Sunday, October 13, 2013

New pterosaurs, new phylogenies


People with an interest in pterosaurs will probably be aware of the recent passing of Wann Langston Jnr, the Texan palaoentologist who was responsible for a lot of the popularizing of Quetzalcoatlus. I was lucky enough to meet Wann at the 2007 Flugsaurier meeting in Munich where he was able to attend, despite being well into his 80s. Wann’s work covered a great many aspects of Mesozoic reptiles and the recent festschrift that has been published in his honour covers a raft of different taxa.

Of interest to us though are three papers on pterosaurs that between them name four new taxa! That’s quite an effort for a single volume that is not even devoted to pterosaurs. However, what I want to talk about here is the phylogeny that appears in the paper by Brian Andres and Time Myers. Many will know that since 2003 pterosaur phylognies can broadly be divided into two camps – those which look more like that of Dave Unwin’s 2003 paper (the one shown here in red and blue is from the Darwinopetus description) and those which resemble Alex Kellner’s effort from the same year (the one shown here is from the Wukongopterus description). They are not actually that different from each other, both have the same general arrangement of taxa but with some difference. Kellner-type phylognies have anurognathids before dimorphodonitds, Unwin the reverse. Unwin-types have ornithocheiroids before the ctenocasmatids, Kellner the reverse. Both have the rhamphorhynchines immediately before Darwinopterus and kin and those coming before the origin of pterodactyloids and both have dsungaripterids close to the azhdarchoids. In short, there’s a way to go to get a consensus and there are some fairly clear and consistent contradictions, but they are not so far apart.


 Interestingly, back in 2007, Brian Andres presented on some of his PhD work where he talked about how the two might be coming together, and Dave Unwin have a similar talk in Beijing in 2010. However, the phylogeny in this paper (below) is really rather different to both of those. This is the latest version of Brian’s analysis which has already popped up in a couple of papers, but frustratingly, the actual core of this (i.e. the actual character list and coding) still isn’t published –owing to the interminable delays on The Pterosauria book - so there’s no way at the moment to see what is causing these shifts.


The contrasts are quite dramatic though. Neither the dimorphodontids nor anurognathids are at the base (or close to it) of the phylogeny, but instead it is the eudimorphodontids and the anurognathids are in fact lying more derived than Darwinopterus and close to the pterodactyloids! The dsungaripterids are also now not sister taxon to the azhdarhoids, but lying within the clade as sister taxon to the thalassodromids and with the tapejarids as a basal clade to these plus the azhdarchids+chaoyangopterids.

In short, if anything, the phylogenies are getting further apart. Now I would expect them to converge again sooner or later: after all, there is only one correct solution. But as often lamented (and in particular by Darren), pterosaur phylogenies are generally rather character poor compared to many analyses of archosaur clades so there is much more to come. That said, I think some of the problem comes from the continued practice of doing analyses that cover the whole of the Pterosauria. Surely we are at the point where the rhamphorhynchoids and pterodactyloids are well separated and there’s no need to repeatedly include both in every analysis – it’s probably not helping the resolution of some trees where large chunks of the characters or states needed to help resolve one clade won’t add anything to the other. Devoting more time to better characters for smaller groups will probably be more productive than continuing to code up large numbers of taxa where large numbers of characters are inappropriate.

Anyway, that’s my 2p on the problem. Certainly the current conflicts are interesting and I look forwards to seeing what characters are supporting some of these unorthodox positions – there are likely to be some interesting convergences and codings in there. Obviously it’s hard to say much without the underlying data, but at face value I’m not overly convinced by some of those positions, but it will be especially interesting to see what these new positions mean for things like character support of branches and if it calibrates better temporally than the other current competing ideas. Now, we just need the book to get finished….

  
Andres, B. & Myers, T.S. 2013. Lone Star Pterosaurs. Earth and Environmental Science Transactions of the Royal Society of Edinburgh. 103: Issue 3-4, p 383-398.


