-states optimize unambiguously to this node (see appendix 8 for diagostic DNA sequence transformations).


Central and South America and the Lesser Antilles, with most species occurring on the eastern slopes of the Andes, throughout the Amazon region, and in the Atlantic forest of Brazil.


Insofar as is known, all species of Aromobatidae lack the ability to sequester alkaloids.

Subfamily: Anomaloglossinae New Subfamily

Anomaloglossinae. Type genus: Anomaloglossus new genus.

Immediately more inclusive taxon


Sister group

Aromobatinae new subfamily.


Anomaloglossus new genus and Rheobates new genus.

Characterization, diagnosis, and support

Branch length  =  45. Bremer support  =  16.

Unambiguously optimized phenotypic synapomorphies are (1) fringe present on preaxial surface of finger II (Character 13, 0 → 1), (2) fringe present on preaxial surface of finger III (Character 15, 0 → 1), (3) toe disc II moderately expanded (Character 32, 1 → 2), (4) fringe on preaxial side of toe I present (Character 36, 0 → 1), (5) distal 1.5 phalanges of postaxial side of toe I free of webbing (Character 37, 2 → 3/4), (6) fringe present on postaxial side of toe V (Character 45, 0 → 1), (7) strong metatarsal fold (Character 46, 1 → 2), (8) male abdomen with irregular (clumped) stippling or faint, diffuse spotting (Character 63, 3 → 4).


Almost exclusively cis-Andean, with most species in eastern Amazonia, the Orinoco drainage, and tepui regions. Three species also occur on the Pacific slopes of Colombia and Ecuador.

Genus: Anomaloglossus New Genus

Anomaloglossus new genus. Type species: Colostethus beebei Noble, 1923.

Immediately more inclusive taxon

Anomaloglossinae new subfamily.

Sister Group

Rheobates new genus.

Content (19 Species)

Anomaloglossus atopoglossus (Grant, Humphrey, and Myers, 1997) new combination; A. ayarzaguenai (La Marca, 1998 “1996”) new combination; A. baeobatrachus (Boistel and Massary, 1999 new combination; A. beebei (Noble, 1923) new combination; A.chocoensis” auctorum (not of Boulenger, 1912; see Grant et al., 1997); A. degranvillei (Lescure, 1975) new combination; A. guanayensis (La Marca, 1998 “1996”) new combination; A. lacrimosus (Myers, 1991) new combination; A. murisipanensis (La Marca, 1998 “1996”) new combination; A. parimae (La Marca, 1998 “1996”) new combination; A. parkerae (Meinhardt and Parmelee, 1996) new combination; A. praderioi (La Marca, 1998 “1996”) new combination; A. roraima (La Marca, 1998 “1996”) new combination; A. shrevei (Rivero, 1961) new combination; A. stepheni (Martins, 1989) new combination; A. tamacuarensis (Myers and Donnelly, 1997) new combination; A. tepuyensis (La Marca, 1998 “1996”) new combination; A. triunfo (Barrio-Amorós, Fuentes, and Rivas, 2004) new combination; A. wothuja (Barrio-Amorós, Fuentes, and Rivas, 2004) new combination.

Characterization, diagnosis, and support

Branch length  =  77. Bremer support  =  35.

The only unambiguously optimized phenotypic synapomorphies of this clade is (1) the unique and unreversed origin of the median lingual process (Character 79, 0 → 1).

Other characteristics include: (1) Dorsal coloration cryptic, brown or gray; (2) pale oblique lateral stripe present or absent; (3) pale dorsolateral stripe present or absent; (4) pale ventrolateral stripe absent; (5) dorsal skin texture posteriorly granular; (6) toe webbing basal to extensive; (7) third finger of adult males swollen or not; (8) finger I shorter than finger II; (9) finger discs weakly expanded; (10) median lingual process present; (11) larval vent tube usually dextral; (12) larval oral disc shape usually “normal” (not umbelliform), variably reduced in endotrophic species; (13) larval oral disc emarginate (variably reduced in endotrophic species); (14) lipophilic alkaloids absent; (15) chromosome number 2N  =  24 (known in Anomaloglossus stepheni); (16) testes unpigmented or medially pigmented; (17) dark throat collar absent.


