Thelodont phylogeny revisited , with inclusion of key scale-based taxa

Knowledge of the Thelodonti has improved greatly in recent years, but phylogenetic relationships remain poorly understood. We revised the data from an earlier phylogenetic study and added 15 scale-based species. Maximum parsimony analysis gives a well-resolved tree in which Archipelepis and Boothialepis form a basal clade, recognized here as the order Archipelepidiformes, sister group to two large clades also recognized as orders. The first is recognized as the order Furcacaudiformes, including Nikolivia, Lanarkia, Phillipsilepis, Pezopallichthys, Drepanolepis (in the new family Drepanolepididae), Barlowodus, Apalolepis, and Furcacaudidae. The second, here recognized as the Thelodontiformes, contains Turinia, Thelodus, Stroinolepis, Loganellia, Longodus, Helenolepis, Phlebolepis, Erepsilepis, Trimerolepis, Eestilepis, Valiukia, Paralogania, and Shielia.


INTRODUCTION
Our understanding of the Palaeozoic vertebrate subclass Thelodonti has improved greatly in recent years as a result of new discoveries (e.g., Wilson & Caldwell 1993, 1998;Märss 1999;Märss et al. 2002) as well as monographic revisions of articulated specimens and scalebased taxa (e.g., Märss & Ritchie 1998;Karatajūtė-Talimaa & Märss 2004;Märss et al. 2006Märss et al. , 2007)).However, in the early years of study of thelodonts, they proved to be a difficult group for workers to deal with.Thelodont scales were first described by L. Agassiz (in Murchison 1838) but their status as a distinct group of jawless vertebrates evolved slowly over many decades (see historical review by, e.g., Turner 1991).Thelodonts preserved as articulated skeletons have been known since Powrie (1870) described 'Cephalopterus' pagei, the generic name of which was subsequently found to be preoccupied.Powrie believed his species to be an acanthodian.This error was corrected by Traquair (1896), who named the genus Turinia to contain Powrie's species and grouped it with other well-preserved Scottish thelodonts as jawless vertebrates (Agnatha).Thelodonti were formally recognized as a higher taxon within Agnatha but distinct from other jawless vertebrates by Kiaer (1932).
In recent decades new discoveries and monographic treatments of thelodonts have included a detailed study of scale-based species from the former Soviet Union and Spitsbergen (Karatajūtė-Talimaa 1978), scale-based species of Estonia and Latvia (Märss 1986), Silurian species based on articulated specimens from Scotland (Märss & Ritchie 1998), Silurian and Devonian species of a new group called Furcacaudiformes by Wilson & Caldwell (1993, 1998), and Silurian and Devonian species based on scales and articulated squamations from Arctic Canada (Märss et al. 2006).Thelodont studies have also been significantly advanced by publication of a reference work on thelodonts from Russia and adjacent countries (Karatajūtė-Talimaa & Märss 2004) and by the thelodont volume of the Handbook of Paleoichthyology (Märss et al. 2007).
However, the phylogenetic relationships within the group have remained problematic, and the question of monophyly of the Thelodonti has also been controversial (e.g., Turner 1991;Janvier 1996;Donoghue & Smith 2001;Wilson & Märss 2004).
The present contribution is our second attempt at resolving the within-group relationships of thelodonts.Our first attempt (Wilson & Märss 2004) was based almost entirely on articulated specimens (in reality, articulated squamations).In that paper 25 species of thelodonts were studied and a preliminary phylogenetic arrangement was proposed.For many of them, data were available for both scale histology and overall body form.For others, however, one of these key sets of features was absent due to imperfect preservation or lack of the needed analysis.The resulting phylogenetic tree (Wilson & Märss 2004, fig. 6; reproduced as Märss et al. 2007, fig. 34) suggested a basal split within Thelodonti between a group consisting of Archipelepis, Phlebolepis, and Erepsilepis and all other thelodonts.The latter group was further divided into one that included Turinia, Loganellia, and Phillipsilepis and a larger separate group of remaining species.This larger group was then divided between a clade consisting of Shielia spp.and Lanarkia spp., collectively sister to the fork-tailed thelodonts or Furcacaudiformes.The basic structure of that tree is reproduced here in Fig. 1A for comparison with new findings.
