Callistomordax, AND THE

Schoch, Rainer R., 2008, A new stereospondyl from the German Middle Triassic, and the origin of the Metoposauridae, Zoological Journal of the Linnean Society 152 (1), pp. 79-113 : 95-104

CALLISTOMORDAX AND THE METOPOSAURIDAE

PREVIOUS CONCEPTS OF METOPOSAURID

RELATIONSHIPS

The origin of the metoposaurids has puzzled workers for more than one and a half centuries, beginning with Hermann von Meyer (in Meyer & Plieninger, 1844; Meyer, 1857) who first described them. After their subsequent discovery in North America, India, and Morocco, metoposaurids turned out to be one of the most clear-cut and well-defined, yet at the same time one of the most isolated, temnospondyl groups.

As knowledge of Mesozoic temnospondyls has grown enormously during the 20th century, any phylogenetic study of metoposaurid origins must include potentially related taxa. In the last nine decades, as many as ten different hypotheses have been proposed as to the origin and evolution of the Metoposauridae .

1. Watson (1919) proposed the first large-scale, highly influential evolutionary scheme of temnospondyls (he referred to them as a subgroup of his Labyrinthodontia ), in which he tied metoposaurids in his ‘grade’ Stereospondyli , separate from trimerorhachids and other ‘Rhachitomi-grade’ temnospondyls. However, he was not explicit about specific relationships to other stereospondyl groups. Watson’s (1919) concept was opposed by Säve-Söderbergh (1935), who classified metoposaurids with other temnospondyls with long postorbital skull tables, notably trimerorhachids.

2. Romer (1947) moved one step ahead in not only classifying metoposaurids among the Labyrinthodontia , as Watson had done three decades before, but also by further specifying that metoposaurids were closely related to brachyopids and plagiosaurids; he used the superfamily name Brachyopoidea to include all three families. In this concept, both trematosaurids and capitosaurids were excluded from that superfamily. Romer’s (1947) concept was followed by Dutuit (1976), who was hesitant whether to rank the almasaurids with Romer’s Brachyopoidea .

3. Shishkin (1973) followed Säve-Söderbergh (1935) in seeking the metoposaurid origin among more primitive temnospondyls, notably trimerorhachids, saurerpetontids, and dvinosaurids. He further considered brachyopids and (nontemnospondyl) colosteids, but not plagiosaurids, as close relatives of metoposaurids.

4. Warren & Black (1985) performed the first phylogenetic analysis ‘by hand’, in which they envisioned a Stereospondyli divided into a trematosaurian and a capitosaurian group. This is the first time that a major dichotomy was explicitly proposed in stereospondyl phylogeny, indeed one with a deep-reaching split: rhinesuchids, lydekkerinids, capitosauroids, almasaurids, and metoposaurids were ranked among the capitosaurian group, contrasted by trematosaurids, rhytidosteids, brachyopids, and chigutisaurids, which together formed their trematosaurian group. Most subsequent studies are variants of this concept, with various permutations in sistergroups relationships in particular.

5. Milner (1990) suggested the most radical alternative to Warren & Black’s (1985) hypothesis. His phylogenetic study was the first to include all major groups of temnospondyls based on Hennigian principles. In Milner’s (1990) cladogram, metoposaurids originated from a particular clade of short-snouted trematosaurians, with ‘latiscopids’ (almasaurids) forming their sister group and indicating the plesiomorphic condition for many metoposaurid traits. In this concept, chigutisaurids and rhytidosteids were derived from lydekkerinids, thus being not immediately related to metoposaurids or even the larger trematosaurian clade. As a further stark contrast to Warren & Black (1985), Milner (1990) ranked brachyopids and plagiosaurids not among the stereospondyls, but sought their ancestry among more ‘primitive’ Permian temnospondyls.

6. Hunt (1993), in revising the family Metoposauridae , discussed two alternative scenarios for deriving the group from other temnospondyls: (1) an origin from ‘primitive’ (trimerorhachid) temnospondyls, and (2) a stereospondyl ancestry for the group. In his trimerorhachid hypothesis, Hunt (1993) placed metoposaurids as a sister group to brachyopids, arguing for both to be nested with the Lower Permian Trimerorhachidae . In his alternative stereospondyl hypothesis, he envisioned the Latiscopidae (Almasauridae) as a sister group of the metoposaurids, together nested with Mastodonsaurus , Eocyclotosaurus , and finally the Capitosauridae .