7 comments:

  1. Gotta disagree with you regarding helping the problem by analyzing less taxa. First, having characters where many taxa are coded 0 or - doesn't hinder resolution, it just doesn't add resolution to most of the tree.

    More importantly, it's very important to get polarity right in cladograms, which is difficult if you have to use suprageneric taxa. In your 'rhamphorhynchoid' analysis, what would Pterodactyloidea be represented by for instance? Pterodactylus is an easy choice, but according to Andres' tree, Germanodactylus would be better. Though Normannognathus would be better still, but that's fragmentary so we can't ONLY have Normannognathus. Then we have the other side with Haopterus, even if we don't delve into his "Ornithocheiroidea". But none of these taxa give us quality three dimensional preservation needed for some characters, etc..

    Ironically, I was going to suggest a lack of appropriate or varied outgroups as a reason why analyses keep finding different pterosaurs to be most basal. Andres used Euparkeria, which isn't particularly close to pterosaurs in anyones' phylogenies. Kellner used Scleromochlus (good according to the consensus), Herrerasaurus (not great) and Ornithosuchus (decent if a good dinosauromorph were used). Unwin just used a hypothetical all zero outgroup, which is worst of all.

    Reducing taxon inclusion just gives crackpots like Peters more justification to gloat about poor quality in published analyses. It's the quick, lazy and ineffective way out, like partly coded supermatrices.

    ReplyDelete
  2. " First, having characters where many taxa are coded 0 or - doesn't hinder resolution, it just doesn't add resolution to most of the tree. "

    Well I didn't actually say it did. I'm well aware this is true, but since large numbers of pterosaur taxa are rather incomplete, we instead have the issue where large numbers of characters are coded as ? when they almost certainly don;t add to the polarity. In other words, we can be really pretty sure no pterodactyloid has an elongate and modified 5th toe, but without a foot, we're coding it as a ?, not a 0.

    "More importantly, it's very important to get polarity right in cladograms, which is difficult if you have to use suprageneric taxa. In your 'rhamphorhynchoid' analysis, what would Pterodactyloidea be represented by for instance? Pterodactylus is an easy choice, but according to Andres' tree, Germanodactylus would be better. Though Normannognathus would be better still, but that's fragmentary so we can't ONLY have Normannognathus. Then we have the other side with Haopterus, even if we don't delve into his "Ornithocheiroidea". But none of these taxa give us quality three dimensional preservation needed for some characters, etc.."

    Well as Brusatte suggested, you could use a number of them. I'm not suggesting you should code a single supragenic "Pterodactyloidea" for a rhamphorhynchoid analysis, but certianly it would be more effective (in terms of time etc.) to code half a dozen or so of the key ones that than 40 or 50 of them where things like azhdarchoids really are not going to add anything.

    "Ironically, I was going to suggest a lack of appropriate or varied outgroups as a reason why analyses keep finding different pterosaurs to be most basal. "

    That's almost certainly linked to the problem, yes.

    " It's the quick, lazy and ineffective way out, like partly coded supermatrices."

    I'm assuming that's an unnecessary are irrelevant swipe at my 2008 paper. I freely admit that is flawed - we all make mistakes - but it's not partially coded.

    As I said quite clearly I think it's a time issue and when not everyone can devote huge amount of time to each new analysis, simply coding up large numbers of taxa for large numbers of characters that are most unlikely to affect the results of the area you are interested in I think is better devoted to trying to create new and better characters to resolve problematic areas/ I'm not suggesting dropping full analysis or just killing taxa for no good reason.

    ReplyDelete
  3. "but frustratingly, the actual core of this (i.e. the actual character list and coding) still isn’t published –owing to the interminable delays on The Pterosauria book - so there’s no way at the moment to see what is causing these shifts"

    The supplementary material of Anders & Myers 2013 includes both character list and data matrix. The analysis has 31 continuous characters (mostly as ratios among skeletal elements re-scaled as vaues along the [0-1] interval) and about 150 "traditional" discrete characters.
    The exclusion of the 31 continuous characters does not alter meaningfully the general topology.