However, the thelodont fossil record includes many very important species that are based solely on isolated scales.Isolated scales derived from acid dissolution of fossiliferous rocks can often be associated with considerable assurance into suites of scales representing different parts of the body of an individual species, using the clues given by intergradations of scale structure within the samples (e.g., Märss 1999).Not only are these scale-based species important for biostratigraphy, but they also often yield the most complete data concerning scale microstructure and histology.Such details tend to be best preserved in carbonate or carbonatecemented rocks that are processed with acetic acid to yield scale-bearing residues.
Fig. 1.Comparison of the results of the earlier analysis (A) of 24 thelodont species from Wilson & Märss (2004) with the results of analysing the same 24 species using the newly revised data matrix (B).The main differences between the two analyses (indicated by shading) are in the position of Erepsilepis plus Phlebolepis, the relationships of Phillipsilepis and Lanarkia, the position of Shielia spp., and the more basal position of furcacaudiforms and Turinia in (B).
The original study (Wilson & Märss 2004) began with data on 25 species and four outgroups scored for 53 characters, but one outgroup and one ingroup species were eliminated because of wildcard behaviour in the analyses (in multiple shortest trees, these species took radically different positions, causing a lack of resolution in the strict and majority-rule consensus trees).The final preferred phylogeny was thus based on 24 species of thelodonts and three outgroups.
In the present study we have augmented the original list of taxa from our earlier study by adding data on representative scale-based species.Some 14 scale-based species, each in a different genus, have been added for the present study, although a few of the new species are assigned to genera included in the earlier analysis.We have also revisited the list of characters and states, eliminating some and substituting others, and recoded the previous set of taxa.Our purpose is to produce a new phylogenetic analysis of 39 species of thelodonts that includes both scale-based and squamation-based species.We also compare the effect of updating the data matrix on an analysis of the original 24 species, and discuss the implications of the new phylogeny for thelodont evolution and classification.

METHODS
Our starting point was the data from our earlier study (Wilson & Märss 2004); we updated the data by editing and recoding of states and by deleting characters that were found to be highly homoplasious in the earlier study (Wilson & Märss 2004) or were rendered uninformative after editing.We modified and reduced the number of states for several characters to simplify them and to attempt to capture the major features of character evolution rather than minor variations.We also reexamined the character coding of every species critically and made a number of changes based on our current understanding of body form and of scale morphology and histology.Finally, we added a small number of new characters to replace deleted characters, including one to reflect the new information available about ultrasculpture of the scale surface (Märss 2006a).The resulting list of 52 characters and states is given here as Appendix 1 and the resulting character-taxon matrix for 42 taxa (39 ingroup species and 3 outgroups) is given as Appendix 2. Key features of thelodont scale morphologies used as the basis for character definitions in Appendix 1 are shown in Fig. 2. Similarly, key features of scale histology used for character definitions are shown in Fig. 3. Thelodont bodies and articulated squamations were illustrated in our earlier paper (Wilson & Märss 2004) and also by Märss et al. (2007).We used the same three outgroups as in our earlier study, including 'Athenaegis Tolypelepis' which is a composite of two of the most primitive undoubted heterostracans so far known, Athenaegis being the main source of body-form character states and Tolypelepis being the main source of histological character states.A study such as this with only three representatives of the many species in outgroup taxa is not designed to test thelodont monophyly.A more rigorous test of monophyly would involve a much larger sampling from among other jawless vertebrates, and a much larger number of characters.In the present study we have assumed thelodont monophyly, using outgroups to root the resulting tree, thereby indicating the direction of evolutionary change within the group.
