RE: Morpho v molecular (was Re: Tinamous: living dinosaurs)

[Mickey Mortimer <mickey_mortimer111@msn.com>]

Tim Williams wrote-

> > Has there ever been a case where a well established molecular-based 
> > relationship was discarded due to morphology? There have been 
> > molecular-based relationships that were later changed by better
> > molecular data which matched morphological ideas (e.g. guinea pigs aren't 
> > rodents), but I don't know of any molecular-based relationship which is 
> > consistantly found and which workers believe is wrong.
>
> I can think of two, off the top of my head.
>
> 1. Whippomorpha
>
> Molecular-based phylogenetic analyses consistently recover a
> Cetacea-Hippopotamidae clade (Whippomorpha or Cetancodonta) nested
> inside Artiodactyla, as a derived artiodactyl clade. By contrast,
> morphology-based phylogenetic analyses (which include fossil taxa)
> fail to recover this "Whippomorpha" clade. Instead, purely
> morphology-based analyses find a more basal position for whales
> (Cetacea), as the sister taxon to Artiodactyla, with hippos closer to
> pigs (Suiformes). (Uhen MD [2010; Annu. Rev. Earth Planet. Sci.
> 38:189–219] shows the two trees side by side, Figure 5).
>
> By and large, paleo workers believe the molecular-based relationship
> (Whippomorpha) is wrong. I tend to agree.

I don't know enough about mammals to have an informed opinion, of course.  But 
Uhen's review you mention accepts Whippomorpha as the most likely hypothesis, 
and he is a paleontologist.  Note too the support of Whippomorpha from O'Leary, 
Spaulding, Geisler, Theodor, Boisserie and other paleontologists.  The most 
recent and largest morphological analysis I know is Spaulding et al. (2010), 
with 661 phenotypic characters (and 46587 
molecular ones) and 81 taxa.  The total evidence analysis finds 
Whippomorpha, but like you say the phenotypic-only one doesn't.  Yet (not 
mentioned 
by the authors, but based on running their matrix in PAUP) Whippomorpha 
is only 4(!) steps longer in the phenotypic-only matrix.  Hardly makes 
Artiodactyla sensu stricto a robustly supported clade morphologically.  Seems 
like usual, the morph matrices just needed more taxa and characters.

> 2. Turtles (Testudines)
>
> A recent morphology-based phylogenetic study of turtle origins (Lyson
> et al. 2010; Biol. Lett. 2010 6: 830-833) found turtles to be
> parareptilies, with Testudines as the sister taxon to the fossil
> _Eunotosaurus_. This Testudines-_Eunotosaurus_ clade directly
> contradicts molecular-based analyses, which found turtles to be inside
> Diapsida, as the sister group to Archosauria.

Lyson et al. just used the Rieppel amniote matrix (with Odontochelys added) and 
added Eunotosaurus and Proganochelys, along with six characters shared by them. 
 I should note adding Proganochelys is odd, since the original Testudines OTU 
already included it.  In any case, what the authors don't mention is that 
constraining turtles to be diapsids is only 7 steps longer.  Makes the number 
of added characters seem pretty convenient.  So again, not a big difference.  
This is especially true given the small number of characters (176 for a group 
as large as Amniota?!) and low taxon sampling.  Archosauromorphs are 
represented by Choristodera, Trilophosaurus, Rhynchosauria, Prolacertiformes 
and Archosauriformes.  What if turtles are closer to Azhendohsaurus, 
Teraterpeton, dosweliids, aetosaurs, Lotosaurus, etc.?  Turtle relationships 
are still equivocal given the current analyses- just a year earlier, two of the 
Lyson et al. coauthors wrote a paper suggesting evidence for archosauromorph 
turtles.  Just because the latest paper supported parareptile turtles doesn't 
make it the morphological consensus.

> > I trust molecular data over morphological any day, since even huge analyses 
> > like Livezey and Zusi's seem doomed by convergence.
>
> Convergence can occur at the molecular level too. As noted by Lyson
> &c (and many others) molecular data is bedevilled by the same issus as
> morphological data: taxon sampling, homplasy/convergence, rate
> heterogeneity, missing data, etc. It's just that these issues are not
> obvious in a molecular tree because the character states are
> effectively invisible.
>
> For example, when ratites emerge as monophyletic in a morphology-based
> analysis, molecular-based workers can (and do) question this topology
> by saying: "This signal is due to convergence, because it's based on
> shared flightlessness-related characters". So any character states
> that might be prone to convergence can be easily identified in a
> morphology-based tree. But if molecular characters are screwing up a
> topology, how would we know?

But at least with molecular data we don't have to worry about human bias in 
choosing characters, small numbers of characters, or miscoding (I'm sure any 
sequencing errors are much less frequent than what is common in morph data 
matrices at least).

Mickey Mortimer