全 文 ：Journal of Systematics and Evolution 46 (3): 274–286 (2008) doi: 10.3724/SP.J.1002.2008.08038
(formerly Acta Phytotaxonomica Sinica) http://www.plantsystematics.com

Animal phylogeny and large-scale sequencing: progress and pitfalls
Henner BRINKMANN* Hervé PHILIPPE**
(Centre Robert Cedergren, Département de Biochimie, Université de Montréal, 2900 Boulevard Edouard-Montpetit, Montréal, Québec, H3T1J4, Canada)
Abstract Phylogenomics, the inference of phylogenetic trees using genome-scale data, is becoming the rule for
resolving difficult parts of the tree of life. Its promise resides in the large amount of information available, which
should eliminate stochastic error. However, systematic error, which is due to limitations of reconstruction meth-
ods, is becoming more apparent. We will illustrate, using animal phylogeny as a case study, the three most effi-
cient approaches to avoid the pitfalls of phylogenomics: (1) using a dense taxon sampling, (2) using probabilistic
methods with complex models of sequence evolution that more accurately detect multiple substitutions, and (3)
removing the fastest evolving part of the data (e.g., species and positions). The analysis of a dataset of 55 animal
species and 102 proteins (25712 amino acid positions) shows that standard site-homogeneous model inference is
sensitive to long-branch attraction artifact, whereas the site-heterogeneous CAT model is less so. The latter model
correctly locates three very fast evolving species, the appendicularian tunicate Oikopleura, the acoel Convoluta
and the myxozoan Buddenbrockia. Overall, the resulting tree is in excellent agreement with the new animal
phylogeny, confirming that “simple” organisms like platyhelminths and nematodes are not necessarily of basal
emergence. This further emphasizes the importance of secondary simplification in animals, and for organismal
evolution in general.
Key words long-branch attraction (LBA) artifact, new animal phylogeny, phylogenomics, random error, sys-
tematic error.
The classical view of animal relationships was
essentially based on morphological considerations,
and was strongly influenced by the assumption of an
evolution towards an ever increasing complexity
(Brusca & Brusca, 1990). It generally focused on the
evolution of internal body cavities (coeloms) and
resulted in a phylogeny grouping mollusks, annelids,
arthropods and deuterostomes to the exclusion of
nematodes and platyhelminths (the Coelomata hy-
pothesis). Nevertheless, after some decades, it was
becoming apparent that morphological studies alone
could not reliably resolve relationships among the
major groups of animals. Great hope was therefore
placed in the newly arising field of molecular phylog-
eny, and its dominant marker 18S-like rDNA. In
contrast to the classical view, the new animal phylog-
eny, initially based on 18S rDNA (Aguinaldo et al.,
1997; Halanych et al., 1995), proposed two major
bilaterian groups, deuterostomes and protostomes, and
the subsequent division of protostomes into Ecdyso-
zoa and Lophotrochozoa (Adoutte et al., 2000). The
new taxonomy led to an intricate mixture of complex
and rather simple organisms, rendering the Coelomata
concept meaningless and implying that several organ-
isms must have become secondarily simplified over
evolutionary time (e.g., loss of coelom).
Small SubUnit (SSU) rDNA, although an excel-
lent phylogenetic marker, contains only ~1,000 align-
able positions, many of which are constant. Its resolv-
ing power is too limited to solidly infer all animal
relationships, because of the preponderance of sto-
chastic (or sampling) errors. An appealing solution lay
in using alternative, possibly longer, markers (essen-
tially proteins). The field was first dominated by
rather easily amplifiable mitochondrial genes like the
12S and 16S rDNA and protein encoding genes like
cox1, cytb, or nad2. Later, genes encoding highly
conserved nuclear proteins, like the translation elon-
gation factors 1 alpha (EF-1α) (Kobayashi et al., 1996;
McHugh, 1997) and 2 (EF-2) (Regier & Shultz, 2001),
the Na/K ATPase (Anderson et al., 2004), the largest
subunit of RNA Polymerase II (Sidow & Thomas,
1994), RAG1 (Groth & Barrowclough, 1999) and
RAG2 (Sullivan et al., 2000) were partially se-
quenced. Although a similar lack of resolution was
observed, results were generally congruent with the
rRNA tree (Halanych, 2004). Some conflicts exist
among these markers, but these can generally be
explained as stochastic errors. However, incongru-
ences can sometimes be well supported because they
are due to other reasons: (i) hidden paralogy (primarily

———————————
Received: 26 March 2008 Accepted: May 2008
* E-mail: .
** Author for correspondence. E-mail: Herve.Philippe@umontreal.ca; Tel.:
1-514-343-6720.
BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


275
in protein markers, but also in rDNA), (ii) horizontal
gene transfer (HGT), (iii) incomplete lineage sorting,
and (iv) systematic errors (tree reconstruction arti-
facts).
A possible way to simultaneously improve re-
solving power and solve the incongruence issue, is to
use multiple genes, especially in form of a concatena-
tion (Kluge, 1989). Concatenation of two or more
genes/gene fragments soon became the standard
approach (Shultz & Regier, 2000), although this only
slightly increases the resolving power. With progress
in sequencing technology, it became possible to
consider complete genome information (especially
mitochondrial genomes) or a large part of it (for
nuclear genomes). This approach was called phyloge-
nomics and we will now discuss in detail its strengths
and weaknesses.
1 Stochastic and systematic errors in phy-
logenomics
The use of a large number of genes (more than
100) provides more primary information, thereby
greatly reducing sampling errors. As a result, statisti-
cal support is greatly increased, compared with studies
based on single or few genes, often leading to trees
with a hundred percent bootstrap support (Blair et al.,
2002; Wolf et al., 2004). However, this unprecedented
level of resolving power (Delsuc et al., 2005) is
unfortunately accompanied by a greatly enhanced
sensitivity to systematic errors (Philippe et al., 2005a).
Systematic errors correspond to cases in which
tree reconstruction methods will recover an erroneous
tree more and more frequently, the more data that are
analyzed. Methods are then said to be inconsistent.
The most famous example is the inconsistency of
maximum parsimony when taxa evolve at a heteroge-
neous rate, the long-branch attraction artifact
(Felsenstein, 1978). Any systematic feature of the
sequences, such as compositional bias, will steadily
accumulate with the addition of more data and can
eventually become dominant. The consequence will
be strongly supported tree reconstruction artifacts,
e.g., grouping of species with similar G+C content
(Phillips et al., 2004). The strong support for artifac-
tual nodes induced by systematic errors in phyloge-
nomics contrasts with random errors in single gene
phylogenies, which are usually not strongly supported,
since they are not accumulative.
Probabilistic methods (either maximum likeli-
hood (ML) or Bayesian inference (BI)) are consistent,
meaning that they recover the correct solution, pro-
vided that the sequences have evolved according to
the process assumed by the model. Therefore, in a
probabilistic framework, tree reconstruction artifacts
can only occur if there is a violation of the underlying
model of sequence evolution by the data. It should
nevertheless be noted that probabilistic methods are
rather robust to model violations (Sullivan & Swof-
ford, 2001).
Model violations are always present, because
even a complex model cannot capture all the subtleties
of the real world. A good example is the composi-
tional heterogeneity of the sequences contained in an
alignment, such as species with a high (or low) GC
content. For example, a strong GC-pressure in a given
organism leads to a preferential mutation of A and T
towards G and C. This biased mutation process will
lead to an increased fixation of A or T to G or C
substitutions, especially at third codon positions. If
another unrelated organism is under the same influ-
ence, the probability is high that these two organisms
independently acquired the same character state at
many homologous positions (homoplasy). However, if
a model assumes that the evolutionary process is
stationary, i.e., is the same in all parts of the tree, it
will predict many fewer nucleotide positions with the
same character state in these two unrelated species
than observed and the probabilistic methods will
erroneously infer that the two GC-rich taxa are closely
related. This illustrates how model violations translate
into tree reconstruction artifacts, when the model
incorrectly interprets multiple substitutions (i.e., the
predictions of the model differ significantly from
reality, see (Lartillot et al., 2007)). In the case of two
GC rich species, disregarding the fast-evolving third
codon positions is often sufficient to overcome com-
positional artifact (Jeffroy et al., 2006).
This also explains why mutational saturation
(i.e., an excess of multiple substitutions) is so delete-
rious to phylogenetic inference. In a perfect world,
saturation would simply erase the ancient phyloge-
netic signal and lead to a lack of resolution in the
deepest nodes of the tree. However, especially in
phylogenomics, a subtle model violation such as
minor variations of amino acid composition may be
sufficient to introduce a bias in the interpretation of
multiple substitutions, hence creating a strong spuri-
ous signal, favoring an incorrect topology (Rodriguez-
Ezpeleta et al., 2007).
2 Detecting systematic errors
One of the main problems of phylogenomics is,
Journal of Systematics and Evolution Vol. 46 No. 3 2008

