Points of View
Syst. Biol. 53(2):333–342, 2004
c Society of Systematic Biologists
Copyright ISSN: 1063-5157 print / 1076-836X online
DOI: 10.1080/10635150490423962
Accepting Partnership by Submission? Morphological Phylogenetics
in a Molecular Millennium
R ONALD A. J ENNER
University Museum of Zoology, University of Cambridge, Downing Street, Cambridge CB2 3EJ, United Kingdom; E-mail: [email protected]
Organismic and morphological approaches to biology
have lately come under attack from a variety of corners
in both teaching and research. However, the main motivations for these attacks are not scientific but are instead
reflective of economic pressures on teaching and the current fashionability of molecular approaches in research.
For example, over the last few decades the number of
students taking courses in invertebrate zoology in North
American universities has declined significantly. This decline has largely been abetted by the abandonment of
classes with small enrolments as increasingly more students are encouraged to take classes in the more fashionable disciplines of molecular and cell biology (Fautin and
Watling, 1999). More recently, the fact that systematics is
a time-consuming science dependent upon the mastering of specialist knowledge has been cited as a motivation
for redirecting research toward DNA taxonomy (Pennisi,
2003; Tautz et al., 2003), and the utility of morphology for
phylogeny reconstruction has recently been questioned
on several grounds (see Hillis and Wiens, 2000; Baker
and Gatesy, 2002).
The most recent challenge to the value of morphology for phylogenetics was published in a recent issue of
this journal (Scotland et al., 2003). After a critical look at
morphological phylogenetics, Scotland et al. (2003:543)
reached a strong conclusion: “We disagree that morphology offers any hope for the future to resolve phylogeny
at lower or higher taxonomic levels.” Instead, they advocated a “more limited role for morphological data in phylogeny reconstruction” (2003:545), which consists chiefly
of studying fewer morphological characters in the context of molecular phylogeny. Because the validity of these
arguments would impinge upon morphological phylogenetic research ranging from algae to elephants, a detailed response to Scotland et al. ( hereinafter SEA) is
necessary.
SEA adduced several arguments in support of their
contention that morphology has little if anything to offer to the future of phylogeny reconstruction and contended that we should be investing our time and effort
into obtaining more molecular phylogenetic data. Here, I
evaluate the main arguments of SEA and expose several
weaknesses in their reasoning, the correction of which
leaves the importance and future promise of morphology for phylogenetics fully intact.
NUMBERS OF CHARACTERS , PHYLOGENETIC ACCURACY,
AND CLADE S UPPORT
The coordinating themes of the critique of SEA are
their concerns for phylogenetic accuracy and quantitative clade support. SEA merged these two aspects of
phylogenetic analysis into a universal yardstick that they
used to assess the relative value of molecules and morphology. SEA (2003:539–540) claimed that in general both
accuracy and support are positively correlated with the
number of phylogenetic characters. First, SEA (2003:539)
concluded that “the number of characters needed in simulation studies to recover accurate trees is an order of
magnitude greater than that available from morphology.” Second, SEA (2003:540) claimed that “the low character/taxon ratio in many morphological studies itself
precludes high support values.” Do these findings provide an irrefutable basis for discarding morphological
evidence from phylogeny reconstruction?
Phylogenetic Accuracy
SEA judged the relative worth of molecular and morphological data with respect to the accuracy of phylogeny reconstruction. They asserted that increasing
phylogenetic accuracy generally depends on increasing
the number of characters, as is summarized in their
figure 1a. As SEA made clear throughout their article, their main contention is that morphological data
are inferior to molecular data in contributing accurate
phylogenetic signal.
Despite comparing the performance of morphology
versus that of molecules with respect to phylogenetic accuracy, SEA failed to discuss how phylogenetic accuracy
333
334
SYSTEMATIC BIOLOGY
can be estimated. Phylogenetic accuracy can only be assessed when the true phylogeny is known prior to the
analysis. Phylogenetic accuracy is therefore assessed either by simulation studies or by study of known phylogenies (laboratory strains) (Hillis et al., 1994a, 1994b;
Hillis, 1995; Givnish and Sytsma, 1997). Consequently,
given the typical limitations of available molecular or
morphological evidence for a particular set of taxa for
which we do not know the phylogeny, we simply cannot
know whether our phylogenetic estimate is accurate, i.e.,
represents the true phylogeny. This realization seriously
compromises the value of phylogenetic accuracy as an
empirical guide in most phylogenetic studies (Siddall
and Kluge, 1997; Siddall, 1998; Farris, 1999).
SEA’s arguments about phylogenetic accuracy are
based on simulation studies, summarized in their
figure 1a. SEA (2003:542) stated that “DNA sequence data
at least offer the unique potential of scoring large numbers of unambiguous characters and character states.”
SEA stated (2003:541) that in contrast to this bounty of
sequence data, simulations show that “the number of
unambiguously coded morphological characters for any
study is finite . . . and less than the number typically required to accurately reconstruct phylogenies in simulation studies.” However, simulation studies can reliably
assess phylogenetic accuracy only when the data evolve
precisely according to the assumed model of evolution
(Hillis et al., 1994a; Hillis, 1995). If the data do not evolve
according to the specified model, then the phylogeny
may not be accurately reconstructed, and the number of
characters necessary to accurately reconstruct the phylogeny cannot be correctly estimated. The number of
characters needed to accurately reconstruct a phylogeny
varies according to the nature of the data at hand, the
number of taxa, and the chosen model of evolution (Hillis
et al., 1994a, 1994b, 2003; Hillis, 1996, 1998; Rosenberg
and Kumar, 2003). Therefore, it is very difficult to generalize the number of characters needed to accurately
reconstruct phylogeny in given instances. Models for sequence evolution have been selected as much (or more)
for their computational convenience as for their approximation to reality (Farris, 1999:201, 202; but see Sullivan and Swofford, 2001). SEA wrote (2003:541) that “the
exact processes underlying nucleotide substitution are
more complex than the simple models used in phylogenetic reconstruction.” This admission makes SEA’s appeal to accuracy largely a moot point because widely
employed methods cannot guarantee phylogenetic accuracy. With respect to morphology, SEA (2003:542) emphasized “our current inability to incorporate models
of morphological evolution into phylogeny reconstruction methods,” which makes it extremely difficult to determine the number of morphological characters necessary to accurately reconstruct phylogeny in specific
cases. However, the development of new models for
dealing with morphological data in a likelihood context (Lewis, 2001) may improve our ability to combine
morphological and molecular data in single phylogenetic
analyses.
VOL. 53
In repeatedly contrasting molecules and morphology,
SEA assumed that these are two distinct and clearly
alternative categories of data for phylogeny reconstruction. The greater potential of molecules to contribute to
increased phylogenetic accuracy is then chiefly a result of
the genome representing a much bigger reservoir of potentially useful unambiguous phylogenetic data. However, is this anything more than a truism?
Millions or billions of genomic nucleotides offer a
vastly bigger pool of potentially useful phylogenetic
variation than the phenotype of even the morphologically most complex eukaryotes. By the same argument,
one might as well conclude that because even entire mitochondrial genomes are typically dwarfed by the size
of the nuclear genome, they are therefore inferior phylogenetic indicators. However, the quality of a data set
and its size are two distinct attributes, and even limited
data sets may contribute phylogenetic accuracy. In practice, molecular phylogenetic studies are typically limited to one or a few genes, the analysis of which is far
more problematic than hinted at by SEA. Such limited
data frequently fail to provide enough information for
a robust and accurate phylogenetic estimate. For example, despite intense taxon sampling, the 18S ribosomal
RNA locus has so far yielded no stable and maximally
accurate hypothesis of metazoan relationships (Peterson
and Eernisse, 2001; Giribet, 2002). This result is not surprising because different genes may evolve at different
rates and may consequently be suited best for resolving
different levels of a phylogeny. In a recent study, Rokas
et al. (2003) made this potential particularly clear. They
performed a phylogenetic analysis of 106 orthologous
genes from seven species of yeast from the genus Saccharomyces. Their results (Rokas et al., 2003:798) suggested
that “data sets consisting of single or a small number of
concatenated genes have a significant probability of supporting conflicting topologies.” Thus, single molecular
data sets are potentially equally fallible with respect to
accurate phylogeny reconstruction as are morphological
data sets. Moreover, “the support for a given branch was
strongly dependent on the gene analysed” (Rokas et al.,
2003:799), which indicates that different genes are complementary in providing support for different parts of the
phylogeny. In line with these findings, workers increasingly have adopted multigene approaches to phylogeny
reconstruction to benefit from complementing data partitions that provide resolution at different levels in the
phylogeny (Colgan et al., 1998; Edgecombe et al., 2000,
2002; Baker et al., 2001; Giribet et al., 2001; Giribet, 2002;
Collin, 2003; Rokas et al., 2003), whereas each data set
alone, be it molecular or morphological, may have only
limited resolving power. So far, having a large proportion of the genome sequenced is an ideal that has only
been realized for a few species.
However, even a multigene approach may not necessarily solve all problems when the number of genes
is limited. Different genes may yield conflicting phylogenetic signals on all taxonomic levels across the tree
of life for a variety of biological and analytical reasons,
2004
POINTS OF VIEW
including choice of optimality criterion, comparison of
paralogous genes, taxon sampling, mutational saturation and long branch attraction, changing substitution
rates of a site across a gene and/or across a phylogeny,
compositional biases, lateral gene transfer, similar selective regimes leading to molecular convergence, and
lineage sorting, and the same gene may yield conflicting phylogenies when analyzed on the nucleotide or
amino acid levels (Miyamoto and Fitch, 1995; Maddison,
1997; Naylor and Brown, 1998; Hillis and Wiens, 2000;
Lockhart et al., 2000; Sota and Vogler, 2001; Sullivan
and Swofford, 2001; Taggart et al., 2001; Gribaldo and
Philippe, 2002; Lopez et al., 2002; Rydin et al., 2002; Shaw,
2002; Simmons et al., 2002a, 2002b; Machado and Hey,
2003; Rokas et al., 2003).
Consequently, not all parts of the genome necessarily uniformly and positively contribute to phylogenetic
accuracy, and their combined use in a single phylogenetic analysis may therefore not lead to increased accuracy. This makes SEA’s dichotomy between all molecular sequence data on the one hand and morphology
on the other not particularly meaningful. SEA’s conclusion (2003:539) “that a main constraint of morphologybased phylogenetic inference concerns the limited number of unambiguous characters available for analysis”
is equally applicable to many available molecular data
sets. Clearly, the well-known limitations of individual
molecular data sets have not led to the general dismissal
of molecular evidence. The acknowledged limitations of
morphological data sets should therefore no more be
used as an argument to dismiss the value of morphology
for phylogeny reconstruction. Even considering the potential of large amounts of genomic information, we do
not know whether we can reasonably expect genomes
to furnish the required amount of data to resolve all
phylogenetic problems on all phylogenetic levels. SEA
(2003:543) acknowledged as much: “an honest observer
would have to agree that even whole genomes for all
species will probably not yield a fully resolved, highly
confident tree.”
Quantitative Clade Support
SEA (2003:540) argued that morphological phylogenetic studies generally have “too few morphological
characters to provide confidence in any given estimate
of phylogeny.” SEA generalized the findings of Bremer
et al. (1999) and a few other studies in their figure 1b
and claimed (2003:540) that these results “demonstrated
explicitly that the character/taxon ratio for morphological studies is such that bootstrap percentages are
likely to be low.” This statement holds true for the
morphological data sets of two plant families analyzed
by Bremer et al. (1999). These data sets had low numbers
of characters: 3.2 and 2.6 characters/taxon for Rubiaceae
and Apocynaceae, respectively. Bremer et al. (1999) cited
the low number of available characters as the cause of
the low support measures for the phylogenetic analyses
of these data sets. The bigger molecular data sets had
335
better support values. I appreciate the potential value
of statistical support measures, leaving aside difficulties
of interpreting commonly used support values such as
bootstrap percentages (reviewed by Sanderson, 1995;
Siddall, 2002) and difficulties of justifying the use of
probabilistic arguments in historical inference in the
first place (Siddall and Kluge, 1997; Swofford et al., 2001;
Kluge, 2002). Morphology may often provide more
coarse grained phylogenetic resolution than molecular
data because of lower character/taxon ratios, but this
does not necessarily lead to lower quantitative support
measures. It is not only the number of characters but the
distribution of homoplasy that will determine how well
a given data set supports a phylogeny. Therefore, one
cannot uncritically generalize the findings of Bremer et
al. (1999) that small morphological data sets are poorer
performers than larger molecular data sets.
Kress et al. (2001) analyzed a morphological data set
for the Zingiberales (a group of monocotyledons including bananas and gingers) with 2.8 characters/taxon and
found that 57% of the ingroup nodes were supported
with bootstrap values >75%. In contrast, two of three
analyzed molecular data sets for this group generated
lower percentages of ingroup nodes with bootstrap values >75%, despite a much larger number of characters. It is unwarranted to generalize from a few morphological data sets discussed by SEA to the conclusion
(2003:540) that high support values are precluded “in
many morphological studies.” Whether high support
values are generated depends entirely on the morphological data set under consideration. In a recent comprehensive morphological analysis of snake phylogeny, Lee
and Scanlon (2002) found that >70% of the ingroup nodes
had bootstrap values of >75%. Similarly, in a morphological phylogenetic analysis of arthropod relationships
Edgecombe et al. (2000) also found that >70% of the ingroup nodes had bootstrap support values of >75%, and
high quantitative support values for morphological analyses are certainly not rare (Wiens and Hollingsworth,
2000; Damgaard and Sperling, 2001; Kress et al., 2001;
Gatesy et al., 2003). SEA presented no convincing evidence that morphological phylogenetic evidence is generally undesirable because “the low character/taxon
ratio in many morphological studies itself precludes high
support values” (2003:540). Data sets of single molecules
may provide contributions to clade support as limited
at that provided by morphological data sets. The important point is that the available character space should be
sampled as comprehensively as possible, and morphology may contribute valuable evidence in combination
with molecular data.
Although SEA acknowledged (2003:545) the potential
value of combined analyses of both molecular and morphological data, they failed to mention that in the study
of Bremer et al. (1999), from which they adapted their
Figure 1a, the support measures of the combined analyses of molecules and morphology were significantly
better than those of either the morphological or molecular data sets alone. Similar positive contributions of
336
SYSTEMATIC BIOLOGY
morphology to quantitative clade support measures in
combined analyses have been observed for a host of
other taxa across the tree of life (Flores-Villela et al., 2000;
Kress et al., 2001; Winterton et al., 2001; Edgecombe and
Giribet, 2002; Giribet et al., 2002; Stach and Turbeville,
2002; Jeffery et al., 2003; Schulmeister, 2003; Wahlberg
and Nylin, 2003). The positive contribution of morphology to combined analyses may also become apparent
from other measures, such as partitioned branch support values (Baker et al., 1998; Gatesy and Arctander,
2000; Damgaard and Sperling, 2001; Murrell et al.,
2001; Gatesy, 2002; Meier and Baker, 2002; Remsen and
O’Grady, 2002; Gatesy et al., 2003; Wahlberg and Nylin,
2003).
Even though morphological data sets may contain relatively small numbers of characters in comparison to
the total amount of included molecular data, morphology may contribute more support and stability to the
combined analysis when considered per character (Baker
et al., 1998; Gatesy, 2002; Meier and Baker, 2002). For example, Baker et al. (1998) and Meier and Baker (2002)
found that morphology may be up to twice as informative as molecules per included character in a combined
analysis across a wide range of different taxa.
Morphology also may contribute to combined data
analyses in the form of hidden clade support that becomes apparent only when data sets are combined
(Gatesy and Arctander, 2000; Damgaard and Sperling,
2001; Murrell et al., 2001; Gatesy, 2002; Gatesy et al.,
2003). Thus, morphology may be partially incongruent with molecular data when analyzed separately, but
when combined the total evidence may generate a wellsupported phylogeny. For the same reason combination
of apparently conflicting molecular data sets may lead to
a single well-supported phylogeny (Rokas et al., 2003).
In some cases, morphology and molecules supply
complementary evidence, where each type of data may
contribute to different parts of the phylogeny (Giribet
and Wheeler, 2002). A strength of morphology is that
characters informative at many different levels in the
phylogeny are habitually included in a single data set,
whereas the average substitution rate of a given gene
may limit its utility to particular levels in a phylogeny.
One or a few genes may not provide sufficient phylogenetic signal across all levels in a phylogeny, especially
when the included taxa span a broad range of divergence dates. In other cases, morphology may contribute
unique phylogenetic signal where molecules are completely mute (Mattern and McLennan, 2000). In some
combined analyses, morphology may increase support
of certain nodes while lowering that of others ( Klompen
et al., 2000; Stach and Turbeville, 2002), and in other cases
morphology may not add any significant support to combined data analyses (Sorhannus, 2001). Although conflict
between morphological and molecular evidence may occur and some workers remain skeptical about combining
morphological and molecular data into a single phylogenetic analysis (SEA did not object to combined analysis
of morphological and molecular data; 2003:545), these
VOL. 53
examples clearly show that morphological data may improve clade support values.
