The Importance of Being Extinct (was: Re: phyletic bracketing]

["Jonathan R. Wagner" <jonathan.r.wagner@mail.utexas.edu>]

I don't have much time to contribute to this disucssion, but I couldn't
resist making a few points from a systematic perspective. My apolgies if
this is disorganized, I have spent too much time on this already. As usual,
please cc me in responding to this post, as I do not check the digest
thoroughly.

First, I should point out that many of the studies Dr. Carpenter referred to
have not been published (except in abstract form). Their widespread
acceptance is, in fact, a fault in the way science is done... why is
everyone taking the lips off of thereopods when no paper has been published?
Anyway, because they are "out there," I will address them, but I do think
we, and the media, should give their authors time to formulate their
arguments completely.


David Marjanovic wrote:
> Eh, no. Then _it is most parsimonious_, in other words _most scientific_,
to
> assume that nest guarding occurred in non-neornithean dinosaurs. As long
as
> we don't have direct evidence for the contrary.

To paraphrase, parsimony is the beginning of wisdom, not the end.Yes,
parsimony is good as a first cut, a non-parametric estimate. However, a
probabalistic approach may be more appropriate, especially under the
circumstances under discussion. Parsimony does not equal science, it is
simply a more scientific approach than unbridled speculation.


Ken Carpenter wrote (comments are reordered):
> Phyletic bracketing analysis (PBA)

Would you care to comment on your choice of "phyletic" over "phylogenetic?"
I'm not objecting, I'd just like to know your reasoning.


>I offer as examples: the predentary bone of all ornithischians, the rostral
bone of all ceratopsians, etc. Neither of which occur in either birds or
crocs

    Phylogenetic methods are meant to assist in the inference of
*unpreservable* (or unpreserved) aspects of fossil animals... in these
cases, these structures are preserved... a more interesting comparison might
be the beaks of ornithischians (which are sometimes preserved, but what if
they never were?).

> The same principle is used to interpret the function of dinosaur
structures.

There are several approaches to reconstructing unpreservable aspects of
extinct animals. The current favorite among American paleontologists is a
phylogenetic method, wherein parsimony-based character reconstruction is
used to infer the ancestral state at a the first node ancestral to the taxon
of interest which has at least two extant descendant lineages in which the
presence or absence of the unpreservable (or simply unpreserved) trait can
be established under parsimony. This can be used to establish a null
hypothesis, but cannot test whether or not the feature actually occurs in
the fossils in question. This is Mallon, Marjanovic, and Williams's point.

Witmer's "Extant Phylogenetic Bracket" method goes a bit further than this,
by positing direct correlations between unpreservable features and
osteological features which are (potentially) preserved in the fossil
record. Presence of these "osteological correlates," and their configuration
in the fossils of interest, is used as a test for the presence/ absence of
the unpreservable trait. This is an explicit method of ranking the
likelihood of hypotheses. Contra Mallon, this is where the examination of
novel structures come into play (see 2e, below). An important point: the
"EPB" is NOT, as some suggest, "just parsimony analysis."

A more traditional approach involves examining organismal analogues (usually
organisms considered to be "morphologically primitive") within the
least-inclusive, often paraphyletic, taxon that includes the fossil in
question. In some cases, analogues are sought from distantly related taxa
which nonetheless exhibit morphology or behaviour similar to that inferred
for the fossil. This technique also involves the exploration of
"osteological correlates," although these are often hedged by an
understanding of the morphology and biology of the fossil group, and/or by
mechanical analogues (see below).

Another popular technique is the "morphologie-konstruction" method,
generally useful in the case of functional morphology, wherein a mechanical
analogue is contructed, and comparison of the mechnical properties of the
model guides interpretation.

The traditional paleontological approach, at least in the U.S., has been a
somewhat haphazzard combination of these last two elements, sometimes with
implicit phylogenetic control. These, taken together, were part of the
traditional discipline of comparative anatomy. The current move towards
phylogenetic techniques can be seen as an outgrowth of the modern
"Comparative Method," which seeks to integrate phylogeny into comparisons.