Most species are cis-Andean; among these, none occurs west or south of the region of Manaus, and there is a large number of tepui species. Three species also occur on the Pacific slopes of Colombia and Ecuador.


Anomaloglossus, formed from the Greek anomalos (irregular, unusual) and glossa (tongue), in reference to the unusual tongue bearing the median lingual process. Gender masculine. (This name should not be mistaken for Anomaloglossa Percival, 1978, which is a genus of brachiopod.)


Anomaloglossus is most simply diagnosed by the synapomorphic occurrence of the median lingual process (Grant et al., 1997). Owing to the shared occurrence of the median lingual process (MLP) in the potential sister taxa specified by the Old World ranoid hypothesis of dendrobatid origins (e.g., Ford and Cannatella, 1993) and its absence in all hyloids, Grant et al. interpreted the MLP as symplesiomorphic in dendrobatids. However, Frost et al. (2006) showed decisively that dendrobatoids are not closely related to Old World ranoids, and their MLP is independently derived.

La Marca (1998 “1996”) did not note the presence or absence of the MLP in A. ayarzaguenai, A. guanayensis, A. murisipanensis, or A. parimae, and we have not examined these species; as such, their inclusion in this genus is a prediction based on distribution and their resemblance to MLP-possessing species, and it must be confirmed. The presence of the MLP is confirmed for all other species referred to this genus.

Within Anomaloglossus there are basically two “flavors” of frogs: small, slender frogs with minimal toe webbing (e.g., A. stepheni), and larger, more robust frogs with moderate to extensive toe webbing (e.g., A. tepuyensis). The former group is strictly cis-Andean, whereas the latter occurs east of the Andes and on the Pacific slopes of Colombia and Ecuador. In the present analysis these two groups are reciprocally monophyletic, but greater taxon sampling is required to thoroughly test this result. Similarly, the trans-Andean MLP-possessing species must be included explicitly in phylogenetic analysis to corroborate their placement in Anomaloglossus.

Genus: Rheobates New Genus

Rheobates new genus. Type species: Phyllobates palmatus Werner, 1899.

Immediately more inclusive taxon

Anomaloglossinae new subfamily.

Sister group

Anomaloglossus new genus.

Content (2 Species)

Rheobates palmatus (Werner, 1899) new combination; R. pseudopalmatus (Rivero and Serna, 2000 “1995”) new combination.

Characterization, diagnosis, and support

Branch length  =  28. Bremer support  =  28.14 Insofar as this genus is represented by a single species in this analysis, we cannot distinguish between autapomorphies and synapomorphies and therefore do not report the apomorphic states.

Other characteristics include: (1) Dorsal coloration cryptic, brown or gray; (2) pale oblique lateral stripe present or absent, often more conspicuous in juveniles; (3) pale dorsolateral stripe absent; (4) pale ventrolateral stripe absent; (5) dorsal skin texture posteriorly granular; (6) toe webbing extensive; (7) third finger of adult males not swollen; (8) finger I shorter than finger II; (9) finger discs weakly expanded; (10) median lingual process absent; (11) larval vent tube dextral; (12) larval oral disc shape “normal” (not umbelliform); (13) larval oral disc emarginate; (14) lipophilic alkaloids unknown (presumed absent); (15) chromosome number 2N  =  24 (known in Rheobates palmatus); (16) testes unpigmented; (17) dark throat collar absent.


Colombia, eastern amd westerm slopes of the Cordillera Oriental and across the Magdalena valley on the eastern slope of the Cordillera Central. The elevational distribution extends from ca. 400 m to over 2,000 m.


Rheobates, from the Greek rheo (stream, current) and bates (a walker) in reference to the riparian habitat of the type species R. palmatus.


Phylogenetic analysis showed Rheobates palmatus to be the sister taxon of Anomaloglossus, from which it differs most strikingly in lacking the median lingual process. Otherwise, this taxon most resembles several extensively webbed species of Hyloxalus, from which it differs in having an elongate, robust zygomatic ramus of the squamosal and more extensive toe webbing, and Colostethus, from which it differs in lacking the swollen third finger in adult males. It differs from species of Aromobates in lacking a pale dorsolateral stripe, and from species of Mannophryne in lacking a dark throat collar.