HOMOLOGY ASSESSMENT AND CHARACTER CODING
SEA nominated the limited number of unambiguous
characters as the main shortcoming of morphological
phylogenetic data. To make matters even worse, SEA
indicated that any attempt to solve this problem by increasing the number of morphological characters will be
doomed to failure. First, SEA (2003:545) concluded that
“much of the useful morphological diversity has already
been scrutinized.” Although this statement may be true
for well-studied taxa such as the seed plants discussed
by SEA, there is no indication that it holds true for other
poorly studied taxa. The widespread use of new morphological data in phylogenetic analyses of many different taxa leaves no doubt that much phenotypic variation
remains to be explored (Lee, 1995; Klass, 2001; Hooge
et al., 2002; Giribet et al., 2002; Sørensen, 2002; Strong,
2003).
Second, SEA (2003:541) argued that the need to properly code and define morphological characters will only
add “a level of subjectivity and interpretation” to the
analysis, which merely “increases the level of ambiguous
or problematic characters.” SEA (2003:542) concluded:
“For morphological studies comprising a relatively high
number of characters, both character coding and character conceptualization become increasingly important
variables that may have a negative impact on a study
as more characters are added (Fig. 1).” SEA buttressed
their conclusions with figures 1c and 1d, which show that
ambiguity of both homology assessment and character
coding will quickly increase with the addition of more
characters.
These arguments erect nothing but a straw man. SEA
fabricated these graphs without marshalling any evidence to support them. There is nothing to support SEA’s
contention that increasing the number of characters in a
morphological data set will necessarily lead to increasingly ambiguous homology determination and character coding. The only example SEA (2003:542) presented
where an increase in the number of morphological characters “made no significant difference to the results”
scarcely provides the basis for these generalizations applicable across the tree of life. In contrast, I expect an increase in the quality of morphological phylogenetic data
with time as the development of new analytical techniques opens up unexplored sources of data and facilitates the validation of older information. For example,
in metazoan phylogenetics powerful techniques such as
transmission electron microscopy, molecular developmental biology, and cell lineage tracing are contributing to an improvement of the morphological data sets as
new data are added and old mistakes are corrected (e.g.,
Giribet, 2002; Jenner, 2004).
The main problem of SEA’s interpretation here is that
they assumed that ease of homology assessment and
character coding can be regarded as proxies of a single,
2004
POINTS OF VIEW
general phylogenetic signal in a data set and are as such
related as single values to the number of characters in
an analysis. However, data sets do not contain a single
phylogenetic signal. The phylogenetic signal in a data set
is hierarchically structured, and different characters are
significant on different phylogenetic levels. Certain characters are therefore relevant for placing certain taxa but
are irrelevant for placing others. Consequently, there is
no simple relationship between an increase in the number of morphological characters and the ambiguity of
character coding and homology assessment that can be
captured in a graph, such as figures 1c and 1d, and that
is valid throughout the whole phylogeny.
SEA’s arguments about defining and coding characters
are reflective of a deplorable appreciation of the morphological scholarship that is at the heart of phylogenetics.
This attitude mirrors a worrying statement in the chapter on animal phylogeny in the recently published new
edition of the widely used textbook on invertebrate zoology by Brusca and Brusca (2003:874): “The process of a
priori character assessment is perhaps the weakest link
in morphological phylogenetic biology.” However, character assessment is the only empirical anchor of phylogenetics. If properly conducted, comparative morphological study will be the greatest strength of a phylogenetic
analysis, and when performed less than critically it may
be the greatest weakness. The objectivity of morphological phylogenetic analyses lies in intersubjective testing,
in which previously compiled data are reanalyzed and
reevaluated for their congruence with other characters
(Kluge, 1997, 1998; Ax, 1999). SEA’s recommendations
of greater explicitness in character selection and care in
morphological studies will naturally contribute to the
quality of morphological phylogenetic analyses.
With respect to character coding and character conceptualization in molecular phylogenetics, SEA (2003:541)
concluded that for “aligned sequence data, there is no
ambiguity in assessing character states,” and “there is no
ambiguity that the unit of comparison is the nucleotide.”
In contrast, the definition of homologous characters and
the coding of character states in morphological phylogenetics is much more problematic. However, the important issue is not only whether two nucleotides at the
same position are identical but also whether they are
historically identical, i.e. homologous. Nucleotides are
characters of relatively low complexity, and the character state space for nucleotides is much more restricted
than that for morphology. In certain circumstances, this
restriction creates a considerable danger that the same
nucleotide has evolved independently in the same position. This realization has been an incentive to develop
models of evolution that estimate the probability that
the same nucleotides at a site are historically identical,
to explore the value of more complex molecular characters (Swofford et al., 1996; Rokas and Holland, 2000),
and to develop alignment methods that use comparative
molecular anatomical information (secondary and tertiary structure) to improve the probability of recovering
homology (Kjer, 1995; Hickson et al., 2000). In contrast,
337
morphology generally presents a richer space of more
complex characters that allows a more fine-grained comparison of potential homology, which may help explain
why in certain cases morphology may be qualitatively
superior to molecules when considered per character
(Baker et al., 1998; Gatesy, 2002; Meier and Baker, 2002).
The definition and coding of morphological characters is certainly a difficult subject, and for a given set
of taxa alternative approaches to homology assessment
and character coding may indeed lead to different phylogenies. However, as the articles cited by SEA indicate,
there is no easy solution to this problem, and subjective choices are unavoidable. One way to deal with this
issue is to perform a sensitivity analysis with different
character codings to see how the phylogeny changes
with changing character codings (Jenner, 2002; Rieppel
and Kearney, 2002; Simmons and Geisler, 2002). This approach at least allows a better understanding of the relationship between assumptions and hypothesis.
Most important in the context of this critique, however, is that despite the admitted difficulties of morphological character analysis, there is no evidence to suggest
that morphology generally performs more poorly than
molecules in estimating phylogeny (Hillis and Wiens,
2000; Baker and Gatesy, 2002). SEA (2003:543) cited an
example where molecular evidence from three genes is
taken to suggest that phylogenetic analyses of morphology and fossils have led to inaccurate phylogenetic hypotheses of the angiosperms. Although this statement
may be true, we cannot automatically conclude that morphology always suggests the wrong answer whenever
molecules and morphology conflict. In many cases different molecular analyses may conflict with each other
(Sota and Vogler, 2001; Taggart et al., 2001; Rydin et al.,
2002; Shaw, 2002; Machado and Hey, 2003; Rokas et al.,
2003), and morphology may be more reliable than some
molecular data partitions (Wiens and Hollingsworth,
2000; Damgaard and Sperling, 2001). The findings of
Rokas et al. (2003) potently illustrate that data from one
or a small number of different genes may lead to a robustly supported phylogeny that may be in direct conflict with an alternative phylogeny based on another set
of genes, even for relatively closely related species. In
the context of such limited amounts of data we simply
cannot know whether we have accurately reconstructed
the true phylogeny. We can only sample character space
as comprehensively as possible and find the most wellcorroborated hypothesis.
SEA (2003:541) claimed that these problems of “subjectivity and interpretation” are absent from molecular
data, because “areas of ambiguity [in sequence alignment] can be excluded.” As recent research has shown,
to choose this way of least resistance may be thoroughly
misleading, and this short statement seriously underplays the degree of subjectivity and interpretation associated with molecular phylogenetics. Apart from the
choice of included taxa, “subjectivity and interpretation”
in molecular phylogenetics may reside in the choice of
the gene, the decisions of how to determine homology
338
SYSTEMATIC BIOLOGY
when insertion and deletion events or exon shuffling occur, the choice of phylogenetic analysis parameters, the
choice of whether nucleotides or amino acids are analyzed for protein-coding genes, and the choice of the optimality criterion. Different decisions for these variables
may result in different phylogenies (Giribet and Wheeler,
2002; Giribet et al., 2002; Simmons et al., 2002a, 2002b;
Rokas et al., 2003). Horizontal gene transfer and gene
duplication may add further difficulties to homology assessment in molecular phylogenetic analyses. Moreover,
SEA’s suggestion to simply exclude all information associated with ambiguously aligned sequences may also
cause misleading results, as illustrated by the following
example.
For more than a decade, 18S ribosomal DNA (rDNA)
sequences suggested that birds were most closely related
to mammals (hematotherm hypothesis). This hypothesis
was in conflict with results derived from a large amount
of traditional (morphological and paleontological) and
other molecular data, which instead united birds with
crocodilians. After different workers analyzed the 18S
data in various different ways, they concluded that this
was an example of different molecules giving significantly different estimates of phylogeny. However, in a
recent study Xia et al. (2003) convincingly showed that
the conflict between 18S data and the traditional and
other molecular data was an artifact attributable to two
main factors: misalignment of sequences and inappropriate estimation of base frequency parameters. Crucial to
the resolution of this paradox was the incorporation of
information from the most variable regions of the 18S
molecule that were most difficult to align unambiguously. This study clearly showed that restricting the data
set to only the least unambiguous sites might produce
a thoroughly misleading phylogeny. The problem that
“different workers will perceive and define characters in
different ways” (2003:541) is therefore certainly not limited to morphological data. The recent development of
methods to deal with ambiguously aligned regions or
to circumvent multiple sequence alignment altogether
(for review, see Lee, 2001) further undermines the fundamental dichotomy between the value and treatment of
molecular and morphological data as portrayed by SEA.
VALUE OF M ORPHOLOGY IN PHYLOGENETICS
SEA (2003:544) stated that “a continued role for morphology in phylogeny reconstruction seems a reasonable
expectation,” and they envisioned the role of morphology in systematics in one of three ways. In the first form,
which is the one strongly recommended by SEA, fewer
morphological features are studied in depth and are then
mapped onto a molecular phylogeny “on the basis that
morphological characters can be diagnostic for nodes
on molecular trees” (2003:545). Each morphological character is studied separately with respect to a node in the
molecular phylogeny, and characters found to be incongruent “are not incorporated into the phylogenetic hypothesis” (2003:545). SEA (2003:545) claimed that this approach “is akin to Patterson’s (1982) congruence test.”
VOL. 53
If one adheres to cladistics as a falsificationist research
program, the problem here is the claim of congruence.
Patterson (1982) was clearly concerned about the congruence of characters against all others. Standard cladistic
analysis operates on the basis of character congruence,
in which the congruence of all characters is assessed simultaneously. Only this approach will guarantee finding
the globally most-parsimonious cladogram that embodies maximal explanatory power. By comparing individual morphological characters against an already resolved
molecular phylogeny, one is not assessing overall congruence of characters, despite SEA’s confusing invocation of Patterson’s congruence test. If 99 of 100 characters were incongruent with the molecular phylogeny but
entirely congruent among themselves, SEA’s approach
would nevertheless disregard them as homoplasies. This
approach is more akin to clique analysis, the logical basis of which for phylogeny reconstruction was long ago
questioned (Farris, 1983). This approach purposely removes any testing power from the phylogenetic analysis, which is in direct conflict with the falsificationist
premises of cladistics.
Even if morphological characters were not directly
used to reconstruct a phylogeny but, as suggested by
SEA, were merely mapped onto a molecular phylogeny
to “provide evidence for a more limited number of monophyletic taxa” (2003:545), we would still have to confront exactly the same problems of character definition
and character coding that SEA nominated as the greatest weaknesses of morphological phylogenetic studies.
Before characters can be mapped onto a molecular phylogeny as a limited number of additional synapomorphies, they must be defined. Just as decisions about character coding determine on what level in the phylogeny a
character is informative, so too is the mapping of a character in a certain place in a phylogeny inextricably linked
with particular assumptions about character coding. For
example, SEA (2003:541) cited the famous “no tails, red
tails, blue tails” example to illustrate that even for relatively simple characters it may be difficult to choose
an appropriate coding method. For these data, multistate coding does not capture the variation of a tail being present versus absent (see discussion by Hawkins
et al., 1997). Consequently, if we map the presence of
tails as a synapomorphy on a particular level in a molecular phylogeny, we necessarily imply that the observed
variation is not conceptualized as a multistate character.
Thus, even though decisions about character coding are
not explicitly made in a data matrix, implicit decisions
cannot be avoided.
SEA recommended the study of fewer rather than
more characters because rigorous character study is too
time consuming. Today, the high pace of publication in
the field of phylogenetics is principally set by molecular
studies, where automated methods allow much faster
data collection than can be achieved when studying
comparative morphology. Unfortunately, this publication pressure may in some cases lead to the publication
of phylogenetic studies based on hastily and uncritically
compiled morphological data sets of low quality (Jenner,
2004
POINTS OF VIEW
2001, in press). However, we want to understand as
much as we can about the evolution of morphology, and
molecules, including features for which homology assessment is difficult. That difficult science can be time
consuming can hardly be a scientific argument for studying fewer rather than more characters.
SEA’s second and third proposed strategies for incorporating morphology into phylogenetics are actually the
same. As a second approach, SEA advocated a continuation of current practice. That is, morphological and
molecular data both have their value, and they are analyzed both separately and in combination to study evolution. The third strategy is similar except only “those
characters that are unproblematic in terms of homology
assessment and character coding are selected” (2003:545)
and analyzed together with the molecular data. These
two approaches actually provide the same possibilities as
the first recommended approach, and as discussed with
respect to combined analyses, more in addition. SEA realized that careful morphological study is valuable, and
I could not agree more.
However, SEA’s recommendation to select only unambiguously homologous morphological characters is
problematic. All primary homology decisions and character state assignments are provisional and can in principle be corroborated and refuted. By concluding that the
definition and coding of even simple characters can be
problematic (the tails example), SEA themselves show
that prior to a phylogenetic analysis there are no “unambiguously coded morphological characters” (2003:541).
Workers who attempted to separate problematic from
supposedly more reliable characters prior to a phylogenetic analysis confirm this. Their results show that the
phylogenetic utility of characters is very difficult to predict a priori (Jenner and Schram, 1999: Table 3; Schander
and Sundberg, 2001; Collin, 2003; Gatesy et al., 2003;
Strong, 2003).
By concluding (2003:543) that morphology cannot offer “any hope for the future to resolve phylogeny at lower
or higher taxonomic levels,” SEA gave very little consideration to fossils. Although fossils were briefly discussed in connection with taxon sampling, SEA did not
reach any firm conclusion about their value for phylogenetics other than that their contribution to phylogenetic accuracy is uncertain in view of the problems of
homology assessment and character coding. However,
the limited amount of molecular and morphological evidence typically available to us does not allow us to determine whether the true phylogeny has been accurately
reconstructed in the first place. As an additional problem of including fossils in phylogenetic analyses, SEA
(2003:543) cited the “large amounts of missing data.” In
contrast, Wiens (2003:536) concluded on the basis of simulation studies that “the proportion of missing data cells
in the incomplete taxa is a poor indicator of their impact
on phylogenetic accuracy.” Accuracy is more related to
the distribution of missing data in the matrix and the
number of characters that can be scored for the incomplete taxa. As Baker and Gatesy (2002:171) concluded;
“If morphological evidence is ignored, the phylogeny of
339
over 99% of life is ignored.” In certain groups, such as
the certartiodactyls, >85% of the genera are known only
from fossils (Gatesy, 2002). The importance of fossils for
reconstructing phylogeny can perhaps best be illustrated
by the importance of stem groups ( Wills et al., 1998; Budd
and Jensen, 2000; Budd, 2002; Holmer et al., 2002; Lee and
Scanlon, 2002; Mallatt and Chen, 2003; Ruta et al., 2003).
Stem groups exemplify the most basal divergences of extant taxa, and their value in bridging the often considerable morphological gaps between disparate body plans
is of central importance to reconstructing the full diversity of the tree of life. With the exception of ancient DNA,
rigorous morphological phylogenetic analysis is the only
way in which fossils can be incorporated into the phylogeny of life. Failure to do so will leave phylogenies of
extant organisms forever uprooted.
SEA (2003:545) admitted the value of reciprocal illumination (Hennig, 1966) in which independent data can
shed unique light on a problem. However, they apparently fail to realize that by restricting morphological phylogenetic analyses, the power of reciprocal illumination
is crippled. Phylogenetic analyses of different data sets
provide valuable and unique independent perspectives
on both phylogenetic questions and questions of character evolution. The existence of a dialogue between morphological and molecular hypotheses guarantees continued attention and excitement in phylogenetic research
on many taxa on many different phylogenetic levels. For
example, a recent phylogenetic analysis of 18S rDNA sequences in asellote isopods failed to yield results strongly
expected on the basis of careful morphological study
(Wägele et al., 2003). This discordance stimulated a more
detailed look at the molecular data, yielding several interesting insights that may otherwise have remained
invisible. Similarly, Vidal and Hedges (2002) recently analyzed snake phylogeny with nuclear and mitochondrial
genes. They found that the morphologically strongly
supported clade of macrostomatans (Lee and Scanlon,
2002) is contradicted by molecular evidence. Because
some workers have claimed that most apparent conflict
between molecular and morphological phylogenies is
due to weak support for either or both of the estimates
and that strongly supported and misleading morphological phylogenies are relatively rare (Hillis and Wiens,
2000; Wiens et al., 2003), such a potential case for strong
conflict is worth exploring. It allows us to explore the
possibility that in this case morphological phylogenetic
analyses may suffer from the confounding effect of concerted homoplasy, possibly in the characters related to
wide gape size. This reassessment may teach us something important about the extent of parallel evolution in a
complex of adaptive features. These insights into evolution are dependent upon the compilation, analysis, and
comparison of independently compiled molecular and
morphological data sets.
