Sunday, May 21, 2023

Raven et al. 2023 on ankylosaur phylogeny missed the shortest trees and should not have dropped Nodosauridae

A new paper came out analyzing thyreophoran phylogenetics - Raven et al. (2023).  The abstract states "This dataset was analysed using equal- and implied-weights parsimony and Bayesian inference, and further explored using constraint trees and partitioned datasets. Stratigraphical congruence was used to identify a 'preferred tree' and these analyses reveal a novel hypothesis for thyreophoran relationships. The traditional ankylosaurian dichotomy is not supported: instead, four distinct ankylosaur clades are identified, with the long-standing 'traditional' clade Nodosauridae rendered paraphyletic. Ankylosauridae, Panoplosauridae, Polacanthidae and Struthiosauridae have distinct morphotypes..."

Four distinct ankylosaur clades with distinct morphotypes?

It sounds intriguing, but then Polacanthidae as a separate clade or grade has been a viable hypothesis since 1998, so I guess it's the paraphyly of Panoplosaurus and Struthiosaurus versus Ankylosaurus that is new?  Looking back, that was ambiguous in Vickaryous et al. (2001),  Osi (2005) and Parsons and Parsons (2009) but rejected by Osi and Makadi (2009), Thompson et al. (2012) and Arbour et al. (2016), thus it would be surprising to someone like myself who doesn't specialize in ornithischians so is not all that familiar with character support and such.

So let's see what Raven et al. found.  They did five basic analyses (A-D with [edited thanks to David Marjanovic's comment] progressively more different assumed weighting- unweighted, then k = 3, 8 and 12; E as a Bayesian attempt) then some constraint analyses and ones using only certain parts of the skeleton.  I'm going to ignore the latter as I don't think anyone considers "cranial only", "postcranial only" or "armor only" to give better results in Mesozoic dinosaur phylogenetics.  In any case, the first analysis (Analysis A) was equal weights, which is the standard for Mesozoic dinosaur analyses and how basically every prior ankylosaur analysis was run.  As an aside, statements like "The strict consensus tree (Supplementary material, Fig. S75) shows a lack of resolution in Stegosauria" .... "and Ankylosauria is found in an unresolved polytomy with most stegosaur taxa" just means you haven't pruned enough taxa a posteriori to see the underlying structure.  It's annoying the authors never look into what that structure is and leave the tree looking artificially uncertain.  But the main result in Ankylosauria is... "There are two clades within Ankylosauria (Ankylosauridae + Nodosauridae)"!  So the expected usual result.  Hmm.  But what about their novel hypothesis of three nodosaur clades at least? 

They state "Within Nodosauridae, there are three groupings of taxa: 'polacanthid' ankylosaurs, but excluding Polacanthus; a 'panoplosaurid' group typified by Edmontonia and Panoplosaurus; and a 'struthiosaurid' group typified by Struthiosaurus and Hungarosaurus."  But are there?  Just look at the cladogram above (which is only in the supp info). 'Polacanthidae excluding Polacanthus' is Texasetes, the then-unnamed Patagopelta, Sauropelta, Hylaeosaurus, Hoplitosaurus, Tatankacephalus, an undescribed Wessex specimen, Zhejiangosaurus, Antarctopelta and Dongyangopelta.  Besides Hoplitosaurus and Hylaeosaurus (which I don't know has ever been recovered as a polacanthid, just guessed to be there), I don't think any of these have been associated with Polacanthidae/inae before.  The 'core' polacanthids besides Polacanthus itself are Gastonia, Mymoorapelta and maybe Gargoyleosaurus, so I wouldn't say this clade is reflective of Polacanthidae in any sense.  But okay you say, maybe it's not Polacanthidae, but surely the analyses (PLURAL as in the abstract) revealed a 'distinct morphotype' for this Sauropelta-Hylaeosaurus clade of nodosaurs?  Looking at their "preferred tree" (Analysis B, k=3) only Patagopelta, Texasetes, Hoplitosaurus and Hylaeosaurus are shared between 'polacanthid' clades, with five new taxa in there compared to Analysis A.  And in Analysis C (k=8) only Hoplitosaurus and Texasetes are shared (with seven new taxa), while by Analysis D (k=12) all these taxa are scattered to the wind and there's no equivalent at all.  And Analysis E (Bayesian) is one huge polytomy for ankylosaurs.  So I would suggest their published results do not support any clade like this that is robust when analyzed under different criteria.