We prepared and edited the data matrix using MacClade Version 4.08 (Maddison & Maddison 2005) and analysed the matrix under the criterion of Maximum Parsimony using PAUP 4.0b10 (Swofford 2002).Using PAUP, we obtained the set of the shortest trees, the strict and majority-rule consensus trees based on those shortest trees, and we also generated random trees for the purpose of assessing the strength of the phylogenetic signal in the data matrix.Using MacClade again, we graphically displayed the strict and majority-rule consensus trees, mapped character changes onto them, and generated tree-fit and character-specific statistics.
In PAUP, we performed heuristic searches with mostly default options, except for obtaining starting trees by random addition with 100 replicates.Branch-Swapping was by TBR (Tree Bisection-Reconnection), with the steepest descent off, multrees in effect, swapping on the best trees only, and no topological constraints.Rooting was on the single composite outgroup (Athenaegis Tolypelepis), although data for two other outgroups (Rhyncholepis parvula and Tremataspis schmidti) were included in the analysis.
The character-taxon matrix contained 52 characters, all of them informative, all treated as unordered and of equal weight.The number of states per character ranged from 2 to 7, and the total number of apomorphic states was 80. Missing data per character ranged from zero to 83.3%, while inapplicable states per character ranged from zero to 85.7%.
We investigated whether the differences in the resulting phylogenies between our 2004 paper and the current study were caused mostly by changes in the characters and states or whether the difference in results was caused by addition of scale-based taxa.To examine this question, we re-analysed the same set of 24 taxa from the earlier study (Wilson & Märss 2004) but using the updated set of characters and states (Appendices 1 and 2).We then compared the resulting phylogeny with that produced by the earlier study.
We then added the 14 newly coded, scale-based taxa along with one ingroup species, Eestilepis prominens, that had been coded for our 2004 study but omitted because of wild-card behaviour.The complete data set was then analysed using the above-mentioned protocols.We also investigated the possibility of adding a fourth outgroup (Poracanthodes menneri) that we had attempted to include in 2004.

RESULTS
The analysis of the original 24 ingroup and three outgroup taxa using the original 53 characters (Wilson & Märss 2004, fig. 6) had given the preferred result illustrated here in Fig. 1A (majority-rule consensus of the 31 shortest trees of length 153 steps).When analysed with the updated data set, the same 24 species yield the arrangement in Fig. 1B (majority-rule consensus of 212 trees at 156 steps).It is evident that the updated data matrix produces a different phylogenetic arrangement for several important taxa.Erepsilepis and Phlebolepis were grouped with Archipelepis near the base of the tree in the 2004 study (Fig. 1A), but with the updated data they are grouped with Loganellia and Shielia (Fig. 1B).In addition, Shielia itself was grouped with Lanarkia in the 2004 study (Fig. 1A), but is united with Loganellia, Erepsilepis, and Phlebolepis using the updated data (Fig. 1B).Turinia was united with Loganellia in 2004 but takes a much more primitive position using the revised data.
When the 14 scale-based taxa as well as Eestilepis prominens, known from a partially articulated squamation, were added to the analysis, the data matrix contained 39 thelodont species and three outgroup taxa, coded for 52 characters.
As with our previous study (Wilson & Märss 2004), our attempt to include the acanthodian outgroup Poracanthodes menneri was not successsful.The primary character-state homologies of this species, relative to the characters designed for thelodont relationships, are difficult to determine because of the lack of comparability of its morphology and scales with the features of thelodonts.Moreover, when we included our tentative codings for Poracanthodes in the analysis, all structure of the phylogeny was destroyed and a large basal polytomy was generated.We therefore eliminated Poracanthodes as an outgroup and do not include it in our published data because we think it unwise to disseminate those preliminary but unreliable character-state codings.
In our 2004 paper we had preferred the phylogeny that omitted Eestilepis from consideration owing to its wild-card behaviour.Unlike our experience with Poracanthodes menneri, the new data matrix was much more successful in placing Eestilepis prominens with consistency, giving a similar relationship for this species in all shortest trees, despite its large proportion of missing data.Our preferred result, therefore, includes all 39 thelodonts for which we coded data.