    ReplyDelete
    Replies
    1. Oh I'd missed that entirely Andrea, thanks for pointing it out. I must confess I hadn't read the paper in detail because, actually I was one of the referees, and one of my complaints (as I recall, don't have the review now) was that this was missing. Looks like they added it in and then I missed it. My bad!

      I am glad they included some continuous ones for the wing ratios etc., that's been something that has bugged me for a while that I felt needed doing.

      Thanks for pointing out my mistake, though the implication is still that we'd have to wade through the data individually to work out the character distributions and find the key changes etc. Hmm, maybe should reread it and see how different it is from the version I refereed. :)

      Delete
  4. "Well I didn't actually say it did."

    You said that you think part of the reason pterosaur analyses disagree is that "large chunks of the characters or states needed to help resolve one clade won’t add anything to the other". Though now you're talking about including incomplete codings being a problem- "the issue where large numbers of characters are coded as ?". These are two different issues, but both were analyzed by Wiens (2003) who found given a large number of characters and taxa, they didn't adversely affect results.

    "but certianly it would be more effective (in terms of time etc.) to code half a dozen or so of the key ones that than 40 or 50 of them where things like azhdarchoids really are not going to add anything."

    It is indeed a time issue, but phylogenetic analysis is an extreme case of getting back what time you put in. Being time effective is basically useless if your analysis lacks the data needed to test hypotheses.

    A real life example of your proposed solution is Montefeltro et al.'s (2011) analysis of baurusuchid crania, which is relatively small (66 characters, 7 taxa plus 3 outgroups) and uses many new and redefined characters. It gets different results than larger analyses like Pol et al. (2012; 347 characters, 88 taxa), so now what's a non-croc expert to do? Is the difference because of the new/improved cranial characters in MEA11? Is it because of the postcranial characters in PEA13? Is it because MEA11 a priori excluded Pabwehshi based on a single character, whereas it falls within Baurusuchidae in PEA11? Is it because MEA11's outgroups aren't particularly close to baurusuchids in PEA13's tree?

    Without doing a lot of extra work, the reader can't know. All that making a restricted analysis has done is create a new hypothesis which can't be tested against others unless the reader spends that time and effort the author saved.

    And now that I actually look at Andres and Myers (2013), I see the character list and matrix are available. Unlike either Unwin's or Kellner's matrices, it has continuous characters, and unlike Kellner's it has polymorphic characters. Notably, Andres and Myers state it includes "all the valid taxa and characters used by previous analyses." If this is true, then it undermines your thesis here that "the phylogenies are getting further apart" and that this needs some sort of explanation and solution. Instead of being a 'third effort', Andres' matrix is actually a combination of Unwin's and Kellner's data. What it shows is that in a lot of areas, Unwin's and Kellner's data were too weak to counter hidden support in their combined matrices and/or new characters/taxa. The solution to Unwin's and Kellner's trees remaining different even as taxa and characters were added was to include both of their data in one analysis and Andres did just that.

    ReplyDelete
  5. "I'm assuming that's an unnecessary are irrelevant swipe at my 2008 paper. I freely admit that is flawed - we all make mistakes - but it's not partially coded."

    It's good you admit Hone and Benton (2008) was a mistake, but I felt it relevant to bring up since the flaw was due to this same philosophy of taking shortcuts to save time coding. It's fairly obviously partially coded, as e.g. Dinosauria is only coded for 70 of the 324 characters (22%) when it should be basically 100% excepting a few inapplicable codings for structures dinosaurs don't have like centralia.

    "As I said quite clearly I think it's a time issue and when not everyone can devote huge amount of time to each new analysis"

    The good thing about Andres' analysis is now that it exists and combines all prior analyses, is that we can just continue to modify it instead of recoding all twenty thousand entries again. It can become the TWG matrix of pterosaurs. Thus we can get the improved and new characters we both agree will help matters, and the extensive coverage that I think is necessary.

    ReplyDelete