Science derives its greatest strength from its multifaceted nature. Artificial restriction of approaches
can lead only to impoverishment of science. Many
researchers see the merits of analyzing molecular and
morphological data sets separately and in combination to
340
SYSTEMATIC BIOLOGY
compare the results (Littlewood et al., 1999; Edgecombe
et al., 2000; Flores-Villela et al., 2000; Janies, 2001; Kress
et al., 2001; Wheeler et al., 2001; Winterton et al., 2001;
Edgecombe and Giribet, 2002; Giribet and Wheeler, 2002;
Giribet et al., 2002; Stach and Turbeville, 2002; Jeffery
et al., 2003; Schulmeister, 2003; Wahlberg and Nylin,
2003). As SEA admitted, study of morphological characters is necessary anyway because one goal of phylogenetic analysis is to understand character evolution. We
might as well submit the collected data to as many potentially fruitful uses as possible.
By discouraging morphological phylogenetics, we
take away a main formal impetus for morphological
comparative research. Combined with the continuing
marginalization of organismic biology in academic curricula (Fautin and Watling, 1999), this attitude introduces
the real danger that potentially interested students will
not be attracted to systematic biology, and this dearth of
students may seriously undermine the future and versatility of our science for largely nonscientific reasons.
SEA failed to present any convincing scientific arguments for narrowing the focus of phylogenetics to mainly
molecular research. In writing that morphology “may
not be able to resolve the full branch structure of the
tree of life,” SEA (2003:543) merely presented an empty
statement, because no single data set, molecular or morphological, can reasonably be expected to be the Holy
Grail of phylogenetics, as they themselves underscore by
stating (2003:543) that even whole genomes “will probably not yield a fully resolved, highly confident tree.”
The only scientifically valid reason that I can think of for
excluding or severely limiting the use of morphological
evidence for phylogeny reconstruction as proposed by
SEA is when we know a priori that these data will be
positively misleading. SEA failed to show that we have
this knowledge. Instead, they presented a few examples
from botanical phylogenetics where the contribution of
morphological data to resolving the phylogeny was limited. However, as shown here, these examples scarcely
provide enough foundation to question the general value
of morphology for other parts of the tree of life. In the
absence of evidence to the contrary, I am convinced that
an equal partnership between molecular and morphological phylogenetics is our best bet for future progress.
ACKNOWLEDGMENTS
I gratefully acknowledge helpful comments on the manuscript by
Tim Collins, Chris Simon, Gavin Naylor, Richard Olmstead, Jonathan
Bennett, and an anonymous reviewer. A Marie Curie Individual Fellowship of the European Community program Improving Human
Potential under contract number HPMF-CT-2002-01712 supports my
work.
R EFERENCES
Ax, P. 1999. Das System der Metazoa II. Ein Lehrbuch der phylogenetischen Systematik. Gustav Fischer Verlag, Stuttgart, Germany.
Baker, R. H., and J. Gatesy. 2002. Is morphology still relevant? Pages
163–174 in Molecular systematics and evolution: Theory and practice
(R. DeSalle, G. Giribet, and W. Wheeler, eds.). Birkhäuser Verlag,
Basel, Switzerland.
VOL.
53
Baker, R. H., G. S. Wilkinson, and R. DeDalle. 2001. Phylogenetic utility
of different types of molecular data used to infer evolutionary relationships among stalk-eyed flies (Diopsidae). Syst. Biol. 50:87–105.
Baker, R. H., X. B. Yu, and R. DeSalle. 1998. Assessing the relative contribution of molecular and morphological characters in simultaneous
analysis trees. Mol. Phylogenet. Evol. 9:427–436.
Bremer, B., R. K. Jansen, B. Oxelman, M. Backland, H. Lantz, and K.-J.
Kim. 1999. More characters or more taxa for a robust phylogeny—
Case study from the coffee family (Rubiaceae). Syst. Biol. 48:413–435.
Brusca, R. C., and G. J. Brusca. 2003. Invertebrates, 2nd edition. Sinauer,
Sunderland, Massachusetts.
Budd, G. E. 2002. A palaeontological solution to the arthropod head
problem. Nature 417:271–275.
Budd, G. E., and S. Jensen. 2000. A critical reappraisal of the fossil record
of the bilaterian phyla. Biol. Rev. 75:253–295.
Colgan, D. J., A. McLauchlan, G. D. F. Wilson, S. P. Livingston, G. D.
Edgecombe, J. Macaranas, G. Cassis, and M. R. Gray. 1998. Histone
H3 and U2 snRNA DNA sequences and arthropod molecular evolution. Aust J. Zool. 46:419–437.
Collin, R. 2003. The utility of morphological characters in gastropod
phylogenetics: An example from the Calyptraeidae. Biol. J. Linn. Soc.
78:541–593.
Damgaard, J., and F. A. H. Sperling. 2001. Phylogeny of the water strider genus Gerris Fabricius (Heteroptera: Gerridae) based on
COI mtDNA, EF-1α nuclear DNA and morphology. Syst. Entomol.
26:241–254.
Edgecombe, G. D., and G. Giribet. 2002. Myriapod phylogeny and
the relationships of Chilopoda. Pages 143–168 in Biodiversidad,
taxonomı́a y biogeografı́a de artrópodos de México: Hacia una
sı́ntesis de su conocimiento (J. Llorente Bousquets, J. J. Morrone,
and H. Ponce Ulloa, eds.). Prensas de Ciencias, UNAM, Mexico City,
Mexico.
Edgecombe, G. D., G. Giribet, and W. C. Wheeler. 2002. Phylogeny of
Henicopidae (Chilopoda: Lithobiomorpha): A combined analysis of
morphology and five molecular loci. Syst. Entomol. 27:31–64.
Edgecombe, G. D., G. D. F. Wilson, D. J. Colgan, M. R. Gray, and G.
Cassis. 2000. Arthropod cladistics: Combined analysis of histone H3
and U2 snRNA sequences and morphology. Cladistics 16:155–203.
Farris, J. S. 1983. The logical basis of phylogenetic analysis. Pages 7–36
in Advances in cladistics, Volume 2 (N. I. Platnick and V. A. Funk,
eds.). Columbia Univ. Press, New York.
Farris, J. S. 1999. Likelihood and inconsistency. Cladistics 15:199–204.
Fautin, D. G., and L. Watling. 1999. Review of textbooks of invertebrate
zoology. Am. Zool. 39:818–824.
Flores-Villela, O., K. M. Kjer, M. Benabib, and J. W. Sites. 2000. Multiple data sets, congruence, and hypothesis testing for the phylogeny
of basal groups of the lizard genus Sceloporus (Squamata, Phrynosomatidae). Syst. Biol. 49:713–739.
Gatesy, J. 2002. Relative quality of different systematic datasets for
certartiodactyl mammals: Assessments within a combined analysis framework. Pages 45–67 in Molecular systematics and evolution:
Theory and practice (R. DeSalle, G. Giribet, and W. Wheeler, eds.).
Birkhäuser Verlag, Basel, Switzerland.
Gatesy, J., G. Amato, M. Norell, R. DeSalle, and C. Hayashi. 2003. Combined support for wholesale taxic atavism in gavialine crocodylians.
Syst. Biol. 52:403–422.
Gatesy, J., and P. Arctander. 2000. Hidden morphological support for
the phylogenetic placement of Pseudoryx nghetinhensis with bovine
bovids: A combined analysis of gross anatomical evidence and DNA
sequences from five genes. Syst. Biol. 49:515–538.
Giribet, G. 2002. Current advances in the phylogenetic reconstruction of
metazoan evolution. A new paradigm for the Cambrian explosion?
Mol. Phylogenet. Evol. 24:345–357.
Giribet, G., G. D. Edgecombe, and W. C. Wheeler. 2001. Arthropod
phylogeny based on eight molecular loci and morphology. Nature
413:157–161.
Giribet, G., G. D. Edgecombe, W. C. Wheeler, and C. Babbitt. 2002. Phylogeny and systematic position of Opiliones: A combined analysis of
chelicerate relationships using morphological and molecular data.
Cladistics 18:5–70.
Giribet, G., and W. Wheeler. 2002. On bivalve phylogeny: A high-level
analysis of the Bivalvia (Mollusca) based on combined morphology
and DNA sequence data. Invertebr. Biol. 121:271–324.
2004
POINTS OF VIEW
Givnish, T. J., and K. J. Sytsma. 1997. Consistency, characters, and the
likelihood of correct phylogenetic inference. Mol. Phylogenet. Evol.
7:320–330.
Gribaldo, S., and H. Philippe. 2002. Ancient phylogenetic relationships.
Theor. Popul. Biol. 61:391–408.
Hawkins, J. A., C. E. Hughes, and R. W. Scotland. 1997. Primary homology assessment, characters and character states. Cladistics 13:275–
283.
Hennig, W. 1966. Phylogenetic systematics. Univ. Illinois Press, Urbana.
Hickson, R. E., C. Simon, and S. W. Perrey. 2000. The performance of several multiple-sequence aligment program in relation to secondarystructure features for an rRNA sequence. Mol. Biol. Evol. 17:530–539.
Hillis, D. M. 1995. Approaches for assessing phylogenetic accuracy.
Syst. Biol. 44:3–16.
Hillis, D. M. 1996. Inferring complex phylogenies. Nature 383:130–131.
Hillis, D. M. 1998. Taxonomic sampling, phylogenetic accuracy, and
investigator bias. Syst. Biol. 47:3–8.
Hillis, D. M., J. P. Huelsenbeck, and C. W. Cunningham. 1994a. Application and accuracy of molecular phylogenies. Science 264:671–677.
Hillis, D. M., J. P. Huelsenbeck, and D. L. Swofford. 1994b. Hobgoblin
of phylogenetics? Nature 369:363–364.
Hillis, D. M., D. D. Pollock, J. A. McGuire, and D. J. Zwickl. 2003. Is
sparse taxon sampling a problem for phylogenetic inference? Syst.
Biol. 52:124–126.
Hillis, D. M., and J. J. Wiens. 2000. Molecules versus morphology in
systematics: Conflicts, artifacts, and misconceptions. Pages 1–19 in
Phylogenetic analysis of morphological data (J. J. Wiens, ed.). Smithsonian Institution Press, Washington, D.C.
Holmer, L. E., C. B. Skovsted, and A. Williams. 2002. A stem group brachiopod from the Lower Cambrian: Support for a Micrina (halkieriid)
ancestry. Palaeontology 45:875–882.
Hooge, M. D., P. A. Haye, S. Tyler, M. K. Litvaitis, and I. Kornfield.
2002. Molecular systematics of the Acoela (Acoelomorpha, Platyhelminthes) and its concordance with morphology. Mol. Phylogenet.
Evol. 24:333–342.
Janies, D. 2001. Phylogenetic relationships of extant echinoderm
classes. Can. J. Zool. 79:1232–1250.
Jeffery, C. H., R. B. Emlet, and D. T. J. Littlewood. 2003. Phylogeny and
evolution of developmental mode in temnopleurid echinoids. Mol.
Phylogenet. Evol. 28:99–118.
Jenner, R. A. 2001. Bilaterian phylogeny and uncritical recycling of
morphological data sets. Syst. Biol. 50:730–742.
Jenner, R. A. 2002. Boolean logic and character state identity: Pitfalls of
character coding in metazoan cladistics. Contrib. Zool. 71:67–91.
Jenner, R. A. In press. The scientific status of metazoan cladistics: Why
current research practice must change. Zool. Scr.
Jenner, R. A. 2004. Towards a phylogeny of the Metazoa: Evaluating alternative phylogenetic positions of Platyhelminthes, Nemertea, and
Gnathostomulida, with a critical reappraisal of cladistic characters.
Contrib. Zool. 73:3–163.
Jenner, R. A., and F. R. Schram. 1999. The grand game of metazoan
phylogeny: Rules and strategies. Biol. Rev. 74:121–142.
Kjer, K. M. 1995. Use of rRNA secondary structure in phylogenetic
studies to identify homologous positions: An example of alignment
and data presentation from the frogs. Mol. Phylogenet. Evol. 4:314–
330.
Klass, K. D. 2001. Morphological evidence on blattarian phylogeny:
”Phylogenetic histories and stories” (Insecta, Dictyoptera). Dtsch.
Entomol. Z. 48:223–265.
Klompen, J. S. H., W. C. Black IV, J. E. Keirans, and D. E. Norris. 2000.
Systematics and biogeography of hard ticks, a total evidence approach. Cladistics 16:79–102.
Kluge, A. G. 1997. Testability and the refutation and corroboration of
cladistic hypotheses. Cladistics 13:81–96.
Kluge, A. G. 1998. Sophisticated falsification and research cycles: Consequences for differential character weighting in phylogenetic systematics. Zool. Scr. 26:349–360.
Kluge, A. G. 2002. Distinguishing ”or” from ”and” and the case for
historical identification. Cladistics 18:585–593.
Kress, W. J., L. M. Prince, W. J. Hahn, and E. A. Zimmer. 2001. Unraveling the evolutionary radiation of the families of the Zingiberales
using morphological and molecular evidence. Syst. Biol. 50:926–
944.
341
Lee, M. S. Y. 1995. Historical burden in systematics and the interrelationships of ”parareptiles.” Biol. Rev. 70:459–547.
Lee, M. S. Y. 2001. Unalignable sequences and molecular evolution.
Trends Ecol. Evol. 16:681–685.
Lee, M. S. Y., and J. D. Scanlon. 2002. Snake phylogeny based on osteology, soft anatomy and ecology. Biol. Rev. 77:333–401.
Lewis, P. O. 2001. A likelihood approach to inferring phylogeny from
discrete morphological characters. Syst. Biol. 50:913–925.
Littlewood, D. T. J., K. Rohde, and K. A. Clough. 1999. The interrelationships of all major groups of Platyhelminthes: Phylogenetic evidence
from morphology and molecules. Biol. J. Linn. Soc. 66:75–114.
Lockhart, P. J., D. Huson, U. Maier, M. J. Fraunholz, Y. Van de Peer,
A. C. Barbrook, C. J. Howe, and M. A. Steel. 2000. How molecules
evolve in Eubacteria. Mol. Biol. Evol. 17:835–838.
Lopez, P., D. Casane, and H. Philippe. 2002. Heterotachy, an important
process of protein evolution. Mol. Biol. Evol. 19:1–7.
Machado, C. A., and J. Hey. 2003. The causes of phylogenetic conflict
in a classic Drosophila species group. Proc. R. Soc. Lond. B 270:1193–
1202.
Maddison, W. P. 1997. Gene trees in species trees. Syst. Biol. 46:523–
536.
Mallatt, J., and J. Y. Chen. 2003. Fossil sister group of craniates: Predicted
and found. J. Morphol. 258:1–31.
Mattern, M. Y., and D. A. McLennan. 2000. Phylogeny and speciation
of felids. Cladistics 16:232–253.
Meier, R., and R. Baker. 2002. A cladistic analysis of Diopsidae (Diptera)
based on morphological and DNA sequence data. Insect Syst. Evol.
33:325–336.
Miyamoto, M. M., and W. M. Fitch. 1995. Testing the covarion hypothesis of molecular evolution. Mol. Biol. Evol. 12:503–513.
Murrell, A., N. J. H. Campell, and S. C. Barker. 2001. A total-evidence
phylogeny of ticks provides insights into the evolution of life cycles
and biogeography. Mol. Phylogenet. Evol. 21:244–258.
Naylor, G. J. P., and W. M. Brown. 1998. Amphioxus mitochondrial DNA,
chordate phylogeny, and the limits of inference based on comparisons of sequences. Syst. Biol. 47:61–76.
Patterson, C. 1982. Morphological characters and homology. Pages 21–
74 in Problems of phylogenetic reconstruction (K. A. Joysey and A.
E. Friday, eds.). Academic Press, London.
Pennisi, E. 2003. Modernizing the tree of life. Science 300:1692–1697.
Peterson, K. J., and D. J. Eernisse. 2001. Animal phylogeny and the
ancestry of bilaterians: Inferences from morphology and 18S rDNA
gene sequences. Evol. Dev. 3:170–205.
Remsen, J., and P. O’Grady. 2002. Phylogeny of Drosophilinae (Diptera:
Drosophilidae), with comments on combined analysis and character
support. Mol. Phylogenet. Evol. 24:249–264.
Rieppel, O., and M. Kearney. 2002. Similarity. Biol. J. Linn. Soc. 75:59–82.
Rokas, A., and P. W. H. Holland. 2000. Rare genomic changes as a tool
for phylogenetics. Trends Ecol. Evol. 15:454–459.
Rokas, A., B. L. Williams, N. King, and S. B. Carroll. 2003. Genome-scale
approaches to resolving incongruence in molecular phylogenies. Nature 425:798–804.
Rosenberg, M. S., and S. Kumar. 2003. Taxon sampling, bioinformatics,
and phylogenomics. Syst. Biol. 52:119–124.
Ruta, M., M. I. Coates, and D. L. J. Quicke. 2003. Early tetrapod relationships revisited. Biol. Rev. 78:251–345.
Rydin, C., M. Källersjö, and E. M. Friis. 2002. Seed plant relationships
and the systematic position of Gnetales based on nulcear and chloroplast DNA: Conflicting data, rooting problems, and the monophyly
of conifers. Int. J. Plant Sci. 163:197–214.
Sanderson, M. J. 1995. Objections to bootstrapping phylogenies: A critique. Syst. Biol. 44:299–320.
Schander, C., and P. Sundberg. 2001. Useful characters in gastropod
phylogeny: Soft information or hard facts? Syst. Biol. 50:136–141.