7. Yates & Warren (2000) performed the first computer-assisted cladistic analysis of stereospondyls, including many taxa relevant to the present study. Their findings were clearly different from the aforementioned in several points: (1) they suggested plagiosaurids to be nested with other short-skulled stereospondyls (brachyopids, chigutisaurids, and rhytidosteids in particular), and within that assemblage they envisioned the small, carapace-bearing Laidleria as an immediate sister group to the plagiosaurids; (2) they suggested that the latter clade was nested deeply within a group that included trematosaurids, Almasaurus , and metoposaurids, referring to the whole assemblage as the Trematosauria ; and (3) they found Lydekkerina to form a clade with Mastodonsaurus and capitosauroids (a group they termed Capitosauria). By that, Yates & Warren (2000) confirmed the trematosaurian concept of metoposaurid ancestry proposed by Milner (1990), albeit with a large clade of short-skulled stereospondyls also having arisen from a vast clade they termed Trematosauria .

8. Schoch & Milner (2000) attempted to form a phylogenetic frame for a higher-ranking taxonomy of stereospondyls, ranking the metoposaurids with Almasaurus , the Platystegidae , and Lyrocephaliscidae in a clade of broad-skulled trematosaurians. They referred to the whole clade of slender-skulled and broad-skulled trematosaurs as Trematosauroidea, which would necessarily include the Metoposauridae , but explicitly excluded from that group all shortskulled temnospondyls such as brachyopids, chigutisaurids, rhytidosteids, and plagiosaurids.

9. Steyer (2002) was the first to restrict a numerical cladistic analysis to trematosaurian in-group relationships, considering the metoposaurids, Almasaurus , and Inflectosaurus to be not intimately related to the trematosaurids proper. He found that these three formed a monophylum nested below a dichotomy of capitosauroids (‘mastodonsauroids’) and trematosaurians s.s., with only the latter referred to as the Trematosauridae by him. In this hypothesis, Lyrocephaliscus and Platystega are nested together with Tertrema within a monophyletic Trematosaurinae, contrasted by a long-snouted sister group Lonchorhynchinae.

10. Most recently, Damiani & Yates (2003) published a more inclusive computer-assisted analysis of trematosaurid phylogeny, again changing the picture in various points. Although not considering brachyopoids or plagiosaurids, they found rhytidosteids to nest with the ‘primitive’ stereospondyl Lydekkerina to form a clade distinct from all trematosaurians, which in their cladogram formed a grade towards Almasaurus and the Metoposauridae . This most closely resembles the hypotheses of Milner (1990) and Schoch & Milner (2000). In particular, their analysis agreed with the latter authors in nesting Lyrocephaliscus with the almasaurid-metoposaurid clade, whereas they found evidence for a separate position of Platystega with Tertrema and the lonchorhynchines.

To summarize, the origin of metoposaurids is still highly controversial, and has been sought among at least three different major nodes or grades within temnospondyl phylogeny: (1) an early origin from Trimerorhachis -like taxa ( Säve-Söderbergh, 1935; Shishkin, 1973; Hunt, 1993, hypothesis A) – this would require an extraordinarily long ghost lineage and would leave most of the character evolution towards metoposaurids entirely hypothetical; (2) a separate origin in a basal stereospondyl grade before the capitosaurian–trematosaurian dichotomy ( Watson, 1919; Romer, 1947; Steyer, 2002), with Almasaurus forming their sister taxon and potentially indicating plesiomorphic states for various metoposaurid autapomorphies; and finally (3) an origin within a large trematosaurian clade, with Lyrocephaliscus and Almasaurus forming successive sister taxa of the Metoposauridae ( Milner, 1990; Schoch & Milner, 2000; Yates & Warren, 2000; Damiani & Yates, 2003).

The last general hypothesis presents the most detailed account of character evolution, as some character states would have evolved prior to Lyrocephaliscus , others with the almasaurid grade, and yet others in the immediate stem of the metoposaurids. On the other hand, the evolution of many features is still unclear in the several variants of that concept, in particular the postcranium for which little articulated material is known. Callistomordax forms a good opportunity to tackle both problems, and by doing so review the case of metoposaurid phylogeny within stereospondyls, especially after the Upper Triassic Rileymillerus , a new, Almasaurus -like taxon, was recently described by Bolt & Chatterjee (2000).

PHYLOGENETIC ANALYSIS

The present analysis is based on 19 taxa and 100 characters from all parts of the skeleton ( Fig. 10 ; Appendix). Most character states were taken from the recent literature (all authors are cited in the character list), but some states expressed in Callistomordax were added. Multistate characters were treated as unordered throughout. All variants of the present analysis were performed in the branch-and-bound mode of PAUP 3.1 ( Swofford, 1991), and characters were traced by making use of MacClade 2.0 ( Maddison & Maddison, 1992). The analysis was run in the ACCTRAN mode.

Taxa

Three successively more derived outgroups were included to root the ingroups.