276
therefore, estimating whether systematic errors have
had an effect on the usually well-supported tree. First,
one looks at the properties of the alignment. For
instance, the level of saturation can be evaluated by
comparing the number of inferred substitutions with
the number of observed differences (Philippe et al.,
1994b), the heterogeneity of nucleotide/amino acid
composition (Rodriguez-Ezpeleta et al., 2007), and the
level of heterotachy (rate heterogeneity within a
position) (Baele et al., 2006; Lopez et al., 2002).
These analyses are only the starting point for further
experiments, such as the removal of the species with
the highest compositional bias. Indeed, sensitivity of
phylogenetic inference to taxon sampling is currently
the most efficient way to detect systematic errors. For
instance, using two different species to represent red
algae lead to two incongruent, yet strongly supported,
eukaryotic phylogenies (Rodriguez-Ezpeleta et al.,
2007). Or, in animals, platyhelminths clustered with
nematodes within Ecdysozoa, but are grouped with
Lophotrochozoa when nematodes were not included in
the analysis (Philippe et al., 2005b). Having a rich
taxon sampling therefore provides an effective way to
evaluate whether systematic errors are playing a role
in a phylogenomic analysis.
The first phylogenomic studies of animals (Blair
et al., 2002; Philip et al., 2005; Wolf et al., 2004) were
based on a very limited number of species (i.e., Homo,
Drosophila and Caenorhabditis) and focused on the
Coelomata (nematodes basal) versus Ecdysozoa
(nematodes together with insects) question. Since only
a single distant outgroup was available, the probability
of a long-branch attraction (LBA) artifact due to a
systematic error was high. Not surprisingly, these
early studies all placed the nematodes as basal with
solid support and came invariably to the conclusion
that the Coelomata hypothesis is correct (Blair et al.,
2002; Dopazo et al., 2004; Wolf et al., 2004). We will
use this example to illustrate the efficiency of three
methods (improved model, improved taxon sampling,
and removal of fast-evolving data) for overcoming
systematic errors.
Although LBA artifact was initially characterized
as a problem specific to parsimony (due to the statis-
tical inconsistency of maximum parsimony (MP) as a
reconstruction method), very similar phenomena are
often observed in probabilistic reconstructions as well.
In the latter, however, the causes are violations of one
or more of the hypotheses of the underlying models of
evolution, thus preventing a correct interpretation of
multiple substitutions that are concentrated along
branches of fast-evolving taxa. In the following, we
will use, for convenience, the term LBA artifact also
in the context of model violation problems.
3 Avoiding systematic errors by improving
the model of sequence evolution
Detecting systematic errors is important, but the
ultimate goal is to avoid them in the first place. One
way to do this is to improve the models of sequence
evolution to make them more realistic, thus reducing
model violations. The most useful improvements for
phylogenetic accuracy are those that allow a better
detection of multiple substitutions, not obligatorily the
ones that most improve the fit of the model to the data
(Yang, 1997). Numerous improvements since the
original, completely homogeneous, Jukes and Cantor
model (Jukes & Cantor, 1969) have been made to
include evolutionary heterogeneities.
The first model that assumed the positions of an
alignment to be heterogeneous (Yang, 1994) consti-
tuted a major improvement. The evolutionary rate of
all amino acid positions is estimated, and then divided
into discrete categories (often 4—here described with
+Γ4 to 16). The resulting distribution is assumed to
follow a gamma law and its shape parameter, alpha, is
estimated from the data and used to obtain a greatly
improved approximation of the evolutionary process.
The assumption of gamma distribution is especially
important because interpretation of the fastest evolv-
ing positions, which are the most prone to generate
misleading signals, are improved. Application of the
discrete gamma correction largely reduced the likeli-
hood of LBA-artifacts (Yang, 1996). More recently,
models incorporating qualitative heterogeneities in
addition to quantitative ones among positions consti-
tute another important advancement (Lartillot &
Philippe, 2004; Pagel & Meade, 2004).
In a Markovian model, the substitution process is
described by a so-called instantaneous rate matrix,
specifying the rate of replacement between each pair
of amino acids (or nucleotides). Standard models
assume that the same rate matrix can describe every
amino acid position of a protein. The exchange rates
of this matrix are obtained by empirical means, or are
directly estimated from the data. In general they are
biochemically meaningful, in that probabilities of
substitutions between biochemically similar amino
acids are much higher than between non-related pairs
of residues. For instance, models based on the WAG
matrix (Whelan & Goldman, 2001), which was estab-
lished empirically (based on alignments of globular
BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