Schulmeister, S. 2003. Simultaneous analysis of basal Hymenoptera
(Insecta): Introducing robuts-choice sensitivity analysis. Biol. J. Linn.
Soc. 79:245–275.
Scotland, R. W., R. G. Olmstead, and J. R. Bennett. 2003. Phylogeny
reconstruction: The role of morphology. Syst. Biol. 52:539–548.
Shaw, K. L. 2002. Conflict between nuclear and mitochondrial DNA
phylogenies of a recent species radiation: What mtDNA reveals and
conceals about modes of speciation in Hawaiian crickets. Proc. Natl.
Acad. Sci. USA 99:16122–16127.
342
SYSTEMATIC BIOLOGY
Siddall, M. E. 1998. Success of parsimony in the four-taxon case: Longbranch repulsion by likelihood in the Farris zone. Cladistics 14:209–
220.
Siddall, M. E. 2002. Measures of support. Pages 80–101 in Techniques
in molecular systematics and evolution (R. DeSalle, G. Giribet, and
W. Wheeler, eds.). Birkhäuser Verlag, Basel, Switzerland.
Siddall, M. E., and A. G. Kluge. 1997. Probabilism and phylogenetic
inference. Cladistics 13:313–336.
Simmons, M. P., H. Ochoterena, and J. V. Freudenstein. 2002a. Amino
acid vs. nucleotide characters: Challenging preconceived notions.
Mol. Phylogenet. Evol. 24:78–90.
Simmons, M. P., H. Ochoterena, and J. V. Freudenstein. 2002b. Conflict
between amino acid and nucleotide characters. Cladistics 18:200–206.
Simmons, N. B., and J. H. Geisler. 2002. Sensitivity analysis of different methods of coding taxonomic polymorphism: An example from
higher-level bat phylogeny. Cladistics 18:571–584.
Sørensen, M. V. 2002. On the evolution and morphology of the rotiferan
trophi, with a cladistic analysis of Rotifera. J. Zool. Syst. Evol. Res.
40:129–154.
Sorhannus, U. 2001. A ”total evidence” analysis of the phylogenetic
relationships among the photosynthetic stramenopiles. Cladistics
17:227–241.
Sota, T., and A. P. Vogler. 2001. Incongruence of mitochondrial and nuclear gene trees in the carabid beetles Ohomopterus. Syst. Biol. 50:39–
59.
Stach, T., and J. M. Turbeville. 2002. Phylogeny of the Tunicata inferred from molecular and morphological characters. Mol. Phylogenet. Evol. 25:408–428.
Strong, E. E. 2003. Refining molluscan characters: Morphology, character coding and a phylogeny of the Caenogastropoda. Zool. J. Linn.
Soc. 137:447–554.
Sullivan, J., and D. L. Swofford. 2001. Should we use model-based
methods for phylogenetic inference when we know that assumptions
about among-site rate variation and nucleotide substitution pattern
are violated? Syst. Biol. 50:723–729.
Swofford, D. L., G. J. Olsen, P. J. Waddell, and D. M. Hillis. 1996. Phylogenetic inference. Pages 407–514 in Molecular systematics, 2nd
edition (D. M. Hillis, C. Moritz, and B. K. Mable, eds.). Sinauer,
Sunderland, Massachusetts.
Swofford, D. L., P. J. Waddell, J. P. Huelsenbeck, P. G. Foster, P. O.
Lewis, and J. S. Rodgers. 2001. Bias in phylogenetic estimation and its
VOL. 53
relevance to the choice between parsimony and likelihood methods.
Syst. Biol. 50:525–539.
Taggart, T. W., B. I. Crother, and M. E. White. 2001. Palm-pitviper (Bothriechis) phylogeny, mtDNA, and consilience. Cladistics 17:355–370.
Tautz, D., P. Arctander, A. Minelli, R. H. Thomas, and A. P. Vogler. 2003.
A plea for DNA taxonomy. Trends Ecol. Evol. 18:70–74.
Vidal, N., and S. B. Hedges. 2002. Higher-level relationships of snakes
inferred from four nuclear and mitochondrial genes. C. R. Biol.
325:977–985.
Wägele, J.-W., B. Holland, H. Dreyer, and B. Hackethal. 2003. Searching
factors causing implausible non-monophyly: ssu rDNA phylogeny
of Isopoda Asellota (Crustacea: Peracarida) and faster evolution in
marine than in freshwater habitats. Mol. Phylogenet. Evol. 28:536–
551.
Wahlberg, N., and S. Nylin. 2003. Morphology versus molecules: Resolution of the positions of Nymphalis, Polygonia, and related genera
(Lepidoptera: Nymphalidae). Cladistics 19:213–223.
Wheeler, W. C., M. Whiting, Q. D. Wheeler, and J. M. Carpenter. 2001.
The phylogeny of the extant hexapod orders. Cladistics 17:113–169.
Wiens, J. J. 2003. Missing data, incomplete taxa, and phylogenetic accuracy. syst. Biol. 52:528–538.
Wiens, J. J., P. T. Chippindale, and D. M. Hillis. 2003. When are phylogenetic analyses misled by convergence? A case study in Texas cave
salamanders. Syst. Biol. 52:501–514.
Wiens, J. J., and B. D. Hollingsworth. 2000. War of the iguanas: Conflicting molecular and morphological phylogenies and long-branch
attraction in iguanid lizards. Syst. Biol. 49:143–159.
Wills, M. A., D. E. G. Briggs, R. A. Fortey, M. Wilkinson, and P. H. A.
Sneath. 1998. An arthropod phylogeny based on fossil and recent
taxa. Pages 33–105 in Arthropod fossils and phylogeny (G. D. Edgecombe, ed.). Columbia Univ. Press, New York.
Winterton, S. L., L. Yang, B. M. Wiegmann, and D. K. Yeates. 2001.
Phylogenetic revision of Agapophytinae subf.n. (Diptera: Therevidae) based on molecular and morphological evidence. Syst. Entomol.
26:173–211.
Xia, X., Z. Xie, and K. M. Kjer. 2003. 18S Ribosomal RNA and tetrapod
phylogeny. Syst. Biol. 52:283–295.
First submitted 31 July 2003; reviews returned 5 October 2003;
final acceptance 26 November 2003
Associate Editor: Tim Collins
Syst. Biol. 53(2):342–355, 2004
c Society of Systematic Biologists
Copyright ISSN: 1063-5157 print / 1076-836X online
DOI: 10.1080/10635150490423971
Inconsistencies in Arguments for the Supertree Approach: Supermatrices versus
Supertrees of Crocodylia
J OHN G ATESY,1 R ICHARD H. B AKER,2 AND CHERYL HAYASHI 1
1
2
Department of Biology, University of California, Riverside, California 92521, USA; E-mail: [email protected] (J.G.)
Evolutionary Genomics Department, DOE Joint Genome Institute, 2800 Mitchell Drive, Walnut Creek, California 94598, USA
Multiple data sets are now available for examining the
phylogenetic relationships of certain well-studied taxa.
Integrating this information into unified statements of
evolutionary history is a major challenge in the field of
systematics, with supertree and supermatrix methods
emerging as competing strategies for this task. In the supermatrix approach, all relevant character data are joined
into a single data set, and a total evidence analysis of this
large matrix is used to test competing phylogenetic hypotheses (Kluge, 1989). In supertree methods, characters
first are partitioned into separate data sets, these data
sets are analyzed individually, and source trees derived
from the preliminary analyses then are combined to yield
large synthetic trees (Sanderson et al., 1998).
Gatesy et al. (2002) highlighted some problems with
previously published supertrees (Purvis, 1995; BinindaEmonds et al., 1999; Jones et al., 2002) in the context
of a recent supertree analysis of mammals (Liu et al.,
2001): (1) the supertree data set of Liu et al. included
a variety of unnecessary appeals to authority, such as
2004
POINTS OF VIEW
assumptions of monophyly and poorly justified source
phylogenies; (2) this supertree data set also included extensive duplications of data; (3) the supertree method utilized by Liu et al., matrix representation with parsimony
(MRP; Baum, 1992; Ragan, 1992), ignores/misinterprets
hidden character support in different source data sets;
(4) these issues seem to explain much of the topological
incongruence between the trees presented by Liu et al.
and previous supermatrix results (e.g., Matthee et al.,
2001); (5) other published mammalian supertree analyses (Purvis, 1995; Bininda-Emonds et al., 1999; Jones
et al., 2002) also are characterized by these same problems; and (6) such supertrees are not efficient summaries of previous systematic work and should not be
the basis for critical tests of evolutionary hypotheses.
Many of the potential difficulties with supertree analysis listed above had been noted previously by proponents of the supertree method (e.g., Purvis, 1995;
Bininda-Emonds and Bryant, 1998; Sanderson et al.,
1998; Bininda-Emonds et al., 1999; Bininda-Emonds and
Sanderson, 2001; Jones et al., 2002; Pisani and Wilkinson,
2002), but these problems unfortunately characterize the
great majority of published supertrees, up to and including those in the most current studies (e.g., Kennedy
and Page, 2002; Pisani et al., 2002; Salamin et al.,
2002).
Bininda-Emonds et al. (2003) published a response
to Gatesy et al. (2002) that defended the supertree approach and outlined a series of corrections to the problems summarized above. Here, we comment on this
response within the context of a comprehensive supertree/supermatrix analysis of Crocodylia (alligators,
caiman, crocodiles, and gavials). We use an empirical
comparison of supertree and supermatrix approaches
for identical sets of characters and taxa to highlight basic differences between the two competing methods, and
suggest that the solutions presented by Bininda-Emonds
et al. (2003) do not correct for deficiencies in previous implementations of supertree procedures.
S UPERMATRICES AND S UPERTREES OF CROCODYLIA
Crocodylia, the relictual sister group of Aves, is composed of ∼22 extant species and their close fossil relatives. The majority of generic diversity is extinct, and
many extant crocodylian species are highly endangered (Brochu, 2003). A comprehensive phylogeny of
Crocodylia would provide a useful framework for estimation of conservation priorities and for synthesis of
comparative data in this group. Supermatrix analyses
of Crocodylia have been completed (Poe, 1996; Brochu,
1997; Brochu and Densmore, 2001; Gatesy et al., 2003),
but a supertree study has not been attempted. We used
published data and 41 new sequences from four nuclear
genes (∼2,000 base pairs [bp] of recombination activating protein-1 [RAG-1], ∼680 bp of ATP7A, ∼560 bp of
brain-derived neurotrophic factor [BDNF], and ∼635 bp
of lactate dehydrogenase-A [LDH-A]) to construct MRP
supertrees and more comprehensive supermatrix topologies for Crocodylia.
343
Methods
Crocodylian DNA samples were those utilized by
Gatesy et al. (2003). New data for exons from RAG-1,
ATP7A, and BDNF (see Fig. 2) were amplified and sequenced using the PCR methods and primers described
by Gatesy et al. (2003) and Murphy et al. (2001). New
data for intron 7 and exon 8 of LDH-A (see Fig. 2) were
amplified by PCR and sequenced using the primers
LAI7 F1 (5 -TGGCTGAAACTGTTATGAAGAACC-3 )
and LAI7 R1 (5 -TGGATTCCCCAAAGTGTATCTG-3 ).
The PCR conditions were 94◦ for 1 min, 53◦ for 1 min,
and 72◦ for 1 min for 50 cycles.
Data from the literature and the NCBI database and
new DNA sequences were organized into 17 characterbased data sets (Figs. 1–3). These included morphological traits, sections of five nuclear genes, five fragments
of mitochondrial DNA (mtDNA), two restriction fragment length polymorphism (RFLP) matrices, two allozyme data sets, chromosomal morphology, and nest
type (Cohen and Gans, 1970; Greer, 1970; Campbell, 1972;
Densmore, 1983; Desjardins and Morais, 1990; Densmore
and White, 1991; Gatesy and Amato, 1992; Brooks and
McLennan, 1993; Gatesy et al., 1993, 2003, Poe, 1996;
Brochu, 1997, 2002, 2003, in prep.; Harlid et al., 1997;
Janke and Arnason, 1997; Caccone et al., 1999; Groth and
Barrowclough, 1999; Mindell et al., 1999; Brochu and
Gingerich, 2000; Roman and Bowen, 2000; Janke et al.,
2001; White and Densmore, 2001; Ray and Densmore,
2002; Harshman et al., 2003; Wu et al., unpubl., Zhu
et al., unpubl.). The last two data sets included only
three traits and one trait, respectively. However, because
the chromosomal and nest characters did not fit logically into any of the other character sets, they were
each considered separate data sets. DNA sequence alignments were either published (White and Densmore,
2001; Gatesy et al., 2003; Harshman et al., 2003) or implied optimization alignments derived from analyses in
POY (Gladstein and Wheeler, 1997) using the following
commands: oneasis, random 10, slop 5, checkslop 10,
molecularmatrix, impliedalignment, indices, stats, time,
spewbinary, trailinggap 3, internal gaps weighted 3, and
substitutions weighted 2. For alignments of nuclear loci,
some adjacent gaps in POY output were consolidated.
Four data sets that were “uncombinable” with character
data also were considered (Fig. 2). These were published
phenograms based on globin peptide differences and albumin immunological distances (Densmore, 1983) and
two Brooks parsimony analysis trees based on the coevolution of crocodylian parasites (Brooks and McLennan,
1993). Delimitation of data sets loosely followed the protocol outlined by Bininda-Emonds et al. (2003, in press)
for partitioning data in supertree analyses.
Supermatrix construction was performed as described
by Gatesy et al. (2003). Morphological character evidence
was included for extinct crocodyliform taxa (61 operational taxonomic units [OTUs]), and morphological,
molecular, and behavioral character data were incorporated for extant crocodylian taxa (22 OTUs). Outgroups
were exemplars from Testudines (turtles) and Aves
(birds), and monophyly of the ingroup, Crocodyliformes
344
SYSTEMATIC BIOLOGY
VOL. 53
FIGURE 1. Strict consensus tree of minimum length topologies for the morphological data set of 164 parsimony-informative characters. The
number of optimal trees and minimum tree length are indicated. Thick shaded branches connect the 23 extant taxa; all other taxa are extinct.
Dark shaded circles at nodes mark clades that were inconsistent with the strict consensus of optimal topologies for the supermatrix data set of
all character data (see Fig. 3). Subgroups of Crocodylia, as defined by Brochu (2003; see Gatesy et al., 2003), are delimited by brackets to the right
of species names. The strict consensus was rooted with Paleognathae (Aves).
2004
POINTS OF VIEW
345
FIGURE 2. Strict consensus trees derived from minimum-length topologies for 16 character-based data sets of extant crocodylians and four
published topologies based on “uncombinable” data. The number of optimal trees and minimum tree length for each of the character-based
data sets are indicated. The “uncombinable” data sets are either phenetic or based on coevolutionary studies of parasites. Dark shaded circles at
nodes mark groupings that were inconsistent with the strict consensus of optimal topologies for the supermatrix data set (see Fig. 3). Trees with
outgroup taxa were rooted with Testudines or Aves (Neognathae + Paleognathae). Topologies that lacked these taxa were drawn to maximize
congruence with the supermatrix topology (Fig. 3). Genera are abbreviated: C. = Caiman; M. = Melanosuchus; P. = Paleosuchus; A. = Alligator;
Cr. = Crocodylus; O. = Osteolaemus; T. = Tomistoma; G. = Gavialis. Allozymes “A” are allozymes coded for alligatoroids, and allozymes “C” are
the same allozymes coded for nonalligatoroid crocodylians. All data for LDH-A and ATP7A were sequenced for this study. The RAG-1 sequence
for Testudines (Geochelone yniphora) and all BDNF sequences except that for Alligator mississippiensis also are new with this study.
346
SYSTEMATIC BIOLOGY
VOL. 53
FIGURE 3. Strict consensus tree of minimum-length topologies for the supermatrix of all character data (1,464 informative characters; 6,924
total characters). The number of optimal trees and minimum tree length are indicated. Characters in the supermatrix are listed across the top
of the figure. Two segments each of RAG-1 and cytochrome b have been sequenced, and these are denoted as A and B. Open circles to the left
of taxon names indicate characters sampled for each taxon, and the large shaded box shows missing data. Open circles at nodes indicate novel
clades that were supported by the supermatrix analysis but were not implied by any combination of source trees from the separate analyses of
individual data sets (Figs. 1, 2). Dark shaded circles at nodes indicate clades that were not stable to exclusion of the unmapped RFLP character
data, which might not be logically independent (see discussion by Gatesy et al., 2003). Subgroups of Crocodylia to the right of species names
were defined as in Figure 1, but because the clades have phylogenetic definitions (Brochu, 2003), the content of these taxa differed from that in
Figure 1. The strict consenus was rooted with Testudines. The supermatrix data set (SuperMatrixCroc.dat) is available on the Systematic Biology
website.
2004
347
POINTS OF VIEW
(see Brochu, 2003), was assumed in all phylogenetic analyses. The parsimony search of the supermatrix in PAUP∗
4.0b10 (Swofford, 1998) was heuristic, with all characters
unordered, equal weighting of transformations, indels
treated as missing data, 100 random taxon addition replicates, tree bisection–reconnection branch swapping, and
the “amb-” option in effect. A strict consensus tree was
derived from all optimal topologies. Misrepresentation
of homologous sites is unavoidable in restriction fragment data that are not mapped (see Swofford and Olsen,
1990). Therefore, an additional search was executed to
determine the stability of systematic results to exclusion
of the unmapped restriction fragment data (Densmore
and White, 1991).