1. Dendrerpeton acadianum (Holmes, Carroll & Reisz, 1998) is one of the known plesiomorphic temnospondyls.

2. Trimerorhachis insignis ( Case, 1935) was included because the last revising author of the metoposaurids, Hunt (1993), made explicit reference to trimerorhachids, after Shishkin (1973) had suggested all temnospondyls with short faces and long posterior skull tables may have formed a clade.

3. Sclerocephalus haeuseri ( Boy, 1988; Meckert, 1993; Schoch, 2003; R. R. Schoch, pers. observ.) is considered a stereospondylomorph (stem stereospondyl) by all recent authors ( Boy, 1990; Schoch & Milner, 2000; Yates & Warren, 2000).

Ingroups:

4. the Rhinesuchidae , here represented by Rhineceps nyasaensis ( Watson, 1962) and Uranocentrodon senekalensis ( van Hoepen, 1915; R. R. Schoch, pers. observ.);

5. Lydekkerina huxleyi ( Broili & Schröder, 1937; Shishkin, Rubidge & Kitching, 1996; Pawley & Warren, 2005);

6. the chigutisaurid Siderops kehli ( Warren & Hutchinson, 1983) ;

7. the brachyopid Batrachosuchus spp. ( Watson, 1919, 1956; Welles & Estes, 1969);

8. Laidleria gracilis ( Kitching, 1957; Warren, 1998);

9. the Rhytidosteidae , as a terminal taxon (based on Rhytidosteus capensis of Cosgriff, 1965; Deltasaurus kimberleyensis of Cosgriff, 1974; Peltostega erici of Säve-Söderbergh, 1936 and Janvier, 1983; Trucheosaurus major of Marsicano & Warren, 1998);

10. the plagiosaurine Gerrothorax pustuloglomeratus ( Hellrung, 2003) ;

11. the plagiosuchine Plagiosuchus pustuliferus ( Hellrung, 2003; R. R. Schoch, pers. observ.);

12. the capitosauroid M. giganteus ( Schoch, 1999) , as representative of the capitosauroids (single exception: character 43, the derived state of which occurs in Mastodonsaurus , whereas other capitosauroids retain the plesiomorphic state, see Schoch, 2000 and Damiani, 2001),

13. the long-snouted trematosaurid Aphaneramma rostratum ( Wiman, 1917; Säve-Söderbergh, 1936);

14. Trematolestes hagdorni ( Schoch, 2006) ;

15. Lyrocephaliscus euri ( Säve-Söderbergh, 1936; Mazin & Janvier, 1983);

16. Rileymillerus cosgriffi ( Bolt & Chatterjee, 2000) ;

17. Almasaurus habbazi ( Dutuit, 1972, 1976);

18. the Metoposauridae , as repesented by Dutuitosaurus ouazzoui ( Dutuit, 1976) , Buettneria perfecta ( Sawin, 1945; Colbert & Imbrie, 1956; Hunt, 1993), and M. diagnosticus ( Meyer & Plieninger, 1844; Meyer, 1857; Hunt, 1993; Sulej, 2002; Milner & Schoch, 2004);

19. C. kugleri , as based on the present findings.

Results

I shall first report the consensus obtained by all variants of the phylogenetic analysis, then list the differences among variants, and finally discuss the two alternative phylogenetic hypotheses. These alternatives differ in the position where the short-faced stereospondyls (plagiosaurids, Laidleria , chigutisaurids, and brachyopids) nest: a basal split is called the pretrematosaurian hypothesis, a position of the shortfaced clade within the trematosaurs of the trematosaurian hypothesis.

Consensus: (1) Metoposaurids and Callistomordax are always sister groups; (2) metoposaurids and Callistomordax are nested within a larger trematosaurian clade that includes trematosaurids, lyrocephaliscids, and Almasaurus ; (3) within that trematosaurian clade, Aphaneramma and Trematolestes form a monophylum; (4) Mastodonsaurus (as representing capitosauroids) is not nested with Lydekkerina , but forms its own branch; (5) the short-skulled stereospondyls form a monophylum, falling into two distinct clades; (6) a brachyopoid clade including brachyopids ( Batrachosuchus ) and chigutisaurids ( Siderops ); and (7) a clade formed by Laidleria and the plagiosaurids ( Gerrothorax plus Plagiosuchus ). The position of the Rhytidosteidae is suggested to be ambiguous by a bootstrap value below 50, and proved to be critical to the whole analysis.

Variants of analysis: The analysis gave two strikingly divergent results, depending on the inclusion or exclusion of the Rhytidosteidae ( Fig. 11 ). Other variants (exclusion of Trimerorhachis , exclusion of Batrachosuchus , exclusion of one or both plagiosaurids, and exclusion of Laidleria with or without the retention of plagiosaurids) had only a minor effect on the topology. The following sections summarize all those points in which these two alternatives (and all their variants) disagree.

