Tuesday, February 12, 2013

Testing Peters' dinosaur phylogeny

As we discussed before, David Peters' amniote analysis finds an odd topology for dinosauriformes.

--Theropoda including Marasuchus, Herrerasaurus and Trialestes
--- Panphagia+Pampadromaeus
---- a clade of Pisanosaurus, poposaurs and silesaurids
----- Sauropodomorpha including Saturnalia, Thecodontosaurus and Massospondylus
------ Daemonosaurus + standard Ornithischia

Why is this?  I set out to determine the reason.  First, I deleted most taxa Peters finds outside Dinosauria, leaving Turfanosuchus and Gracilisuchus as outgroups, as they are close to the base of Avemetatarsalia's sister group in both his and Nesbitt's phylogenies.  This way I don't have to deal with all the unrelated taxa I'm unfamiliar with.  I also kept in Arizonasaurus, to see if poposaurs would group by it instead.  His same topology resulted when I ran this in PAUP, so my deletions didn't affect that.  Enforcing traditional Dinosauria and Saurischia took 28 more steps.

I added the 15 saurischian and 11 dinosaurian characters from Nesbitt's analysis missing in Peters' (and also one I thought was coded for- the subnarial foramen, but it turns out Peters only codes for a subnarial fenestra).  I used Nesbitt's codings for these, also coding Massospondylus kaalae, Pampadromaeus, Panphagia and Daemonosaurus myself.  The new phylogeny excluded Marasuchus from Dinosauria but was otherwise largely the same.  It took only 5 more steps to make traditional Dinosauria and Saurischia.  A reduction in 23 steps isn't a bad start.

Then I started going through Peters' characters, making them ordered when needed (which sometimes involved rearranging states), and making sure each was only coding for one variable.  Luckily, with the non-dinosauriforms mostly deleted, a lot of Peters' characters had far less states in included taxa, so this was much easier than if I were to try it for his whole analysis.  Still, it's mighty tedious, so I stopped after doing the 115 cranial and dental characters.  During this process of fixing characters, I found 32 miscodings.  Being the first taxon in the list, I decided to code Turfanosuchus for these first 115 characters, plus the three extra characters I had spun off of them to keep each one only coding for one variable.  Turfanosuchus was miscoded for 33 characters, or 28%, which is pretty terrible.  Running the analysis found..... traditional Dinosauria and Saurischia. 

-- outgroups
--- Arizonasaurus
---- poposaurs
----- silesaurs
------ Marasuchus
------- Pisanosaurus
-------- Ornithischia
--------- Sauropodomorpha including Saturnalia, Thecodontosaurus and Massospondylus
------------ Daemonosaurus
------------ Panphagia+Pampadromaeus
--------------- Theropoda including Herrerasaurus and Trialestes

It doesn't perfectly match the consensus, but I didn't add any characters for e.g. Crurotarsi, Ornithischia including Pisanosaurus, Sauropodomorpha including Panphagia and Pampadromaeus, silesaurs+dinosaurs, so the fact I didn't get these clades is hardly surprising.  What was surprising was how soon this happened.  Adding 26 characters and changing 65 cranial states, half of the latter in the basalmost outgroup, and the traditional phylogeny shines through.  So why does Peters' analysis find odd topologies- it lacks enough characters, has poorly formed characters, and has numerous miscodings.  Fix even a fraction of those issues, and it resembles the traditional consensus.

Below are the gritty details of my alterations, for those who want to slog through them.