277
proteins) allow one to predict more convergences and
reversions among amino acids with similar biochemi-
cal properties. However, the process reaches equilib-
rium after ~5 substitutions and is only dominated by
its equilibrium frequency vector over 20 amino acids:
which implies that, at saturated positions, each
amino-acid is supposed, by a site-homogeneous
model, to appear with a probability essentially re-
flecting its relative abundance in the overall alignment
(Lartillot et al., 2007).
The fact that site-homogeneous models are
dominated by equilibrium frequency vector is bio-
logically unsound; in all alignments many positions
seem to be saturated and yet still display very few
different amino acids, such as positions with only D or
E (or with only K or R), when the functional con-
straint is to have a negative (positive) charge at that
location. Standard homogeneous models will underes-
timate multiple substitutions at these positions, be-
cause they expect to see many different amino acids
after a few substitutions have occurred. In other word,
when confronted with highly saturated positions that
experienced multiple substitutions, this leads to an
inadequate behavior of the standard model (Lartillot et
al., 2007). In contrast, a mixture model such as the
CAT model (Lartillot & Philippe, 2004) allows deal-
ing with the amino acid specificity of positions by
sorting them into groups with various profiles of
amino acid stationary frequencies; since the size of the
alphabet is close to the actual value (e.g., two for a
D/E position), the CAT model predicts that about two
amino acids will be present at these positions, even
after numerous substitutions. Due to that property, the
CAT model is able to more correctly evaluate the
possibility of multiple conservative substitutions,
thereby successfully avoiding LBA artifacts where
other models fail, as detailed below.
Starting from a phylogenomic dataset of animals
that strongly rejects the Coelomata hypothesis
(Philippe et al., 2005b), Lartillot et al. (Lartillot et al.,
2007) reduced the species sampling to include only
fungi, arthropods, nematodes and deuterostomes.
Because a single distant outgroup (fungi) is used, a
spurious attraction of the fast-evolving nematodes
towards the base of the tree would be expected and
was indeed observed when using a site-homogeneous
model (e.g., WAG+Γ4). In contrast, under the same
difficult conditions, the CAT+Γ4 model resolves the
fast-evolving nematodes as a sister-group to arthro-
pods. Posterior predictive analysis suggests that the
CAT model is less sensitive to LBA artifacts than
standard site-homogeneous models, because of its
more realistic handling of highly saturated positions,
due to a more accurate estimate of the true alphabet
size per position.
Another interesting case concerning the phy-
logenetic position of the very fast-evolving acoels
represented by Convoluta was recently published. The
classical view of a platyhelminth affiliation is sup-
ported by an analysis of a phylogenomic dataset using
the WAG+Γ4 model (probably because of the rapid
evolutionary rate of platyhelminths). Interestingly, the
CAT+Γ4 model strongly rejects this grouping and
instead favors the acoels as an early part of the deu-
terostome radiation (Philippe et al., 2007), rather than
a sister-group of all other Bilateria, as suggested by
rRNA analysis (Ruiz-Trillo et al., 1999).
There are other heterogeneities of the evolution-
ary process that are now incorporated in models. For
instance, heterotachy can be handled by a covarion-
like model (Galtier, 2001; Huelsenbeck, 2002) or a
mixture of branch length model (Kolaczkowski &
Thornton, 2004). The first appears to have a better fit
(Zhou et al., 2007), whereas the second was shown to
avoid LBA artifacts (Kolaczkowski & Thornton,
2008). Similarly, various non-stationary models
(Blanquart & Lartillot, 2006; Foster, 2004; Galtier &
Gouy, 1995; Yang & Roberts, 1995) deal with com-
positional bias. To date, various model violations have
generally been addressed independently, mainly
because of mathematical difficulties and of computa-
tion time limitations. However, difficult cases, such as
the position of honeybees based on mitochondrial
genomes (Blanquart & Lartillot, 2008), require ad-
dressing several model violations simultaneously. The
use of more complex models could become prohibi-
tive, however, due to their excessive need of memory,
computation power and time.
4 Avoiding systematic errors by improving
the taxon sampling
Solely improving models is often insufficient to
adequately interpret multiple substitutions, because
complex models need much more data to estimate the
additional parameters. These parameters, such as
positional rate heterogeneity, require many species in
order to be accurately estimated. Moreover, increasing
the number of taxa, especially by breaking long
branches, is in itself an excellent way to more effi-
ciently detect multiple substitutions (Hendy & Penny,
1989; Hillis, 1996; Philippe & Douzery, 1994; Zwickl
& Hillis, 2002).
Journal of Systematics and Evolution Vol. 46 No. 3 2008

278
A good example illustrating the strong influence
of an improved taxon sampling in the context of
animal evolution is the taxon-dependent switch be-
tween two significantly supported and mutually
exclusive topologies (Delsuc et al., 2005). Starting
with the three classical ingroup species (i.e., Homo,
Drosophila and Caenorhabditis) and budding yeast as
a distant outgroup, a significant support for the
Coelomata topology (Homo with Drosophila) was
found. However, the addition of three rather slowly
evolving fungi and three closely related outgroups,
i.e., two choanoflagellates and a cnidarian, which
efficiently break the long branch leading to the bud-
ding yeast, leads to a significant support of the Ecdy-
sozoa topology (Drosophila with Caenorhabditis).
Another example of major improvement in phy-
logenetic performance that is associated with better
species sampling can be seen in the progression of
three papers that were all based on the same original
dataset, and using the same phylogenetic methods
(Baurain et al., 2007; Philippe et al., 2005b; Philippe
et al., 2004), with the major difference being the
successive addition of more animal and outgroup
sequences as they became available. The support for
protostomes was only marginal (Bootstrap Value
(BV) of 57%) and that for Ecdysozoa non-existant in
the original dataset, which contained only seven
non-fast-evolving bilaterian sequences and one cho-
anoflagellate (close outgroup) (Philippe et al., 2004).
The second dataset had a 100% BV for protostomes
(18 non-fast bilaterian sequences and four close
outgroups including two diploblasts) but only a 7%
BV for Ecdysozoa (Philippe et al., 2005b). Finally, the
third dataset, which contained 31 non-fast bilaterian
sequences and nine close outgroups, including two
sponges and four cnidarians, showed in addition to a
100% BV for protostomes, a 79 % BV for Ecdysozoa
(Baurain et al., 2007).
Improvement in taxon sampling to reduce the
impact of systematic error can be obtained by in-
creasing the number of species, but also by the exclu-
sion of rogue taxa (Sanderson & Shaffer, 2002).
Rogue taxa are generally fast-evolving but also evolve
differently from the other taxa (e.g., compositionally
biased or a different set of variable positions). They
therefore accumulate problems: numerous multiple
substitutions and numerous model violations. System-
atic errors can often be avoided by changing the
representative species of a group from a rogue taxon
to a “normal” species. For instance, with a limited
taxon sampling (23 species), the simple replacement
of the rogue, fast-evolving nematode, Caenorhabditis,
by a rather slowly evolving one, Xiphinema, was
sufficient to avoid the LBA artifact between nema-
todes and platyhelminths and to recover the mono-
phyly of Ecdysozoa (Baurain et al., 2007). Therefore,
sequencing a large number of taxa has two advan-
tages: breaking (long) branches and allowing the
selection of the slowest evolving representative spe-
cies.
One should also keep in mind that the outgroup
species, which are needed to root molecular phylog-
enies (at least with reversible models), are in essence
rogue taxa, because they diverged earlier and often
evolved under different evolutionary constraints.
Therefore, an easy way to reduce the systematic error
and one that should generally be applied is to perform
an additional analysis without the outgroup species
(Brinkmann et al., 2005).
There is a tradeoff between these two solutions
for a better phylogenetic inference: improved species
sampling versus the use of better models of sequence
evolution. Since with fewer species one needs a better
model to infer multiple changes that could be more
readily deduced from a denser species sampling, an
improved species sampling would in principle permit
the use of more simple models. Taxon sampling in
phylogenomics is nevertheless still too sparse at
present to evaluate this tradeoff in more detail.
5 Avoiding systematic errors by removing
fast-evolving data
Improving taxon sampling is not always possible,
mainly because of extinctions (e.g., no close relatives
are known for Amborella and Latimeria), and we
cannot be sure that models correctly interpret multiple
substitutions. An alternative method to avoid system-
atic error is to remove the data that evolve the fastest
or that violate most a given model. For instance, to
address the issue of compositional bias, one can easily
remove the third codon positions, which accumulate
the bias faster, or recode the sequences into a reduced
alphabet that is more homogeneous, e.g., RY coding
for purine and pyrimydine (Woese et al., 1991).
The most general approach is the selective re-
moval of the fastest evolving positions, since this
reduces the level of saturation and concomitantly
eliminates the positions that violate most a given
model. A simple way to achieve this positional sorting
is to use the affiliation to discrete gamma categories
(Ruiz-Trillo et al., 1999): the higher the number, the
faster the corresponding positions. Unfortunately, this

BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


279
approach is topology dependent, with different to-
pologies having an effect on the affiliation of posi-
tions and a strong impact on further analyses
(Rodriguez-Ezpeleta et al., 2007). Alternatively, one
can use a compatibility approach to improve the
quality of the input data without any a priori knowl-
edge of the phylogenetic relationships (Pisani, 2004;
Qiu & Estabrook, 2008). One identifies alignment
positions that are the least compatible with the other
positions and excludes them from the analysis, based
on the principle that their random behavior is due to
multiple substitutions. In other words, this approach is
based on the assumption that the positions that are the
most compatible in an alignment matrix share a
similar historical signal and evolve slowly. The SF
(slow-fast) method is also largely topology independ-
ent, since the evolutionary rate is inferred within
predefined monophyletic groups for which only the
intergroup-relationships are under study (Brinkmann
& Philippe, 1999). It was recently used to de-saturate
a phylogenomic animal data set, where the distant
outgroup (fungi) was suspected to erroneously attract
the fast-evolving nematodes to the base of Bilateria,
hence supporting the Coelomata hypothesis (Delsuc et
al., 2005). The progressive removal of fast-evolving
positions leads to a continuous decrease in the support
for Coelomata; lack of resolution due to the reduced
amount of data was not an explanation, since the
support for Ecdysozoa monophyly (arthropods +
nematodes) continuously increases.
One can also eliminate genes instead of posi-
tions, based on the idea that more information is
available to determine if they are fast evolving and
that one can selectively pinpoint problematic species
(Brinkmann et al., 2005). A good example is provided
by a phylogenomic analysis that provides strong
support in favor of the strange grouping of platy-
helminths+nematodes, possibly due to an ingroup
LBA artifact (Philippe et al., 2005b). The progressive
elimination of the proteins that evolved the fastest in
platyhelminths and nematodes regularly decreases the
support for platyhelminths+nematodes, and increases
the one for the grouping of platyhelminths with Lo-
photrochozoa.
Although powerful, the removal of fast-evolving
data (positions, genes and species) is potentially
dangerous, because one needs to make at least some
assumptions about the phylogeny and the evolutionary
model to estimate the rate. These assumptions may
bias the output of the data removal (Rodriguez-Ezpe-
leta et al., 2007). This approach should therefore
mainly be reserved to exploratory analyses or when
long branches cannot be broken.
6 Phylogenomics and lack of resolution
Because of systematic errors, a high statistical
support in phylogenomics can be misleading, espe-
cially when few species are considered (Hedtke et al.,
2006; Jeffroy et al., 2006). Interestingly, a weak
support can also be created by systematic errors if the
strength of the phylogenetic and of the artifactual
signal are about the same (Rodriguez-Ezpeleta et al.,
2007). Coming back to the animal phylogeny, it was
proposed that the major lines of bilaterian evolution
can not be resolved by a phylogenetic approach
(Rokas et al., 2005), since they result from a rapid and
massive adaptive radiation, the so-called “Cambrian
explosion”. It is indeed well known that resolving
closely spaced speciation events could require a very
large amount of data (Philippe et al., 1994a; Saitou &
Nei, 1986) but also that noise (i.e., mutational satura-
tion) can seriously reduce the resolving power of
molecular phylogenetics. Therefore the lack of resolu-
tion observed by Rokas et al. (2005) could alterna-
tively be due to mutational saturation. This explana-
tion is suggested by (i) the use of several rogue taxa as
well as a rather poor species sampling and (ii) the use
of too simplistic phylogenetic inference methods. We
recently applied two of the approaches detailed above
simultaneously to reduce systematic error: (i) the use
of an improved taxon sampling (from 13 to 46 ani-
mals, with elimination of some rogue taxa, e.g.
Caenorhabditis or Drosophila), and (ii) the use of a
better method (CAT+Γ4 model). Interestingly, this
resulted in a greatly improved statistical support, most
of the nodes within Bilateria being supported by a
high bootstrap value, e.g., Ecdysozoa and Lophotro-
chozoa (Baurain et al., 2007). This improved resolu-
tion is best explained by the more efficient extraction
of the phylogenetic signal that results in a better signal
to noise ratio. This study illustrates that, even when
considering a very large dataset, interpreting lack of
resolution as evidence of radiation is hazardous. It is
of prime importance to first exclude the possibility
that a systematic error that erased part of the phyloge-
netic signal is not the explanation.
7 An updated phylogenomic tree of animals
The ongoing major sequencing efforts at the level
of both complete genomes and expressed sequence
tags (ESTs) allows the construction of more complete
Journal of Systematics and Evolution Vol. 46 No. 3 2008