Pretrematosaurian hypothesis ( Fig. 11A ): This is based on a matrix including Rhytidosteidae . The main feature of this topology is that the short-skulled stereospondyls branch off before the capitosaurian– trematosaurian dichotomy. The resulting three most parsimonious trees comprise all possible variants of the three-taxon statement Rileymillerus , Almasaurus , and ( Callistomordax plus Metoposauridae ). Each of the three topologies requires 222 steps and has a consistency index of 0.536, a retention index of 0.719, and a rescaled consistency index of 0.386.

Trematosaurian hypothesis ( Fig. 11B ): A matrix that excludes the Rhytidosteidae gives a quite different result. The topology of the trematosaurian clade is very poorly resolved, with a basal polytomy that gives six alternative most parsimonious trees. The differences to the pretrematosaurian hypothesis are as follows: the short-skulled stereospondyls – brachyopids, chigutisaurids, Laidleria , and the two plagiosaurids – are nested with the trematosaurian clade, forming its sister taxon. The trematosaurian clade proper forms a polytomy with the following constituents: (1) a clade including Aphaneramma plus Trematolestes ; (2) Lyrocephaliscus ; (3) Almasaurus ; (4) Rileymillerus ; and finally (5) a clade encompassing the Metoposauridae plus Callistomordax . (In the strict Adams consensus tree variant, Lyrocephaliscus , Almasaurus , the Metoposaridae, and Callistomordax form successive sister groups, whereas Rileymillerus falls outside and is nested with the basal trematosaurian polytomy). In all variants of this analysis, the capitosauroids thus form the sister group of trematosaurians plus the short-skulled stereospondyls. Each of the six topologies requires 219 steps and has a consistency index of 0.543, a retention index of 0.721, and a rescaled consistency index of 0.392.

Preferred phylogeny ( Figs 10 , 11A , 12 ): The hypothesis found to be more plausible here is the pretrematosaurian hypothesis. The reasons for the preference are threefold: (1) it does not exclude ‘problematic’ taxa such as the rhytidosteids; (2) some of the characters supporting the trematosaurian hypothesis are either multiply homoplastic (i.e. they occur even outside stereospondyls), or are inadequately understood at present; and (3) a calibration of the two hypotheses with the fossil record reveals that the pretrematosaurian hypothesis requires markedly shorter ghost lineages than the trematosaurian hypothesis ( Fig. 13 ).

In the following section I discuss all nodes on Figure 10 with regard to their former appearance in the literature, their support from characters, and their relative robustness, as assessed by the decay-testing procedures Bootstrap and Bremer, both of which were performed in PAUP 3.1. To make the results easier to read, I use the operational abbreviation ‘BPR clade’ for the possibly monophyletic assemblage of brachyopids, chigutisaurids, Laidleria , the plagiosaurids, and the rhytidosteids. The name ‘trematosaurian clade’ will be used for the probable monophylum formed by trematosaurids, Lyrocephaliscus , Almasaurus , Rileymillerus , the metoposaurids, and Callistomordax .

Node A: The present analysis supports a monophyletic Stereospondyli . This includes rhinesuchids, Lydekkerina , the BPR clade, the capitosauroids, and the trematosaurian clade. Stereospondyls as such have been proposed by Yates & Warren (2000) and Yates (1999), whereas Schoch & Milner (2000) excluded both the plagiosaurids and brachyopids from the Stereospondyli proper. The present analysis supports this clade robustly, with seven synapomorphies (characters 14, 26, 44, 62, 69, 85, and 97), six steps of Bremer support, and a 96% Bootstrap.

Node B: All postrhinesuchid stereospondyls. Most of the recent phylogenetic studies place Lydekkerina in a slightly advanced position with respect to rhinesuchids, regardless of whether the latter form a clade or a grade. Here, this hypothesis is supported by three synapomorphies (33, 34, and 53), one step of Bremer support, and a 74% Bootstrap.

Node C: Stereospondyls higher than rhinesuchids and lydekkerinids. This is in accordance with Yates & Warren (2000), but contradicts the concept of Schoch & Milner (2000) and Damiani & Yates (2003), who both found (at least) the rhytidosteids to be nested with the Lydekkerinidae . The present hypothesis is firmly supported by four synapomorphies (45, 52, 55, and 59), five steps of Bremer support, and a 99% Bootstrap.

Node D: BPR clade (all short-skulled stereospondyls: Chigutisauridae , Brachyopidae , Laidleria , Plagiosauridae , and Rhytidosteidae ). First suggested by Warren & Black (1985) and underscored with a cladistic analysis by Yates & Warren (2000), the monophyly had been doubted by Milner (1990) and Schoch & Milner (2000). In the present analysis the hypothesis of a monophyletic BPR assemblage is suggested, although it lacks support from unequivocal synapomorphies, and relies on two steps of Bremer support and a Bootstrap of below 50%.