But what about Panoplosauridae and Struthiosauridae?  First of all, just as there are nine 'nodosaurs' that fall outside the Nodosauridae+ Ankylosauridae split, there are one to five nodosaurids that don't fall into either the panoplosaur or struthiosaur clades.  We can't tell how many because Raven et al. do not prune a posteriori to try to resolve any polytomies.  But the panoplosaur group is Dracopelta, Aletopelta, both Edmontonia species, Denversaurus and of course Panoplosaurus.  The latter three have always been grouped together, but the first two would be interesting if they actually were panoplosaurs.  The 'preferred tree' (k=3) takes away Aletopelta and adds Nodosaurus, Anoplosaurus and Tianchisaurus; Analysis C (k=8) has the core three genera plus Dracopelta and Anoplosaurus, while Analysis D (k=12) has everything in A plus Anoplosaurus and Nodosaurus.  So Dracopelta is always a panoplosaur and Anoplosaurus is with any unequal weighting, which are the first surprising, new and widely supported nodosaur placements in this study.  Yet neither is touted in the text or written up with character support, and honestly the idea of Albian-Cenomanian English Anoplosaurus and especially Jurassic European Dracopelta breaking up not just Campanian-Maastrichtian North American Panoplosaurus+Edmontonia, but the genus Edmontonia itself(!) just seems unlikely.  And indeed, a fairly complete Dracopelta specimen (Russo and Mateus, 2023) was recently discovered, and the authors found "D. zbyszewskii [was] consistently recovered as sister taxa of G[argoyleosaurus] parkpinorum, from the Upper Jurassic of Morrison Formation, USA, in a basal ankylosaur group that also includes the other Morrison Formation ankylosaur, M[ymoorapelta] maysi," which matches temporally so much better.

As for Struthiosauridae, the unweighted tree would have this include Borealopelta, Minmi, Niobrarasaurus, Polacanthus, Europelta, Liaoningosaurus, Stegopelta, the Paw Paw Formation juvenile (* see below), Hungarosaurus and all three Struthiosaurus species (not monophyletic, at least Tianchisaurus is closer to S. transylvanicus than to S. austriacus or S. languedocensis).  The 'preferred tree' (k=3) keeps only Struthiosaurus, Hungarosaurus, the Paw Paw juvenile and Europelta, and adds Silvisaurus and Taohelong.  Analysis C (k=8) drops the latter two but adds Tianchisaurus back, and Analysis D (k=12) keeps the two and everything from the unweighted tree plus adds Invictarx and Hoplitosaurus.  So far from dividing the 'distinct morphotypes' of Polacanthidae and Struthiosauridae, two of five trees have Polacanthus as a struthiosaur.  Here the results are besides the Santonian-Maastrichtian Central European Struthiosaurus/Hungarosaurus, the struthiosaur clade includes Europelta (as guessed by its describers), the Early Cretaceous American Paw Paw juvenile and usually the Jurassic Chinese Tianchisaurus.  Yet the authors never mention the Paw Paw juvenile ever falling out here, and have Tianchisaurus as a panoplosaurid because Analysis B is the one time it wasn't the sister to Struthiosaurus transylvanicus.  In fact, the authors falsely* state the Paw Paw juvenile is usually a basal ankylosaur and Figure 2 incorrectly* shows it being one in a supposed "Agreement subtree of the three implied weighing analyses (analyses B–D)."  How did this happen?! ...