The same taxa and characters in the supermatrix were
encoded into an MRP supertree data set. In the MRP
framework, individual characters are not directly interpreted as phylogenetic evidence. Instead, topologies
supported by different data sets (source trees) are encoded into a matrix and used to reconstruct the evolutionary history of a particular group (Baum, 1992; Ragan,
1992). Separate parsimony analyses were conducted for
the 17 character-based data sets. A strict consensus tree
was derived from optimal topologies for each data set,
and each resolved group (Figs. 1, 2) was coded as a separate nodal character in the MRP supertree data set. Published topologies for the four “uncombinable” data sets
also were encoded into the MRP supertree data set of
86 OTUs. To facilitate comparison with the supermatrix analysis, the MRP matrix was first analyzed with
the “uncombinable” data sets excluded. PAUP∗ searches
were conducted under the same conditions, with equal
weighting of all nodal characters.
Results
The supermatrix analysis produced 720 optimal
topologies (minimum length = 4,272) and a wellresolved strict consensus (Fig. 3); when individual gaps
were treated as informative characters, the same set of
trees was obtained. No groups supported by the present
analysis conflicted with the previous, more limited supermatrix analysis of Gatesy et al. (2003). Incorporation
of four nuclear loci, two segments of mt DNA, RFLPs, allozymes, and 16 new ingroup taxa did not alter the basic
structure of the crocodylian tree; relationships among extant species were completely consistent with those presented by Gatesy et al. (2003). Overall, 11 groups that
emerged in the supermatrix analysis (Fig. 3) were not implied by any combination of trees supported by separate
analyses of the 17 data sets in the supermatrix (Figs. 1,
2). Such novel clades are an expected outcome of total
evidence analyses (Barrett et al., 1991).
The MRP supertree analysis of the character-based
data sets yielded 103,633 optimal topologies with a minimum length of 216 steps (Fig. 4). Because the placements
of many extinct taxa were not stable, basal relationships
of Crocodylia were poorly resolved in the strict consensus of minimum-length trees. Five more nodes were resolved in the supermatrix search relative to the MRP supertree search, and six clades supported by the supertree
data set were strictly incongruent with the supermatrix
analysis of the same data (Figs. 3, 4). Some of the best trees
for the supertree data set were very poor fits to the supermatrix. For example, the topology shown in Figure 4,
which was optimal according to the MRP supertree analysis, included 18 nodes that strictly conflicted with supermatrix results and demanded an extra 43 steps beyond the minimum length from the supermatrix data set
(Fig. 3). This difference in number of steps was significant
according to the Wilcoxon signed-rank test of Templeton
(1983), with P ≤ 0.0001. Simply constraining the supermatrix to retain all groups strictly supported by the supertree analysis resulted in a cost of 21 extra steps beyond minimum length for the supermatrix (4,293 steps).
A comparison of minimum-length trees with and without this supertree constraint showed that the difference
in character steps was significant (P ≤ 0.0001). Unlike
the supermatrix analysis, no emergent novel clades were
strictly supported by the MRP supertree data set.
I NCONSISTENCIES IN ARGUMENTS FOR S UPERTREES
Several arguments in defense of the approach used in
previous supertree analyses of mammals have been published recently (Bininda-Emonds and Sanderson, 2001;
Bininda-Emonds et al., 2002, 2003, in press). Within
the context of our supertree/supermatrix analyses of
Crocodylia, we suggest that many of these rationalizations are poorly justified.
Unambiguous Data Duplication
Springer and DeJong (2001), Gatesy et al. (2002),
and Gatesy and Springer (in press) noted that many
published MRP supertrees of mammals (Purvis, 1995;
Bininda-Emonds et al., 1999, Liu et al., 2001; Jones et al.,
2002) included unequivocal duplications of data, i.e., redundant sampling of the same characters for the same
taxa in the same data matrix. Most published supertrees
include clear-cut duplications of primary character evidence, the ultimate source data for both supertree and
supermatrix analyses. However, any unambiguous duplication of characters in a data matrix is problematic
because it entails arbitrary and unjustified character
weighting. Even if duplication of data had no effect on inferred relationships, it would still not be justified because
it would surely influence interpretations of support for
competing topologies.
Bininda-Emonds et al. (2003) partly dismissed the difficulty of data duplication in past supertree analyses by
arguing that supermatrix analyses suffer from similar
problems. They noted (2003:725):
Clearly, data duplication exists [in previously published supertrees].
This duplication violates a key assumption of phylogenetic analysis,
namely that the source data are independent. However, this assumption is routinely violated in analyses based on primary character
data. For instance, several characters are often described for a single
morphological structure. In molecular studies, secondary structure
(e.g., stem regions in tRNAs, protein folding) or codon position in
coding DNA mean that nonindependent compensatory mutations
may accompany primary ones. The combination of phenotypic and
genotypic data (i.e., morphological and molecular, respectively) in
348
SYSTEMATIC BIOLOGY
VOL. 53
FIGURE 4. Strict consensus tree of minimum-length topologies for the MRP supertree data set of 190 informative nodal characters that
included all character-based source trees (left), and one of the minimum-length trees for this MRP matrix that is significantly incongruent with
trees supported by the supermatrix (right). The number of optimal trees and the minimum tree length are indicated. For the topology on the right,
tree length for the supermatrix of all character data (see Fig. 3) also is shown. Dark shaded circles at nodes mark clades that were inconsistent
with the strict consensus for this supermatrix data set. Open circles at nodes indicate novel clades that were not supported by supermatrix
analysis (Fig. 3) and were not implied by any combination of source trees from the separate analyses of individual data sets (Figs. 1, 2). When
the five mtDNA data sets were merged and considered one data set, MRP supertree analysis strictly supported slightly different relationships
within Crocodylinae. Crocodylus cataphractus grouped with other extant Crocodylus and C. palaeindicus to the exclusion of Osteolaemus and all
other crocodylines. Furthermore, C. rhombifer and C. moreletii formed a monophyletic group. When the four “uncombinable” source trees (see
Fig. 2) were included in the analysis, relationships were consistent with the strict consensus shown; the basal polytomy among major crocodylian
taxa remained. However, the immunological data set for albumin is partially redundant with the allozyme data set, which also summarized
molecular variation in albumin. Subgroups of Crocodylia are to the right of species names, and trees were rooted with Testudines. The MRP
supertree data set (SuperTreeCroc.dat) is available an the Systematic Biology website.
a supermatrix approach must represent duplication at some level.
Thus, issues of data duplication and nonindependence are not limited to the supertree approach.
This reasoning is not adequate to justify data duplication in supertree analysis for three reasons. First, the unambiguous duplication of evidence in many published
supertrees (e.g., Purvis, 1995; Bininda-Emonds et al.,
1999; Liu et al., 2001; Jones et al., 2002) is quite different
from the potential phylogenetic correlation of characters
in a supermatrix analysis. The former is an unequivocal,
arbitrary redundancy that is introduced by the investigator. Phylogenetic correlations among characters are not
so easily observed and can be detected only through detailed systematic analysis. For example, replication of the
mitochondrial 12S ribosomal DNA (rDNA) sequences in
our crocodylian supermatrix analysis would arbitrarily
increase the impact of this data set on the final hypothesis
of relationships. This step clearly would be unwarranted
and avoidable. In contrast, phylogenetic character correlation is not trivial to detect and accommodate into the
tree building process (see Yang, 1997; Wollenberg and
2004
POINTS OF VIEW
Atchley, 2000; O’Keefe and Wagner, 2001). For 12S rDNA,
interdependence among sites in the stem regions of this
gene may vary widely from one site to another site, over
time, and among lineages (Kraus et al., 1992; Gatesy et al.,
1994).
Second, the unambiguous duplication of evidence in
many published supertrees (e.g., Purvis, 1995; BinindaEmonds et al., 1999; Liu et al., 2001; Jones et al., 2002)
also is quite different from the potential redundancy between some genotypic and phenotypic characters in a
supermatrix analysis. The former is an unequivocal redundancy. By contrast, it generally is not obvious when
specific phenotypic characters are redundant with particular genotypic characters in an analysis.
Third, the problem of clear-cut data duplication that is
evident in most published supertrees exists in addition
to rather than instead of any nonindependence among
primary character data and any redundancies between
phenotypic and genotypic characters. Because the source
trees used in supertree analysis are generated from primary character data, supertrees are affected by the same
character correlations and redundancies that are present
in supermatrix analysis. The explicit data duplications
in many published supertree analyses simply add an
extra, more egregious form of nonindependence among
characters. Such redundancies are not justified and are
easily avoided. For example, our character-based MRP
supertree data set for Crocodylia had no clear-cut duplications of evidence (see Fig. 4), and other published supertree analyses also have avoided redundancy (Daubin
et al., 2001).
Definition of Pseudoindependent Phylogenetic Hypotheses
In both supertree and supermatrix analysis, primary
character data (hypotheses of homology) are critically
important; the definition of characters and character
states by the systematist determines final systematic results (Patterson, 1982). However, in supertree analysis,
delimitation of data sets also is crucial. Contrasting definitions of data set have the potential to influence systematic results in the supertree framework but not in the
supermatrix approach. For example, when all mtDNA
sequences were considered one data set, as opposed to
the five data sets outlined in our MRP supertree analysis
of Crocodylia, different phylogenetic conclusions were
obtained (see Fig. 4).
In recent publications, Bininda-Emonds et al. (2003,
in press) outlined an “explicitly defined set of rules” to
identify “phylogenetic hypotheses that can reasonably
be viewed as being independent” (2003:725) and therefore appropriate for supertree analysis.
1. “Delimiting pseudo-independent evolutionary ‘packets’ based on genes may be defensible given
the recognition of the gene tree/species tree dichotomy. . . . Non-overlapping data sets (e.g. different genes) are considered to be independent, even if
they appear on a single heritable unit like mt DNA”
(2003:725).
349
2. “Different portions of the same gene are not independent for an overlapping set of taxa, even if these gene
portions do not overlap at all” (2003:725).
3. “We also hold unique combinations of genes to be independent sources and independent from data sets
containing subsets of all the genes in the combination”
(in press).
4. “Different morphological data sets are equivalent to
novel combinations of genes and so are considered to
be independent of one another unless one data set is
contained completely within another” (in press).
5. “Trees for non-overlapping taxon sets, even if they are
derived from the same set of characters, are independent by practical necessity” (2003:725).
These five rules allow for extensive duplications of
character data in supertree analyses and are not logically
consistent for the reasons listed below.
The recognition that a given gene tree may differ from
the species tree is not sufficient justification for asserting the primacy of genes as independent evolutionary
units. In any particular case study, there could be substantial independence among characters within genes
as well as nonindependence among genes. This point
has been made by some of the authors themselves, who
stated (Bininda-Emonds et al., 2002:284),
The nonindependence of molecular data owing to linkage associations between genes is often underestimated. For example, it is often
argued that mt DNA with its many genes constitutes a single phylogenetic data source because it forms a single heritable unit that is
not normally subject to recombination.
This view is incompatible with rule 1.
Rules 1 and 3 are inconsistent with each other. Rule
1 states that separate genes are the independent units
of analysis, but rule 3 states that the same gene can be
included multiple times in the same analysis because
of novel phylogenetic signals that might emerge when
different genes are analyzed together. These conflicting
rules make the protocol for defining independent phylogenetic hypotheses arbitrary and prone to extensive data
duplication. If all novel combinations of genes were allowed into supertree analysis, it would have been acceptable for us to include the combined molecular tree from
Gatesy et al. (2003) in our crocodylian MRP supertree
analysis. The tree from Gatesy et al. (2003) overlaps with
∼3,000 nucleotides and four genes in our supertree data
set. We contend that such duplications of genetic data
are not justified in phylogenetic analysis.
According to our view of systematics, it is assumed
that each character state in a morphological data set is
encoded by at least one independent genetic difference.
Thus, by proxy many “genes” are included in a single
morphological data set. The primacy of genes as the independent units of analysis in supertree studies (rule 1)
is, therefore, not compatible with the fact that morphological matrices also were considered independent units
by Bininda-Emonds et al. (2003, in press).
Rule 4 permits extensive duplications of morphological characters. For example, Brochu (1997) analyzed a
350
SYSTEMATIC BIOLOGY
VOL. 53
FIGURE 5. Three minimum-length topologies based on a fragment of the RAG-1 gene (∼2,000 bp). There is no overlap in taxa between trees a
and b. Tree c includes all of the taxa in trees a and b. Thirteen identical nucleotide substitutions unequivocally supported the grouping of Caiman
+ Alligator in each of these trees. These changes were at the following positions: 55, C → T; 310 C → T; 311, A → G; 445, G → A; 467, C → T; 593,
A → C; 706, A → G; 940, T → C; 1370, G → A; 1691, T → C; 1741, A → G; 1891, G → A; 1906, A → G. Despite the lack of taxonomic redundancy
in trees a and b, there is redundancy in the character data.
morphological data set of 164 characters for 61 crocodyliform taxa. In our supertree analysis, we used a recently
revised version of Brochu’s data set that included 164
morphological characters for 83 crocodyliform taxa. The
data sets shared 162 characters and 60 taxa in common,
so each data set had some unique characters and unique
taxa. According to rule 4, these grossly overlapping data
sets are pseudoindependent, and it would be valid to include each source tree into the same supertree analysis
because neither data set “is contained completely within
another” (Bininda-Emonds et al., in press). Morphological data sets for Crocodylia with slightly less overlap in
characters (e.g., Norell, 1989; Wu et al., 1996; Buscalioni
et al., 2001) also would be considered pseudoindependent according to rule 4.
Together, rules 3 and 4 basically allow any set of characters to be independent from all other data sets, as
long as rule 2 is not broken. By our calculation, if these
rules were taken literally, minimally eight previously
published analyses that included 12S rDNA data for
crocodylians would be considered independent source
trees (Poe, 1996; Brochu, 1997; Brochu and Densmore,
2001; Gatesy et al., 2003, this study). We do not think it is
prudent to arbitrarily octuplicate characters in systematic studies.
Data sets for nonoverlapping taxon sets, even if they
are derived from the same set of characters, are considered independent by practical necessity (rule 5), but
such nonoverlapping taxon sets can include duplications of character evidence. For example, a source tree
for the nuclear gene RAG-1 that included only Caiman
crocodilus + Alligator mississippiensis + Crocodylus rhombifer + Paleognathae would be phylogenetically redundant with a source tree for RAG-1 that included only
Caiman latirostris + Alligator sinensis + Crocodylus cataphractus + Neognathae (see Fig. 5). These data sets have
no overlap of taxa, yet each matrix supports Caiman +
Alligator, and there is duplication of character evidence;
the same nucleotide substitutions in each of these two
data sets unequivocally support the same grouping of
genera (Fig. 5). Complete nonoverlap of taxa does not
guarantee avoidance of data duplication in supertree
analysis; this data duplication is not a practical necessity and is completely avoided in supermatrix analyses.
Source Data Quality, Quantity, and Availability
Gatesy et al. (2002) pointed out that some published supertrees included source trees based on dubious data (see
discussions by Purvis, 1995; Bininda-Emonds et al., 1999;
Springer and DeJong, 2001; Jones et al., 2002; Gatesy and
Springer, in press) and that incorporation of problematic source topologies makes these published supertrees
weak phylogenetic statements. Bininda-Emonds et al.
(2003:725) countered that data quality is a more general issue in systematics and is not specific to supertree
2004
POINTS OF VIEW
analysis: “The use of poor data may compromise the results in any phylogenetic analysis (i.e. including a supermatrix analysis), and researchers should ensure that all
data used are of the highest achievable quality.” We agree
with this statement but still think that in an effort to incorporate numerous source trees and to achieve broad taxonomic sampling many previous supertree analyses (e.g.,
Purvis, 1995; Bininda-Emonds et al., 1999; Liu et al., 2001;
Jones et al., 2002; Kennedy and Page, 2002) failed to exclude dubious data, especially in comparison with published supermatrix studies. For example, Liu et al. (2001)
employed source trees based on reviews by authorities,
Kennedy and Page (2002) utilized a morphological tree
that was not derived from an explicit data set/analysis,
and other supertree studies have included topologies
based on informal taxonomies (e.g., Jones et al., 2002).
Purvis (1995) and Bininda-Emonds et al. (1999) downweighted suspect source trees in their mammalian supertree analyses and noted that this procedure had little
effect on topology. However, a more stringent test for
the effect of weak source trees is to give these topologies zero weight in the supertree analysis. When such
searches were executed by Purvis (1995), resolution in
his supertree for Primates decreased significantly. Thus,
many of the relationships in the preferred supertree of
Purvis (1995) had no basis in terms of explicit hypotheses
of homology and were the result of dubious source trees.
In the construction of our crocodylian supermatrix/supertree data sets, we were confronted by a number of data quality issues. For example, how should we
code ambiguous regions of sequence alignments for mitochondrial rDNA genes? Are the chromosomal traits
characterized in enough detail to warrant inclusion? Are
unmapped RFLPs valid evidence? Are outlier sequences
PCR contaminants (e.g., the unpublished Crocodylus porosus sequence for mitochondrial cytochrome b [NCBI
AF306452] apparently is a Homo sapiens sequence)? We
suspect that some researchers will disagree with our decisions regarding which data to include or exclude. However, each of these data quality issues was dealt with at
the character level, and we suggest that future supertree
studies also should engage in character analysis at this
fine scale instead of simply recording topologies reported
by other researchers. The supermatrix approach naturally facilitates the reassessment of primary character
data because these data actually are analyzed in this
framework.