> The problem of PBA is that it cannot explain novel structures, and indeed
in the strict adherence to the techniques does not allow for such

    I agree that one of the weaknesses of the "EPB" method, specifically, is
that its reliance on osteological correlates weakens inferrence in the case
of novel structures (here defined as structures not present in bracketing
organisms). However, it should be noted that a lot of Witmer's recent
research seems to be focused on the question of interpreting anatomical
novelties.

    In no particular order, these are my criticisms of the method, based on
work I plan to publish at some point:

    1) "EPB" relies on obligatory character correlation. "Osteological
correlates" are characters, as are the structures with which they are
presumeably correlated. In our current understanding (or lack thereof, if
you prefer) of morphological evolution, many supposedly correlated features
of organisms appear to evolve at different times, and some seemingly tight
correlations were much more labile in the past (see the supposed obligate
predator-prey relationship between prairie dogs and the black-footed
ferret). By assuming character correlation a priori, you limit your universe
of possibilities. Over short divergence times, this may not be a problem
(since you presumeably have good evidence that the characters are strongly
correlated today). See below for long divergences.
    It should be noted that the "EPB" method is not alone in its reliance on
character correlation. Many traditional approaches also posit character
correlations (e.g., fins and swimming).

    2) Extant animals are often quite morphologically dissimilar to their
last common ancestor, especially when their lineages diverged long in the
past. This leads to a number of considerations:

        a) Many EPB studies ONLY consider the immediate "extant bracket" of
the fossils in question. However, it makes more sense to "outgroup" one's
bracket; increasing your sample of extant taxa can help account for long
divergence times (see below). For example: consider the putative liver of
ankylosaurs... crocs have a liver and birds have a liver, but (given the
arguments below), this could be homoplastic. Add in lepidosaurs, turtles,
mammals, lungfish, coelocanths, ray-finned fish and sharks, and the odds of
homoplaisy in the liver drop dramatically, from a probabalistic perspective.
For a less trite example, reference to other vertebrates shows that lips are
widespread in vertebrates, and may even be ancestral for saurians (see 2c,
below). Why this additional information in other extant vertebrates is not
used is beyond me, but it is not a flaw in the method, just in the
execution.

        b) Over long divergences (such as bird/croc), many changes may have
occurred among sister taxa. The vulnerability of parsimony analysis to the
problem of long branches has been extensively documented, and is especially
problematic in cases of poor taxon sampling (see 2a above). The rostra of
birds and crocodylians are exceptionally modified relative to those of their
last common ancestor; indeed, some of the hard-tissue modifications have
occurred in parallel (e.g., extensive development of a solid bony palate).
>From a probabalistic perspective, the longer the branches leading to these
taxa from their last common ancestor, the more likely it is that homoplastic
character change has occurred.

        c) Despite the phrasing of this discussion, it is not clear that
"characters" have objective reality. Animals do not come in descrete chunks,
and our perceptions and our coding can greatly affect the outcome of an
analysis... If you simply code, say, "lips/ no lips," you can optimize the
ancestral archosaur as lacking lips. However, if you note that the presence
of a full beak pretty much obviates the possibility of lips, and code it as
a three-state unordered character, "lips, no lips, beak," or make two
characters, "labial integumnet, no labial integument" and "lips, lips and
beak, beak" (ordered), the ancestral state for archosaurs is "lips" under
DELTRAN, and "beak" under ACCTRAN.

        d) Combining points 1, b and c, historical transformations, such as
would be expected over long branches, may severaly complicate
interpretations, and may even result in the loss of "osteological
correlates." For example, amphibians without a lacrimal bone will lack a
lacrimal foramen... does this mean they lack the lacrimal duct? The rostra
of crocs and birds are highly modified, and their lifestyles have diverged
greatly from that of the ancestral archosaur... would we EXPECT to find a
vomeronasal organ in either group?

        e) Using the "EPB" to interpret novel structures is fine, but it
must be remembered that, because the structure IS novel, there will not be
osteological correlates of that structure in the bracketing organisms. This
is not catastrophic, becauce Witmer et al. are much more sophisticated than
that: they look for osteological correlates of structures that should be
present in all relevant taxa (e.g., the nasal capsule), then use the
position of those structures to infer other aspects of the fossil's anatomy.
However, with novel structures, we already KNOW that something is VERY
different, and reliance on the morphology of other taxa (especially those
that are themselves highly modified) may bias our outcome against more novel
interpretations.