(*) I figured it out- the trees in the supp info switched Pawpawsaurus and the Paw Paw juvenile, while figures 1-3 in the paper are correct.  So actually their trees have Pawpawsaurus as the struthiosaur sister to Hungarosaurus, which has the same chronostratigraphic issues as the juvenile from the same formation.

If we go back to the big picture in Ankylosauria, the unweighted Analysis A gave us a large Nodosauridae with some mostly Jurassic taxa basal to the Nodo-Ankylo split, and a few 'nodosaurs' (Kunbarrasaurus, Peloroplites, Liaoningosaurus) as basal ankylosaurids.  'Preferred' Analysis B (k=3) has polacanthids basal to the split, then panoplosaurs and struthiosaurs sister to each other, so again is pretty standard.  Analysis C (k=8) has struthiosaurs, panoplosaurs and polacanthids successively closer to ankylosaurids, so at least that has the paraphyletic nodosaurs the article touts.  Finally, Analysis D (k=12) has struthiosaurs further from ankylosaurids than panoplosaurs while polacanthids cease to really exist (their internal specifier Gastonia [see below] is an ankylosaurid but Polacanthus and Hoplitosaurus are struthiosaurs).

So I guess if I had a takeaway from their published results (**, see below), it would be that Dracopelta (probably incorrectly) and usually Anoplosaurus are panoplosaurs; Europelta, Pawpawsaurus and usually Tianchisaurus (not even supported by the authors) are struthiosaurs; polacanthids are not strongly supported in any form; and nodosaurs become increasingly paraphyletic with more weighting greater values of k (corrected again thanks to David Marjanovic's comment), although you need to get to k=8 for anything really novel.  And it's the weighting that is one of my major issues with this paper, because why is k=3 the preferred tree?  Because "The stratigraphically most congruent topology, as identified by the four stratigraphical congruence metrics (SCI, RCI, MSM and GER), was Analysis B, and so this was selected as the 'preferred tree'."  But if you look at their Table 2 (below), the Bayesian analysis destroys the others at SCI (0.929 vs. 0.438-0.500), but we never get to know what those results are since the authors just leave it as a huge polytomy without further analysis.  And in the other three, Analysis B is 16.954, .023 and .005 better respectively, which seems increasingly less important.  I have no idea how any of these measures work, but it seems incredibly arbitrary to say whichever k value is best in a majority of four methods wins, ignoring anything quantitative.

(**) Raven et al. didn't find the shortest trees

But now we get to the part where I reveal nothing I said above matters, because Raven et al. didn't get the shortest trees.  EDIT BELOW Not even close.  Instead of producing "eight MPTs with lengths of 1508 steps", their unweighted matrix Analysis A results in >99999 MPTs of 1464 steps.  Here's the real strict consensus with 13 taxa pruned a posteriori for resolution-

As you can see, it's not the same as Raven et al.'s Figure S75.  For one, stegosaurs resolve, although weirdly with Toujiangosaurus+Paranthodon as ankylosaurs.  For two it's WAY less resolved in Ankylosauria.  All those taxa from Zhejiangosaurus through Ahshishlepelta never form a consistent clade with each other or the four ankylosaur clades, and if you prune all nine genera, struthiosaurs, ankylosaurids and 'panoplosaurs'+polacanthids are still a trichotomy.  So their matrix doesn't actually show how these taxa relate (besides polacanthids being sister to 'panoplosaurs').  Isn't it ironic though that we do get Raven et al.'s three nodosaur clades including a classic Polacanthinae that includes Gastonia, Gargoyleosaurus and Mymoorapelta in addition to Jurassic Sarcolestes (and basally some parankylosaurs, but the paper was too late to include Stegouros)?  Struthiosauridae is only Struthiosaurus spp. plus Hungarosaurus and Tianchisaurus, which again is funny because the latter was not included in the clade by the authors due only to their Analysis B.  'Panoplosauridae' includes a lot more taxa, and yes those basal ones don't clade with each other or successively to the core group no matter how many are pruned, so that's another real polytomy.  As for the pruned taxa-