Several authors have argued that an advantage of supertree analysis is that systematic data, which cannot
be combined into a supermatrix, can be analyzed in a
supertree analysis (e.g., Purvis, 1995; Sanderson et al.,
1998; Bininda-Emonds et al., 1999, 2002, 2003; Kennedy
and Page, 2002). However, among the wide range of
phylogenetic evidence available to modern systematists,
only coevolutionary parasite trees, reviews that are based
on the opinions of authorities, classifications that have
no clear empirical basis, DNA/DNA hybridization data
that do not distinguish paralogy from orthology (Marks
et al., 1988), immunological distances that are nearly
obsolete, and other vague phenetic measures such as
351
genomic fingerprints are excluded from supermatrices.
For our crocodylian supertree data set, inclusion of the
four “uncombinable” source trees (Fig. 2) had little effect
on the supertree topology. The polytomy at the base of
Crocodylia remained, and relationships generally were
consistent with the strict consensus tree shown in Figure
4. Exclusion of the vague uncombinable data from our
supermatrix did not seem like much of a loss. Overall, the
amount of available quality data that cannot be incorporated into a supermatrix is so small that this can hardly
serve as a major justification for the use of supertrees.
Proponents of supertrees have argued that extended
taxonomic coverage is one of the primary benefits of using source trees that are incompatible with supermatrix
analysis. For example, the use of taxonomies in supertree
analysis can yield comprehensive phylogenetic hypotheses that include all extant species in a group (e.g., Jones
et al., 2002), but the empirical basis for a particular taxonomy might not be clearly stated. Many classifications
do not result from explicit analyses of documented data
matrices and are based simply on the cumulative knowledge (or whims) of particular authorities (see Gatesy and
Springer, in press). In the 21st century, phylogenetic analyses of primary character data from hundreds of taxa are
quite feasible (e.g., Soltis et al., 1999). Therefore, we suggest that if no explicit character information exists for a
particular taxon, the most useful response from a systematist would be to collect new character data for that
taxon, not to include that taxon in a supertree analysis.
Until a framework of relationships based on actual character data is presented, broad comparative evolutionary
studies seem premature.
Hidden Support, Novel Clades, and the MRP
Supertree Method
An interesting property of some supertree methods
and the supermatix approach is that novel clades can
emerge with the combination of diverse data sets. Such
novel clades either contradict all separate analyses of individual data sets or are not implied by any combination
of trees supported by the separate analyses. In supermatrix analysis, the basis for this hidden support is well
characterized (Barrett et al., 1991; Gatesy et al., 1999; Lee
and Huggal, 2003) and is consistent with the most basic
principles of cladistic parsimony analysis (Farris, 1983)
and likelihood calculations (Felsenstein, 1981). In MRP
supertree analysis, the phenomenon is considered an artifact of an illogical methodology (Goloboff and Pol, 2002;
Pisani and Wilkinson, 2002).
Bininda-Emonds and Sanderson (2001:574) suggested
that “MRP is necessarily an approximation of a total evidence solution, intended primarily for those situations
when the latter technique cannot be applied. . . . At least
in theory, MRP supertree construction might also be expected to be a reasonable approximation.” To us, an important issue is whether MRP, or any supertree method,
can evaluate secondary phylogenetic signals that are
hidden in separate data sets. If this is not the case, how
can an MRP supertree study approximate a supermatrix
352
SYSTEMATIC BIOLOGY
analysis where a large percentage of the character support is hidden?
Several recent supermatrix studies have shown that
significant character support can be hidden in different
data sets. For example, in an analysis of five data sets
for Bovidae (cattle, sheep, and antelope), ∼85% of the
character support was emergent (Gatesy and Arctander,
2000), and in an analysis of 17 data sets for Cetacea
(whales), ∼45% of the support was hidden in the separate character partitions (Gatesy et al., 1999). Likewise,
a recently published supermatrix analysis of three data
sets showed extensive hidden support for relationships
among the four major extant clades of Crocodylia. Approximately 70% of the character support was emergent,
and several novel clades were recovered in the combined
analysis (Gatesy et al., 2003).
Our supermatrix analysis of Crocodylia was topologically consistent with the previous, more limited analysis
of Gatesy et al. (2003) and resolved 11 novel clades (Fig. 3)
that were not implied by any combination of trees supported by the separate analyses of individual data sets.
For example, a grouping of Paratomistoma courtii closer
to Gavialis gangeticus than to Tomistoma schlegelii was favored in the supermatrix analysis, but this relationship
was contradicted by the separate analysis of the morphological partition and was not implied by any combination of topologies derived from the individual data sets
(Figs. 1, 2).
In contrast to the supermatrix analysis, MRP supertree
analysis of the same data strictly supported no emergent
groupings (Fig. 4). This result could be interpreted as
an indication that hidden signals in the primary character data were ignored and/or distorted. Examination
of optimal topologies for the supertree data set showed
that individual trees expressed many novel clades, but
these were not necessarily the emergent relationships
supported by supermatrix analysis. Some of the optimal
supertrees were significantly incongruent with topologies supported by the supermatrix, and the basal fossil
taxa were placed in a bewildering array of equally parsimonious alternative resolutions (Fig. 4). In this empirical
test case, MRP supertree analysis clearly did not distill
the same hidden signals that emerged in supermatrix
analysis (Fig. 3), and MRP was not a good approximation of total evidence.
Almost all published supertree analyses have utilized
the MRP method (e.g., Purvis, 1995; Bininda-Emonds
et al., 1999; Daubin et al., 2001; Liu et al., 2001; Jones et al.,
2002; Kennedy and Page, 2002; Salamin et al., 2002), but
the logical basis for this procedure is unclear. In parsimony and likelihood analyses of supermatrices, the fit of
each character to competing topologies can be calculated
precisely, and there is a clear connection between implied
evolutionary character transformations and the choice
among competing phylogenetic hypotheses (Farris, 1983;
de Queiroz and Poe, 2001). MRP and all other previously
described supertree methods lack these critical properties. In MRP analysis, detailed heterogeneous information encoded by the character data is lost (Pisani and
Wilkinson, 2002), and support that is hidden in different
VOL. 53
data sets by conflicting patterns of homoplasy is completely ignored or misinterpreted. Sometimes MRP analysis yields novel emergent clades that also are supported
by supermatrix analysis (Bryant, in press), sometimes
MRP mimics strict consensus results and conflicts with
supermatrix results (Pisani and Wilkinson, 2002), and in
other situations MRP analysis can produce novel systematic hypotheses that conflict with all data sets that compose the supertree data set and that are not supported
by supermatrix analysis of all data sets. Thus, MRP supertree analysis is an unpredictable systematic “black
box” (Rodrigo, 1993, 1996; Goloboff and Pol, 2002; Pisani
and Wilkinson, 2002; Bryant, in press). Wilkinson et al.
(2001:300) concluded that MRP supertrees “resolve conflicts without promoting any understanding. . . . In the
light of the known behavior of MRP, it remains to be
demonstrated that there is any convincing justification
for this approach to supertree construction.”
Bininda-Emonds and Sanderson (2001) suggested that
MRP is well grounded in basic graph and network theory, but MRP is not well grounded in phylogenetic theory. MRP contrasts with certain supertree methods, such
as gene tree analysis, that reconcile source tree incompatibility in terms of evolutionary events such as gene duplication, gene loss, and horizontal transfer (Maddison,
1997). Slowinski and Page (1999:818) argued that MRP “is
flawed because homoplasy in this context has no obvious
biological meaning.” Thus, even if there is no duplication
of character data, the resolution of incongruence among
different source trees in an MRP data set has not been
justified adequately (see Rodrigo, 1993, 1996; Slowinski
and Page, 1999; Wilkinson et al., 2001; Goloboff and Pol,
2002; Bryant, in press).
Given the lack of a clearly articulated evolutionary
framework for MRP supertree analysis, proponents of
this method have turned to computer simulations to support the general utility of the MRP method. These simulations have shown that under certain circumstances MRP
supertree analysis can approximate total evidence supermatrix results and produce accurate phylogenetic trees
(Bininda-Emonds and Sanderson, 2001). It is important
to recognize the specific assumptions of the simulations,
however, and to compare these assumptions with conditions in empirical supertree studies.
In the simulations, all component data sets had the
same number of characters, the rate of evolution for
each data set on each branch was identical, all characters were multistate nucleotides, and a single model
of evolution was utilized (Bininda-Emonds and Sanderson, 2001). None of these conditions were duplicated in
published MRP supertree data sets (e.g., Purvis, 1995;
Bininda-Emonds et al., 1999; Liu et al., 2001; Jones
et al., 2002; Kennedy and Page, 2002). More importantly, unlike published supertrees, the simulated data
of Bininda-Emonds and Sanderson (2001) did not include character redundancies, poor data, unnecessary
assumptions of monophyly, and other appeals to authority (see Springer and de Jong, 2001; Gatesy et al.,
2002; Gatesy and Springer, in press). Until these inconsistencies between theoretical and actual supertree data
2004
POINTS OF VIEW
sets are sorted out, by simulating informal taxonomies,
review papers, and extensive character duplications on
a computer, or by removing these problems from published MRP supertree data sets, the relevance of the simulations to empirical studies is questionable. Therefore,
we disagree with the conclusion of Bininda-Emonds and
Sanderson (2001:575), based on their simulation results,
that, “published supertrees should be judged and used
with about the same degree of confidence that a total
evidence tree might.” We see no logical relationship between most published supertrees and these simulations.
Supertrees as Summaries of Previous Work
Gatesy et al. (2002) noted that it would be difficult
to argue that a relatively crude supertree data set is a
better description of past systematic research than a detailed supermatrix. In contrast, Bininda-Emonds et al.
(2003:727–728) recently countered that
Gatesy et al. held supertrees to be “imprecise summaries of previous work.” It is unclear to us why this might be the case. We
contend that supertrees are extremely precise summaries of previous work. . . . Because the supertree approach works at the level of
the source trees, it can easily compare the phylogenetic stability of
a taxon or a species with the amount of research effort it has received. Thus unlike supermatrix approaches, supertrees can more
easily highlight taxa that have been well studied, but the relationships within which remain controversial.
The logical basis of phylogenetic robustness measures
in supertree analysis is unclear (see Purvis, 1995; Pisani
et al., 2002; Pisani and Wilkinson, 2002). Alternatively,
in a supermatrix context interpretation of stability, as
measured by a standard support index (e.g., decay index; Bremer, 1994), relative to research effort, as indicated
by the amount of character data for a taxon (see Fig. 3),
is extremely straightforward. If a supermatrix contains
many character data for certain taxa and a clade that contains those taxa has a low decay index, then that relationship might reasonably be called controversial. For supermatrices, interpretations of research effort and stability
are quite straightforward, and unlike supertrees the distribution of character support among data sets can be
assessed in this framework (for examples, see Baker and
DeSalle, 1997; Gatesy et al., 1999, 2003; Lee and Hugall,
2003).
A supermatrix contains all of the information in a supertree data set but without confusing, imprecise data
redundancies and without inclusion of dubious phylogenetic information. Our character-based MRP data set
for Crocodylia can be derived directly from our supermatrix of Crocodylia, because all of the information in
the supertree data set is contained in the supermatrix
data set. However, the reverse does not hold. Supertree
data sets do not encode as much information as do supermatrices, and supermatrices always will be more general/efficient summaries of primary systematic data, i.e.,
characters scored for taxa.
Trees based on supermatrix analysis are accountable
because the actual characters that support particular relationships are directly accessible to all researchers. In
many previous supertree analyses, phylogenetic results
353
are not accountable for three reasons. First, the primary
character data (hypotheses of homology), which are the
ultimate source data for phylogenetic trees, are not represented in supertree data sets. Second, many source
trees encoded into previously published supertree data
sets have no recorded empirical basis and are not accountable themselves. For example, taxonomies that lack
documented support from any data source have been
included in several published supertree matrices (e.g.,
Jones et al., 2002). Third, the MRP method, which has
been utilized in most published supertree analyses, does
not resolve conflicts among source trees in any justifiable
way, again making support for clades unaccountable (see
Rodrigo, 1993, 1996; Wilkinson et al., 2001; Goloboff and
Pol, 2002; Pisani and Wilkinson, 2002).
CONCLUSIONS
We reassessed several difficulties in previously published supertree analyses that were noted by Gatesy et al.
(2002) and contend that solutions to these problems, as
presented by Bininda-Emonds et al. (2003), are not valid
for various reasons.
Unaccountable data, such as taxonomies with no empirical basis, should not be considered acceptable phylogenetic data in past or future supertrees. Given the ease
with which character data can be collected, it is unnecessary to use taxonomies and other outdated information
in modern systematic analyses. If character data for the
taxa of interest do not yet exist, the most reasonable response from a systematist would be to collect new character data, not to make a supertree that includes the taxa
of interest.
Homologous characters are the ultimate source data
for both supermatrix and supertree studies. Unnecessary
duplications of these character data are not justified in
supertree analysis. The definition of independent source
trees according to conflicting criteria (i.e., the gene is the
independent unit of analysis, and any combination of
genes or data sets also is an independent unit) is illogical.
This framework (Bininda-Emonds et al., 2003, in press)
has no safeguards for reducing duplications of evidence
that are present in many published supertree studies.
Bininda-Emonds et al. (2003) correctly noted that systematic data collection protocols can eliminate duplication of
characters in supertree analysis, but the framework outlined by these authors did not achieve this goal.
In supermatrix analysis, the delimitation of characters and character states determines final systematic results. In supertree analysis, the delimitation of characters, character states, and data sets determines final
systematic results (see Fig. 4). Thus, the definition of
data sets (pseudoindependent phylogenetic hypotheses
of Bininda-Emonds et al., 2003) is critical to the supertree approach. This is not the case for the supermatrix approach and makes the supertree approach more
assumption laden relative to the supermatrix approach.
Genes are linked in linear, continuous arrays (chromosomes); specifying individual genes as independent
units of analysis in systematics is thus unjustified and
354
SYSTEMATIC BIOLOGY
at the minimum controversial (see Miyamoto and Fitch,
1995; Kluge, 1997; Maddison, 1997; Siddall, 1997).
MRP, by far the most popular method for building
large supertrees, ignores/misinterprets hidden character support in different data sets and resolves conflicts among source trees in erratic and unjustified
ways (Wilkinson et al., 2001). All methods of supertree
construction that have been proposed thus far ignore/misinterpret secondary phylogenetic signals in different data sets.
Unrealistic simulations have shown that MRP supertree analyses can mimic total evidence analyses
(Bininda-Emonds and Sanderson, 2001). However, the
conditions of published simulations do not match the
conditions of most published MRP supertree studies
(e.g., Purvis, 1995; Bininda-Emonds et al., 1999, Liu et al.,
2001; Jones et al., 2002; Kennedy and Page, 2002; Pisani
et al., 2002; Salamin et al., 2002). Duplications of data,
informal taxonomies, and reviews were not modeled,
therefore the simulations have few implications for assessing the accuracy of most published supertrees.
Published mammalian supertrees (Purvis, 1995;
Bininda-Emonds et al., 1999, Liu et al., 2001; Jones et al.,
2002) are not efficient summaries of previous systematic work. For a comparable sample of primary character
data, a supermatrix contains all of the information in a
supertree data set and much more. Many previously published supertrees are both imprecise, i.e., not minutely
exact, and unaccountable and should not be the basis for
critical tests of evolutionary hypotheses.
ACKNOWLEDGMENTS
NSF grants to J.G. (EAR-0228629, DEB-9985847, DEB-0213171, and
DEB-0212572) provided funding for this work, and R.B. was supported
by an NIH National Research Service Award #1F32GM67463-01. O.
Bininda-Emonds, R. Page, D. Pol, C. Simon, and M. Steel offered helpful
comments. We are especially grateful to C. Brochu, L. Densmore, J.
Harshman, S. Poe, and P. White for rapidly sending us data matrices
by e-mail. G. Amato, P. Brazaitis, Miami Metro Zoo, and the Wildlife
Conservation Society provided DNA and tissue samples. We also thank
M. Lee for suggesting that the BDNF and ATP7A primers from Murphy
et al. (2001) might be useful for resolving crocodylian relationships.
R EFERENCES
Baker, R., and R. DeSalle. 1997. Multiple sources of character information and the phylogeny of Hawaiian drosophilids. Syst. Biol. 46:654–
673.
Barrett, M., M. Donoghue, and E. Sober. 1991. Against consensus. Syst.
Zool. 40:486–493.
Baum, B. 1992. Combining trees as a way of combining data sets for
phylogenetic inference, and the desirability of combining gene trees.
Taxon 41:3–10.
Bininda-Emonds, O., and H. Bryant. 1998. Properties of matrix representation with parsimony analyses. Syst. Biol. 47:497–508.
Bininda-Emonds, O., J. Gittleman, and A. Purvis. 1999. Building large
trees by combining phylogenetic information: A complete phylogeny
of the extant Carnivora (Mammalia). Biol. Rev. 74:143–175.
Bininda-Emonds, O., J. Gittleman, and M. Steel. 2002. The (super)tree
of life: Procedures, problems, and prospects. Annu. Rev. Ecol. Syst.
33:265–289.
Bininda-Emonds, O., K. Jones, S. Price, R. Grenyer, M. Cardillo, M.