280
datasets with less missing positions and also with a
largely improved species sampling (Philippe & Tel-
ford, 2006). Here we updated alignments used in three
recent studies (Delsuc et al., 2006; Jimenez-Guri et al.,
2007; Philippe et al., 2007). A phylogenomic dataset
assembled using SCAFOS (Roure et al., 2007) (55
species, 102 genes and 25712 amino acid positions)
contains the major metazoan lineages and two inde-
pendent and closely related outgroups, choanoflagel-
lates and ichthyosporeans. For the first time, three
very fast-evolving species, the appendicularian tuni-
cate Oikopleura dioica, the acoel Convoluta pulchra
and the myxozoan Buddenbrockia plumatellae, are
simultaneously included. These are all rogue species,
which could induce LBA artifacts. We therefore used
a rich taxon sampling and a complex model of se-
quence evolution (CAT+Γ4).
The Bayesian tree shown in Fig. 1, inferred by
Phylobayes using a CAT+Γ4 model, is in good agree-
ment with current knowledge. First, in contrast to
what would be expected if LBA plays a dominant
role, the three very fast evolving species are not
clustered together. Second, the tree is in agreement
with the new animal phylogeny, especially the mono-
phyly of Ecdysozoa and Lophotrochozoa is recovered.
Poriferans are the most basal monophyletic group of
metazoans and cnidarians emerge shortly thereafter.
Cnidarians are deeply divided into two major and
highly discrete monophyletic groups, the anthozoans
represented here by sea anemones (Nematostella) and
stony corals (Acropora) and two additional cnidarian
groups hydrozoans (Hydra and Hydractinia) and
scyphozoans (Cyanea) as well as the divergent se-
quence of the myxozoan Buddenbrockia plumatellae,
in agreement with a recent work (Jimenez-Guri et al.,
2007). However, even with the site-heterogeneous
CAT model, which is less sensitive to LBA (Lartillot
et al., 2007), the area surrounding the position of a
very fast evolving species is usually associated with
an enhanced uncertainty that is reflected here in the
low posterior probability (PP) value of 0.88. Alterna-
tively this phenomenon may be seen as being related
to the much greater error margin associated with the
estimate of its very long branch.
The bilaterian part of the metazoan tree (Fig. 1)
has a pronounced basal branch that separates it from
the other metazoans, this may indicate either a long
evolutionary period and/or a shorter phase of acceler-
ated evolution potentially associated with the origin of
bilaterian animals. This tree supports a division of
bilaterians into the two major groups deuterostomes
and protostomes. The location of the very fast evolv-
ing acoel Convoluta and of the chaetognath Spadella
at the base of protostomes is interesting since their
phylogenetic position is a longstanding and ongoing
dispute. The chaetognaths were originally considered
to be part of the deuterostomes, essentially based on
morphological characters and on their deuterostome-
like development. Our analysis confirms the pro-
tostome affinity of chaetognaths, as the sister-group to
Ecdysozoa + Lophotrochozoa (Marletaz et al., 2006),
and not to Lophotrochozoa (Matus et al., 2006). If
chaetognaths are the most basal protostomes, one
could consider their unusual features as an indication
for the potential ancestry of the deuterostome devel-
opmental program and of deuterostome-like organ-
isms within bilaterians (Marletaz et al., 2006).
The evolutionary position of the acoel flatworms
is much more disputed. Classically they were consid-
ered to be a part of the platyhelminths, i.e. flatworms.
However, several molecular phylogenetic analyses
(rRNA, myosin, and mitochondrial genome) found
strong support for a much more basal position within
the metazoan tree as the sister-group to all extant
bilaterian animals (Baguna & Riutort, 2004; Ruiz-
Trillo et al., 2002; Ruiz-Trillo et al., 2004; Ruiz-Trillo
et al., 1999). A major problem of these studies is the
very fast evolutionary rate of the acoels,which could
easily lead to an LBA artifact, with acoels being
attracted by the distantly related non-bilaterian out-
group. An EST based phylogenomic analysis using
the CAT+Γ4 model significantly rejects a sister-group
relationship to the platyhelminths, but did not recover
acoels as the most basal bilaterian, suggesting instead
that they are rather close to, or included within, deu-
terostomes (Philippe et al., 2007). The current analysis
(Fig. 1) confirms the rejection of the classical platy-
helminths link, but suggests a slightly different posi-
tion, as a sister-group of the protostomes. The moder-
ate support (PP 0.88) argues in favor of an improved
taxon sampling within acoels before making any firm
conclusion.
Although, the monophyly of both deuterostomes
and chordates is recovered in our analysis (Fig. 1), the
basal and the subsequent branches of deuterostomes
are very short suggesting that the diversification of
major lineages, such as chordates or cephalochordates,
could have happened rapidly after the origin of the
deuterostomes themselves (at least more rapidly than
within protostomes, where the branches at the base of
Lophotrochozoa and Ecdysozoa are longer). Accord-
ingly, a very similar analysis (same inference method
and similar set of genes and of species) did not re-
cover the monophyly of deuterostomes, and the
BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


281




Fig. 1. The tree was inferred based on a phylogenomic dataset (102 genes and 25712 amino acid positions) using the program phylobayes (Bayesian
inference) and the CAT model with four discrete gamma categories (CAT+Γ4) after elimination of the constant positions. The methods are the same
as in (Lartillot & Philippe, 2008). Posterior probability values are only indicated if they are lower than 1.


Journal of Systematics and Evolution Vol. 46 No. 3 2008

282
chordates were inferred as the first emerging bilaterian
group with the remaining deuterostomes forming the
sister-group to the protostomes (Lartillot & Philippe,
2008). This high sensitivity to species/gene sampling
indicates that the monophyly of deuterostomes is an
open question, requiring all the improvements in
phylogenomic inference discussed above to be solved
(in particular, improved sampling within Xenambu-
lacraria and improved models). An important reason
explaining this difficulty is that only a distant out-
group is available, hence the deuterostome monophyly
could be due to an attraction of protostomes by the
non-bilaterian outgroup (Lartillot & Philippe, 2008).
The topology within “core” protostomes does not
reveal any surprises. Within Ecdysozoa the arthropods
represent the sister-group of nematodes and tardi-
grades (water bears); within arthropods there is a basal
division between chelicerates and the pan-crustaceans
with paraphyletic crustaceans, i.e., with the mala-
costracan Litopenaeus basal and the phyllopod Daph-
nia as the sister-group to insects, here represented by
four species of winged insects (Pterygota). Among the
three lophotrochozoan groups represented here, the
traditional view of a sister-group relationship between
the two trochozoans annelids and mollusks (Neotro-
chozoa) is rejected by the data. Instead a grouping
between annelids and platyhelminths is supported (PP
0.99), suggesting that the flatworms are also trocho-
zoans. This unorthodox relationship was also recov-
ered in several recent analyses (Lartillot & Philippe,
2008; Lavrov & Lang, 2005; Passamaneck & Ha-
lanych, 2006). However, caution is necessary using
this interpretation since many protostome lineages
(e.g. priapulids, onychophorans, rotiferans, acantho-
cephalans, nematomorphans, bryozoans, phoronids,
brachiopods, sipunculans, nemertines, pogonophores
and echiurids) are not present in our analysis (see
(Dunn et al., 2008) for an improvement of taxon
sampling).
Finally, we studied both the impact of the model
of sequence evolution and of the heuristic search. In
addition to phylobayes (CAT+Γ4; Fig. 1), we per-
formed standard phylogenetic analyses with a site-
homogeneous model (WAG+Γ4), employing widely
used software, including MrBayes (Ronquist & Huel-
senbeck, 2003), PhyML (Guindon & Gascuel, 2003),
PhyML-SPR (subtree pruning and regrafting) (Hordijk
& Gascuel, 2005), Treefinder (Jobb et al., 2004),
RaxML (Stamatakis et al., 2005) and IQPNNI (Vinh
le & Von Haeseler, 2004). Twelve different topologies
(Table 1) were obtained and compared, assuming the
WAG+Γ4 model, using the program Tree-Puzzle
(Schmidt et al., 2002) to perform likelihood ratio tests,
including the Shimodaira-Hasegawa and the Expected
Likelihood Weight (ELW) test. The most obvious

Table 1 Estimation of the quality of the model of sequence evolution and of the heuristic search
Software Observed differences ΔLnL S-H ELW
IQPNNI (C(E((Ac,P),(M,A)))) 4.60 0.94 0.34
MrBayes (E((C,P),(M,A))) Ac+My basal to Bilateria 83.07 0.36 0.02 #
“ (C(E(((Ac,My)P),(M,A)))) 69.38 0.48 0.02 #
“ (E((P,C),(M,A))) Ac basal to Bilateria 59.00 0.52 0.03 #
“ (E(C(((Ac,My)P),(M,A)))) 72.10 0.46 0.03 #
“ (C(E(((Ac,My)P),(M,A)))) 69.38 0.48 0.02 #
“ ((Ac,C),(E(P(M,A)))) 55.59 0.55 0.02 #
Treefinder ((E,C),((((Ac,My)T)P),(M,A))) 817.21 0.00 # 0.00 #
PhyML (C(E((Ac,P),(M,A)))) My+T basal to Meta 787.95 0.00 # 0.00 #
PhyML-SPR (E(((Ac,P)C),(M,A))) 0.00 1.00
RAxML-MP (E(((Ac,P)C),(M,A))) 0.00 1.00
RAxML-6xRan (E(((Ac,P)C),(M,A))) 0.00 1.00
RAxML-1xRan (Cni(C(E((M,A) P)))) T basal to Xamb; D sister-group to other Meta;
Ac basal to Meta
1322.69 0.00 # 0.00 #
RAxML-1xRan (C(E(Ac,P),( M,A)))) T basal to Bilateria 368.97 0.00 # 0.00 #
RAxML-2xRan (E(((Ac,P)C),(M,A))) Agnaths not mono 249.09 0.02 0.00 #
Phylobayes (Ac(C(E(M(A,P))))) 126.54 0.18 0.00 #
Abbreviations: A, annelids; Ac, acoels; C, chaetognaths; Cni, Cnidarians; D, deuterostomes; E, Ecdysozoa; M, mollusks; Meta, metazoa; mono,
monophyletic; My, myxozoans; P, platyhelminths; T, Tunicate; Xamb, Xenambulacraria.
Six independent chains were run for MrBayes and produced six different topologies. Ten random trees were tested for RAxML. Note that all compu-
tations were done with the WAG+Γ4 model, except for phylobayes (CAT+Γ4 model). In that case, topological difference is likely due to the model,
and not to tree search, because four independent chains gave the same topology. # indicates that the difference is significant (P<0.05).
BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