Node E: Brachyopoidea plus ( Plagiosauridae plus Laidleria ). This clade has not been found by previous authors; Yates & Warren (2000) suggested a similar grouping but included rhytidosteids, which they conceived more derived than plagiosaurids and Laidleria , forming the sister taxon of the brachyopoids. The group proposed here is supported by one synapomorphy (character 21), two steps of Bremer support, and a 80% Bootstrap.

Node F: Brachyopoidea ( Chigutisauridae plus Brachyopidae ). This clade was proposed by Warren & Hutchinson (1983), and found by Warren & Black (1985) and Yates & Warren (2000) to be monophyletic. In the present hypothesis it is supported by one synapomorphy (60), three steps of Bremer support, and a 70% Bootstrap.

Node G: Plagiosauridae plus Laidleria . This clade was first found by Yates & Warren (2000). Here it is supported by five synapomorphies (characters 30-2, 68-2, 77, 99, and 100), three steps of Bremer support, and an 82% Bootstrap.

Node H: Plagiosauridae ( Gerrothorax plus Plagiosuchus ). Plagiosaurids have mostly been considered a robust monophylum ( Watson, 1919; Romer, 1947; Panchen, 1959; Yates & Warren, 2000). In the present hypothesis, plagiosaurid monophyly is firmly supported by four synapomorphies (characters 1, 72-2, 82, 84), seven steps of Bremer support, and a 100% Bootstrap.

Node I: Capitosauroidea plus trematosaurian clade. This clade was recognized by Milner (1990) and Schoch & Milner (2000) but not by Warren & Black (1985) and Yates & Warren (2000), who found the BPR clade nested within their Trematosauria . The case for a capitosaurian–trematosaurian ‘superclade’ is not particularly strong, being supported by only two synapomorphies (characters 12 and 98), one step of Bremer support, and a Bootstrap of below 50%.

Node J: Trematosaurian clade. A very similar group has been suggested by Milner’s (1990) phylogenetic study, followed by Schoch & Milner (2000) and Damiani & Yates (2003). Steyer (2002) disagrees, excluding Almasaurus and the Metoposauridae from the trematosaurian clade, whereas Warren & Black (1985) and Yates & Warren (2000) added the entire BPR clade to their Trematosauria . The trematosaurian clade as defined here is supported by three synapomorphies (characters 4, 73, and 95), three steps of Bremer support, and a 62% Bootstrap.

Node K: Aphaneramma plus Trematolestes . The monophyly of these two well-preserved trematosaurians is supported by two synapomorphies (characters 83-2, 91), two steps of Bremer support, and a 60% Bootstrap.

Node L: Lyrocephaliscus plus ( Almasaurus , Rileymillerus , Metoposauridae , and Callistomordax ). This new grouping is supported by one synapomorphy (character 86-2), one step of Bremer support, and a Bootstrap below 50%.

Node M: Almasaurus , Rileymillerus , plus ( Metoposauridae plus Callistomordax ). This unresolved trichotomy is supported by one synapomorphy (character 74), three steps of Bremer support, and a Bootstrap of 56%.

Node N: Metoposauridae plus Callistomordax . This clade is supported by four synapomorphies (characters 29, 61, 65, and 75), two steps of Bremer support, and a 73% Bootstrap.

DISCUSSION

Before the two divergent hypotheses are discussed, it should be emphasized that the present analysis is necessarily constrained and incomplete. Limited by the need to keep the number of taxa at an operational level, various interesting taxa, such as the derwentiids (see Schoch & Milner, 2000 for a definition), lonchorhynchine trematosaurids ( Welles, 1993), and other trematosaurians ( Damiani & Yates, 2003; Damiani, 2004), had to be left out. However, Steyer (2002), Damiani & Yates (2003), and Schoch (2006) have recently worked on a phylogeny of trematosaurids, where no clear consensus could be reached.

Likewise, the relationship of the dvinosaurians – a clade suggested by Yates & Warren (2000) to encompass trimerorhachids, saurerpetontids, Dvinosaurus , and tupilakosaurids – was not the focus of interest here. The present study accepts Yates & Warren’s (2000) concept of dvinosaurians being relatively primitive temnospondyls, as opposed to brachyopoids (brachyopids plus chigutisaurids), which here nested deeply within a monophyletic Stereospondyli . Among these, I included only the brachyopoids that were also found to be monophyletic here.

The stem of the Stereospondyli has been focused on by Boy (1990), Gubin (1997), Yates & Warren (2000), Schoch & Milner (2000), and Witzmann & Schoch (2006), who came to broadly similar conclusions. S. haeuseri , the best known among the most primitive stereospondylomorphs, is here considered the most closely related outgroup of the analysed set of taxa.