Habib, A. Purvis, and J. Gittleman. 2003. Supertrees are a necessary
not-so-evil: A comment on Gatesy et al. Syst. Biol. 52:724–729.
Bininda-Emonds, O., K. Jones, S. Price, R. Grenyer, M. Cardillo, and
A. Purvis. In press. Garbage in, garbage out: Some practical issues
VOL.
53
in supertree construction. In Phylogenetic supertrees: Combining information to reveal the tree of life (O. Bininda-Emonds, ed.). Kluwer
Academic, Dordrecht, The Netherlands.
Bininda-Emonds, O., and M. Sanderson. 2001. Assessment of the accuracy of matrix representation with parsimony analysis supertree
construction. Syst. Biol. 50:565–579.
Bremer, K. 1994. Branch support and tree stability. Cladistics 10:295–
304.
Brochu, C. 1997. Morphology, fossils, divergence timing, and the phylogenetic relationships of Gavialis. Syst. Biol. 46:479–522.
Brochu, C. 2002. Thecachampsoides minor and early gavialoid history:
Coastal Atlantic origins of longirostrine crocodylians. J. Vertebr. Paleontol. 22:39A.
Brochu, C. 2003. Phylogenetic approaches toward crocodylian history.
Annu. Rev. Earth Planet. Sci. 31:357–397.
Brochu, C., and L. Densmore. 2001. Crocodile phylogenetics: A summary of current progress. Pages 3–8 in Crocodilian biology and evolution (G. Grigg, F. Seebacher, and C. Franklin, eds.). Surrey Beatty
and Sons, Chipping Norton, New South Wales.
Brochu, C., and P. Gingerich. 2000. New tomistomine crocodylian from
the middle Eocene (Bartonian) of Wadi Hitan, Fayum Province,
Egypt. Contrib. Mus. Paleontol. Univ. Mich. 30:251–268.
Brooks, D., and D. McLennan. 1993. Parascript: Parasites and the language of evolution. Smithsonian Institution Press, Washington, D.C.
Bryant, H. In press. The cladistics of matrix representation with parsimony analysis. In Phylogenetic supertrees: Combining information
to reveal the tree of life (O. Bininda-Emonds, ed.). Kluwer Academic,
Dordrecht, The Netherlands.
Buscalioni, A., F. Ortega, D. Weishampel, and C. Jianu. 2001. A revision
of the crocodyliform Allodaposuchus precedens from the Upper Cretaceous of the Hateg Basin, Romania. Its relevance in the phylogeny
of Eusuchia. J. Vertebr. Paleontol. 21:74–86.
Caccone, A., J. Gibbs, V. Ketmaier, E. Suatoni, and J. Powell. 1999. Origin
and evolutionary relationships of giant Galapagos tortoises. Proc.
Natl. Acad. Sci. USA 96:13223–13228.
Campbell, H. 1972. Ecological or phylogenetic interpretations of
crocodilian nesting habits. Nature 238:404–405.
Cohen, M., and C. Gans. 1970. The chromosomes of the order
Crocodilia. Cytogenetics 9:81–105.
Daubin, V., M. Gouy, and G. Perriere. 2001. Bacterial molecular phylogeny using supertree approach. Genome Inf. 12:155–164.
Densmore, L. 1983. Biochemical and immunological systematics of the
order Crocodilia. Pages 397–465 in Evolutionary biology, Volume 16
(M. K. Hecht, B. Wallace, and G. T. Prance, eds.). Plenum Press, New
York.
Densmore, L., and P. White. 1991. The systematics and evolution of
the Crocodilia as suggested by restriction endonuclease analysis of
mitochondrial and nuclear ribosomal DNA. Copeia 1991:602–615.
De Queiroz, K., and S. Poc. 2001. Philosophy and phylogenetic inference: A comparison of likelihood and parsimony methods in the
context of Karl Popper’s writings on corroboratio Syst. Biol. 50:305–
321.
Desjardins, P., and R. Morais. 1990. Sequence and gene organization
of the chicken mitochondrial genome: A novel gene order in higher
vertebrates. J. Mol. Biol. 212:599–635.
Farris, J. 1983. The logical basis of phylogenetic analysis. Pages 7–36
in Advances in cladistics, Volume 2 (N. Platnick and V. Funk, eds.).
Columbia Univ. Press, New York.
Felsenstein, J. 1981. Evolutionary trees from DNA sequences: A maximum likelihood approach. J. Mol. Evol. 17:368–376.
Gatesy, J., and G. Amato. 1992. Sequence similarity of 12S ribosomal
segment of mitochondrial DNAs of gharial and false gharial. Copeia
1992:241–244.
Gatesy, J., G. Amato, M. Norell, R. DeSalle, and C. Hayashi. 2003. Combined support for wholesale taxic atavism in gavialine crocodylians.
Syst. Biol. 52:403–422.
Gatesy, J., and P. Arctander. 2000. Hidden morphological support for
the phylogenetic placement of Pseudoryx nghetinhensis with bovine
bovids: A combined analysis of gross anatomical evidence and DNA
sequences from five genes. Syst. Biol. 49:515–538.
Gatesy, J., R. DeSalle, and W. Wheeler. 1993. Alignment-ambiguous nucleotide sites and the exclusion of systematic data. Mol. Phylogenet.
Evol. 2:152–157.
2004
POINTS OF VIEW
Gatesy, J., C. Hayashi, R. DeSalle, and E. Vrba. 1994. Rate limits for
mispairing and compensatory change: The mitochondrial ribosomal
DNA of antelopes. Evolution 48:188–196.
Gatesy, J., C. Matthee, R. DeSalle, and C. Hayashi. 2002. Resolution of
a supertree/supermatrix paradox. Syst. Biol. 51:652–664.
Gatesy, J., P. O’Grady, and R. Baker. 1999. Corroboration among data
sets in simultaneous analysis: Hidden support for phylogenetic relationships among higher level artiodactyl taxa. Cladistics 15:271–
313.
Gatesy, J., and M. Springer. In press. A critique of matrix representation with parsimony supertrees. In Phylogenetic supertrees: Combining information to reveal the tree of life (O. Bininda-Emonds, ed.).
Kluwer Academic, Dordrecht, The Netherlands.
Goloboff, P., and D. Pol. 2002. Semi-strict supertrees. Cladistics 18:514–
525.
Gladstein, D., and W. Wheeler. 1997. POY: Phylogeny reconstruction
via direct optimization of DNA data, version 2.0 American Museum
of Natural History, New York.
Greer, A. 1970. Evolutionary and systematic significance of crocodilian
nesting habits. Nature 227:523–524.
Groth, J., and G. Barrowclough. 1999. Basal divergences in birds and
the phylogenetic utility of the nuclear RAG-1 gene. Mol. Phylogenet.
Evol. 12:115–123.
Harlid, A., A. Janke, and U. Arnason. 1997. The mtDNA sequence of the
ostrich and the divergence between paleognathous and neognathous
birds. Mol. Biol. Evol. 14:754–761.
Harshman, J., C. Huddleston, J. Bollback, T. Parsons, and M. Braun.
2003. True and false gharials: A nuclear gene phylogeny of
Crocodylia. Syst. Biol. 52:386–402.
Janke, A., and U. Arnason. 1997. The complete mitochondrial genome
of Alligator mississippiensis and the separation between recent Archosauria (birds and crocodiles). Mol. Biol. Evol. 14:1266–1272.
Janke, A., D. Erpenbeck, M. Nilsson, and U. Arnason. 2001. The mitochondrial genomes of the iguana (Iguana iguana) and the caiman
(Caiman crocodylus): Implications for amniote phylogeny. Proc. R. Soc.
Lond. B 268:623–631.
Jones, K., A. Purvis, A. MacLarnon, O. Bininda-Emonds, and N.
Simmons. 2002. A phylogenetic supertree of the bats (Mammalia:
Chiroptera). Biol. Rev. 77:223–259.
Kennedy, M., and R. D. M. Page. 2002. Seabird supertrees: Combining
partial estimates of procellariiform phylogeny. Auk 119:88–108.
Kluge, A. 1989. A concern for evidence and a phylogenetic hypothesis
of relationships among Epicrates (Boidae, Serpentes). Syst. Zool. 38:7–
25.
Kluge, A. 1997. Testability and the refutation and corroboration of
cladistics hypotheses. Cladistics 13:81–96.
Kraus, F., L. Jarecki, M. Miyamoto, S. Tanhauser, and P. Laipis. 1992.
Mispairing and compensational changes during the evolution of mitochondrial ribosomal DNA. Mol. Biol. Evol. 9:770–774.
Lee, M., and A. Hugall. 2003. Partitioned likelihood support and the
evaluation of data set conflict. Syst. Biol. 52:15–22.
Liu, F., M. Miyamoto, N. Freire, P. Ong, M. Tennant, T. Young, and K.
Gugel. 2001. Molecular and morphological supertrees for eutherian
(placental) mammals. Science 291:1786–1789.
Maddison, W. 1997. Gene trees in species trees. Syst. Biol. 46:523–536.
Marks, J., C. Schmid, and V. Sarich. 1988. DNA hybridization as a guide
to phylogeny: Relations of the Hominoidea. J. Hum. Evol. 17:769–
786.
Matthee, C., J. Burzlaff, J. Taylor, and S. Davis. 2001. Mining the mammalian genome for artiodactyl systematics. Syst. Biol. 50:367–390.
Mindell, D., M. Sorenson, D. Dimcheff, M. Hasegawa, J. Ast, and T. Yuri.
1999. Interordinal relationships of birds and other reptiles based on
whole mitochondrial genomes. Syst. Biol. 48:138–152.
Miyamoto, M., and W. Fitch. 1995. Testing species phylogenies and
phylogenetic methods with congruence. Syst. Biol. 44:64–76.
Murphy, W., E. Eizrik, W. Johnson, Y. Zhang, O. Ryder, and S. O’Brien.
2001. Molecular phylogenetics and the origins of placental mammals.
Nature 409:614–618.
Norell, M. 1989. The higher level relationships of the extant Crocodylia.
J. Herpetol. 23:325–335.
355
O’Keefe, F., and P. Wagner. 2001. Inferring and testing hypotheses
of cladistic character dependence by using character compatibility.
Syst. Biol. 50:657–675.
Patterson, C. 1982. Morphological characters and homology. Pages 21–
74 in Problems of phylogenetic reconstruction (A. Joysey and A. Friday, eds.). Academic Press, London.
Pisani, D., and M. Wilkinson. 2002. Matrix representation with parsimony, taxonomic congruence, and total evidence. Syst. Biol. 51:151–
155.
Pisani, D., A. Yates, M. Langer, and M. Benton. 2002. A genus-level
supertree of the Dinosauria. Proc. R. Soc. Lond. B 269:915–921.
Poe, S. 1996. Data set incongruence and the phylogeny of crocodilians.
Syst. Biol. 45:393–414.
Purvis, A. 1995. A composite estimate of primate phylogeny. Philos.
Trans. R. Soc. Lond. B 348:405–421.
Ragan, M. 1992. Phylogenetic inference based on matrix representation
of trees. Mol. Phylogenet. Evol. 1:53–58.
Ray, D., and L. Densmore. 2002. The crocodilian mitochondrial control
region: General structure, conserved sequences, and evolutionary
implications. J. Exp. Zool. 294:334–345.
Rodrigo, A. 1993. A comment on Baum’s method for combining phylogenetic trees. Taxon 42:631–636.
Rodrigo, A. 1996. On combining cladograms. Taxon 45:267–274.
Roman, J., and B. W. Bowen. 2000. The mock turtle syndrome: Genetic
identification of turtle meat purchased in the south-eastern United
States of America. Anim. Conserv. 3:61–65.
Salamin, N., T. Hodkinson, and V. Savolainen. 2002. Building supertrees: An empirical assessment using the grass family (Poaceae).
Syst. Biol. 51:136–150.
Sanderson, M., A. Purvis, and C. Henze. 1998. Phylogenetic supertrees:
Assembling the trees of life. Trends Ecol. Evol. 13:105–109.
Siddall, M. 1997. Prior agreement: Arbitration or arbitrary. Syst. Biol.
46:766–770.
Slowinski, J. B., and R. D. M. Page. 1999. How should species phylogenies be inferred from sequence data. Syst. Biol. 48:814–825.
Soltis, P., D. Soltis, and M. Chase. 1999. Angiosperm phylogeny inferred from mutiple genes as a tool for comparative biology. Nature
402:402–404.
Springer, M., and W. DeJong. 2001. Which mammalian supertree to
bark up? Science 291:1709–1711.
Swofford, D. 1998. PAUP*: Phylogenetic analysis using parsimony (∗ and other methods), version 4.0b8. Sinauer, Sunderland,
Massachusetts.
Swofford, D., and G. Olsen. 1990. Phylogeny reconstruction. Pages 411–
501 in Molecular systematics, 1st edition (D. Hillis and C. Moritz,
eds.). Sinauer, Sunderland, Massachusetts.
Templeton, A. 1983. Phylogenetic inference from restriction endonuclease cleavage site maps with particular reference to the evolution
of humans and the apes. Evolution 37:221–244.
White, P., and L. Densmore. 2001. DNA sequence alignments and
data analysis methods: Their effect on the recovery of crocodylian
relationships. Pages 29–37 in Crocodilian biology and evolution
(G. Grigg, F. Seebacher, and C. Franklin). Surrey Beatty and Sons,
Chipping Norton, New South Wales.
Wilkinson, M., J. L. Thorley, D. T. J. Littlewood, and R. A. Bray. 2001.
Towards a phylogenetic supertree of Platyhelminthes? Pages 292–
301 in Interrelationships of the Platyhelminthes (D. T. J. Littlewood
and R. A. Bray, eds.). Taylor and Francis, London.
Wollenberg, K., and W. Atchley. 2000. Separation of phylogenetic and
functional associations in biological sequences by using the parametric bootstrap. Proc. Natl. Acad. Sci. USA 97:3288–3291.
Wu, X., D. Brinkman, and A. Russell. 1996. A new alligator from the Upper Cretaceous of Canada and the relationships of early eusuchians.
Palaeontology 39:351–375.
Yang, Z. 1997. PAML: A program package for phylogenetic analysis by
maximum likelihood. Comput. Appl. Biosci. 13:555–556.
First submitted 30 March 2003; reviews returned 12 June 2003;
final acceptance 20 November 2003
Associate Editor: Mike Steel
356
SYSTEMATIC BIOLOGY
VOL. 53
Syst. Biol. 53(2):356–359, 2004
c Society of Systematic Biologists
Copyright ISSN: 1063-5157 print / 1076-836X online
DOI: 10.1080/10635150490440396
Trees Versus Characters and the Supertree/Supermatrix “Paradox”
O LAF R. P. B ININDA-EMONDS
Lehrstuhl für Tierzucht, Technical University of Munich, Alte Akademie 12, 85354 Freising-Weihenstephan, Germany;
E-mail: [email protected]
In a pair of recent articles, Gatesy and colleagues
(Gatesy et al., 2002, 2004; also Gatesy and Springer, 2004)
have strongly criticized several recent supertree studies.
In so doing, they have pointed out important, but correctable, shortcomings in how the supertree approach
was applied in specific instances, and have helped to
fine-tune the methodology of this comparatively young
field.
However, their equally strong critiques of the MRP
method, if not the supertree approach as a whole,
derive from a faulty basis for comparison. Gatesy
et al. (2004:XXX) state (correctly) that primary character data are “the ultimate source data for both supertree and supermatrix analyses” (also p. XXX), and
use this statement to justify comparing both approaches
on this level. However, because any connection between the primary character data and the supertree
analysis is highly indirect—a feature of supertree construction that they also criticize—it is invalid to judge
supertrees according to criteria designed for characterbased phylogenetic reconstruction. Instead, the (MRP)
supertree approach should be judged with respect to
the data that it uses directly, namely the phylogenetic
hypotheses presented in the source trees. As I hope to
show, recognizing that supertree and supermatrix analyses operate at different levels blunts most of Gatesy
et al.’s criticisms of the supertree approach, thereby
resolving the “paradox” they mention in their earlier
paper.
S OURCE TREE COLLECTION AND D ATA D UPLICATION
As part of our efforts to construct a supertree for all extant species of mammal, we drew up a list of guidelines to
help us decide which source trees were suitable for inclusion (summarized in Bininda-Emonds et al., 2003, 2004).
These guidelines were based on the same two major
issues raised independently by Gatesy and colleagues:
data duplication and source tree quality.
As noted by Gatesy et al. (2004), our rules still allow
for the duplication of the primary character data among
source trees. However, we do not hold this to necessarily
be problematic. Duplication can occur at this level and
still result in independent phylogenetic hypotheses because a phylogenetic tree is composed of more than the
data going into it (Bininda-Emonds et al., 2003, 2004). All
assumptions made in the analysis (e.g., the alignment,
any weighting schemes, the model of evolution used)
as well as the form of the analysis itself (i.e., the opti-
mization criterion used) can impact on the resultant phylogeny. We raised the example previously where different assumptions of rooting for virtually the same data set
gave very different hypotheses about the phylogenetic
relationships among cetaceans (see Bininda-Emonds et
al., 2003). Another cogent example of the effect any auxiliary assumptions can have on our phylogenetic hypotheses is the detailed study of Maddison et al. (1999) on the
phylogeny of carabid beetles, where different manipulations of the same base data set produced very different
trees. Even the large molecular supermatrix of Madsen
et al. (2001) yielded a different set of relationships when
reanalyzed under a different set of assumptions by Malia
et al. (2003).