283
result is the major impact of heuristic search; only two
software (PHYML-SPR and RAxML) found the same
topology. Some software (Treefinder and PHYML)
have been trapped in distant local minima (~800 log
likelihood values of difference). This confirms that
heuristic search strategies are potentially problematic
when very long sequences are used, even if the num-
ber of taxa (55) is relatively small: the potential
barriers between local minima are high in phyloge-
nomics and difficult to overcome. MrBayes clearly
illustrates this problem, because six independent
MCMCMC chains remain trapped (during ~500,000
generations) in different, albeit close, local minima. It
should be noted that our best tree, despite being found
twice independently, is not necessarily the ML tree,
because an exhaustive search was not performed.
This best ML tree under the WAG+Γ4 model,
however, contains nodes that are very likely the result
of an LBA artifact, e.g. the node that unites the acoel
Convoluta, platyhelminths, and the chaetognath
Spadella. These three taxa correspond to long, unbro-
ken, branches that are not grouped with the CAT+Γ4
model (Fig. 1). Moreover, to our knowledge, there are
no morphological characters to support the grouping
of chaetognaths with acoels or platyhelminths. The
CAT topology is, under the WAG model, far from
being the best tree, since it has a log likelihood value
more than 125 units worse and is even rejected by the
ELW test. Since several studies have demonstrated a
much better fit for the CAT than for the WAG model
(Lartillot et al., 2007; Lartillot & Philippe, 2004;
Lartillot & Philippe, 2008), the CAT topology should
be preferred, even if we cannot be sure that this
topology is the best under that model (although it has
been found in four chains independently).
8 Biological implications
The new animal phylogeny that assumes a basal
division in deutero- and protostomes (with Lophotro-
chozoa and Ecdysozoa) proposed by rRNA analysis is
now confirmed, in some detail, by phylogenomics and
can be used with confidence. Comparison with the
older view of metazoan evolution essentially repre-
sented in the Coelomata hypothesis illustrates the
strong influence of the “great chain of being” of
Aristotle on scientists (Adoutte et al., 1999). Almost
all of the major differences between the two classifi-
cation schemes involve simple organisms, of which
the best known groups are platyhelminths (acoelo-
mates) and nematodes (pseudocoelomates) which
were considered to be the two oldest lineages within
Bilateria under the Coelomata hypothesis. These
groups are not primitively simple, but are the product
of secondary simplification, which transforms for-
merly rather complex organisms into more or less
simple ones that look at least superficially primitive.
Interestingly, many “simple” animals (e.g., nema-
todes, platyhelminths, acoels, myzostomids, myxozo-
ans) are often fast-evolving from a molecular point of
view, suggesting that their secondary simplification,
which implies gene losses, would have relaxed func-
tional constraints. The same phenomenon is observed
in eukaryotes, especially when mitochondria are lost
(Philippe et al., 2000). The absence of simple and
early-branching bilaterian animals strongly suggests
that the urbilaterian ancestor was rather complex. This
does not imply that it appears suddenly from a simpler
ancestor, but instead that, among the intermediate,
simpler, ancestral species, only one lineage survives.
The strong influence of the “great chain of being”
is harmful. The complex thinking of Aristotle, elabo-
rated long before the Darwinian theory, was unfortu-
nately too often reduced to the implicit assumption
that evolution invariantly leads towards more com-
plexity. Molecular phylogenetics refutes this hypothe-
sis, by demonstrating that “simple” organisms almost
invariantly emerge from within “complex” organisms,
not at their base, hence being secondarily simplified.
Unfortunately, the erroneous “great chain of being”
ideology remains influential, especially thanks to the
incorrect use of the adjectives “higher” and “lower” to
characterize taxa, a practice that should be prohibited
(Mogie, 2007). Instead of uselessly searching for
simple and primitive organisms to understand evolu-
tion, one should rather look for simple evolutionary
mechanisms that could explain the characters of
complex and simple extant organisms.
9 Perspectives
The guidelines we suggest to obtain reliable
phylogenomic trees (Philippe et al., 2005a) are, not
surprisingly, very similar to those recommended to
obtain a reliable single gene phylogeny (Sanderson &
Shaffer, 2002): (1) increase the taxon sampling, with
at least two species per group of interest (although the
evolutionary process renders it often inherently scarce
by the extinction of related species), (2) improve the
models of sequence evolution (especially the proper-
ties that enhance detection of multiple substitutions),
and (3) remove the fastest-evolving data, especially
species (although the last approach is potentially
dangerous, and intellectually not satisfying, since the
Journal of Systematics and Evolution Vol. 46 No. 3 2008

