The Phylogenetic Relationships of Eucynodontia (Amniota: Synapsida)
- First Online:
- Cite this article as:
- Liu, J. & Olsen, P. J Mammal Evol (2010) 17: 151. doi:10.1007/s10914-010-9136-8
- 361 Views
The phylogeny of Eucynodontia is an important topic in vertebrate paleontology and is the foundation for understanding the origin of mammals. However, consensus on the phylogeny of Eucynodontia remains elusive. To clarify their interrelationships, a cladistic analysis, based on 145 characters and 31 species, and intergrating most prior works, was performed. The monophyly of Eucynodontia is confirmed, although the results slightly differ from those of previous analyses with respect to the composition of both Cynognathia and Probainognathia. This is also the first numerical cladistic analysis to recover a monophyletic Traversodontidae. Brasilodon is the plesiomorphic sister taxon of Mammalia, although it is younger than the oldest mammals and is specialized in some characters. A monophyletic Prozostrodontia, including tritheledontids, tritylodontids, and mammals, is well supported by many characters. Pruning highly incomplete taxa generally has little effect on the inferred pattern of relationships among the more complete taxa, although exceptions sometimes occur when basal fragmentary taxa are removed. Taxon sampling of the current data matrix shows that taxon sampling was poor in some previous studies, implying that their results are not reliable. Two major unresolved questions in cynodont phylogenetics are whether tritylodontids are more closely related to mammals or to traversodontids, and whether tritylodontids or tritheledontids are closer to mammals. Analyses of possible synapomorphies support a relatively close relationship between mammals and tritylodontids, to the exclusion of traversodontids, but do not clearly indicate whether or not tritheledontids are closer to mammals than are tritylodontids.
The origin of mammals is one of the key transitions in vertebrate evolution. [In this paper, Mammalia includes Adelobasileus and Mammalia sensu Luo et al. (2002), which is defined as a group including the common ancestor of Sinoconodon, living monotremes, and living therians, plus all its descendants; and equals to Mammaliformes of Rowe (1988: Fig. 4) but not Rowe (1993: fig. 10.2), because Sinoconodon is not included in the latter definition.] This transition is one of the best documented examples in the fossil record of an evolutionary sequence connecting two major structural grades, and thus is an appropriate case for studying macroevolutionary models (e.g., Kemp 2007). To study this transition, the phylogenetic position of mammals must be established first. This is a historically important topic in evolutionary studies and an area of active current research. Mammals were viewed as having a polyphyletic origin from mammal-like reptiles (Simpson 1928, 1929; Olson 1944, 1959). Later, mammals were considered to have evolved from cynodonts (Hopson and Crompton 1969), but controversy persisted regarding the interrelationships within cynodonts and especially within Eucynodontia, i.e., those cynodonts more derived than Thrinaxodon. However, the monophyletic Eucynodontia proposed by Kemp (1982) has been corroborated by all subsequent students of cynodont phylogeny apart from Battail (1991). Major questions relating to the interrelationships of eucynodonts include: Which taxon is the sister group of mammals? Can distinct monophyletic carnivorous and herbivorous lineages be recognized? What is the phylogenetic position of Cynognathus?
Historically, six cynodont groups have been proposed as particularly close relatives of mammals: Thrinaxodontidae (Hopson 1969; Hopson and Crompton 1969; Barghusen and Hopson 1970; Fourie 1974), Probainognathidae (Romer 1970; Crompton and Jenkins 1979), Dromatheriidae (Hopson and Kitching 1972), Tritylodontidae (Kemp 1983; Rowe 1988, 1993; Wible 1991; Wible and Hopson 1993), Tritheledontidae (Hopson and Barghusen 1986; Shubin et al. 1991; Crompton and Luo 1993; Hopson 1994; Luo 1994; Hopson and Kitching 2001), and Brasilodontidae (Bonaparte et al. 2005; Abdala 2007; Martinelli and Rougier 2007). The Thinaxodontidae and Probainognathidae hypotheses have been virtually abandoned in recent studies.
Dromatheriids (Dromatherium and Microconodon) had been regarded as mammals in the 19th century (Owen 1871; Osborn 1886, 1887), although they were later referred to Cynodontia (Simpson 1926a, b). Their phylogenetic positions were uncertain, but they were regarded as possibly having mammalian affinities by Hopson and Kitching (1972). Dromatheriidae was subsequently redefined to include Dromatherium, Microconodon, Pseudotricodon, Therioherpeton, Tricuspes, and Meurthodon (Hahn et al. 1984, 1994). Therioherpeton was excluded from Dromatheriidae by Battail (1991), although retained as sister taxon to the group. Recently, several isolated teeth from India were described as belonging to a new dromatheriid cynodont, Rewaconodon (Datta et al. 2004). Fossil dromatheriid remains are scarce and typically restricted to isolated teeth, though some dentary fragments are also known. Poor material limits current understanding of the phylogenetic relationship between dromatheriids and mammals and among non-mammalian eucynodonts (Sues 2001; Datta et al. 2004).
Tritylodontidae is a herbivorous group ranging stratigraphically from the Late Triassic/Early Jurassic to the Late Cretaceous (Kühne 1956; Kamiya et al. 2006). The monophyly of this group is universally accepted. Tritylodontids were once thought to be mammals, but the lack of a dentary-squamosal articulation challenged this interpretation (Watson 1942). Watson (1942) and Kühne (1956) stressed the similarity between the skull of tritylodontids and cynodonts and concluded that tritylodontids were derived from cynodonts, but no more precise statement could be made. Tritylodontids have been classified as cynodonts since Haughton and Brink (1954), and were subsequently suggested to have been derived from Traversodontidae (Crompton and Ellenberger 1957). This opinion, which implies that Traversodontidae is not monophyletic, has also found favor among more recent authors (Hopson and Kitching 1972, 2001; Sues 1985). Battail (1991: fig. 8) accepted the idea of a close relationship between Traversodontidae and Tritylodontidae, but suggested that traversodontids were monophyletic. Tritylodontids did not represent close relatives of mammals in either case. This implies that many features of the orbital wall and sphenoid region shared by tritylodontids and early mammals as well as several features of the rest of the skull and the postcranium would be tritylodontid-mammal homoplasies (Luo 1994). Kemp (1983) was the first to propose that tritylodontids were more closely related to mammals than is Probainognathus, and that there was no close relationship between Traversodontidae and Tritylodontidae. However, he acknowledged that Tritheledontidae might be even more closely related to mammals than Tritylodontidae.
