Monday, July 23, 2012

Resolving why Confuciusornis has a modified wireframe nearly identical to juvenile Alligator

In my last post, I took Bhullar et al. (2012) to task for 1) their choice of Confuciusornis as being a close match in analyzed skull shape to juvenile Alligator; 2) Bhullar's own online statement it wouldn't matter which basal bird they used, since the Catalan enantiornithine embryo and others would come out very similarly; and 3) why their wireframes vary so much from the specimens themselves.  After communicating with Bhullar, we have answers to these issues.

Number three is easiest, and works to explain the rest.  As Bhullar explained (and Rodriguez noted on the DML), the wireframes in the paper are not actually the raw wireframes that result from placing landmarks on photos.  Instead, they are a combination of principal components.  Now morphometrics is not my thing, but as I understand it, Principal Component 1 is the horizontal variation in landmarks and Principal Component 2 is the vertical variation in landmarks, once the skulls have been resized and rotated to best agree with each other.  As their figure S6 below shows, each component has a hypothetical extreme on each end.  On each axis, a certain line will change from one extreme to another.  So on PC1, you can see the naris moving further back going left to right.  However, some features weren't common or consistent enough to show over those features that explain most of the variation.  So you'll notice nothing in any extreme has a large naris, since that didn't vary consistently with the main horizontal or vertical changes.  And that's why the Confuciusornis and juvenile Alligator both have modified wireframes with small, partially retracted nares- most of their horizontal measurements fall out to average a PCI of ~0.08-0.10, and at that (and every) point in the graph, skulls have small and partially retracted nares.  So you take the average wireframe of skulls on a certain point of the PC1 axis, and mix it with the average wireframe of skulls on a certain point of the PC2 axis, and ta-da!, you have your modified wireframe that doesn't really resemble the original much at all.

Bhullar said he should have noted this in their paper (especially in the oft copied figure 4 which purported to show the "nearly identical skull configuration" of Confuciusornis and juvenile Alligator, with almost complete overlap of every line) and I said I shouldn't have called his data seriously flawed before confirming his methods.  We've both apologized and are happy with how things turned out.  Bhullar sent me his Confuciusornis and juvenile Alligator raw wireframes, and while I would place a few points differently, they're accurate overall.  The above discussion also shows what Bhullar meant when saying the choice of Confuciusornis over e.g. the Catalan nestling wouldn't matter, since it's very close in PCA morphospace.  The reduced antorbital area of Confuciusornis doesn't factor in, since on average, taxa in that area of the graph don't have such reduction, so the PCA average wireframe the computer uses for Confuciusornis doesn't either.

So then, despite their wildly varying anatomy, Confuciusornis and juvenile Alligator are getting similar PCAs.  What actual anatomy is controlling most of the variation?  We both hypothesized that most of the variation is simply due to the general skull outline.  You'll remember from before that the snout length and slope match almost perfectly, and the braincase is just a little more expanded in Confuciusornis (though perhaps due to crushing). 
To test this, I suggested Bhullar eliminate all but the major landmarks outlining the skull- 1, 2, 3, 4, 16, 23, 24, 25, 29, 40 and 43.  He ran this reduced analysis based purely on the skull outline and found very similar results.
PCA plot by Bhullar using only 11 landmarks, all defining the lateral skull outline.  Figure shown upside down to more easily compare to their figure S6 above.  Juvenile Alligator highlighted in blue, Confuciusornis in brown.  Other taxon-number correlates given below.
Note Confuciusornis and juvenile Alligator as about as close as they are when all 45 points are analyzed, showing Bhullar et al.'s analysis mostly compared gross skull shape.

And that's that.  As I said at the beginning on my first post, I agree with their main thesis- that bird skulls are paedomorphic.  Before ending, I'd like to thank Bhullar for how gracious he has been toward me and my critique.  As he said in an email, such criticisms and responses are how science should work.

Supplementary data- 

Numbers representing each skull in Bhullar's reduced PCA plot.
ID=01 Alligator adult
ID=02 Alligator embryo
ID=03 Allosaurus adult
ID=04 Anchiornis adult
ID=05 Archaeopteryx adult
ID=06 juv Archaeopteryx
ID=07 Bambiraptor
ID=08 Byronosaurus adult
ID=09 Ceratosaurus adult
ID=10 Citipati adult
ID=11 Citipati embryo
ID=12 Coelophysis juvenile
ID=13 Coelophysis adult
ID=14 Compsognathus adult
ID=15 Confuciusornis adult
ID=16 Dilong adult
ID=17 Dilophosaurus adult
ID=18 Dromaius adult
ID=19 Eoraptor
ID=20 Erlikosaurus adult
ID=21 Eudromia adult
ID=22 Euparkeria adult
ID=23 Gallus adult
ID=24 Gallus juvenile
ID=25 Garudimimus adult
ID=26 Gorgosaurus adult
ID=27 Guanlong adult
ID=28 Herrerasaurus adult
ID=29 Incisivosaurus adult
ID=30 Majungasaurus adult
ID=31 Monolophosaurus adult
ID=32 Scipionyx juvenile
ID=33 Suchomimus adult
ID=34 Tarbosaurus adult
ID=35 Tarbosaurus juvenile
ID=36 therizinosaurid embryo
ID=37 Tyrannosaurus adult
ID=38 Tyrannosaurus subadult
ID=39 Tyrannosaurus large adult
ID=40 Velociraptor adult
ID=41 Zanabazar adult
ID=42 Haplocheirus adult
ID=43 enantiornithine juvenile
ID=44 Pengornis adult
ID=45 Shenqiornis adult
ID=46 Struthio adult
ID=47 Struthio juvenile
ID=48 Yixianornis adult
ID=49 Pterocles adult
ID=50 Chauna adult
ID=51 Opisthocomus adult
ID=52 "Byronosaurus" perinate