In short, our guidelines specified a level of primary
data duplication that we held still resulted in reasonably
independent phylogenetic hypotheses. Others will undoubtedly disagree, including Gatesy et al., for whom
all primary data duplication is problematic. In the end,
what is important is for the researcher to assess data independence in the supertree analysis at the appropriate
level, and this is at the level of the source tree and not
the primary character data. Moreover, as we stressed, the
rules were not designed to be applied literally and inflexibly, but to be interpreted according to the data at hand
and the specific question being asked (Bininda-Emonds
et al., 2004:277). This is in line with conventional phylogenetic analyses, where hard-and-fast rules with respect
to which data to include, how to process them (e.g., aligning molecular data or scoring morphological data), and
how to weight or analyze them are extremely rare.
T HE T HEORETICAL B ASIS OF MRP
S UPERTREE CONSTRUCTION
Gatesy et al. (2004; also Gatesy and Springer, 2004)
argued that MRP lacks a logical basis and, as such, constitutes a systematic “black box” that is inappropriate
for phylogeny reconstruction. In part, their perception
of the lack of a logical basis to MRP derives from their
attempts to judge it according to inappropriate criteria.
However, they also reiterate previous criticisms (e.g.,
Rodrigo, 1993, 1996; Slowinski and Page, 1999) that
the use of parsimony as an optimization criterion in
MRP is unfounded because any “homoplasy” on a supertree cannot be interpreted in a biologically meaningful way (i.e., as instances of convergence, parallelism, or
reversal).
2004
POINTS OF VIEW
However, incongruence in a supertree analysis is simply that, and there is no reason to equate it with homoplasy. In its purest form, the principle of parsimony
makes no statements regarding either homoplasy or incongruence having to be biologically interpretable. It
merely asserts that the preferred hypothesis is the one
that minimizes the number of ad hoc assumptions (i.e., the
simplest possible solution, loosely speaking). As such,
the use of parsimony in MRP has the same logical basis as that for analyzing character data, namely to find
the solution with the minimum amount of incongruence
(as measured by the objective function of a parsimony
analysis) to the data being analyzed. Homoplasy is instead a post hoc explanation that biologists use to explain incongruence in character data, the same as when
specific instances of incongruence are held to represent
faulty hypotheses of homology on the part of the investigator. Because (MRP) supertree analysis does not
analyze character data, there is no need to invoke the
idea of homoplasy, nor require incongruence to have a
biological meaning (although it can in supertree methods such as gene-tree parsimony; Slowinski and Page,
1999).
Gatesy et al. (2004) noted that MRP supertrees at times
variously resemble or contradict the results of either supermatrix or taxonomic congruence analyses, and use
this “inconsistent” behavior as evidence for the blackbox nature of MRP. The flaw in the argument is seen
easily: one could use it to show that parsimony is also
a black box because it produces results that are sometimes closer to phenetic methods like NJ and sometimes
to probabilistic methods like ML or Bayesian analysis.
The reality is that different methods will converge on the
same answer at different times because of the nature of
the data being analyzed and not because of any black-box
qualities to the method.
Nor does the fact that most conventional characterbased support measures (e.g., bootstrap frequencies or
Bremer support) are invalid when applied to MRP supertrees invalidate the entire approach or cast its logical basis into doubt (as implied by Gatesy et al., 2004).
Instead, it merely argues that appropriate supertreespecific support measures be developed that operate
at the level of trees and not characters. Several such
measures already exist: triplet- and quartet-fit similarity
measures (Page, 2002; Piaggio-Talice et al., 2004), or the
QS index (Bininda-Emonds, 2003).
HIDDEN S UPPORT
The inability of all supertree methods to account fully
for hidden support in the character data is an accepted
limitation, but a necessary tradeoff, of the combining of
tree topologies in a supertree approach. As such, the
validity of any novel clades in a supertree analysis is
open to question (Pisani and Wilkinson, 2002; Gatesy
et al., 2004). Fortunately, however, such clades appear
to be exceptionally rare, at least for MRP supertrees.
Simulation results indicate that novel clades occurred
predominantly, but still at a frequency of <0.2%, when
357
just two source trees were combined (Bininda-Emonds,
2003). This merely indicates that phylogenies should not
be constructed from limited data, be they source trees or
character data. Most supertrees have been constructed
from many more source trees than under these limiting
conditions, and no novel clades have been reported for
any of the major supertree studies (see Bininda-Emonds,
2003), including that of Gatesy et al. (2004).
However, I would argue against the assertion of
Gatesy et al. (2004) that novel clades in a supermatrix
analysis are always justified because they derive from
hidden support in character data. Consider the example
they cited with approval regarding the novel clade associated with Paratomistoma courtii. In the supermatrix
tree, this species clusters as the sister group to the remaining species of Gavialinae. However, this position
conflicts with its placement deep within Tomistominae
from the analysis of the morphological data set, the only
one to specify the position of this fossil species. Although
a subsignal within this data set does cluster P. courtii
within gavialines (J. Gatesy, pers. comm.), it is unclear
why hidden support is preventing this species in the
supermatrix tree from remaining as sister to the clade
comprising Gavialosuchus eggenbergensisi, Tomistoma lusitanica, and Tomistoma schlegeli, a clade present in both the
morphological and supermatrix trees. Instead, the novel
placement of P. courtii appears to be an artifact of missing data arising from this case of “taxon sampling.” The
lack of other, largely molecular, data for P. courtii means
that its final position is determined largely by being optimized on the scaffold imposed on it by the much more
numerous remaining data, where 13 of the 17 data sets
favor a sister group relationship between gavialines and
tomistomines (in contrast to the morphological data set).
In a sense, P. courtii is being “left behind” while the extant
species with more data sort themselves out. The end result is that its final position does not reflect the only data
available for it. Such artifacts need not be limited to fossil
species either, but to any poorly sampled species. Given
the patchy distribution of molecular data (see Sanderson
et al., 2003), it is likely that some novel clades arising because of hidden support in purely molecular data sets
might be equally suspect, and should not be accepted
uncritically.
D ATA Q UALITY
Gatesy et al. (2002, 2004) counter that the greater degree of taxonomic completeness that supertrees make
possible comes at the cost of having to include what they
hold to be source trees of poor quality (taxonomies in
particular), whether as a result of “dubious data” or invalid analytical techniques. They therefore question the
utility of several supertree studies as a framework under
which to study evolutionary phenomena. The potential
corollaries of this criticism to supertree construction are
twofold: 1) that trees derived from poor data or analyses are necessarily inaccurate, and 2) that any inaccuracy
is detrimental to the resulting supertree. However, evidence suggests otherwise in both cases.
358
SYSTEMATIC BIOLOGY
The history of phylogenetic analysis is arguably one of
broadly congruent results rather than widespread disagreement. Specific exceptions abound, of course, but
I would suggest that the relative amount and degree
of conflict has been overplayed. For example, I have
shown elsewhere that estimates of phylogeny within the
mammalian order Carnivora are indistinguishable statistically for the most part (Bininda-Emonds, 2000). This
included phylogenies derived from good and poor data
or analyses (including taxonomies and other “data-free”
phylogenies). Naturally, this finding applies only to the
Carnivora. However, it implies that source trees derived
from “dubious” data or techniques should not be excluded automatically, but subjected to the same process
of data assessment that Gatesy et al. (2002, 2004) advocate, and which is possible in a supertree framework
(contrary to their claims).
An implicit assumption in phylogenetic analysis is
that phylogenetic signal is coherent and will outweigh
any non-phylogenetic “signals,” which are random or at
least less coherent. This is, in fact, the principle underlying signal enhancement and hidden support. However,
it also explains why the inclusion of poor source trees
need not be detrimental to a supertree analysis: any inaccurate information from such source trees should be
outweighed by the coherent phylogenetic signal from the
remaining source trees. Indirect support for this derives
from the observation that heavily downweighting poor
source trees had little appreciable effect in most supertree
studies (e.g., Purvis, 1995; Bininda-Emonds et al., 1999;
Jones et al., 2002; Stoner et al., 2003). Altogether, differential weighting schemes and other sensitivity analyses
would seem to therefore be reasonable counterstrategies
to ascertain any effects owing to the inclusion of poor
data.
On a more practical note, Gatesy et al. (2004) argue
that the lack of good data for a species should be taken
as a sign to collect some. I agree wholeheartedly. But,
what do we do in the meantime, especially in the face of
the looming biodiversity crisis? In view of the increasing importance of phylogenetic studies to conservation
biology (Purvis et al., in press), the stand of Gatesy et al.
that poorly known species be excluded from phylogenetic (super)trees and the comparative analyses based
on them has serious consequences. Most methods that
can identify threatened, species-poor clades, and therefore possible factors correlating with this threat, rely on
complete taxon sampling (see Gittleman et al., 2004). Yet,
threatened species are precisely those for which good
data are often lacking (see McKinney, 1999; Mace et al.,
2003; Gittleman et al., 2004), and this will likely be true for
some time to come, especially for less charismatic organisms (e.g., most invertebrate groups). Surely it is better
for conservation biologists to draw inferences now, and
ones that are based on phylogenies that only might be inaccurate as a result of using poor data, rather than to wait
and start conservation efforts after enough good phylogenetic data has been amassed (when it might be too
late). Gatesy et al. (2004) rightly note that their complete
VOL.
53
phylogeny of Crocodylia will have important implications for the conservation of this group, but complete
phylogenies such as this, even for other equally small
groups, remain very much the exception.
T HE FUTURE: G LOBAL CONGRUENCE
AND D IVIDE-AND -CONQUER
The recognition that the supertree and supermatrix
approaches analyze different data using different assumptions and methods has an important consequence.
Contrary to Gatesy et al. (2004:XXX), these approaches
should be seen as being complementary, and not competing, strategies for phylogenetic reconstruction in a
manner akin to the global congruence approach (sensu
Lapointe et al., 1999). Where these different approaches
both support the same set of relationships for a comparable set of studies, we can have increased confidence in
the reality of those relationships. By contrast, relationships upon which the approaches disagree, especially
poorly supported relationships, should be examined
more closely for possible causes of this conflict (e.g., data
used or assumptions made in either approach, or true
conflict), and targeted for additional data collection and
analysis.
Gatesy and colleagues have done much to improve
the phylogenetic database by generating and collating a
tremendous amount of character data on two very different taxonomic groups. However, they have also now
produced two sets of twin supertree-supermatrix analyses that could be profitably compared to help elucidate
the phylogenetic relationships of the respective groups,
especially outstanding areas of uncertainty. Together, the
global solution provided by the supertree and supermatrix approaches is stronger than the solution from either
analysis alone.
Even so, the full promise of this complementarity
has yet to be realized. As part of a divide-and-conquer
strategy—whereby a large phylogenetic problem is broken down into many smaller, computationally easier
ones, the results of which are later combined—supertree
construction could play a vital role in the analysis of very
large supermatrices. Preliminary results indicate this to
be the case (Sanderson et al., 2003; Roshan et al., 2004).
Therefore, the complementary nature of the supertree
and supermatrix approaches will become increasingly
important as we tackle ever-larger portions of the Tree of
Life for analysis.
ACKNOWLEDGMENTS
I thank John Gatesy, John Gittleman, Kate Jones, Andy Purvis, and
David Williams for their helpful comments, and the German research
program BMBF, through the “Bioinformatics for the Functional Analysis of Mammalian Genomes” (BFAM) project, for financial support.
R EFERENCES
Bininda-Emonds, O. R. P. 2000. Factors influencing phylogenetic inference: A case study using the mammalian carnivores. Mol. Phylogenet. Evol. 16:113–126.
2004
POINTS OF VIEW
Bininda-Emonds, O. R. P. 2003. Novel versus unsupported clades: Assessing the qualitative support for clades in MRP supertrees. Syst.
Biol. 52:839–848.
Bininda-Emonds, O. R. P., J. L. Gittleman, and A. Purvis. 1999. Building
large trees by combining phylogenetic information: A complete phylogeny of the extant Carnivora (Mammalia). Biol. Rev. 74:143–175.
Bininda-Emonds, O. R. P., K. E. Jones, S. A. Price, M. Cardillo, R.
Grenyer, and A. Purvis. 2004. Garbage in, garbage out: Data issues
in supertree construction. Pages 267–280 in Phylogenetic supertrees:
Combining information to reveal the Tree of Life (O. R. P. BinindaEmonds, ed.). Kluwer Academic, Dordrecht, the Netherlands.
Bininda-Emonds, O. R. P., K. E. Jones, S. A. Price, R. Grenyer, M.
Cardillo, M. Habib, A. Purvis, and J. L. Gittleman. 2003. Supertrees
are a necessary not-so-evil: A comment on Gatesy et al. Syst. Biol.
52:724–729.
Gatesy, J., R. H. Baker, and C. Hayashi. 2004. Inconsistencies in arguments for the supertree approach: Supermatrices versus supertrees
of Crocodylia. Syst. Biol. 53:XXX–XXX.
Gatesy, J., C. Matthee, R. DeSalle, and C. Hayashi. 2002. Resolution of
a supertree/supermatrix paradox. Syst. Biol. 51:652–664.
Gatesy, J., and M. S. Springer. 2004. A critique of matrix representation with parsimony supertrees. Pages 369–388 in Phylogenetic supertrees: Combining information to reveal the Tree of Life
(O. R. P. Bininda-Emonds, ed.). Kluwer Academic, Dordrecht, the
Netherlands.
Gittleman, J. L., K. E. Jones, and S. A. Price. 2004. Supertrees: Using complete phylogenies in comparative biology. Pages 439–460 in Phylogenetic supertrees: Combining information to reveal the Tree of Life
(O. R. P. Bininda-Emonds, ed.). Kluwer Academic, Dordrecht, the
Netherlands.
Jones, K. E., A. Purvis, A. MacLarnon, O. R. P. Bininda-Emonds, and
N. B. Simmons. 2002. A phylogenetic supertree of the bats
(Mammalia: Chiroptera). Biol. Rev. 77:223–259.
Lapointe, F.-J., J. A. W. Kirsch, and J. M. Hutcheon. 1999. Total evidence, consensus, and bat phylogeny: A distance based approach.
Mol. Phylogenet. Evol. 11:55–66.
Mace, G. M., J. L. Gittleman, and A. Purvis. 2003. Preserving the Tree
of Life. Science 300:1707–1709.
Maddison, D. R., M. D. Baker, and K. A. Ober. 1999. Phylogeny of
carabid beetles as inferred from 18S ribosomal DNA (Coleoptera:
Carabidae). Syst. Entomol. 24:103–138.
Madsen, O., M. Scally, C. J. Douady, D. J. Kao, R. W. DeBry, R. Adkins,
H. M. Amrine, M. J. Stanhope, W. W. de Jong, and M. S. Springer.
359
2001. Parallel adaptive radiations in two major clades of placental
mammals. Nature 409:610–614.
Malia, M. J., Jr., D. L. Lipscomb, and M. W. Allard. 2003. The misleading
effects of composite taxa in supermatrices. Mol. Phylogenet. Evol.
27:522–527.
McKinney, M. L. 1999. High rates of extinction and threat in poorly
studied taxa. Conserv. Biol. 13:1273–1281.
Page, R. D. M. 2002. Modified mincut supertrees. Pages 537–552 in Algorithms in Bioinformatics, Second International Workshop, WABI,
2002, Rome, Italy, September 17–21, 2002, Proceedings (R. Guigó and
D. Gusfield, eds.). Springer, Berlin.
Piaggio-Talice, R., J. G. Burleigh, and O. Eulenstein. 2004. Quartet supertrees. Pages 173–191 in Phylogenetic supertrees: Combining information to reveal the Tree of Life (O. R. P. Bininda-Emonds, ed.).
Kluwer Academic, Dordrecht, the Netherlands.
Pisani, D., and M. Wilkinson. 2002. Matrix representation with parsimony, taxonomic congruence, and total evidence. Syst. Biol. 51:151–
155.
Purvis, A. 1995. A composite estimate of primate phylogeny. Philos.
Trans. R. Soc. Lond. B 348:405–421.
Purvis, A., J. L. Gittleman, and T. M. Brooks. In press. Phylogeny and
conservation. Cambridge University Press, Cambridge.
Rodrigo, A. G. 1993. A comment on Baum’s method for combining
phylogenetic trees. Taxon 42:631–636.
Rodrigo, A. G. 1996. On combining cladograms. Taxon 45:267–274.
Roshan, U., B. M. E. Moret, T. L. Williams, and T. Warnow. 2004. Performance of supertree methods on various data set decompositions.
Pages 301–328 in Phylogenetic supertrees: Combining information
to reveal the Tree of Life (O. R. P. Bininda-Emonds, ed.). Kluwer
Academic, Dordrecht, the Netherlands.
Sanderson, M. J., A. C. Driskell, R. H. Ree, O. Eulenstein, and
S. Langley. 2003. Obtaining maximal concatenated phylogenetic
data sets from large sequence databases. Mol. Biol. Evol. 20:1036–
1042.
Slowinski, J. B., and R. D. M. Page. 1999. How should species
phylogenies be inferred from sequence data? Syst. Biol. 48:814–
825.
Stoner, C. J., O. R. P. Bininda-Emonds, and T. M. Caro. 2003. The
adaptive significance of coloration in lagomorphs. Biol. J. Linn. Soc.
79:309–328.
First submitted 31 December 2003; final acceptance 29 January 2004
Associate Editor: Mike Steel