We refer Rheobates pseudopalmatus to this genus provisionally based on Rivero and Serna's (2000 “1995”) assertion that R. palmatus and R. pseudopalmatus are sister species. Nevertheless, we caution that the diagnostic characters provided by Rivero and Serna are inadequate to validate their claim and exclude R. pseudopalmatus from Hyloxalus or Aromobates. That said, the diagnostic differences given by Rivero and Serna are also inadequate to distinguish this species from R. palmatus, and given that the type locality of Amalfi lies within the known distribution of R. palmatus on the eastern slope of the Cordillera Central, it is likely that two taxa are conspecific.

Bernal et al. (2005) recently examined Rheobates palmatus on the Amazonian and western flanks of the Cordillera Oriental and reported bioacoustic and genetic evidence of lineage differentiation between populations on either side of the Andes. Should sampling of additional localities corroborate this finding, names are available: Phyllobates (Hypodictyon) palmatus Werner, 1899 was named from Fusagasugá and would therefore apply to the western slope populations, whereas Hyloxalus granuliventris Boulenger, 1919 was described from “Bogotá” and, although that is a vague locality, a first reviser could apply the name to the eastern slope populations.

Subfamily: Aromobatinae New Subfamily

Aromobatinae new subfamily. Type genus: Aromobates Myers, Paolillo, and Daly, 1991.

Immediately more inclusive taxon

Aromobatidae new family.

Sister group

Anomaloglossinae new subfamily.

Content (2 Genera)

Aromobates and Mannophryne.

Characterization, diagnosis, and support

Branch length  =  54. Bremer support  =  38. All unambiguously optimized synapomorphies for this clade are from DNA sequences (see appendix 8).


Mérida Andes of Venezuela and adjacent Cordillera Oriental of Colombia, Cordillera de la Costa, and Peninsula de Paría, Trinidad and Tobago.


Aromobates and Mannophryne form a morphologically and geographically compact clade.

Genus: Aromobates Myers, Paolillo, And Daly, 1991

Aromobates. Type species Aromobates nocturnus Myers, Paolillo, and Daly, 1991 by original designation.

Nephelobates La Marca, 1994. Type species: Phyllobates alboguttatus Boulenger, 1903 by original designation.

Immediately more inclusive taxon

Aromobatinae new subfamily.

Sister group

Mannophryne La Marca, 1991.

Content (12 Species)

Aromobates alboguttatus (Boulenger, 1903) new combination; A. capurinensis (Péfaur, 1993) new combination; A. duranti (Pefaur, 1985) new combination; A. haydeeae (Rivero, 1978 “1976”) new combination; A. leopardalis (Rivero, 1980 “1978”) new combination; A. mayorgai (River, 1980 “1978”); A. meridensis (Dole and Durant, 1972) new combination; A. molinarii (La Marca, 1985) new combination; A. nocturnus Myers, Paolillo, and Daly, 1991; A. orostoma (Rivero, 1978 “1976”) new combination; A. saltuensis (Rivero 1980 “1978”) new combination; A. serranus (Péfaur, 1985) new combination.

Characterization, diagnosis, and support

Branch length  =  61. Bremer support  =  25. All unambiguously optimized synapomorphies for this clade are from DNA sequences.

Other characteristics include: (1) Dorsal coloration cryptic, brown or gray; (2) pale oblique lateral stripe present or absent; (3) pale dorsolateral stripe present; (4) pale ventrolateral stripe absent; (5) dorsal skin texture posteriorly granular; (6) toe webbing basal to extensive; (7) third finger of adult males not swollen; (8) finger I shorter than, equal to, or longer than finger II; (9) finger discs weakly to moderately expanded; (10) median lingual process absent; (11) larval vent tube dextral; (12) larval oral disc shape “normal” (not umbelliform); (13) larval oral disc emarginate; (14) lipophilic alkaloids absent; (15) chromosome number 2N  =  24 (known in Aromobates leopardalis); (16) testes unpigmented; (17) dark throat collar absent.


Mérida Andes of Venezuela and adjacent Cordillera Oriental of Colombia.


The inclusion of A. capurinensis in this genus is provisional because we have not examined specimens and osteological data (e.g., length of zygomatic ramus) have not been published. Nevertheless, Péfaur's (1993) description called attention to the resemblance of this species to the other species here included in Aromobates, and its distribution at approximately 2,400 m in the Mérida Andes lends indirect support to this relationship.