Tritheledontids are small, presumably insectivorous forms occurring from the Late Triassic to the Early Jurassic (Lucas and Hunt 1994). This family initially included only the species Tritheledon riconoi (Broom 1912); later Diarthrognathus broomi and Pachygenelus monus were referred to this family although the former was viewed as a junior synonym of the latter (Hopson and Kitching 1972). Gow (1980) showed Diarthrognathus broomi is a valid taxon. Shubin et al. (1991) listed four dental features as diagnostic of Tritheledontidae. On this basis, they included only Tritheledon, Diarthrognathus, and Pachygenelus in Tritheledontidae; Chalimina, Riograndia, Irajatherium, and Elliotherium were referred to this family later (Martinelli et al. 2005; Sidor and Hancox 2006), but Riograndia sometimes was excluded from this family (Martinelli and Rougier 2007). In the hypothesis of Martinelli and Rougier (2007), Tritheledontidae is a monophyletic group including Chalimina, Irajatherium, Elliotherium, Tritheledon, Diarthrognathus, and Pachygenelus; Ictidosauria includes Tritheledontidae and Riograndia. However, Ictidosauria is more inclusive in Abdala (2007)’s usage, it includes not only Tritheledontidae but also Tritylodontidae. Tritheledontids were combined with Prozostrodon, Therioherpeton, Brasilitherium, and Brasilodon as the more inclusive taxon Tritheledonta (Kemp 2005).
Bonaparte et al. (2003, 2005) proposed that brasilodontids, including Brasilodon and Brasilitherium from the Late Triassic of Rio Grande do Sul together constitute the sister taxon to mammals. This opinion was supported by the study of Martinelli and Rougier (2007). Sidor and Hancox (2006) found Prozostrodon to be the sister taxon of mammals.
Recently, Abdala (2007) published a phylogenetic analysis of eutheriodonts (including traditional therocephalians, cynodonts and their descendants—the mammals) based on 95 cranial and dental characters. In his results, Ecteninion grouped with Cynognathus and Gomphodontia (not including Tritylodontidae). Traditionally, four gomphodont groups have been recognized: diademodontids, trirachodontids, traversodontids, and tritylodontids (Seeley 1895; Hopson and Kitching 1972), but recently tritylodontids are excluded from Gomphodontia (Abdala and Ribeiro 2003; Hopson 2005). Platycraniellus lay between Thrinaxodon and the remaining Eucynodontia. Pachygenelus was the sister group of Tritylodontidae, and they formed a monophyletic Ictidosauria. Moreover, Brasilitherium was the sister-taxon of mammals but Brasilodon was far more basal. Martinelli and Rougier (2007) also published a tree of eucynodonts. They did not include tritylodontids in their analysis, and their result was similar to the trees of Martinelli et al. (2005) and Sidor and Hancox (2006).
Differences in taxonomic sampling across several studies of eucynodont phylogeny, including the present paper
Martinez et al. (1996)
Hopson and Kitching (2001)
Bonaparte et al. (2005)
Martinelli and Rougier (2007)
Most previous analyses have sampled few taxa, usually only one genus for each major group. For example, traversodontids were represented only by Exaeretodon in Wible (1991), and by Massetognathus in Martinez et al. (1996). Furthermore, Morganucodon was the sole representative of mammals in some analyses, including those of Martinez et al. (1996), Hopson and Kitching (2001), and Bonaparte et al. (2003). Bonaparte et al. (2003) did not include in their analysis any gomphodont cynodonts or tritylodontids. Recent works, however, have tended to sample increased numbers of taxa.
Wible (1991) used 66 dental and cranial characters in his analysis. Luo (1994) used 82 characters, 11 of which came from the temporomandibular joint, and a further 20 from the petrosal. Martinez et al. (1996) used 68 characters, including 13 from the dentition. Hopson and Kitching (2001) used 101 characters, of which 29 were dental and 19 postcranial. Bonaparte et al. (2005) used 80 characters, including 20 dental ones and 12 postcranial ones. Martinelli et al. (2005) included 63 characters, 13 of which were postcranial. Martinelli and Rougier (2007) included 93 characters, 32 of which were related to the dentition and 17 of which were postcranial. Abdala (2007) used 98 craniodental characters.
Important conflicts among the various hypotheses are the convergent appearance of some characters in traversodontids, tritylodontids, and mammals on the one hand, and among tritheledontids, tritylodontids, and mammals on the other (Luo 1994; Kemp 2005). Partly because of this widespread convergence, character selection has a crucial impact on the conclusions of any analysis of eucynodont phylogeny. However, no author other than Wible (1991) presented explicit criteria for accepting or rejecting characters. No analysis to date has compiled and integrated all previous used anatomical data into a data matrix. Another factor confounding analyses of interrelationships within Eucynodontia is the limited information available for some taxa.
Here we present a compilation of published morphological data and augment it with new taxa and characters. This data set is the largest ever to be simultaneously analyzed for Eucynodontia. Cladistic analyses of these data are performed here in order to (1) determine the phylogenetic position of Tritylodontidae, of Tritheledontidae, and of Cynognathus; and (2) examine the effects of missing data and increasing taxonomic sample size.
BMNH, Natural History Museum, London, UK; BP, Bernard Price Institute for Palaeontological Research, University of the Witwatersrand, Johannesburg, South Africa; CUP, Catholic University of Peking, now housed in the Field Museum, Chicago, USA; UFRGS-PV, Setor de Paleovertebrados, Instituto de Geociências, Universidade Federal do Rio Grande do Sul, Porto Alegre RS, Brazil.