Pretrematosaurian origin of brachyopoids and plagiosaurids

In the hypothesis preferred here, brachyopoids, Laidleria , and plagiosaurids originated before the trematosaurian–capitosaurian dichotomy. On a general scale, this was suggested by Milner (1990) and Schoch & Milner (2000), but these authors sought the origin of both brachyopids and plagiosaurids outside the stereospondyls, and linked chigutisaurids with lydekkerinids.

In the present scenario, brachyopids and chigutisaurids are sister taxa nested with Laidleria and the plagiosaurids. The rhytidosteids form the most primitive branch of that short-skulled clade, or equally likely a grade at their stem. Several clearly plesiomorphic character states of rhytidosteids highlight the plausibility of that concept, as do the primitive features shared by plagiosaurids and rhytidosteids.

Palatine ramus of pterygoid (character 53 and 58): In temnospondyls, the anterior ramus of the pterygoid was subject to manifold modifications, in most cases involving a thinning or foreshortening of the ramus ( Milner, 1990). In capitosauroids and the large trematosaurian clade defined here, the pterygoid fails to reach both the vomer and the palatine. The only exceptions are taxa in which an apomorphic posteromedial projection of the palatine contacts the pterygoid. In brachyopoids, Laidleria , and plagiosaurids, a similar morphology is established, but always without a posteromedial process of the palatine. This potential synapomorphy of capitosauroids, trematosaurs, and short-faced stereospondyls is weakened by the rhytidosteid Deltasaurus , which possesses a wellestablished contact between the pterygoid and palatine. This character distribution suggests shortskulled stereospondyls evolved the reduced pterygoid in parallel with capitosauroids and trematosaurs. The frequent reduction of the pterygoid in taxa as far apart as dissorophoids, zatracheids, plagiosaurids, and capitosauroids makes a convergent acquisition of a reduced pterygoid plausible.

Shagreen of palatal denticles: The retention of additional tooth patches on the vomer, palatine, and ectopterygoid is a plesiomorphic character state retained by rhytidosteids. These palatal bones are covered by small teeth in rhinesuchids and the more primitive stereospondylomorphs ( Schoch & Milner, 2000; Witzmann, 2006). In combination with other characters, this suggests the clade branched before the capitosaurian–trematosaurian dichotomy.

Anterior palatal vacuity (character 43): The palatal vacuity of lydekkerinids, capitosauroids, brachyopoids, rhytidosteids, and plagiosaurids is unpaired, and the very similar outline of the anterior palatal depression in rhinesuchids suggests this to be the primitive condition. (In rhinesuchids the anterior palate may be perforated, but then differs from all other cases in being tiny and deeply emplaced within a much larger unpaired depression, see Schoch, 2000.) Early in the evolution of the trematosaurian clade a wide medial bridge separated the anterior palatal opening in the midline, which is retained throughout the trematosaurian clade as defined here. In capitosauroids, such a medial subdivision evolved at least twice: once in mastodonsaurids and a second time within the cyclotosaurids (Schoch, 2000). The primitive condition for capitosauroids was a heartshaped unpaired opening.

Knife-edged cultriform process (character 50): Trematosaurians are readily recognized by this derived feature, and even highly derived taxa such as Almasaurus and Callistomordax retain this. Metoposaurids lack this condition, instead having a greatly expanded and flattened parasphenoid that obviously evolved after Callistomordax separated from the metoposaurid stem line. None of the short-skulled stereospondyls has a knife-edged cultriform process.

Trematosaurian origin of brachyopoids, Laidleria , and plagiosaurids

This concept was outlined and developed by Warren & Black (1985) and Yates & Warren (2000). In the present analysis, it is only supported in those variants of the analysis in which rhytidosteids are excluded from the data matrix. The reason for this is that it prevents the plesiomorphic character states of rhytidosteids to outweigh the derived character states shared between some or all taxa of the trematosaurian clade and the short-skulled stereospondyls. Laidleria , which has previously been considered as a rhytidosteid relative ( Schoch & Milner, 2000), was not found to be closely related with the ATM clade, and its exclusion does not affect the resulting topology. I have not dealt with the in-group phylogenies of brachyopids, chigutisaurids, or rhytidosteids. The various papers of Anne Warren and coworkers have covered these questions in depth, and suggested that the relationship may be more complicated than a simple ‘two families – two clades’ solution ( Warren & Hutchinson, 1983; Warren & Black, 1985; Damiani & Warren, 1996; Marsicano & Warren, 1998; Warren, 1998; Warren & Marsicano, 2000; Yates & Warren, 2000).