The result for the inclusive analysis was 2558 shortest trees of length 204 steps.The strict consensus tree (Fig. 4A) has a high degree of resolution.The inclusive majority-rule tree (Figs 4B, 5) is only slightly more resolved than the strict consensus tree (Fig. 4A).For the shortest trees, the Consistency Index (C.I.) was 0.39, the Retention Index (R.I.) was 0.72, and the Rescaled Consistency Index (R.C.) was 0.28.
Comparing the inclusive tree for 39 thelodonts with that based on the original 24 species indicates that the inclusive majority-rule tree (Fig. 4B) is about as well resolved as the tree based only on the original 24 species (Fig. 1B).The two trees differ in that furcacaudiforms are a distinct clade in the analysis of 24 taxa (Fig. 1B) but they are united with Lanarkia, Phillipsilepis, and Nikolivia when the additional scale-based taxa are included (Fig. 4B).
was 346 steps (compared to the shortest overall trees of 204 steps).The random trees had a mean length of 436 steps, a standard deviation of 12.4, and a skewness of -0.30; the shortest trees found by the heuristic procedure (Fig. 4) are thus more than 18 standard deviations shorter than the average of the random trees and we conclude that the phylogenetic signal in the data is strong.
The character changes are mapped onto the majorityrule consensus tree in Fig. 5.Note that only unequivocal changes are shown; the lack of changes adjacent to polytomies (e.g., the clade Archipelepis spp.and Boothialepis; the polytomy among furcacaudid species; the polytomy among Shielia species) is a product of this restriction.Different, arbitrary resolutions of these polytomies would give different suggested synapomorphies at adjacent nodes.For example, when any one of the three constituent species of the clade Boothialepis + Archipelepis is placed as sister to the other two species, at least four synapomorphies are mapped unequivocally on the node: 11(1), 16(0), 17(3), and 22(2).These four synapomorphies are joined by state 4(2) if either of the two species of Archipelepis is placed as sister to the other two members of the clade.Characters evolving with the minimum number of Fig. 4. Results of the present analysis (maximum parsimony, heuristic analysis, 100 random-addition starting-tree replicates) using the combined data matrix of 39 species of thelodonts and three outgroup taxa.A, the strict consensus obtained from all 2558 shortest trees of length 204 steps (C.I. = 0.39, R.I. = 0.72, R.C. = 0.28).B, the majority-rule consensus tree from the same 2558 trees of length 204 steps.All of the clades in this tree have a majority-rule percentage of 100 and appear in the strict consensus except the five clades for which percentage values are shown.Dark grey shading: the 14 scale-based species that were not included in the analysis of Wilson & Märss (2004).Light grey shading: Eestilepis prominens, which was eliminated from the earlier study because of wild-card behaviour but included in the present study because it did not display wild-card behaviour.

DISCUSSION
We consider that the present results represent a significant advance over our earlier attempt (Wilson & Märss 2004) both because of the improvements made in the data matrix and because of the more inclusive set of species in this new study.
Our analysis of the original 24 thelodont species using the updated data matrix shows that most of the changes in the relationships of these original species, compared with their relationships in the earlier study (Fig. 1A vs Figs 1B, 4), are caused by revisions to characters and states.Some other changes result from the addition of 15 mostly scalebased species to the analysis.The revised data matrix was able to place Eestilepis prominens with confidence; that species had acted as a wild-card and had been eliminated from the earlier analysis (Wilson & Märss 2004).
These results suggest to us that there is no fundamental barrier to the future analysis of additional, scale-based taxa despite their lack of data concerning body morphology.Similarly, the relationships of some of the articulated, squamation-based species for which ultrastructural data are currently lacking were not greatly changed by adding the scale-based taxa.This is good news for the future of thelodont phylogenetics.