Materials and methods
The following supra-generic groups have been recognized within Cynodontia: Procynosuchidae, Galesauridae, Cynognathidae, Diademodontidae, Trirachodontidae, Traversodontidae, Chiniquodontidae, Probainognathidae, Tritylodontidae, Tritheledontidae, Brasilodontidae, and Mammalia (Hopson and Kitching 1972; Battail 1991). However, some of these groupings are not monophyletic. Some of the mentioned groups are monotypical and therefore monophyletic, for example, Cynognathidae and Probainognathidae. Tritylodontidae and Mammalia can be considered as well-established monophyletic groups at the present time (Hopson and Barghusen 1986; Rowe 1993; Luo 1994; Luo et al. 2002). The different groups also vary greatly in the number of species that they contain. Most proposed suprageneric groups include only a few genera and species, and indeed some are monogeneric. However, Traversodontidae and Tritylodontidae have more than ten genera and 20 species, and Mammalia of course includes a far greater number.
Recent papers have suggested that adding more species to a cladistic analysis can greatly reduce phylogenetic error, increasing the accuracy of phylogenetic estimates generated by computer simulations (Pollock et al. 2002; Zwickl and Hillis 2002; Debry 2005). However, it is impractical to include all or most relevant species in analyses attempting to resolve relationships among higher taxa (Donoghue 1994; Rice et al. 1997), leading to the problem of selecting particular species as exemplar for higher taxa. Wiens’ (1998) simulation showed that sampling a single randomly chosen species per higher taxon yields low accuracy under many conditions. Nevertheless, mammals were represented in some studies by only a single taxon, Morganucodon (Hopson and Kitching 2001), and other studies have used either Exaeretodon or Massetognathus as the sole representative of Traversodontidae (Wible 1991; Martinez et al. 1996). A close relationship between tritylodontids and traversodontids has been suggested on the basis of comparisons between members of the tritylodontids and either Exaeretodon, a derived traversodont (Sues 1985), or Scalenodon hirschoni (Hopson and Kitching 2001). Accordingly, both Exaeretodon and Scalenodon are important taxa that should be included. Nonetheless, if only one species is selected to represent a particular group, it should be the most basal member of that group that is available. If only derived taxa are included, the morphological gaps among clades will be exaggerated, and problems with long-branch attraction could possibly result, as in the example involving iguanid lizards given by Wiens and Hollingsworth (2000).
Specific strategies have sometimes been proposed for taxonomic sampling in phylogenetic analyses. Prendini (2001) advocated choosing at least two species per non-monotypic higher taxon, with preference given to type taxa, basal taxa, and sets of taxa that capture as much morphological disparity as possible within the clades they represent. Luo et al. (2002) suggested criteria such as morphological informativeness, within-group morphological diversity, within-group geological age (early members of respective lineages), and consideration of anatomical transformation (morphologically distinctive taxa, particularly those with a potential bearing on structural transformations).
Taxa were selected for the present analysis with attention to all these criteria, and we also tried to incorporate taxa used in previous analyses. We selected 31 cynodont species, two of which were used as outgroups (Table 1). Individual species were used as terminal taxa in order to avoid a priori assumptions of monophyly within large, suprageneric clades. Some taxa represented in previous analyses were excluded from the current study. Within mammals, only Adelobasileus, Sinoconodon, and Morganucodon were selected because our study is not intended to consider the interrelationships of early mammals. Among primitive cynodonts, Dvinia was excluded, whereas Procynosuchus and Galesaurus are used as outgroups. Probelesodon was removed because it is regarded as a junior synonym of Chiniquodon (Abdala and Giannini 2002). Aleodon was original described as a gomphodont cynodont, but Hopson and Kitching (1972) reassigned this taxon to Chiniquodontidae based on undescribed specimens that they may have subsequently used (Hopson and Kitching 2001) to code its morphological characters. Based on BMNH 9390 and 10048. Abdala and Giannini (2002) excluded Aleodon from Chiniquodontidae and concluded that the evidence was insufficient to determine its true taxonomic position. The known material is poorly preserved, so this genus is not included in the present analysis. We regard Chiniquodontidae as a monogeneric taxon, containing two species of Chiniquodon, one of which we included in our analysis. Gomphodontosuchus was excluded, because the only known specimen is a juvenile (Hopson 1985) and the analysis already incorporated six other traversodontids.
Character and corresponding number in their original character list
Cusp d in lower postcanines
Prootic and ophistotic (56)
Petrosal promontorium (57)
Separation of perilymphatic foramen from jugular foramen (60)
Length of secondary palate related to tooth row (36)
Selection and coding of the characters
The morphological characters used in this study were taken from several sources. An initial character list was generated by combining anatomical characters used in the following studies: (1) Rowe (1988), with corrections noted by Wible (1991); (2) Lucas and Luo (1993), most characters were adopted by Luo (1994), and some characters modified by Luo et al (2001); (3) Luo and Crompton (1994) (on the quadrate); (4) Martinez et al. (1996); (5) Hopson and Kitching (2001); (6) Bonaparte and his colleagues (Bonaparte et al. 2003, 2005; Martinelli et al. 2005); and (7) Abdala (2007).
The definitions of all characters were examined and some were revised for the present study. Following initial compilation, the character set was examined carefully to identify and remove redundant or covariant characters. Elimination of redundant characters ensured that all characters in the analysis were logically independent. For clearly covariant characters, the definition of the characters was revised to ensure the biological independence of the character. However, the biological independence or non-independence of characters is uncertain in the vast majority of cases, so that covariant characters cannot be completely eliminated. Single characters that were created by combining multiple characters from different sources are identified as such in the character list, and all contributing sources are cited accordingly (Appendix I).
Because of differences in taxonomic sampling, some characters used by previous authors are clearly uninformative in the context of this analysis. Characters in this category were excluded. Only some characters of Adelobasileus, Sinoconodon, and Morganucodon were included, because the monophyly of mammals is well supported by several other characters. The synapomorphies of ingroups were partially included too.
This process of character selection was intended to ensure that the analysis was based on data that were as accurate as possible, and secondarily to maximize the information content of our data matrix. Inapplicable characters were coded as dashes (missing data) rather than as numerical character states. Although this method has its disadvantages, it is preferred here because it has been shown to produce trees that best reflect the information content of the observations (Strong and Lipscomb 1999).