Location of orbit (character 5): In stereospondyls the lateral placement of the orbits is shared by all taxa of the BPR clade, and also by all representatives of the trematosaurian clade. However, this state evolved in parallel at least once outside the Stereospondyli , namely in the vast dvinosaurian clade ( Milner, 1990; Warren, 1999).

Tusks keeled (character 35): Carinate fangs are present in brachyopoids ( Warren & Davey, 1992), and in some taxa of the trematosaurian clade. Within the metoposaurids, keeled tusks and marginal teeth have been reported in Metoposaurus ( Milner & Schoch, 2004) , but poor preservation or inadequate preparation of other material precludes an assessment of this character in other metoposauids at the moment. However, the occasional presence of carinate teeth in other temnospondyls ( Cyclotosaurus, Kuhn, 1942 ; Sclerocephalus, R. R. Schoch , pers. observ.) suggests that this state must have arisen at various times in parallel.

Characters of unclear significance

Lacrimal presence (characters 7 and 8): This is a critical character to any phylogenetic study of stereospondyls. Brachyopoids, rhytidosteids, and Laidleria have no lacrimal, and in some trematosaurids, some metoposaurids, Rileymillerus , and in Callistomordax the element is in an unusual position and/or has a peculiar morphology. Bolt & Chatterjee (2000) suggested that in Rileymillerus the tiny element wedged in between the jugal and prefrontal is a laterally exposed palatine (LEP) rather than a lacrimal. If this be the case, such a structure would have formed convergently to the condition in dissorophoids ( Bolt, 1974), saurerpetontids ( Sequeira, 1998), and tupilakosaurids ( Marsicano & Warren, 1998). In an LEP, the palatine forms an ornamented dorsal projection set into the rim of the orbit. However, in Callistomordax the palatine and its supposed dorsal projection are separated by disruption along a horizontal plane, suggesting that they were attached by means of a suture rather than forming a unit. This does indeed indicate that the small element may be a lacrimal that failed to expand anteriorly, or alternatively – as is probably the case in Trematolestes ( Schoch, 2006) – it was overplated by the prefrontal and maxilla, respectively. The absence of a lacrimal in rhytidosteids, brachyopoids, and Laidleria represents a shared derived state, but the condition in plagiosaurids is uncertain. Despite having been figured as possessing a lacrimal in the plesiomorphic position, I found no unequivocal evidence of a lacrimal being present in any of the skulls I examined (contra Hellrung, 2003). This leaves the question unsettled, and poses an additional problem to the already difficult homology question just outlined. Hence, without any substantial new data on the plagiosaurids, or the sutural structure of the ‘lacrimal’ region in brachyopids, rhytidosteids, and many trematosaurids, there will be no clarification of this point.

Ribs with uncinate spines (character 81): This feature is very interesting yet confined to the few taxa of which articulated finds are available, in particular Trematolestes and Callistomordax . As all other derived character states suggest that these two genera are not intimately related, the possession of such elongated spines must be a more widespread feature, probably characterizing a grade within the trematosaurian clade, similar to the possession of a knife-edged parasphenoid.

Narrow intercentra (character 73): This feature is again restricted to a few trematosaurian taxa, and, among the available material, it is best exemplified by the Trematolestes and Callistomordax .

Relationships of the Plagiosauridae

The plagiosaurids have not formed the focus of the present study, but their inclusion is believed to be critical to the analysis. The findings of the present analysis corroborate the data of Yates & Warren (2000), in which the small (yet unfortunately imperfectly known) Laidleria appears to form a plausible sister taxon. A critical synapomorphy, the carapace of dorsal osteoderms, is perhaps not as convincing, as dermal ossicles are known from several other, more distant temnospondyls: dissorophids ( DeMar, 1966), trematopids ( Berman, Reisz & Eberth, 1985), Peltobatrachus ( Panchen, 1959) , and Sclerothorax ( Schoch et al., 2007) . On the other hand, none of these taxa has a laterally widened carapace and correlated rib cage (character 100), tiny postfenestral windows (character 30-2), or lacks pleurocentra, whereas the intercentra are greatly enlarged and tightly fitting (character 77). Thus, although the transformation of numerous characters remains unknown, Laidleria appears to be a rather good candidate in looking for the origin of plagiosaurids. Resolution of this question requires further study of plagiosaurids, particularly the rich material of Plagiosuchus and Plagiosternum from Germany, as well as a thorough reconsideration of Peltobatrachus pustulatus , which Panchen (1959) suggested as a plagiosaurid relative.