4. Changed to ordered. Turfanosuchus (Wu and Russell, 2001- fig. 1) has a flat skull table.
5.  Changed to ordered, states rearranged.
7. Turfanosuchus (Wu and Russell, 2001- pg. 41) lacks a firm premaxilla-maxilla suture.
10. Changed to ordered, states rearranged.
11. The state "flares over nares" was deleted, only being coded in Arizonasaurus, which shows no flaring (Nesbitt, 2005, figure 2A).  It was recoded as "widest at midpoint".  Turfanosuchus (Wu and Russell, 2001- fig. 1A) has nasals widest at midpoint. 
13. Pampadromaeus is reconstructed with a smoothly curved dorsal snout edge (Cabreira et al., 2011).  Panphagia doesn't preserve a premaxilla, so can't be coded (Martinez and Alcober, 2009).  Herrerasaurus has a smoothly curved left side (Sereno and Novas, 1993, figure 1A), and is the only remaining taxon coded as "convex, angled over naris", so the only two useful states are straight and convex. Turfanosuchus (Wu and Russell, 2001- fig. 1B, 2A) does not preserve the anterodorsal corner of the snout.
14. Changed to ordered, states rearranged.  Heterodontosaurus is the only taxon coded "notch for fang", but that's unrelated to the angle of its subnarial notch, and indeed other taxa like Coelophysis also have dentary fangs which enter their notch.  As Heterodontosaurus' notch has a high angle of 124-130 degrees (Norman et al., 2011, figures 4 and 5), it is recoded as having a notch over 45 degrees. Turfanosuchus (Wu and Russell, 2001- fig. 1B) has a notch of <25 degrees.="" span="">
15. Changed to ordered, states rearranged.
21. Turfanosuchus (Wu and Russell, 2001- fig. 1B) has a naris less than rwice as long as tall.
22. Composite character- examine taxa. Turfanosuchus (Wu and Russell, 2001- fig. 1A) has a dorsolaterally oriented external naris.
23. The state "snout tip or elongated" combines naris placement, with naris elongation, the latter of which was covered by character 21.  It is redefined as "snout tip" only.  This character also combines anteroposterior naris placement with dorsolateral naris placement, the latter which is already coded for by character 22.  So all taxa coded "snout tip but elevated" are recoded "snout tip". Turfanosuchus (Wu and Russell, 2001- fig. 1B; anterior edge 10% down preorbital length) has nares placed near the snout tip.
24. Thecodontosaurus is the only taxon coded as lacking a posterolateral premaxillary process (presumably based on Pantydraco), but the posterior premaxilla and anterior maxilla are unpreserved (Galton and Kermack, 2011).  Thus the state is removed and the taxon recoded unknown. 
25. Changed to ordered. 
27. Turfanosuchus (Wu and Russell, 2001- fig. 1B) has maxilla-orbit contact.
28. The state "descending anteriorly" actually indicates a ventrally concave maxilla.  The character is thus changed to ordered, and the states rearranged so that "straight" is in the middle. 
37. This ratio is so close in Turfanosuchus (Wu and Russell, 2001- fig. 1B) that it should be coded as unknown, since the paroccipital process would add an unknown amount to the skull length.
38. Changed to ordered, states rearranged.
41. This character (Frontal-parietal suture: (0) not; (1) straight and longer than nasal-frontal suture) involves both length and straightness.  Yet taxa like Turfanosuchus have frontoparietal sutures which are bent and longer than their frontonasal suture.  Thus character 41 is changed to only code for straightness, and new character 258 codes for length. 
43. Turfanosuchus (Wu and Russell, 2001- pg. 42) has maxilla-prefrontal contact.
45. deBraga and Rieppel (1997) define the frontal lateral process as "a distinct lateral lappet (process) that projects out from the frontal margins at an abrupt 70° to 90° angle."  This is lacking in Turfanosuchus (Wu and Russell, 2001- fig. 1A).
48. This character coded for postparietal orientation, and all included taxa are coded as state 3 "absent".  But character 47 covered absence of postparietals, so the taxa are recoded inapplicable for 48.
50. Peters defines this character as "Temporal ledge (distinct overhang at upper temporal area): (0) no ledge; (1) made from squamosal; (2) made from supratemporal and tabular. [Gauthier et al. 1988, Laurin & Reisz 1995, Hill 2005]."  Yet none of those references include such a character.  If he meant Gauthier et al.'s character 10 " Temporal musculature originates ventrally (0), or dorsally (1), on the parietal table" and Hill's character 91 " Temporal muscles on parietal table: (0) originate ventrolaterally – parietal margin straight; (1) originate dorsolaterally – parietal lateral margin embayed",  that involves the medial edge of the supratemporal fenestra, not the lateral edge. 
52. Changed to ordered, states rearranged.