Mongolostegus can at least be sister to Chungkingosaurus or a struthiosaurine.
Adratiklit is part of the Dacentrurus+Stegosaurus clade.
Anodontosaurus and Scolosaurus are part of the Euoplocephalus+Pinacosaurus clade.
Tarchia kielanae is part of the Ankylosaurus+Euoplocephalus clade.
Acantholipan can be a 'panoplosaur', a struthiosaurine or outside (by which I mean a taxon closer to Ankylosaurus than Toujiangosaurus but not part of the struthiosaur, ankylosaurid or polacanthid+'panoplosaur' clades shown, though it could be e.g. sister to any of these clades and thus fall under their definitions).
Borealopelta can be a struthiosaurine, a 'panoplosaur' or outside(?).
Europelta can be a struthiosaurine or outside.
Invictarx can be a struthiosaurine, a 'panoplosaur' or outside.
Nodosaurus is always closer to Panoplosaurus than Silvisaurus.
Patagopelta can be a basal 'panoplosaur' or outside.
Pawpawsaurus can be a struthiosaurine or 'panoplosaur'.
Stegopelta can be a struthiosaurine or outside.

EDIT ADDED 5-21: Thanks to Andrea Cau in the comments for pointing out Raven et al. didn't include their character ordering settings in their txt file.  I wrongly assumed it would have a ccode line or a ctype line below the matrix but never scrolled all the way down.  I should have been suspicious when I had to manually choose the outgroup instead of them just making Lesothosaurus the first taxon in the matrix.  On the one hand, my bad.  On the other hand, it's surely best practice to not force your readers to modify the settings of the file you provided to the journal.

In any case, it still doesn't matter because Raven et al. STILL didn't find the shortest trees.  Instead of producing "eight MPTs with lengths of 1508 steps", their unweighted matrix Analysis A results in >99999 MPTs of 1506 steps.  Here's the real REAL strict consensus with 13 taxa pruned a posteriori for resolution-

Two steps doesn't sound like much, but it's enough to make 'polacanthids' a grade of basal ankylosaurids, make struthiosaurines nodosaurids as in traditional phylogenies and kick Dracopelta and Anoplosaurus out of panoplosaurs.  And yes, Nodosaurus is still closer to Panoplosaurus than Silvisaurus or Struthiosaurus.  The toplogy still has Tuojiangosaurus+Paranthodon as ankylosaurs and a struthiosaur Tianchisaurus too.

What are these clades named?

Raven et al. propose new definitions for their three nodosaur families-

All ankylosaurs more closely related to Panoplosaurus than to Ankylosaurus, Struthiosaurus austriacus or Gastonia burgei

All ankylosaurs more closely related to Gastonia burgei than to Ankylosaurus, Panoplosaurus or Struthiosaurus austriacus

All ankylosaurs more closely related to Struthiosaurus austriacus than to Ankylosaurus, Panoplosaurus or Gastonia burgei"

Tim Williams has already rightfully complained on the DML that their Polacanthidae definition needs to use Polacanthus foxii.  Why is this so hard in 2023?!  Do Arbour and the "three anonymous referees" not know the basics of phylogenetic nomenclature?  PhyloCode Article 11.10 states "when a clade name is converted from a preexisting name that is typified under a rank-based code or is a new or converted name derived from the stem of a typified name, the definition of the clade name must use the type species of that preexisting typified name or of the genus name from which it is derived (or the type specimen of that species) as an internal specifier."  We've been complaining about it since Sereno 24 years ago, surely every dinosaur worker knows by now.

Another obvious issue is that Nodosauridae has priority over Panoplosauridae, Struthiosauridae and Polacanthidae, so where's Nodosauridae?  Raven et al. explain "Nodosaurus is recovered outside of
Panoplosauridae in Analyses A and C, further suggesting that application of the name Nodosauridae would add confusion."  But as noted above, Nodosaurus is actually always a 'panoplosaur' when Analysis A is run correctly, and the fact it's supposedly an ankylosaurid sister to Dyoplosaurus when k=8 (fig. S77) should be close to worthless in my view.  Unless one of the authors wants to claim whatever character is being weighed eight times more than others could realistically mean this Cenomanian taxon is really within a Campanian species complex that are so similar they were all placed under Euoplocephalus tutus until recently?  Also, Polacanthus doesn't fall in their definition of Polacanthidae in two of their four trees, so why is that still allowed as a family name?