284
first two points should be sufficient).
We have only explored phylogenomic inference
based on primary sequences (in fact, only on su-
per-matrices, but super-trees generally give similar
results), because this approach is the most advanced.
This offers the unique opportunity to explore the
strengths and weaknesses of genome-based inference.
Several fundamentally different approaches have been
proposed (for review see Delsuc et al., 2005), such as
similarity in oligonucleotide frequencies (not depend-
ent on homology), gene presence/absence, or gene
order. Although they are promising (especially gene
order because character space is huge, greatly reduc-
ing the possibility of homoplasy), the inference meth-
ods need to be improved. The use of intron positions
is illustrative, by supporting either Coelomata (Zheng
et al., 2007) or Ecdysozoa (Roy & Gilbert, 2005),
depending only on the methods used. In the long run,
these alternative phylogenomic approaches are ex-
pected to furnish an important corroboration, since
they are derived from completely independent charac-
ters that evolve according to their proper rules.
Acknowledgements This work was supported by
operating funds from the Natural Sciences and Engi-
neering Research Council and the Canada Research
Chairs Program. I wish to thank Nicolas Lartillot for
discussions and helpful comments.
References
Adoutte A, Balavoine G, Lartillot N, de Rosa R. 1999. Animal
evolution. The end of the intermediate taxa? Trends in
Genetics 15: 104–108.
Adoutte A, Balavoine G, Lartillot N, Lespinet O, Prud’homme
B, de Rosa R. 2000. The new animal phylogeny: reliability
and implications. Proceedings of the National Academy of
Sciences USA 97: 4453–4456.
Aguinaldo AM, Turbeville JM, Linford LS, Rivera MC, Garey
JR, Raff RA, Lake JA. 1997. Evidence for a clade of
nematodes, arthropods and other moulting animals. Nature
387: 489–493.
Anderson FE, Cordoba AJ, Thollesson M. 2004. Bilaterian
phylogeny based on analyses of a region of the
sodium-potassium ATPase beta-subunit gene. Journal of
Molecular Evolution 58: 252–268.
Baele G, Raes J, Van de Peer Y, Vansteelandt S. 2006. An
improved statistical method for detecting heterotachy in
nucleotide sequences. Molecular Biology and Evolution
23: 1397–1405.
Baguna J, Riutort M. 2004. The dawn of bilaterian animals: the
case of acoelomorph flatworms. Bioessays 26: 1046–
1057.
Baurain D, Brinkmann H, Philippe H. 2007. Lack of resolution
in the animal phylogeny: closely spaced cladogeneses or
undetected systematic errors? Molecular Biology and
Evolution 24: 6–9.
Blair JE, Ikeo K, Gojobori T, Hedges SB. 2002. The
evolutionary position of nematodes. BMC Evolutionary
Biology 2: 7.
Blanquart S, Lartillot N. 2006. A Bayesian compound
stochastic process for modeling nonstationary and
nonhomogeneous sequence evolution. Molecular Biology
and Evolution 23: 2058–2071.
Blanquart S, Lartillot N. 2008. A Site- and Time-Heterogeneous
Model of Amino-Acid Replacement. Molecular Biology
and Evolution 25: 842–858.
Brinkmann H, Giezen M, Zhou Y, Raucourt GP, Philippe H.
2005. An empirical assessment of long-branch attraction
artefacts in deep eukaryotic phylogenomics. Systematic
Biology 54: 743–757.
Brinkmann H, Philippe H. 1999. Archaea sister group of
Bacteria? Indications from tree reconstruction artifacts in
ancient phylogenies. Molecular Biology and Evolution 16:
817–825.
Brusca RC, Brusca GJ. 1990. Invertebrates. Sunderland, MA:
Sinauer Associates.
Delsuc F, Brinkmann H, Chourrout D, Philippe H. 2006.
Tunicates and not cephalochordates are the closest living
relatives of vertebrates. Nature 439: 965–968.
Delsuc F, Brinkmann H, Philippe H. 2005. Phylogenomics and
the reconstruction of the tree of life. Nature Reviews
Genetics 6: 361–375.
Dopazo H, Santoyo J, Dopazo J. 2004. Phylogenomics and the
number of characters required for obtaining an accurate
phylogeny of eukaryote model species. Bioinformatics 20:
i116–i121.
Dunn CW, Hejnol A, Matus DQ, Pang K, Browne WE, Smith
SA, Seaver E, Rouse GW, Obst M, Edgecombe GD,
Sorensen MV, Haddock SH, Schmidt-Rhaesa A, Okusu A,
Kristensen RM, Wheeler WC, Martindale MQ, Giribet G.
2008. Broad phylogenomic sampling improves resolution
of the animal tree of life. Nature 452: 745–749.
Felsenstein J. 1978. Cases in which parsimony or compatibility
methods will be positively misleading. Systematic
Zoology 27: 401–410.
Foster PG. 2004. Modeling compositional heterogeneity.
Systematic Biology 53: 485–495.
Galtier N. 2001. Maximum-likelihood phylogenetic analysis
under a covarion-like model. Molecular Biology and
Evolution 18: 866–873.
Galtier N, Gouy M. 1995. Inferring phylogenies from DNA
sequences of unequal base compositions. Proceedings of
the National Academy of Sciences USA 92: 11317–11321.
Groth JG, Barrowclough GF. 1999. Basal divergences in birds
and the phylogenetic utility of the nuclear RAG-1 gene.
Molecular Phylogenetics and Evolution 12: 115–123.
Guindon S, Gascuel O. 2003. A simple, fast, and accurate
algorithm to estimate large phylogenies by maximum
likelihood. Systematic Biology 52: 696–704.
Halanych KM. 2004. The new view of animal phylogeny.
Annual Review of Ecology, Evolution, and Systematics
35: 229–256.
Halanych KM, Bacheller JD, Aguinaldo AM, Liva SM, Hillis
DM, Lake JA. 1995. Evidence from 18S ribosomal DNA
that the lophophorates are protostome animals. Science
267: 1641–1643.
BRINKMANN & PHILIPPE: Animal phylogeny and large-scale sequencing