53. This character codes for the presence of postorbital-parietal contact, but dinosaurs and shuvosaurs are coded as state 3 "Postfrontal-Postorbital fusion", which does not relate to whether the postorbital contacts the parietal.  Postfrontal fusion is coded for character 54 anyway.  All taxa coded as state 3 are recoded.  Herrerasaurus (Sereno and Novas, 1993- figure 1G), Coelophysis (Colbert, 1989- figure 29), possibly Pampadromeus (Cabreira et al., 2011- figure 2; though reconstructed), Massospondylus kaalae (Barrett, 2010- figure 4), Pantydraco (Yates, 2003- figure 2C), Scelidosaurus (Owen, 1861) and Heterodontosaurus (Norman et al., 2011- figure 12) lack contact.  Lesothosaurus (Sereno, 1991- pg. 179) and Shuvosaurus (Lehane, 2005- figure 6) have narrow contact.  Gracilisuchus (Romer, 1972- figure 1), Turfanosuchus (Wu and Russell, 2001- figure 1), Agilisaurus (Peng, 1992- figure 1B) and Effigia (Nesbitt, 2007- figure 11) have broad contact. Tawa does not preserve a published parietal (Nesbitt et al., 2009), and the medial surface of the postorbital has not been described, so should be coded unknown.  Similarly, Daemonosaurus cannot be coded.  I can't code Lotosaurus from published data.
54. Included taxa lacking postfrontals are coded as state 3- postfrontal fused to intertemporal, which is wrong as these taxa lack intertemporals as well.  They are recoded as merely lacking postfrontals.
57. This character codes for the angle of the frontonasal suture.  A separate state is assigned to Gracilisuchus and Arizonasaurus for a zig-zag suture, but the straightness of each suture is a different variable than the angle between sutures.  The suture in Gracilisuchus is roughly transverse (Romer, 1972- figure 1), while in Arizonasaurus  the "anterior edge of the frontal is incompletely preserved, so the articulation to the nasal is not known".  Gracilisuchus is thus recoded as transverse, and Arizonasaurus recoded unknown. 
59. Changed to ordered.  Turfanosuchus (Wu and Russell, 2001- fig. 1A, based on parietal) has paroccipital processes angled more than 40 degrees posteriorly.
60. Most taxa are coded "postfrontal fused to frontal", so this is another character correlated with the postfrontal absence character 54.  As this character codes for postfrontal-supratemporal fenestra contact, taxa without a postfrontal are recoded inapplicable.
61. Turfanosuchus (Wu and Russell, 2001- fig. 2A) has more jugal posterior to the postorbital process than anterior to it.
62. Changed to ordered. 
64. Changed to ordered. 
65. Turfanosuchus (Wu and Russell, 2001- fig. 4B) has a concave posterior quadrate edge. 
67. Changed to ordered.  Turfanosuchus (Wu and Russell, 2001- fig. 1B) seems to have postorbital-jugal contact.
69. This character is coding for two things- shape of the posterior indentation of the laterotemporal fenestra, and squamosal-quadratojugal contact.  The latter is separated here as character 256.
72. Changed to ordered. 
73. Changed to ordered. 
74. The disarticulation in Turfanosuchus (Wu and Russell, 2001) makes it uncertain how much of the quadrate was exposed laterally. 
75. Changed to ordered, states rearranged. Turfanosuchus (Wu and Russell, 2001- fig. 1A and 2A) had quadrate condyles far posterior to its occiput.
76. Changed to ordered, states rearranged. Turfanosuchus (Wu and Russell, 2001- fig. 2A) had a jaw articulation roughly on level with its maxilla. 
79. Turfanosuchus (Wu and Russell, 2001- pg. 43) has a basioccipital fused to its opisthotic-exoccipital. 
82. This character codes for supratemporal-squamosal contact, with all scorable taxa coded "st absent or fused", but supratemporal absence/fusion was coded by character 81.  Thus the taxa are recoded inapplicable.
83. This character codes for supratemporal-postorbital contact, with all scorable taxa coded "Supratemporal and squamosal are fused", but supratemporal absence/fusion was coded by character 81.  Thus the taxa are recoded inapplicable.  Oddly, Arizonasaurus and Saturnalia were coded as having supratemporal absence/fusion for this character despite not being coded for the previous two.  The disarticulated nature of Arizonasaurus' skull means it can't be coded, while the posterior skull of Saturnalia has yet to be described in detail.  Both are recoded as unknown.
86. Disarticulation and partial preservation prevent knowing whether the sides of the occiput converged dorsally in Turfanosuchus (Wu and Russell, 2001).
87. Turfanosuchus (Wu and Russell, 2001- pg. 43, fig. 5) does not preserve the vomer or anterior palatine, so cannot be coded for a vomeronasal fenestra. 
88. This character codes for palatine contact with the vomeronasal fenestra (vno), but all included taxa lack the latter (character 87), so should be coded inapplicable, not "palatine does not approach vno". 