Each of the three basic clades of thelodonts according to the present results has distinctive features indicated by the character-state changes mapped onto the phylogeny in Fig. 5, although the most basal clade (a trichotomy containing two species of Archipelepis and one of Boothialepis) has no unambiguous synapomorphies unless the trichotomy is arbitrarily resolved (see Results for details).This clade is distinguished from the other two clades by the following synapomorphies that the two larger clades share: scale base from wide to narrow (23:2-0), although the base configuration changes in some descendants to medium and/or wide, and pulp canals from absent to one (28:0-1), again with some subsequent changes to absent or more than one in certain descendants.
The large clade containing Nikolivia, Lanarkia, Phillipsilepis, and Furcacaudiformes (including Barlowodus and Apalolepis) is united by a rounded as opposed to a quadrangular scale base (21:2-0) except Furcacauda fredholmae and the clade consisting of all of these taxa except Nikolivia is united by a crown that overlaps posterior scales greatly (16:1-2), again with a few exceptions.
The largest clade, which includes Turinia, Thelodus, Loganellia, Phlebolepis, Shielia, and relatives is united by three synapomorphies.The first is that basal members of the clade share a moderately deep scale base (22:0-1), although according to our topology subsequent evolution leads to more variation in this character.The second is that most of the clade shares (homoplasiously with Lanarkia) a moderate to strong anterior process on the scale base (24:0-1).The third synapomorphy uniting the clade is a long, straight intestine (43:0-1), although only in a few species can probable gut endocasts be seen.In others, one can estimate gut proportions by the length of the post-branchial, pre-caudal trunk, which is rather long and slender in most species.However, based on a similar argument about body proportions, one would also expect a long, straight intestine in Lanarkia spp.based on their body proportions (Märss & Ritchie 1998), and if that were confirmed, the character might be optimized differently.
The entire clade apart from Turinia shares sinuous or branching dentine tubules, as opposed to straight ones, in the mid-crown of scales (31:0-1), although this feature is also seen in furcacaudiforms.All members of the clade except Turinia and Thelodus share two unreversed synapomorphies: presence of a pulp depression in the scale base (25:0-1), and absence of a pulp cavity in the scale (26:1-0).
The new phylogeny makes interesting predictions about unseen features of both scale-based and squamationbased taxa.For example, the scale-based Thelodus laevis might be expected to have a body form something like those of Turinia and Loganellia.Stroinolepis maenniki is predicted to have a Loganellia-like body form.Longodus and Helenolepis are predicted to share additional features with Phlebolepis and Shielia.Trimerolepis, Valiukia, and Paralogania are predicted to share features with Shielia.Valiukia and Paralogania had already been united with Shielia in Shieliidae (Märss et al. 2006(Märss et al. , 2007)), but the inclusion of Trimerolepis is a novel result, the genus previously being classified in Katoporodidae (Märss et al. 2007).Similarly, Barlowodus and Apalolepis are predicted to share body-form features with Furcacaudiformes rather than with the Thelodontiformes.Barlowodus had been included in the Furcacaudiformes by Märss et al. (2002Märss et al. ( , 2006) ) but it, along with Apalolepis, was not classified with Furcacaudiformes by Märss et al. (2007).________________________________________________________________________________________________________ Fig. 5.The preferred, majority-rule consensus tree of this study, based on 39 species of thelodonts with three outgroup taxa, showing unambiguous character changes mapped onto the tree using MacClade 4.08 (Maddison & Maddison 2005).For each change, the character is presented in the form 'X:Y-Z', where X is the character number, Y is the previous (plesiomorphic) state, and Z is the new (apomorphic) state for that clade.
Similarly, Boothialepis is expected to share morphological features with Archipelepis.When articulated specimens of these taxa are discovered, their morphologies will test the predictions of the phylogeny presented here.
Predictions concerning squamation-based taxa are focused mainly on those species for which limited data are currently available concerning histological or other ultrastructural features.Eestilepis prominens for example, known from a partial squamation that does not allow coding of most morphological features (Märss et al. 2002(Märss et al. , 2006)), is predicted to share histological features with Paralogania and Shielia, and morphological features with Shielia.