In any case, the clades have already been officially defined with PhyloCode registrations, by Madzia et al. (2021).  Raven et al. was "Received 7 February 2022", so I don't know why Arbour (a coauthor on Madzia et al.!) or the reviewers would let that stay in the paper.  Reading through, Raven et al. actually cite Madzia et al. and state "the underlying philosophy of the latter study is based on the PhyloCode (de Queiroz & Cantino, 2020) and offers an alternative hypothesis to our study, which is framed by the traditional principles of the International Commission on Zoological Nomenclature (1999), and so is not discussed further." Hahahaha  I hate to tell you guys, but defining clades based on phylogenetic relationships has nothing to do with the ICZN.  And if you were following the "traditional principles" of the 1999 ICZN, you couldn't just throw Nodosauridae away while stating "In the 'preferred' tree Panoplosauridae consists of ... Nodosaurus" and "A clade of generally Late Cretaceous North American taxa is also recovered here and named Panoplosauridae. As well as Denversaurus, Edmontonia spp., Nodosaurus and Panoplosaurus..."  Instead you would follow ICZN Article 65.2.3 - "by the discovery that the type genus was, when established, based on a type species then misidentified, the author may fix as the type species a nominal species as prescribed in Article 70.3. If the threat cannot be overcome by the fixation of a type species under the provisions of Article 70.3 the case is to be referred to the Commission for a ruling."  But that's not happening because I bet they think it's unlikely Nodosaurus is outside Nodosauridae.  "... is not discussed further" is short here for "... we know it makes no sense but we don't want to address it."

And Madzia et al. do a good job because they actually follow the rules.  Except in Raven et al.'s topology Stegosauridae ends up being the Chungkingosaurus+Eurypoda clade due to Huayangosaurus' weird position outside Eurypoda (which supposedly happens in the Bayesian analysis too), which could be saved by adding Ankylosaurus magniventris as an external specifier.  Similarly, Struthiosaurini might benefit from an Ankylosaurus magniventris external specifier due to the polytomy.  So here's the actual unweighted results with official clade names-

1. Ankylosauria; 2. Ankylosauridae; 3. Ankylosaurinae; 4. Ankylosaurini; 5. Eurypoda; 6. Huayangosauridae; 7. Nodosauridae; 8. Nodosaurinae; 9. Polacanthinae; 10. Shamosaurinae; 11. Stegosauria; 12. Struthiosaurini. Note due to Nodosaurus' possible positions, Panoplosaurini cannot be placed precisely in this tree.

For a couple final thoughts, Raven et al.'s definition for Shamosaurinae ("All ankylosaurid ankylosaurs more closely related to Shamosaurus than to Ankylosaurus") is better than the official definition ("max ∇ (Gobisaurus domoculus Vickaryous et al., 2001 & Shamosaurus scutatus Tumanova, 1983 ~ Ankylosaurus magniventris Brown, 1908") because Gobisaurus has no reason to be involved.  Also, Kirkland et al.'s (2013) definition of Struthiosaurinae is just horrible- "the most inclusive clade containing Europelta but not Cedarpelta, Peloroplites, Sauropelta or Edmontonia."  Doesn't separate it from ankylosaurids or polacanthines, includes Cedarpelta and Peloroplites which have no agreed upon positions or competing family-level names, AND doesn't use Struthiosaurus austriacus as the internal specifier.  Oof.  And why have "Bayesian analysis' as a keyword in Raven et al. (2023), run the Bayesian analysis for a week, mention it in the abstract, and despite having by far the best SCI score just leave it as a huge ankylosaur polytomy that has no effect on its conclusions?