90. Turfanosuchus (Wu and Russell, 2001- pg. 43) has premaxilla-choana contact. 
92. Included taxa are coded as having one of three states for choanal position- lateral (0); medial vomers narrow (1); medial rostral pinch (2).  Herrerasaurus and Coelophysis are the two taxa coded as having a rostral pinch.  In Herrerasaurus, only the posteriormost portion of the vomer is preserved, and is only viewable laterally (Sereno and Novas, 1993- pg. 458).  It can't be distinguished in CT scans either, so I guess Peters scored it from the dotted hypothetical vomer in Sereno and Novas' figures 7 and 8.  The preserved palatines visible in the CT scan do indicate the choanae were placed medially though.  Colbert (1989- figure 43A) reconstructed Coelophysis as having a rostral pinch, but his figures are widely derided as inaccurate, no photographed specimen shows transverse width of the vomer or palatine, and indeed he indicates how tentative this area's interpretation is with " the palatine probably forms the posterior-medial border of the naris (sharing this border with the vomer)".  Even if Coelophysis does have the pinch, it would not share it with other included taxa, making the state useless.  So states 1 and 2 are combined as "medial".  The only taxon coded as having laterally placed choanae is Pantydraco, but that doesn't even preserve the vomer or anterior palatine (Yates, 2003- fig. 2B), so cannot be coded.
93. Turfanosuchus (Wu and Russell, 2001- pg. 43, fig. 5) does not preserve the vomer or anterior palatine, so cannot be coded for choanal size. 
95.  Included taxa are either coded as having an interpterygoid vacuity that tapers sharply (0) or "palatines are parallel" (4), which doesn't even involve pterygoid vacuity shape.  Of taxa coded 4, Herrerasaurus' pterygoids are preserved pressed together (Sereno and Novas, 1993- figure 1F), which would be state 2 (absent, no interpterygoid vacuity).  This was assumed to be taphonomic by the authors who reconstructed it as having state 0 (figure 7D), but it could just as easily be state 3 (open anteriorly).  So I would code it 0/2/3.  Turfanosuchus was also coded 4, but has a sharply tapered vacuity (Wu and Russell, 2001- figure 5).  Gracilisuchus has state 3 (Romer, 1972- fig. 3).
96. Turfanosuchus (Wu and Russell, 2001- pg. 93) does not preserve the vomer, so cannot be coded for vomeral teeth.
98. Changed to ordered, states rearranged.
101. Turfanosuchus (Wu and Russell, 2001- pg. 43) lacks teeth on the pterygoid transverse flange.
104. Turfanosuchus (Wu and Russell, 2001- fig. 5) has a sharply angled pterygoid transverse flange.
105. You define this character as "Pterygoid shape: (0) broad triangle; (1) narrow; ... [other states]" and credit it to Rieppel (1998) and Mueller and Reisz (2006).  But the only similar characters in those analyses are 44 in the former "Pterygoid flanges well developed (0) or strongly reduced (1)" and 18 in the latter " Pterygoid: transverse flange broad-based and distinctly angular (0) or narrow and tonguelike (1) in ventral view."  The covered anterior portion of the flange in Turfanosuchus (Wu and Russell, 2001- fig. 5) makes it uncodable. 
106. Turfanosuchus (Wu and Russell, 2001- fig. 5) has prominent basipterygoid processes.
107. This codes for the presence of palatal teeth, which are already covered by character 100 and 101.  Thus it is deleted.
108. Changed to ordered, states rearranged.  Also, the state "one, two or three" was divided into three states, one for each different number of premaxillary teeth. Turfanosuchus (Wu and Russell, 2001- pg. 44) has five premaxillary teeth.  Trialestes only preserves the posterior part of the premaxilla bearing two teeth (Reig, 1963- fig. 6), so can only be coded as having at least two teeth.  Daemonosaurus has three teeth (Sues et al., 2011).  Scelidosaurus has five teeth (Norman et al., 2004- pg. 335).
110. This codes for premaxillary teeth being robust (0) or tiny or absent (1).  As character 108 already coded for premaxillary toothlessness, state 1 is changed to "tiny", and taxa without premaxillary teeth are coded inapplicable.
114. Changed to ordered.
115. This codes for maxillary tooth morphology.  State 4 "toothless" is deleted as character 114 codes for that.  Taxa without maxillary teeth are recoded inapplicable.  Tooth basal constriction is made into a new character- 257. Teeth being sharp vs. blunt is better expressed as serrations being small vs. large, as taxa coded as having blunt teeth like Pantydraco actually have sharp tips (Yates, 2003- fig. 9) but large serrations.
256. Turfanosuchus (Wu and Russell, 2001- pg. 42) does not show definite evidence of quadratojugal-squamosal contact.