The final character list included a total of 145 characters. Of all characters, 81 are from the skull, 10 from the lower jaw, 28 from the dentition, and 26 from the postcranial skeleton. Character states were then scored either from first-hand observations of specimens in museum collections, or from original published descriptions and photographs. When two previous analyses gave conflicting information on a character, an assessment of the correct character state was made based on first-hand observations. We did not treat any one published analysis as being more reliable a priori than the others. In addition to morphological characters gleaned from published analyses, we introduced three new cranial characters (characters 12, 18, and 124; Appendix I). The relative large amount of missing and inapplicable data in the matrix largely results from the fact that only fragmentary material is available for many species.
Multistate characters generally should be treated as unordered in cladistic analysis except when they represent a transformation series based on prior knowledge (Hauser and Presch 1991; Slowinski 1993). Hopson and Kitching (2001) also showed that ordered multistate characters resulted in a different topology from unordered characters. We apply different strategies: some multistate characters are treated as both unordered and ordered in our analyses. Postcranial characters were excluded from many previous analyses. To direcly compare with these studies and to evaluate the effect of postcranial characters on the phylogenetic relationships of Eucynodontia, the postcranial characters were excluded in some of our analyses.
Hypotheses of tooth homology
The cheek teeth of most cynodonts are usually classified as either gomphodont or sectorial, but those of tritylodontids are distinctive enough that they are usually placed in a separate category distinct from the gomphodont type. Because cynodonts had intensely modified their cheek teeth in some groups, it is difficult to deduce the homologies of individual cusps (and cingula) among different tooth forms. In general, gomphodont teeth are thought to have originated from sectorial teeth by widening of both the crown and the root (e.g., Abdala and Ribeiro 2003). Based on a comparison between the postcanines of Scalenodon and Oligokyphus, Crompton and Ellenberger (1957) suggested that tritylodontid teeth can be derived from traversodontid teeth. Hopson and Kitching (2001) held the same opinion, so they presumed that cingula occurring on the same side (lingual or labial) of the tooth row are homologous across all cynodonts (e.g., their characters 61 and 62) and that similarly positioned individual cusps are homologous between traversodontids and tritylodontids (e.g., their characters 67 and 69). By contrast, Rowe (1986) used evidence from Trirachodon to support the view that gomphodont teeth evolved through the counter-clockwise rotation of teeth in their dentary sockets, possibly to allow more teeth to be packed into the available space. He further proposed that tritylodontid teeth did not evolve through rotation, implying that no individual cusps can be homologized between tritylodontid and gomphodont teeth. In the tritheledontid Diarthrognathus, which has transversely widened teeth, all newly erupting teeth are oriented with their long axes parallel to that of the jaw, so that rotation must have occurred later in ontogeny in order to allow the teeth to become functional (Gow 1994). This clearly demonstrated the reality of the rotation mechanism in cynodonts. Because of the conflicting interpretations of homology, we applied characters on cusps only for the same type of teeth, i.e., no hypothetical homology between sectorial and gomphodont teeth, and no assumed correspondence of cusps between tritylodontid and other gomphodont teeth. However, correspondence of cusps is assumed within Gomphodontia excluding Tritylodontidae (diademodontids, trirachodontids, plus traversodontids) following previous works such as Abdala and Ribeiro (2003) and Hopson (2005).
Analyses and results
To evaluate the impact of synonymizing Probelesodon with Chiniquodon, and Brasilitherium with Brasilodon, these taxa were coded separately and analyzed with other taxa. The results are almost the same as those of the above analysis.
The impact of fragmentary taxa
The effect of pruning various taxa from data matrix in Appendix II on the most parsimonious trees obtained with phylogenetic analysis
Deleted taxon (taxa)
Prozostrodon plus Therioherpeton
The relationships among eucynodont taxa undergo little change after deletion of Sinognathus, Scalenodon hirschoni, or Adelobasileus. If only the most complete taxa (coded for more than 80% of the characters) are included in the analysis, the result is a subtree of Fig. 1. This agrees with the simulations of Wiens (2003), which also shows that the inclusion of highly incomplete taxa tends to have little impact on the recovered pattern of relationships among more complete ones. However, when Scalenodon angustifrons is excluded, a monophyletic Traversodontidae disappears from some trees; when Prozostrodon is deleted from the analysis, Riograndia and Pachygenelus become closer to mammals than tritylodontids in half of the most parsimonious trees (MPTs), and Chiniquodon forms a monophyletic clade with Probainognathus in some trees. Exclusion of Therioherpeton results in a consensus tree similar to Fig. 2.
The impact of selecting taxa
To ascertain the effect of taxon sampling on the eucynodont interrelationships, the composed data matrix (Appendix II) was run following the previous taxon samplings.
Using only the taxa considered by Hopson and Kitching (2001), the consensus tree is nearly identical to the tree shown in Fig. 1. Tritylodontidae always groups within Probainognathia, rather than within Traversodontidae, and Lumkuia lies in a basal position within Probainognathia in some of the most parsimonious trees.
Discussion and Conclusion
Platycraniellus was regarded as more derived than Thrinaxodon by Abdala (2007), but the position of Platycraniellus is equivocal here. Kemp (1982) defined Eucynodontia as all cynodonts closer to extant mammals than Thrinaxodon, so Platycraniellus could be a basal member of Eucynodontia.
Eucynodontia is a robust clade with a Bremer support value of 5, and is supported by six unambiguous characters. Four of the most parsimonious trees have three additional unambiguous synapomorphies. A dichotomy within most species of Eucynodontia also is recovered here, although the membership of each branch differs slightly from what previous analyses have proposed. Lumkuia is a basal eucynodont. This possibility was implicit in the cladograms of Martinelli et al. (2005), in which Lumkuia forms a trichotomy with cynognathians and probainognathians. This position contrasts with proposals by Hopson and Kitching (2001) and Abdala (2007) that Lumkuia is a basal probainognathian. Cynognathia (not including Tritylodontidae) has a Bremer support value of 3, and seven unequivocal synapomorphies, while Probainognathia has a support value of 2 and six unequivocal synapomorphies.
