Tuesday, September 25, 2012

Bennett's new paper with non-archosaurian pterosaurs

So you all know the basic history of pterosaur affinities.  Viewed as avemetatarsalians since the 1980s, this result has been found in basically every analysis.  Yet Peters (2000) argued they were actually related to non-archosauriforms like Longisquama, Cosesaurus and Sharovipteryx.  Indeed, he now thinks all of these taxa are lepidosauromorphs based on his analysis with too few characters, correlated characters and his ever un-trusty digital segregation technique of determining anatomy.  Peters rightly points out that no analysis has included the above three genera, many use suprageneric taxa, and most only examine archosaurs, so pterosaurs nesting with dinosaurs is an artifact.  Thus so far we have flawed analysis versus flawed analysis, so Bennett's new paper finding pterosaurs to be basal archosauriforms between proterosuchids and erythrosuchids has a chance to help.

Unfortunately, I don't think it does.  The basic premise is that Bennett is checking his 1996 analysis to see if hindlimb characters are causing pterosaurs and dinosaurs to clade together, and reviewing the cursorial characters (the Revised Data Set), then adding more characters and taxa (the Updated Data Set).  One thing I enjoyed about the paper is the speaking style, where Bennett goes against convention and speaks in the first person.  I also like the amount of discussion about coding choice and miscoding, which makes things more transparent.  As an example of both-

"I recently determined that the data set used in the analyses described in that paper differed from that in the published data matrix in that Scleromochlus was coded as 0 for Char. 43, whereas it was coded as missing in the published data matrix. From my present perspective, 14 years later, I cannot determine whether this was a typographical error or an intentional change during revision that was incompletely propagated through the manuscript, but here the published coding (i.e. – for missing) is used."

To cut to the chase, Bennett found the 14 cursorial characters were incongruent with other body areas and responsible for pterosaurs being avemetatarsalians.  Thus, "9 of 14 characters are found to be non-independent of other characters, functioning as inappropriate additional weighting of the underlying characters, and so are subsumed into the remaining characters as they are reformulated."  Once this was done, pterosaurs ended up as basal archosauriforms, with or without the new taxa and characters.

On the surface, this sounds great.  I was just complaining about correlated characters earlier.  But closer examination shows reason to worry.  Let's look at an example, character 112 from the 1996 analysis- "Bird-like distal end of femur – prominent anterior and posterior intercondylar grooves with latter constricted by prominent external tibial condyle. States: 0 = absent; 1 = present".  Well, that's pretty terrible in itself.  If I were to try to code Herrerasaurus for instance, it has a prominent posterior groove, but no anterior groove, and the lateral condyle doesn't constrict the posterior groove.  So we have a composite character that should be split into three.  Let's see how Bennett handles it.  He says-

"Comparison of the knees of ornithodirans sensu Gauthier reveals that they fall into two discrete types. [snip] Because of the presence of two discrete types of double condyle knees, the character needs to be reformulated:
Double condyle knee. States:
0 = absent;
1 = distal femur suboval, somewhat less than twice as broad transversely as long anteroposteriorly, with medial, lateral, and fibular condyles, and the posterior intercondylar groove constricted by the prominent external tibial condyle, proximal tibia and fibula not fused, and proximal tibia subtriangular and its anteroposterior length exceeds its transverse breadth;
2 = distal femur roughly D-shaped, roughly twice as broad transversely as long anteroposteriorly, with only two condyles and medial condyle considerably larger than lateral, posterior intercondylar groove not constricted, and the proximal tibia and fibula fused to form a suboval to subrectangular articular surface roughly twice as broad transversely as long anteroposteriorly,
with the coding as before except that the Pterosauria are coded 2. Because there is no evidence that one state evolved from the other, the character should be unordered for analysis."

That sound you just heard was the cladist in me choking and dying.  Bennett just made the character MORE composite.  In fact, I don't think I've ever seen such a composite character in my entire life.  I don't even know where to begin.  Herrerasaurus again, I suppose.  I guess the distal femur is kind of oval, but the front is flat.  Is that D-shaped as in 2, or does Bennett mean another side is flat in D-shaped femora?  It's 18% wider than deep, which is closer to "somewhat less than twice" than "roughly twice", but that state still doesn't sound right, nor are they quantified to help me if coding a femur e.g. 90% wider than deep.  It does have a fibular condyle, so that's like 1.  The posterior groove is unconstricted though, so that's like 2.  Also note state 1 doesn't define the comparative sizes of the medial and lateral condyles, though state 2 does.  The tibia and fibula are unfused, but what does that have to do with the femur?  It's quite the assumption that pterosaur tibiofibular fusion evolved at the same time as these other features.  The tibia is triangular because it has a cnemial crest.  It's longer than wide proximally, but state 2 compares that proportion in the tibia+fibula, so is not measuring the same thing.  It's not just that Herrerasaurus is some outlier either.  Heterodontosaurus lacks a posterior groove, but has a fibular condyle.  Scleromochlus lacks a cnemial crest, but also lacks tibiofibular fusion. Dromomeron romeri has a femur 71% wider than deep, but a tibia much longer than wide proximally.  This is why composite characters are terrible, and Bennett goes on to make equally horrible characters for the ankle and foot.  Note too that states 1 and 2 each contain plesiomorphic states that are shared with state 0, like the lack of tibiofibular fusion in state 1. 