  1. Mickey, Thank you for your efforts.

    Please send me the MacClade file and your tree.

    Miscodings galore in Turfanosuchus? I'd like to see those and your references. That was a tough one to code. I downloaded your comments and will be able to reply as I digest them.


  2. I honestly believe that this analysis should be published in a peer review academic journal (as many other excellent posts coming from the archive section of TDBB).


  3. Mickey,

    Just ran your MacClade file through PAUP and it recovers 11 trees as you indicated with Turfanosuchus and Gracilisuchus (carnivores) at the base giving rise to Poposaurids (herbivores), then Silesaurids (herbivores), then Marasuchus (carnivore), then Pisanosaurus (herbivore), then a split between Ornithischians (herbivores) and Saurischians led by Sauropodomorphs on one branch (herbivores) and Daemonosaurus (?vore) at the base of Panphagia + Pampadromaeus (herbivores) and theropods (carnivores). As you can see this is a varied mix of herbies and carnies, which is not the case in the Large Reptile Tree, which you were testing. In the LRT the carnies were basal, giving rise via PP&D to herbies. It is also the case in the LRT that removal of any taxa or clades does little to nothing to shift the remaining taxa. It's a stable tree. However, in your tree, if you remove the poposaurids, then the topology reverts to the LRT with carnies at the base. Try it and replicate the experiment, then get back to me with your thoughts.

    Dave Peters

    1. Poposaurus was carnivorous, as Parker and Nesbitt (2013) show with their cranial material, so your tree has to deal with that too. Daemonosaurus also shows no reason to doubt carnivory- its lateral teeth were recurved and compressed with small serrations like most carnivorous archosaurs. The anterior teeth were rounded in section, but that's true of things like megalosaurs (e.g. Duriavenator) and Compsognathus too. It doesn't matter though, as your tree is based on miscodings. It can look as neat and tidy as it wants, but if the input's wrong, the output has no credibility. Ditto for how well your tree survives taxon deletion (note "my" tree is still mostly yours- I haven't changed any postcranial characters or codings, and only corrected two taxa for all cranial characters when I sent you the file; so the matrix you're complaining is unstable is really yours with half the character formulations fixed and 1.4% of the entries changed).

      Take your Theropoda with Marasuchus, Trialestes and SMNS 12352, for instance. Of cranial characters diagnosing it... one is a narrow skull, but how could you code this for Marasuchus or Trialestes? We only have lateral views of both. Another is maxillary palatal process present, but again how could you code Trialestes which only has a lateral view published? You only code the character as lacking in three "phytodinosaurs"- "Thecodontosaurus"(=Pantydraco) and Pisanosaurus which don't preserve the anterior maxilla so cannot be coded, and Lesothosaurus which actually has it (Sereno, 1991 figure 5D). Another is an anteriorly pinched interchoanal rod, but that's unknown for Herrerasaurus (you used the dotted hypothetical vomer in Sereno and Novas' figure), again uncodable for Trialestes which has no palatal information available, which leaves only Coelophysis coded for it (itself questionable). Going further into Theropoda, Marasuchus was joined with Coelophysis, Tawa and SMNS 12352 based on characters like nasals with parallel sides, when Marasuchus doesn't preserve a nasal and SMNS 12352's nasal is expanded posteriorly, making both of your codings wrong. For the same node, you code them as having a naris angled less than 30 degrees downwards, but SMNS 12352 doesn't preserve any narial border, while Trialestes' preserved ventral border suggests a highly angled naris. And again for that node, paroccipital processes angled >40 degrees posteriorly, but Marasuchus' are transverse. I could go on. Gracilisuchus has 32 miscodings, Arizonasaurus has 53, Trialestes has 26, Marasuchus has 24, SMNS 12352 has 17. Each out of 115.

      So yes your trees are stable, but only because your data is wrong. Yet as I correct it, your tree's looking more and more like the consensus. After I recoded the above taxa for cranial characters, Trialestes and SMNS 12352 are kicked out of Dinosauria, and we still have traditional Dinosauria and Saurischia. I'd love to debate the differences between our data, but what's important are the codings we put in, not the characters of the trees that come out.

  4. PS I encourage you to take another look at the many shared traits of Gracilisuchus and Herrerasaurus, which are close kin in the LRT, but are widely separated in your recovered tree.