Hopson first proposed a sister relationship between Cynognathus and gomphodonts (Hopson and Barghusen 1986; Hopson 1991). This relationship is corroborated here. Gomphodontia emerged as a natural group in the tree of Hopson and Kitching (2001). Hopson (1991) proposed a basic pattern of ((Diademodontidae, (Trirachodontidae, Traversodontidae))); essentially the same pattern was recovered here, although a monophyletic Trirachodontidae was absent. Tritylodontidae was thought to be derived from traversodontids (Hopson 1991; Hopson and Kitching 2001), but it was found to be closely related to mammals here. The species of Trirachodontidae used in our analysis showed considerable variation in skull morphology, but they may form a monophyletic group in an analysis including more dental characters. The detailed relationships among these groups will be discussed elsewhere.
Recovered relationships within Probainognathia in our hypotheses were slightly different from those suggested by Hopson and Kitching (2001). The position of Ecteninion varies in different studies: it was originally proposed to be a cynodont more derived than Chiniquodon (Martinez et al., 1996), a basal probainognathian (Hopson and Kitching 2001; Martinelli and Rougier 2007), or the sister taxon of cynognathians (Abdala 2007). The present analysis placed Ecteninion at a basal position within Probainognathia.
Probainognathus is placed closer to mammals than is Chiniquodon in most studies, e.g., Martinez et al. (1996), Hopson and Kitching (2001), Abdala (2007), and Martinelli and Rougier (2007); this relationship is absent only in the trees of Bonaparte et al. (2003, 2005) and Martinelli et al. (2005).
Clade E is a stable clade in all results. It includes “tritheledontans”, tritylodontids, and mammals. This clade is more inclusive than Mammaliamorpha of Rowe (1988), and is named as Prozostrodontia here. This group is defined as the least inclusive clade containing Prozostrodon brasiliensis, Tritylodon langaevus, Pachygenelus monus, and Mus musculus. Salient synapomorphies of this clade include the reduction of the prefrontal, postorbital, and postorbital bar, presence of sphenopalatine foramen (convergently appearing in traversodontids), unfused dentary symphysis, posterior extension of sagittal crest at the same level as the posteriormost part of the lambdoidal crest, neural spines of posterior thoracic vertebrae posterodorsally inclined, convex iliac blade, reduced posterior iliac spine, acetabular notch on ischium, and lesser trochanter located near the level of femoral head on the medial surface of the femoral shaft.
Tritheledonta is not monophyletic, because as defined by Kemp (2005) it excludes mammals, which contrast with the result of the present analysis. Furthermore, Tritheledonta remains non-monophyletic, even if mammals are not considered, in the cladogram of Abdala (2007) and in the present study. Hence, Tritheledonta in its original meaning cannot be used as a formal taxonomic name. The monophyly of Tritheledontidae has never been seriously doubted; it has been supported in those analyses that have tested it by including multiple tritheledontids (Martinelli et al. 2005; Martinelli and Rougier 2007). The monophyly of Tritheledontidae (sensu Martinelli and Rougier 2007) cannot be tested in this study. Riograndia does not group with Pachygenelus in Fig. 1, but a monophyletic Ictidosauria (sensu Martinelli and Rougier 2007) appeared in 12 of the 24 most parsimonious trees recovered when the postcranial characters were excluded. There is only one unambiguous synapomorphy for Riograndia plus Pachygenelus [93(1)], but nine possible synapomorphies for this clade. The possibility of a monophyletic Ictidosauria (sensu Martinelli and Rougier 2007) cannot be excluded, and needs additional testing.
Pachygenelus is more closely related to mammals than Prozostrodon in all most parsimonious trees recovered in this study, and closer to mammals than Therioherpeton in most of the trees. This result agrees with the findings of Martinelli et al. (2005) and Martinelli and Rougier (2007), but differs from those of Bonaparte et al. (2005) and Sidor and Hancox (2006).
In all most parsimonious trees, Brasilodon falls between other non-mammalian cynodonts and Mammalia, and all four species of tritylodontids used in our analysis group together as a robust monophyletic clade that is supported by 34 unequivocal characters. Even though Tritylodontidae is consistently more closely related to mammals than Prozostrodon, the position of Tritylodontidae varies under different character sets. It appears as the sister taxon of Brasilodon plus mammals using all characters (Fig. 1), but is found to be the sister taxon of a grouping that includes most “tritheledontans” if postcranial characters are excluded (Fig. 2). Abdala (2007) did not include postcranial characters in his matrix, but tritylodontids form a clade with Pachygenelus in his tree.
The impact of fragmentary taxa
In general, incompletely known taxa are usually associated with a large number of equally parsimonious cladograms and poorly resolved consensus trees that fail to reveal strictly supported relationships (Gauthier 1986; Wilkinson 2003). This is not true in the present analysis. Thus, it is clear that the proportion of missing characters for a particular taxon should not be used as a criterion for its inclusion or exclusion within an analysis (Kearney and Clark 2003; Wilkinson 2003). Taxa with the most missing features in this study are the basal traversodontid Scalenodon angustifrons, basal prozostrodontians Prozostrodon and Therioherpeton, and the basal mammaliaform Adelobasileus. Basal members of clades are likely to have an important impact on phylogenetic analysis, and should generally be included in the analysis even if they are known only from fragmentary material.
The impact of selecting taxa
In summary, the results of previous analyses are partly congruent with the most parsimonious trees found in this study. Because these simulated analyses mentioned above were based on our data matrix, their different results should be due to taxon sampling. When taxa are poorly sampled, the phylogenetic trees obtained are not reliable.
Different hypothesis and supporting characters
Alternative hypotheses regarding the relationships of these taxa conflict mainly in the placement of tritylodontids, which are sometimes within Cynognathia (including Traversodontidae) and sometimes within Probainognathia. In this data set, the placement of tritylodontids within Cynognathia requires an extra 14 steps if multistate characters are treated as unordered.
