Re: Cladistics in Science

[Jason Anderson <jander3@po-box.mcgill.ca>]

gmbra@cygnus.uwa.edu.au wrote:
> 
> A recent issue of Science (11th June, Vol 284, 1999) has an article by Fox,
> Fisher and Leighton on a comparison of cladistics vs stratocladistics. It
> claims stratocladistics is much better at recovering the "real" phylogeny
> than conventional cladistics is. Trouble is, I don't understand enough
> cladistics to evaluate the conclusions. It would be most informative if one
> or more of the cladistic experts on this list could post a potted summary
> of this paper and their response to the conclusions and implications (if
> any) for dinosaur phylogenies. I'm sure I'm not the only one who would be
> interested in your views. Any takers?
> Thanks
> Graeme Worth

I'll take a shot at it.

Cladistics and phenetics are methods which makes observations of
character states in a given group of animals and clusters the animals
based upon shared possession of these traits.  Unlike phenetics, which
considers the total number of shared character states, cladistics
determines ancestral vs. derived states, and making the clusters by the
shared number of the derived states only.  These derived states, or
synapomorphies, arrange the animals into a nested heirarchy--the more
synapomorphies, the closer the common ancestry of the given animals in
question.

Different characters may give different trees, though.  Character 1, for
instance, could suggest the cluster A(B+C), while characters 2 and 3
might suggest the cluster B(A+C).  To solve this problem systematists
use parsimony--the answer which requires the fewest appeals to
additional ad hoc hypotheses.  If A(B+C) were to be accepted, we would
have to assume that characters 2 and 3 were giving false
information--say, reverting to a more primitive state, or had achieved
this state convergently.  Whereas the cluster B(A+C) only requires
character 1 to be in error--thus the second cluster is accpeted because
it requires only one additional ad hoc hypothesis--called homoplasy (the
character being homoplastic).

Neither cladistics nor phenetics considers information beyond the
anatomical (or molecular) characters, but paleontologists recognize that
there is information to be had from the fossil record as well.  This
only makes sense--ancestors must preceed their descenants, thus animals
found lower in the stratigraphic column *should* be more primitive.  

Stratocladistics is an attempt to include stratigraphic information into
the character analysis.  The stratigraphic column is divided into units,
and periods of representation are coded and the results evaluated using
parsimony.  Here though, the additional ad hoc hypotheses are not
homoplasies but are instances of non-preservation.  So, the phylogenetic
hypothesis which will be accepted is not the one which minimizes the
number of additional ad hoc anatomocal changes (reversals or
convergences) but the one which minimizes the number of anatomical
homoplasies AND stratigraphic hypotheses of non-preservation (termed
"Total Parsimony Debt")--so the shortest character tree may not be the
shortest stratocladistic tree.  (Fisher wrote a chapter on the
particulars of this method in the MacClade manual, if you have access to
it.  Also, Clyde and Fisher in Paleobiology 23:1-19 give a further
review beyond Fox et al.)

This last point is the crux of the sometimes heated debate of the use of
stratigraphic information in phylogenetic reconstruction.  Yes, the
fossil record is very imcomlpete, but the proponents of stratocladistics
have recognized this fact by removing periods of non-representation in
the fossil record--times when there are simply no fossil-bearing
rocks--from the analysis at the outset.  To see opinions on all of the
sides of this arguement the Nature web site had an on-line forum which
should be archived.  See Fisher's, Wagner's and Siddal's contributions
for a taste of the various opinions.  And delivery styles ;)

So, with this background, what was accomplished by Fox et al. was rather
simple.  They got a computer to generate random phylogenies upon a given
time scale, deleted 10-90% of the information, and compared which
method--strict character or stratocladistics--recovered the known, true
phylogeny most frequently.

As to implications for dinosaur phylogeny, well, the tree(s) presented
by Sereno already follow the stratigraphic column pretty closely, so
application of stratocladistics is unlikely to radically change our
views of dinosaur evolution.  There are problems with large periods of
non-preservation throughout the Mesozoic, which SHOULD not be a problem
for stratocladistics.  This, in my opinion, is the biggest question
mark.  I work on tetrapods from the Paleozoic--I can count the number of
windows into the fauna on my fingers and toes.  When you only have
fairly derived end points to
analyze, convergences may be falsely interpreted as synapomorphies.  I
would like to see further work demonstraiting that stratigraphic
information has the ability to break down these "long-branches".  I
suspect that this method is at its most effective when the fossil record
is most complete--Cenozioc mammals and marine invertebrates, which is
exclusively where the method has been employed to date.

I hope this helps,

Jason Anderson
Redpath Museum, McGill University

jander3@po-box.mcgill.ca
htt[://www.mcgill.ca/redpath/jsahome2.htm