Genus: Mannophryne La Marca, 1991

Mannophryne La Marca, 1992. Type species: Colostethus yustizi La Marca, 1989 by original designation.

Immediately more inclusive taxon

Aromobatinae new subfamily.

Sister group

Aromobates Myers, Paolillo, and Daly, 1991.

Content (12 Species)

Mannophryne caquetio Mijares-Urrutia and Arends R., 1999; M. collaris (Boulenger, 1912); M. cordilleriana La Marca, “1994” 1995; M. herminae (Boettger, 1893); M. lamarcai Mijares-Urrutia and Arends R., 1999; M. larandina (Yustiz, 1991); M. neblina (Test, 1956); M. oblitterata (Rivero, 1986 “1984”); M. olmonae (Hardy, 1983); M. riveroi (Donoso-Barros, 1965 “1964”); M. trinitatis (Garman, 1887); M. yustizi (La Marca, 1989).

Characterization, diagnosis, and support

Branch length  =  72. Bremer support  =  40.

Phenotypic synapomorphies that optimize unambiguously to this node are (1) presence of a dermal collar (Character 59, 0 → 1) and (2) male abdomen color evenly stippled (Character 63, 3 → 2).

Other characteristics include: (1) Dorsal coloration cryptic, brown; (2) pale oblique lateral stripe present; (3) pale dorsolateral stripe present; (4) pale ventrolateral stripe absent; (5) dorsal skin texture posteriorly granular; (6) toe webbing moderate to extensive; (7) third finger of adult males not swollen; (8) finger I shorter than finger II; (9) finger discs narrow to moderately expanded; (10) median lingual process absent; (11) larval vent tube dextral; (12) larval oral disc shape “normal” (not umbelliform); (13) larval oral disc emarginate; (14) lipophilic alkaloids absent; (15) chromosome number 2N  =  24 (known in Mannophryne herminae, M. olmonae, M. neblina, M. trinitatis); (16) testes unpigmented; (17) dark throat collar present.


Andes, Cordillera de la Costa, and Peninsula de Paría in Venezuela; Trinidad and Tobago.


The content of Mannophryne does not change with this study.

Subfamily: Allobatinae New Subfamily

Immediately more inclusive taxon

Aromobatidae new family.

Sister group

Aromobatinae new subfamily.

Characterization, diagnosis, and support

Branch length  =  56. Bremer support  =  33.

Unambiguously optimized phenotypic synapomorphies of this clade are (1) finger IV reaching distal half of distal subarticular tubercle of finger III (character 4, 0 → 1/2), (2) finger III swollen in adult males (Character 20, 0 → 1), (3) webbing absent on postaxial side of toe I (Character 37, 2 → 0), (4) webbing absent on preaxial side of toe II (Character 38, 1 → 0), (5) webbing absent on postaxial side of toe II (Character 39, 1 → 0), (6) webbing absent on preaxial side of toe III (Character 40, 2 → 0), (7) pale paracloacal mark present (Character 49, 0 → 1), (8) oblique lateral line diffuse (Character 57, 0 → 2), (9) male abdomen pale, free or almost free of melanophores (Character 63, 3 → 0), (10) palatines absent (Character 132, 1 → 0).


South America and the Lesser Antilles. South American species are cis-Andean with two exceptions: (1) Allobates talamancae occurs in the Pacific lowlands of Colombia and Ecuador and north through Central America to Nicaragua, (2) its undescribed sister species A. “Magdalena” from this study, which occurs in the Magdalena Valley of Colombia. A single species (A. chalcopis) occurs in Martinique.


This name is currently redundant with Allobates. However, Allobates is a large, broadly distributed, and heterogeneous clade. Current knowledge is inadequate to name additional genera and assign species not included explicitly in the present analysis, and the need for a functional taxonomy outweighs the need to name additional clades. Given the rapid accumulation of data over the last few years, we anticipate that the paucity of knowledge will be remedied quickly and this large genus will be partitioned as knowledge of its phylogeny accumulates, making Allobatinae an informative name (for recognized species groups, see Comments for Allobates, below). Recognition of Allobatinae is necessitated by the recognition of Aromobatinae for the clade of Aromobates and Mannophryne.