Characters supporting tritylodontids within Cynognathia
Maxillary platform lateral to the tooth row
Zygomatic arch high
Posteroventral process of jugal high
Squamosal groove for external auditory meatus deep
Posterior extension of the squamosal dorsal to squamosal sulcus in zygomatic arch well developed
Nerve V2&3 exit via two foramina between prootic and epipterygoid
Posteriorly directed power stroke during occlusion for mandibles
Bilateral, interdigitating occlusion between multiple cusps on each postcanine tooth
One or two transverse and crescentic wear facets on multiple cusps
Upper postcanines bucco-lingually expanded
Upper postcanines bear multiple cusps in multiple rows
Sues (1985) listed 11 characters as possible synapomorphies supporting the inclusion of tritylodontids within Cynognathia. Some of his characters are ambiguous. For example, his character 5, lack of the ectopterygoid, is problematic because this bone is also absent in both Probainognathus and prozostrodontians. Similarly, his character 7, ventral margin of the basicranium distinctly sigmoid, is not developed in traversodontids to the same degree as in tritylodontids. Most traversonodontids in fact have a relatively flat basisphenoid. Even in Exaeretodon, the basisphenoid is far less sigmoid than in tritylodontids, so that the presence of a sigmoid basisphenoid is best considered as an apomorphy of Tritylodontidae. The condition of the basicranial process of the prootic is unclear in basal probainognathians. If Oligokyphus exemplifies the ancestral condition of tritylodontids, characters 10 and 11 of Sues are perhaps not synapomorphies for this group because of the presence of anapophysis (Kühne 1956: fig. 45) and the absence of expanded apices on neural spines in Oligokyphus.
If tritylodontids are constrained to form a monophyletic clade with traversodontids, the sister taxon of tritylodontids is found to be Exaeretodon, not Scalenodon hirschoni as suggested by Hopson and Kitching (2001). The following characters support the relationship of tritylodontids plus Exaeretodon: 7(1), 52(1), 84(1), 118(2), 125(0), and 133(1). Meanwhile, only three of them unequivocally support the postulated sister-group relationship between Exaeretodon and tritylodontids: 52(1), space for trigeminal ganglion partially floored by prootic; 84(1), angle of dentary close to jaw joint; and 118(2), upper tooth series extends posteriorly beyond anterior border of subtemporal fenestra.
A close relationship between tritylodontids and traversodontids was originally proposed mainly on postcanine morphology (Crompton and Ellenberger 1957; Crompton 1972). Most of the dental characters used by Hopson and Kitching (2001) are included in the current list, although a few characters on cusp pattern are excluded because the homologies of the cusps cannot be ascertained. In previous studies, postcranial characters typically played a key role in supporting a close relationship between tritylodontids and the tritheledontids-mammals clade, to the exclusion of traversodontids (abbreviated as: Tri-M/Tra) (Kemp 1983). The present analysis persists in placing tritylodontids within Probainognathia even when postcranial characters are excluded (Fig. 2), requiring three additional steps in order to place tritylodontids within traversodontids for the data set without postcranial characters.
The cranial characters that support Tri-M/Tra mainly come from the orbital region, the palatal complex, the prootic, and the quadrate, including 10(1), 12(1), 13(1), 14(2), 15(2), 37(1), 38(2), 39(1), 49(1), 51(1), 54(1), 58(2), 59(1), 60(1), 61(1), 62(1), 64(1), 65(1), 71(1), 73(1), 74(1), 75(1), 76(3), 77(1), 83(3), and 86(1). The completely divided roots of the postcanines constitute a potential synapomorphy for Tri-M/Tra. Postcranial characters that support Tri-M/Tra include 120(1), 121(1), 122(1), 123(1), 131(1), 132(1), 138(2), 139(1), 140(1), 141(1), 142(1), 143(1), 144(1), and 145(1). On balance, the postcranial skeleton of tritylodontids can be regarded as more mammal-like than that of traversodontids.
Sues and Jenkins (2006) questioned the value of postcranial characters as synapomorphies for Tri-M/Tra, citing Luo (1994: 104) for support. However, Luo (1994) discussed the relationships of tritylodontids with mammals compared with tritheledontids rather than traversodontids, and is therefore not directly relevant. According to Sues and Jenkins (2006), “most of the alleged postcranial similarities are only superficial in nature” and “certain mammal-like features of the postcranial skeleton of the Tritylodontidae (e.g., large, ossified olecranon process) appear to represent autapomorphies for this group and thus are not useful for determining its phylogenetic relationships”. In discussing the relationships of Tritylodontidae, Sues (1985) argued that autapomorphic features should not be emphasized in the context of phylogenetic analysis. Cladistic analysis proceeds by identifying similarities between potentially homologous structures in different taxa. Accepting hypotheses of primary homology only when structures are comparable in minute detail would reduce most anatomical data to lists of uninformative autapomorphies. Although the coding of characters is still an art, workers attempt to maintain a consensus that avoids dismissing too many structures as autapomorphic. Most postcranial characters used in the present study have been widely accepted by different scholars.
A predominantly preacetabular iliac blade has evolved in Exaeretodon (Bonaparte 1963), Therioherpeton (Bonaparte and Barberena 2001), Tritylodontidae (Sues and Jenkins 2006), and Morganucodon (Jenkins and Parrington 1976). Even if a posterior process is present in Tritylodontidae, it nevertheless lies entirely anterior to acetabulum. Even differences in the lesser trochanter of the femur were emphasized by Sues and Jenkins (2006), but these osteological details do not obviate the fact that the gross body plan of tritylodontids is unquestionably more mammal-like than is the body plan of basal cynodonts. We believe that the postcranial characters listed are well established as synapomorphies for Tri-M/Tra.
Identifying the mammal sister-group and Tritylodontidae–Tritheledontidae–mammals relationship (TTMR) and the impact of Brasilodon
Traditionally, the two predominant hypotheses relating to the sister-group of mammals have been the tritylodontid-mammal hypothesis (TYMH) and the tritheledontid-mammal hypothesis (TRMH). Each of these alternatives is supported by a large number of putative synapomorphies and contradicted by a substantial amount of opposing anatomical evidence. Luo (1994) analyzed the support for each hypothesis in detail, and showed that it is difficult to conclusively choose between them on the basis of available evidence. These two hypotheses have different implications for the phylogenetic transformations of important mammalian characters. Although no consensus on TTMR can be easily obtained, new findings have shed additional light on the problem of identifying the sister taxon of mammals. Brasilodon from Rio Grande do Sul, Brazil, shares more synapomorphies with mammals than does any other non-mammalian cynodont. Brasilodon has been recovered as the sister taxon of mammals in all subsequent cladograms that have included it, provided that Adelobasileus is accepted as a basal mammal (Bonaparte et al. 2003, 2005; Abdala 2007; Martinelli and Rougier 2007; this paper).