So does nobody test run these analyses before they are accepted?

I began this post just planning to harp on peoples' seeming inability to learn Phylocode Article 11.10, but it ended up so much worse.  It's not like I searched and scoured to find a better cladogram in the data, TNT popped it out in (checking) under 30 seconds.  And then there's the Pawpawsaurus vs. Paw Paw juvenile switch that made the supp info wrong that I guess nobody noticed?  I'm just a person looking over the article on a random Saturday, not a journal editor or a professional whose job description woefully includes peer reviewing.  And sure mistakes are made, maybe the authors are anti-Phylocode despite proposing phylogenetic definitions (what organization is going to lend yours any validity?) and maybe nobody on the team knows how to use TNT, but I also just found the basic results quote from the top of my post to be misleading.  The published results did NOT show that nodosaurs fall into three "distinct morphotypes" past the standard Latest Cretaceous Central European struthiosaurines and North American panoplosaurins except for maybe 2-5 additional genera, NOR that they were more paraphyletic than normally thought and basically nothing was shown about polacanthines, with the statement "'polacanthid' ankylosaurs, but excluding Polacanthus" in Analysis A being plain wrong.  My best guess is that after getting pretty inconclusive results the team went with the "Nodosauridae isn't real" gimmick despite the hypocrisy in keeping Polacanthidae.  I think the real message of the analysis (assuming accurate scorings) is that ankylosaurs need a lot more characters analyzed to determine their basal relationships. the basic topology from twenty years ago is still most parsimonious.

References- Vickaryous, Russell, Currie and Zhao, 2001. A new ankylosaurid (Dinosauria:
Ankylosauria) from the Lower Cretaceous of China, with comments on ankylosaurian relationships. Canadian Journal of Earth Sciences. 38(2), 1767-1780.

Osi, 2005. Hungarosaurus tormai, a new ankylosaur (Dinosauria) from the Upper Cretaceous of Hungary. Journal of Vertebrate Paleontology. 25(2), 370-383.

Osi and Makadi, 2009. New remains of Hungarosaurus tormai (Ankylosauria, Dinosauria) from the Upper Cretaceous of Hungary: Skeletal reconstruction and body mass estimation. Palaontologische Zeitschrift. 83, 227-245.

Parsons and Parsons, 2009. A new ankylosaur (Dinosauria: Ankylosauria) from the Lower Cretaceous
Cloverly Formation of central Montana. Canadian Journal of Earth Sciences. 46(10), 721-738.

Thompson, Parish, Maidment and Barrett, 2012. Phylogeny of the ankylosaurian dinosaurs (Ornithischia: Thyreophora). Journal of Systematic Palaeontology. 10, 301-312.

Kirkland, Alcala, Loewen, Espılez, Mampel and Wiersma, 2013. The basal nodosaurid ankylosaur Europelta carbonensis n. gen., n. sp. from the Lower Cretaceous (Lower Albian) Escucha Formation of northeastern Spain. PLoS ONE. 8, e0080405.

Arbour, Zanno and Gates, 2016. Ankylosaurian dinosaur palaeoenvironmental associations were influenced by extirpation, sea-level fluctuation, and geodispersal. Palaeogeography, Palaeoclimatology, Palaeoecology. 449, 289-299.

Madzia, Arbour, Boyd, Farke, Cruzado-Caballero and Evans, 2021. The phylogenetic nomenclature of ornithischian dinosaurs. PeerJ. 9, e12362.

Raven, Barrett, Joyce and Maidment, 2023. The phylogenetic relationships and evolutionary history of the armoured dinosaurs (Ornithischia: Thyreophora). Journal of Systematic Palaeontology. 21(1), 2205433.

Russo and Mateus, 2023. Review of Dracopelta zbyszewskii, an ankylosaur from the Upper Jurassic of Portugal. 14th Symposium on Mesozoic Terrestrial Ecosystems and Biota. The Anatomical Record. 306(supp. 1), 221-223.