        f) Another issue with some comparative analysis, not just those in
the "EPB" framework, is the practice of assessing soft-tissue homologies
before hard tissue homologies, and hard tissue data. For example, Dr.
Carpenter's point about the cheek armor of ankylosaurs. Witmer's anteorbital
fenestra monograph seems to be a good example of a study that does not fall
into this trap; Witmer carefully considers the hard-tissue homologies
throughout fossils, then makes his interpretations. Because one can usually
find fossils more closely related to the taxon of interest than are the
"bracketing taxa," and these fossils should be on much shorter branches (see
below), it seems clear that assessment of hard-tissue homologies should
preceed soft-tissue homologies. Also, because of other variables (e.g.,
degree of ossification), we may be able to identify potential "osteological
correlates" in some extinct taxa not present in existing relatives (e.g., in
amphibians). We wouldn't want to fall into a version Patterson's Fallacy,
and believe that only extant taxa can be informative.

        g) A similar objection can be raised with regard to the hypotheses
of some "EPB" studies: very often they focus on soft-tissue expalantion to
the exclusion of other possibilities (e.g. biomechanics). Witmer has pointed
out the importance of soft-tissues in shaping hard tissues, apparently as
justification for this approach. I would not contest that soft tissues play
an integral role in shaping bone. However, hard tissues play a major role in
the structure and function of vertebrates, and themselves have a strong
influence on, and correlation with, the arrangement of soft tissues. There
are a number of archosaurs that lack evidence of pneumaticization of the
anteorbital fenestra. Is it not possible that this structure could have a
biomechanical function, and was subsequently invaded by pneumatizing
tissues?

    For my money, the solution to these issues is to consider the
reconstruction of unpreservable features in a phylogenetic context, but a
context cast a broadly as possible, deeply outgrouping your analysis (2a),
examining hard-tissue homologies before soft. Coding osteological correlates
separately (1) as part of a sophisticated character analysis(2c), supported
by reference to possible natural and mechanical analogous, and an
appreication for the morphological integration of organisms (2b, 2d, 2e, 2f,
2g), should account for many of the issues I raised above. Detailed
understanding of the fossils in question, and their extinct relatives, can
be used to generate assessments of hard-tissue homology, and can guide
interpretation further. I suspect that Dr. Witmer and his students are
actively engaged in at least some of these activities, which may explain the
long delay in publication of their results.

    I further believe that a probabilistic, parametric approach to ancestral
state reconstruction is more appropriate to this sort of analysis than is
parsimony. Likelihood-based approaches account for branch lengths directly,
accepting more homoplaisy among longer branches (2b). Such an approach is
widely accepted in modern systeamtics; has been used to reconstruct the
ancestral visual pigment of archosaurs, and is being applied to a number of
other problems. The "EPB" method is already essentially probabalistic...,
the "types" of inference are (IMHO) rankings of different probability
classes. Models are now available that incorporate most of the same
assumptions as parsimony analysis in dealing with discrete character data,
but provide for a more sophisticated approach to character analysis in a
likelihood framework. Ancestral states can be estimated as probabilities
rather than point-estaimates.

    Note that this still places the emphasis squarely on phylogeny.
Phylogenetic relationships (when they can be established) must be understood
prior to any interpretation of fossil organisms, because they can inform as
to the appropriate analogous organisms, or the extent to which
generalizations may be applied or must be qualified. Bryant and Russell
(1992, Phil. Trans. R. Soc. Lond. B, 337:405-418) published an often-ignored
study that presaged Witmer's work. B&R advocate a mixed approach,
incorporating more outgroups, as well as "extrapolatory anlaysis" (e.g.,
biomechanical models). They accept the primacy of phylogeny in comparative
interpretation, but they acknowledge some of the limitations of bracketing
noted above. They offer the possibility of rejecting phylogenetic arguments
(to be fair, Witmer's "EPB" method does not discount this). B&R point out
the conservative nature of phylogenetic approaches, and their limitiations
regarding novelties. They specifically cite non-avian ornithosuchians as
vulnerable due to the degree of transformation in modern birds.

    End of rant.

    Wagner