Synapomorphies of Brasilodon and morganucodontids from Bonaparte et al. (2003)
Reduced postdentary bones
Low position of Meckelian groove
Presence of three anteriorly directed lower incisors
Canines reduced to near the size of the last incisor
Presence of cusp g in lower postcanines
Greatly reduced mandibular symphysis
Expansion of braincase in parietal region
Synapomorphies of Brasilodon and mammals recognized in this study
Anteroventral corner of zygomatic arch lying at same level as postcanine line
Snout longer than temporal region
Basioccipital overlapping medial side of promontorium
Nerve V2&3 exiting via separate foramina, some enclosed by anterior lamina of prootic (petrosal)
Presence of foramen and passage of prootic sinus on lateral trough
Four upper incisors present
Lower canine reduced
Delayed postcanine tooth replacement
The present study does not conclusively resolve the problem of TTMR. Luo (1994) listed the synapomorphies for each hypothesis. The following paragraphs present alternative interpretations of some of these characters, and present the possible synapomorphies found in this study.
For the orbital region, Luo (1994) listed four characters as synapomorphies shared by tritylodontids and mammals. However, all four characters are invalid or at least problematic. A large ascending process of the palatine and orbitosphenoid contributing to the orbital wall is present not only in tritylodontids but also in Prozostrodon, Therioherpeton (Bonaparte and Barberena 2001: figs. 1, 9), Riograndia (Soares, 2004), and Brasilodon (Bonaparte et al., 2005: fig. 14). Among tritylodontids, the palatine participates in the subtemporal border of the orbit only in Kayentatherium; the state of this character is unclear in Oligokyphus based on the original reconstruction of Kühne (1956: text-fig. 18). In the holotype of Bienotherium (personal observation), the palatine is close to the subtemporal border but does not not participate in it. Young (1947) did not illustrate a clear border between the palatine and the pterygoid. In Tritylodon (BP/1/4778), Brink (1988) showed different states on the two sides of the skull. Luo (1994) also listed “separate orbital openings for greater and lesser palatine nerves”. However, the homology of the formina for the greater and lesser palatine nerves is hard to understand in tritylodontids and mammals. Within Tritylodontidae, two separated foramina are known only in Kayentathrium (Sues 1986); it is unclear whether this condition is shared by other genera.
The pterygoplatine ridges include a middle ridge and intermediate ridges. The middle ridge is absent in Pachygenelus (Allin and Hopson 1992: fig. 28.4H) and Riograndia (Soares 2004); the middle ridge is present and reaches the basisphenoid in most tritylodontids, including Bienotherium (Young 1947: fig. 3), Tritylodon (BP/1/4778), and Yunnanodon (Luo 2001: fig. 1), but not Bienotheroides (Sun 1984: fig. 4); and the same condition is present in Brasi1odon (Bonaparte et al. 2005: fig. 11, UFRGS-PV 0929T), Adelobasileus (Lucas and Luo 1993: fig. 9), Sinoconodon (Crompton and Luo 1993: fig. 4.10), and Morganucodon (Kermack et al. 1981: fig. 98). The presence of a middle pterygoplatine ridge [character 38(2)] optimizes as a synapomorphy for tritylodontids and mammals. The intermediate ridges extend posteriorly to the anterior border of the basisphenoid in both tritheledontids (Pachygenelus and Riograndia) and basal mammals, but not tritylodontids. The description of this character in Luo’s (1994) table 6.2 is not entirely correct; the character should be given as “intermediate pterygopalatine ridges reach basisphenoid”. This character is correlated with the width of the anterior part of the basisphenoid , as intermediate pterygopalatine ridges only can extend to the basisphenoid if the basisphenoid is broad.
Character 75(1), round dorsal margin of the dorsal plate of the quadrate, was found to be a synapomorphy of tritheledontids and mammals by Luo (1994) and also in the present analysis. However, despite that tritylodontids primitively have a peg-like dorsal process, such as in Oligokyphus (Kühne 1956; Luo and Crompton 1994), the rounded margin also occurs convergently in some derived tritylodontids, such as Kayentatherium (Luo and Crompton 1994: fig. 9) and Bienotherium (CUP 2241).
The putative synapomorphies shared by tritheledontids and mammals found in this study also include: moderate expansion of the braincase in the parietal region [11(1)], presence of an interpterygoid vacuity in the adult [31(0)], and basisphenoid wing (parasphenoid ala) much shorter and overlapping prootic pars cochlearis (cochlear housing) [46(2)]. The potential synapomorphies shared by tritylodontids and mammals are fusion of the prootic and the opisthotic at early ontogenetic stage [49(1)], presence of foramen “X” (Rougier et al. 1992) in the posterior part of the lateral flange of the prootic [54(1)], lateral flange vascular canal present for route of the venous drainage exiting from the back of the cavum epiptericum [56(2)] (Crompton and Luo 1993), and completely divided postcanine roots [106, 107] (modified in advanced tritylodontid Bienotheroides (Cui and Sun 1987; Luo 1994)).
Most postcranial characters offer equal support to the TRMH and TYMH interpretations, rather than unequivocally favoring TYMH. Only two postcranial characters favor TYMH: absence of an ectepicondylar foramen in the humerus [132(1)] and presence of the longitudinal ridge dividing the lateral surface of iliac blade into dorsal and ventral portions [138(2)]. However, the result would change if the postcranial skeleton of Adelobasileus and Sinoconodon were shown to differ from that of Morganucodon. Only one known postcranial character, elongation of the scapula between the acromion and glenoid, supports TRMH more strongly than TYMH.