But besides their terrible, terrible formation, what's really worrying about these characters is how they affect the analysis.  Bennett's basically taking characters that support avemetatarsalian pterosaurs (and in the case of ankle and foot characters, reducing their number), then combining them with perhaps functionally related but indepedent characters that differ between dinosaurs and pterosaurs, so that they can't group pterosaurs with avemetatarsalians.  You could do this with any character or set of characters.  Sure both caenagnathids and oviraptorids have a sliding mandible that lacks teeth and has a coronoid process, but oviraptorids have a mandible that's shorter, lacks symphyseal fusion, lacks the ASC complex, and has a surangular prong.  And I'm sure the sliding glenoid worked in conjunction with the coronoid process somehow and that it's related to their toothlessness.  Time to make that into two unordered states of a single character.  Ugh.  So Bennett eliminated most characters supporting avemetatarsalian pterosaurs and *gasp* didn't recover avemetatarsalian pterosaurs.

Just to round out my complaints, the characters were all ran unordered although many should be ordered (e.g. Parietal foramen. States: 0 = large; 1 = small; 2 = absent.).  Well, actually he states he ran the bistate characters as ordered, despite the fact ordering has no affect on bistate characters.  Also "most loss characters and a few gain characters (Char. 6, 14, 19, 20, 23–26, 28, 36, 41, 53–55, 68, 97, 118, 135, 136) were defined as irreversible".  I guess the humerus can never surround the median nerve and branchial artery again once the entepicondylar foramen is lost (24).  And hey, the antorbital fenestra can never be lost (41).  Who knew alligators were impossible?  In his Discussion, Bennett frankly admits his philosophical differences from most cladists-

"The general trend in phylogenetic analyses of diapsids has been towards analyses with more taxa and more
characters, as if more necessarily equated to better, and Nesbitt’s (2011) analysis is but the most recent. Unfortunately, the usual way to get more characters is to atomise the morphology of structures, coding for more and more insignificant features of structures, which leads to multiple unintentional weightings of the underlying character. Nesbitt’s analysis has 30 characters from the astragalus and calcaneum including, for example the shape of the anteromedial corner of the astragalus: acute or obtuse. It is absurd that the angle of the corner, a few degrees this way or that has the same weight in an analysis as the presence of an antorbital fenestra, a calcaneal tuber or a pteroid bone. I reject such atomisation of morphology, and point out again that most effort should go into formulation and testing of characters, rather than the analysis of the data and manipulation of trees and that one should use characters only if the ‘hypotheses of homology cannot be refuted’ (Gaffney et al. 1991)."

I wholeheartedly agree that we need to spend more time formulating and testing characters, but find the rest laughable.  First, there have been numerous papers showing adding taxa and characters usually results in a more accurate tree, though there are eventual limits to the payoff for both.  Bennett's 19 taxa and 144 characters is nowhere near those limits, needless to say.  What's truly absurd is that Bennett thinks he knows the angle of the anteromedial astragalar corner is less likely to change than the presence of an antorbital fenestra.  How would you even determine that?  Homoplasy?  Well, the corner character never converges or reverses in Nesbitt's matrix, so that can't be the reason.  The number of mutations it takes?  We have no clue about that, and given recent studies showing single genes can have huge phenotypic effects, I think it would be foolish to equate the apparent significance of a character with its phylogenetic worth.  The same issue surrounds his choice to make some characters irreversable.  How does he know which reversals are impossible genetically?  What Bennett is advocating is a return to the good ol' days of "key characters", or in this case key character complexes, which are even worse.

Oh, and Bennett doesn't include Longisquama, Cosesaurus or Sharovipteryx, instead adding two drepanosaurs and splitting Prolacertiformes into Prolacerta and Tanystropheus.  So he doesn't even really test Peters idea, since Peters has been saying since 2000 that those taxa are important for getting pterosaur affinities correct.  And of course he keeps Lepidosauromorpha as the outgroup, so Peters' new idea of lepidosauromorph pterosaurs isn't tested either.  Nor does it include important taxa like lagerpetids, silesaurids and Doswellia.  Bennett's analysis was fine in 1996, but doesn't cut it 16 years later.

So basically Bennett uses an outdated analysis, unfairly eliminates most characters supporting avemetatarsalian pterosaurs, doesn't include the best candidates for non-archosaurian pterosaurs, and espouses a horrible, subjective cladistic philosophy.  I'm still waiting for that good test of pterosaur relationships.

References- Peters, 2000. A reexamination of four prolacertiforms with implications for pterosaur phylogenesis. Rivista Italiana di Paleontologia e Stratigrafia. 106(3), 293-336.

Bennett, 2012. The phylogenetic position of the Pterosauria within the Archosauromorpha re-examined.  Historical Biology. iFirst article, 19 pp.