Genus: Allobates Zimmermann And Zimmermann, 1988

Allobates Zimmermann and Zimmermann, 1988. Type species Prostherapis femoralis Boulenger, 1884, by original designation.

Immediately more inclusive taxon

Aromobatidae new family.

Sister group

Aromobatinae new subfamily.

Content (42 Species)

Allobates alagoanus (Bokermann, 1967); A. alessandroi (Grant and Rodríguez, 2001) new combination; A. bromelicola (Test, 1956) new combination; A. brunneus (Cope, 1887) new combination; A. caeruleodactylus (Lima and Caldwell, 2001) new combination; A. capixaba (Bokermann, 1967) new combination; A. carioca (Bokermann, 1967) new combination; A. cepedai (Morales, 2002 “2000”) new combination; A. chalcopis (Kaiser, Coloma, and Gray, 1994) new combination; A. conspicuus (Morales, 2002 “2000”) new combination; A. craspedoceps (Duellman, 2004) new combination; A. crombei (Morales, 2002 “2000”) new combination; A. femoralis (Boulenger, 1883); A. fratinescus (Morales, 2002 “2000”) new combination; A. fuscellus (Morales, 2002 “2000”) new combination; A. gasconi (Morales, 2002 “2000”) new combination; A. goianus (Bokermann, 1975) new combination; A. humilis (Rivero, 1980 “1978”) new combination; A. insperatus (Morales, 2002 “2000”) new combination; A. juanii (Morales, 1994) new combination; A. kingsburyi (Boulenger, 1918) new combination; A. mandelorum (Schmidt, 1932) new combination; A. marchesianus (Melin, 1941) new combination; A. masniger (Morales, 2002 “2000”) new combination; A. mcdiarmidi (Reynolds and Foster, 1992) new combination; A. melanolaemus (Grant and Rodríguez, 2001) new combination; A. myersi (Pyburn, 1981) new combination; A. nidicola (Caldwell and Lima, 2003) new combination; A. olfersioides (Lutz, 1925) new combination; A. ornatus (Morales, 2002 “2000”) new combination; A. picachos (Ardila-Robayo, Acosta-Galvis, and Coloma, 2000 “1999”); A. pittieri (La Marca, Manzanilla, and Mijares-Urrutia, 2004) new combination; A. ranoides (Boulenger, 1918) new combination; A. rufulus (Gorzula, 1990 “1988”); A. sanmartini (Rivero, Langone, and Prigioni, 1986) new combination; A. sumtuosus (Morales, 2002 “2000”) new combination; A. talamancae (Cope, 1875) new combination; A. trilineatus (Boulenger, 1884 “1883”) new combination; A. undulatus (Myers and Donnelly, 2001) new combination; A. vanzolinius (Morales, 2002 “2000”) new combination; A. wayuu (Acosta, Cuentas, and Coloma, 2000 “1999”) new combination; A. zaparo (Silverstone, 1976).

Characterization, diagnosis, and support

As for Allobatinae, above.

Other characteristics include: (1) Dorsal coloration cryptic in most species (brighter in A. femoralis group); (2) pale oblique lateral stripe present in most (but not all) species; (3) pale dorsolateral stripe present or absent; (4) pale ventrolateral stripe present or absent; (5) dorsal skin texture posteriorly granular except in A. femoralis group, which is strongly granular; (6) toe webbing absent to moderate (basal in most species); (7) third finger of adult males swollen or not swollen; (8) finger I longer than finger II in most species (equal or shorter in some); (9) finger discs weakly expanded; (10) median lingual process absent; (11) larval vent tube dextral; (12) larval oral disc shape “normal” (not umbelliform); (13) larval oral disc emarginate; (14) lipophilic alkaloids absent; (15) chromosome number 2N  =  24 (known in A. femoralis, A. olfersioides, A. talamancae) and 2N  =  22 (known in A. nidicola, A. caeruleodactylus, A. chalcopis); (16) testes unpigmented; (17) dark throat collar absent.


As for Allobatinae, above.


With 42 nominal species, Allobates includes nearly half of the species previously referred to the polyphyletic genus Colostethus. Although the monophy