These character conflicts impede resolution of the interrelationships within Mammaliamorpha. One possibility for overcoming this problem is to recover more information from known taxa, whereas another is to discover new basal taxa within the clade. For example, Prozostrodon and Therioherpeton are undoubtedly important taxa in this context, but they are so fragmentary (more than 50% missing data), that more information on their morphology is essential if their potential for helping to elucidate phylogenetic relationships is to be realized.
As stated by Bonaparte et al. (2005), mammalian characters emerged in a mosaic fashion across different non-mammalian cynodont clades appearing alongside persistent primitive features. Diagnostic characters of Mammalia include the presence of craniomandibular joint comprising of dentary condyle and squamosal glenoid, the presence of a petrosal promontorium, the extensive development of a petrosal floor for the cavum epiptericum, the presence of a separate tympanic aperture for the prootic canal, the separation of the hypoglossal foramen from the jugular foramen, and the presence of four lower incisors (Luo et al. 2002). These authors also included the loss of the thickened rim of the fenestra vestibuli in their diagnosis, but this formulation of the character differs from that given in the work they cited (Lucas and Luo 1993). The character should be changed to “loss of the basisphenoid contribution to the thickened ring of the fenestra vestibuli”. An incipient dentary/squamosal joint may exist in tritheledontids, although this was doubted by Gow (1981). The promontorium is the most distinctive feature of the mammalian basicranium (Rowe 1988; Wible 1991; Luo 1994; Luo et al. 2002), but this feature also occurs in Adelobasileus (Lucas and Luo 1993) and Brasilodon (Bonaparte et al. 2005). A distinctive cochlear canal is discovered in the tritylodontid Yunnanodon (Luo 2001). The space for the trigeminal ganglion is partially floored by the prootic in Exaeretodon (Bonaparte 1966), Bienotherium (Hopson 1964), and Tritylodon (Gow 1986). A separate tympanic aperture for the prootic canal also occurs in Probainognathus and Massetognathus (Wible and Hopson 1995). The hypoglossal foramen is completely separated from the jugular foramen in Riograndia (UFRGS-PV 0833T), Brasilodon (UFRGS-PV 0628T, Bonaparte et al. 2005: fig. 7), and Tritylodon (personal observation on Hopson’s collection), but is positioned on the sidewall of the jugular foramen in Oligokyphus (Crompton 1964: fig. 2). Four lower incisors are also found in Prozostrodon (Bonaparte and Barberena 2001).
In conclusion, the monophyly of Eucynodontia is confirmed in this study, although the results differ slightly from those of previous analyses with respect to the composition of both Cynognathia and Probainognathia. Pruning highly incomplete taxa has little effect on the inferred pattern of relationships among the more complete taxa, although this pattern can change according to the inclusion or exclusion of basal fragmentary taxa. Taxon sampling of the current data matrix shows that taxon sampling was poor in some previous studies, implying that their results are not reliable.
Two major unresolved questions in cynodont phylogenetics are whether tritylodontids are more closely related to mammals or to traversodontids, and whether tritylodontids or tritheledontids are closer to mammals. Analyses of possible synapomorphies support a relatively close relationship between mammals and tritylodontids, to the exclusion of traversodontids, but do not clearly indicate whether or not tritheledontids are closer to mammals than are tritylodontids.
We thank James Hopson, Hans-Dieter Sues, Marina Bento Soares, and especially Fernando Abdala for fruitful discussions. The comments from Zhe-Xi Luo, Fernando Abdala, Agustín G. Martinelli, and John R. Wible greatly improved this paper. The cooperation and hospitality of the staff of various museums and institutions greatly facilitated our comparative studies. We would like to thank Tom Kemp (Oxford University Museum of Natural History, UK); Ray Symonds (University Museum of Zoology, Cambridge, UK); Sandra Chapman (Natural History Museum, London, UK); Fernando Abdala and Bruce Rubidge (Bernard Price Institute for Palaeontological Research, Johannesburg, South Africa); Jennifer Botha and Elize Butler (National Museum, Bloemfontein, South Africa); Roger Smith and Sheena Kaal (Iziko Museums–South African Museum, Cape Town, South Africa); Johann Neveling (Council for Geosciences, Pretoria, South Africa); Stephany Potze (Transvaal Museum, South Africa); Ana Maria Ribeiro (Museu de Ciências Naturais, Fundação Zoobotânica do Rio Grande do Sul, Porto Alegre, Brazil); Maria C. Malabarba (Museu de Ciências e Tecnologia, Pontifïcia Universidade Católica do Rio Grande do Sul, Porto Alegre, Brazil); Marina B. Soares and Cesar L. Schultz (Universidade Federal do Rio Grande do Sul, Porto Alegre RS, Brazil); Jaime E. Powell (Universidad Nacional de Tucumán, Argentina); Alejandro Kramarz and Agustín G. Martinelli (Museo Argentino de Ciencias Naturales “Bernardino Rivadavia”, Buenos Aires, Argentina); Guillermo F. Vega (Museo de Antropología, Universidad Nacional de La Rioja, Argentina); Ricardo Martinez (Museo de Ciencias Naturales, Universidad Nacional de San Juan, Argentina); Marcelo Reguero and Rosendo Pascual (Museo de La Plata, Argentina); Charles R. Schaff and Wu Shaoyuan (Museum of Comparative Zoology, Harvard University, Cambridge, Massachusetts, USA), Hans-Dieter Sues and Matthew Carrano (National Museum of Natural History, Washington, D.C., USA), James A. Hopson (University of Chicago, Chicago, USA), Olivier Rieppel, Elaine Zeiger, and William F. Simpson (Field Museum of Natural History, Chicago, USA); and John Flynn (American Museum of Natural History, New York, USA). Special thanks to Corwin Sullivian for reading the manuscript and greatly improving the writing.
Financial support for this project was provided by Columbia University through a Faculty Fellowship, the Climate Center of Lamont-Dohert Earth Observatory, Theodore Roosevelt Memorial Fund of AMNH, and Chinese Academy of Sciences (KZCX2-YW-BR-07). The Field Museum provided grants that make possible a study visit to Chicago.