The origin of metoposaurids

All analyses performed for the present study firmly place Callistomordax as the sister taxon of metoposaurids, nested above Almasaurus and Rileymillerus . This provides further evidence for a trematosaurian origin of metoposaurids, a concept first suggested by Milner (1990), expanded by Schoch & Milner (2000), and essentially confirmed by a cladistic analysis in Damiani & Yates (2003). The divergent topology proposed by Steyer (2002), placing the metoposaurids plus Almasaurus outside the Trematosauria proper, is based on a more restricted set of characters, most of which were included in the present analysis. Steyer’s (2002) analysis did not consider short-skulled stereospondyls (thereby causing homoplasies shared between these and trematosaurians to be treated as synapomorphies), and included only two postcranial characters. Postcranial characters shared between the almost completely known Trematolestes ( Schoch, 2006) and Callistomordax have turned out to be shared derived states. Along with other postcranial data also shared with Aphaneramma and Lyrocephaliscus , these postcranial data have a strong impact on the placement of Callistomordax , and consequently of the metoposaurids. Most of these postcranial features are not synapomorphic, because they are reversed or modified in metoposaurids, which evolved flattened trunks and stereospondylous intercentra in parallel to capitosauroids and brachyopoids.

The incompletely known Microposaurus casei may also be a close relative of Almasaurus and the metoposaurid stem. Unfortunately, its lack of sutures resulting from large-scale co-ossification prevents a more definitive assignment of the taxon, although Damiani (2004) has convincingly argued for a platystegid relationship of Microposaurus . Skull outline and general proportions of Microposaurus match those of Almasaurus almost as well as those of Inflectosaurus and Platystega . It is quite possible that platystegids and almasaurids formed a grade within which Microposaurus and Almasaurus were slightly more advanced towards the metoposaurid condition. Many other features are derived character states shared with some trematosaurians and/or brachyopoids, such as the configuration of the posterior skull table, its ornamentation, the foreshortened preorbital region, and the apomorphic condition of the lacrimal region. Almasaurus , Callistomordax , and the metoposaurids share only one unequivocal synapomorphy, the quadrangular ventral surface of the intercentrum.

Unlike Almasaurus and Rileymillerus , Callistomordax had already acquired several important metoposaurid synapomorphies: among these, the more massive medial trochlea of the quadrate (character 61), the occiput with its deep sloping postparietals and tabulars (character 29), and the anteriorly convex intercentrum (character 75-1), are the most outstanding.

On the other hand, the palate of Callistomordax retains numerous plesiomorphic features, such as the structure of the basicranium, mixed with autapomorphies like the dentition and many aspects of its postcranial anatomy. The elongation of the trunk, paralleling the situation in dvinosaurians, was accomplished in a unique way: the intercentra formed bulbous, ventrally elongated wedges that despite the rather conventional presacral count of 26–28 amounted to a trunk three times the length of the skull. (This feature may also characterize Almasaurus , but the existing postcranial data of that taxon are too poor to permit clarity).

The stereospondyl condition (character 72-1) probably evolved in parallel, once in capitosauroids ( Mastodonsaurus ), a second time in plagiosaurids where the intercentrum became cylindrical (character 72-2), and a third time in metoposaurids. The mid-trunk intercentra of Mastodonsaurus and the metoposaurids – which usually form the most fully ossified centra – differ in that those of Mastodonsaurus are higher, forming near-perfect circles in transverse outline, as contrasted by transverse ovals in metoposaurids ( Dutuit, 1976: fig. 35; Schoch, 1999: figs 28, 31). The neural arches are also substantially higher in capitosauroids as compared with metoposaurids.

In Dvinosaurus , Trimerorhachis , and Kourerpeton , the elongation of the body was produced by an increase in the number of vertebrae ( Bystrow, 1938; Olson & Lammers, 1976), and in tupilakosaurids this led to the evolution of diplospondylous, discshaped vertebrae ( Shishkin, 1973; Warren, 1999), which effectively doubled the number of elements and thereby increased flexibility. This degree of flexibility was obviously not reached by Callistomordax .

Figures

Figure 10. Cladogram, depicting results of the preferred phylogeny (pretrematosaurian hypothesis). Nodes are defined and discussed in the text. See the Phylogenetic analysis for details.

Figure 11. Comparison of alternative hypotheses (consensus trees) derived from variant cladistic analyses. A, pretrematosaurian hypothesis (including Rhytidosteidae); B, trematosaurian hypothesis (excluding Rhytidosteidae).

Figure 12. Supporting synapomorphies, mapped onto a cladogram of the preferred phylogeny. Only unequivocal character distribution is counted here as synapomorphy.

Figure 13. Stratigraphical calibration of the two phylogenies discussed in the text. A, pretrematosaurian hypothesis (preferred phylogeny); B, trematosaurian hypothesis. Bold dashed lines mark ghost lineages that are substantially longer than in the alternative scenario, pentagons mark origin of major clades.