285
Hedtke SM, Townsend TM, Hillis DM. 2006. Resolution of
phylogenetic conflict in large data sets by increased taxon
sampling. Systematic Biology 55: 522–529.
Hendy MD, Penny D. 1989. A framework for the quantitative
study of evolutionary trees. Systematic Zoology 38:
297–309.
Hillis DM. 1996. Inferring complex phylogenies. Nature 383:
130–131.
Hordijk W, Gascuel O. 2005. Improving the efficiency of SPR
moves in phylogenetic tree search methods based on
maximum likelihood. Bioinformatics 21: 4338–4347.
Huelsenbeck JP. 2002. Testing a covariotide model of DNA
substitution. Molecular Biology and Evolution 19:
698–707.
Jeffroy O, Brinkmann H, Delsuc F, Philippe H. 2006.
Phylogenomics: the beginning of incongruence? Trends in
Genetics 22: 225–231.
Jimenez-Guri E, Philippe H, Okamura B, Holland PW. 2007.
Buddenbrockia is a cnidarian worm. Science 317: 116–118.
Jobb G, von Haeseler A, Strimmer K. 2004. TREEFINDER: a
powerful graphical analysis environment for molecular
phylogenetics. BMC Evolutionary Biology 4: 18.
Jukes TH, Cantor CR. 1969. Evolution of protein molecules.
In: Munro HN ed. Mammalian protein metabolism. New
York: Academic Press.
Kluge AG. 1989. A concern for evidence and a phylogenetic
hypothesis of relationships among Epicrates (Boidae,
Serpentes). Systematic Zoology 38: 7–25.
Kobayashi M, Wada H, Satoh N. 1996. Early evolution of the
Metazoa and phylogenetic status of diploblasts as inferred
from amino acid sequence of elongation factor-1 alpha.
Molecular Phylogenetics and Evolution 5: 414–422.
Kolaczkowski B, Thornton JW. 2004. Performance of
maximum parsimony and likelihood phylogenetics when
evolution is heterogeneous. Nature 431: 980–984.
Kolaczkowski B, Thornton JW. 2008. A mixed branch length
model of heterotachy improves phylogenetic accuracy.
Molecular Biology and Evolution. (in press)
Lartillot N, Brinkmann H, Philippe H. 2007. Suppression of
long-branch attraction artefacts in the animal phylogeny
using a site-heterogeneous model. BMC Evolutionary
Biology 7 Suppl 1: S4.
Lartillot N, Philippe H. 2004. A Bayesian mixture model for
across-site heterogeneities in the amino-acid replacement
process. Molecular Biology and Evolution 21: 1095–1109.
Lartillot N, Philippe H. 2008. Improvement of molecular
phylogenetic inference and the phylogeny of Bilateria.
Philosophical Transactions of the Royal Society B:
Biological Sciences 363: 1463–1472.
Lavrov DV, Lang BF. 2005. Poriferan mtDNA and animal
phylogeny based on mitochondrial gene arrangements.
Systematic Biology 54: 651–659.
Lopez P, Casane D, Philippe H. 2002. Heterotachy, an
important process of protein evolution. Molecular Biology
and Evolution 19: 1–7.
Marletaz F, Martin E, Perez Y, Papillon D, Caubit X, Lowe CJ,
Freeman B, Fasano L, Dossat C, Wincker P, Weissenbach
J, Le Parco Y. 2006. Chaetognath phylogenomics: a
protostome with deuterostome-like development. Current
Biology 16: R577–578.
Matus DQ, Copley RR, Dunn CW, Hejnol A, Eccleston H,
Halanych KM, Martindale MQ, Telford MJ. 2006. Broad
taxon and gene sampling indicate that chaetognaths are
protostomes. Current Biology 16: R575–576.
McHugh D. 1997. Molecular evidence that echiurans and
pogonophorans are derived annelids. Proceedings of the
National Academy of Sciences USA 94: 8006–8009.
Mogie M. 2007. Drop “higher” and “lower” to raise descriptive
standards. Nature 448: 865.
Pagel M, Meade A. 2004. A phylogenetic mixture model for
detecting pattern-heterogeneity in gene sequence or
character-state data. Systematic Biology 53: 571–581.
Passamaneck Y, Halanych KM. 2006. Lophotrochozoan
phylogeny assessed with LSU and SSU data: evidence of
lophophorate polyphyly. Molecular Phylogenetics and
Evolution 40: 20–28.
Philip GK, Creevey CJ, McInerney JO. 2005. The
Opisthokonta and the Ecdysozoa may not be clades:
Stronger support for the grouping of plant and animal than
for animal and fungi and stronger support for the
Coelomata than Ecdysozoa. Molecular Biology and
Evolution 22: 1175–1184.
Philippe H, Brinkmann H, Martinez P, Riutort M, Baguna J.
2007. Acoel flatworms are not Platyhelminthes: evidence
from phylogenomics. PLOS One 2: e717.
Philippe H, Chenuil A, Adoutte A. 1994a. Can the Cambrian
explosion be inferred through molecular phylogeny?
Development 120: S15–S25.
Philippe H, Delsuc F, Brinkmann H, Lartillot N. 2005a.
Phylogenomics. Annual Review of Ecology, Evolution,
and Systematics 36: 541–562.
Philippe H, Douzery E. 1994. The pitfalls of molecular
phylogeny based on four species, as illustrated by the
Cetacea/Artiodactyla relationships. Journal of Mammalian
Evolution 2: 133–152.
Philippe H, Germot A, Moreira D. 2000. The new phylogeny of
eukaryotes. Current Opinion in Genetics and Development
10: 596–601.
Philippe H, Lartillot N, Brinkmann H. 2005b. Multigene
analyses of bilaterian animals corroborate the monophyly
of Ecdysozoa, Lophotrochozoa, and Protostomia.
Molecular Biology and Evolution 22: 1246–1253.
Philippe H, Snell EA, Bapteste E, Lopez P, Holland PW,
Casane D. 2004. Phylogenomics of eukaryotes: impact of
missing data on large alignments. Molecular Biology and
Evolution 21: 1740–1752.
Philippe H, Sörhannus U, Baroin A, Perasso R, Gasse F,
Adoutte A. 1994b. Comparison of molecular and
paleontological data in diatoms suggests a major gap in
the fossil record. Journal of Evolutionary Biology 7:
247–265.
Philippe H, Telford MJ. 2006. Large-scale sequencing and the
new animal phylogeny. Trends in Ecology & Evolution
21: 614–620.
Phillips MJ, Delsuc F, Penny D. 2004. Genome-scale
phylogeny and the detection of systematic biases.
Molecular Biology and Evolution 21: 1455–1458.
Pisani D. 2004. Identifying and removing fast-evolving sites
using compatibility analysis: An example from the
arthropoda. Systematic Biology 53: 978–989.
Qiu Y-L, Estabrook GF. 2008. Inference of phylogenetic
relationships among key angiosperm lineages using a
Journal of Systematics and Evolution Vol. 46 No. 3 2008

286
compatibility method on a molecular data set. Journal of
Systematics and Evolution 46: 130–141.
Regier JC, Shultz JW. 2001. Elongation factor-2: a useful gene
for arthropod phylogenetics. Molecular Phylogenetics and
Evolution 20: 136–148.
Rodriguez-Ezpeleta N, Brinkmann H, Roure B, Lartillot N,
Lang BF, Philippe H. 2007. Detecting and overcoming
systematic errors in genome-scale phylogenies. Systematic
Biology 56: 389–399.
Rokas A, Kruger D, Carroll SB. 2005. Animal evolution and
the molecular signature of radiations compressed in time.
Science 310: 1933–1938.
Ronquist F, Huelsenbeck JP. 2003. MrBayes 3: Bayesian
phylogenetic inference under mixed models.
Bioinformatics 19: 1572–1574.
Roure B, Rodriguez-Ezpeleta N, Philippe H. 2007. SCaFoS: a
tool for Selection, Concatenation and Fusion of Sequences
for phylogenomics. BMC Evolutionary Biology 7 Suppl 1:
S2.
Roy SW, Gilbert W. 2005. Resolution of a deep animal
divergence by the pattern of intron conservation.
Proceedings of the National Academy of Sciences USA
102: 4403–4408.
Ruiz-Trillo I, Paps J, Loukota M, Ribera C, Jondelius U,
Baguna J, Riutort M. 2002. A phylogenetic analysis of
myosin heavy chain type II sequences corroborates that
Acoela and Nemertodermatida are basal bilaterians.
Proceedings of the National Academy of Sciences USA
99: 11246–11251.
Ruiz-Trillo I, Riutort M, Fourcade HM, Baguna J, Boore JL.
2004. Mitochondrial genome data support the basal
position of Acoelomorpha and the polyphyly of the
Platyhelminthes. Molecular Phylogenetics and Evolution
33: 321–332.
Ruiz-Trillo I, Riutort M, Littlewood DT, Herniou EA, Baguna
J. 1999. Acoel flatworms: earliest extant bilaterian
Metazoans, not members of Platyhelminthes. Science 283:
1919–1923.
Saitou N, Nei M. 1986. The number of nucleotides required to
determine the branching order of three species, with
special reference to the human-chimpanzee-gorilla
divergence. Journal of Molecular Evolution 24: 189–204.
Sanderson MJ, Shaffer HB. 2002. Troubleshooting molecular
phylogenetic analyses. Annual Review of Ecology and
Systematics 33: 49–72.
Schmidt HA, Strimmer K, Vingron M, von Haeseler A. 2002.
TREE-PUZZLE: maximum likelihood phylogenetic
analysis using quartets and parallel computing.
Bioinformatics 18: 502–504.
Shultz JW, Regier JC. 2000. Phylogenetic analysis of
arthropods using two nuclear protein-encoding genes
supports a crustacean + hexapod clade. Proceedings of the
Royal Society: Biological Sciences 267: 1011–1019.
Sidow A, Thomas WK. 1994. A molecular evolutionary
framework for eukaryotic model organisms. Current
Biology 4: 596–603.
Stamatakis A, Ludwig T, Meier H. 2005. RAxML-III: a fast
program for maximum likelihood-based inference of large
phylogenetic trees. Bioinformatics 21: 456–463.
Sullivan J, Swofford DL. 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? Systematic
Biology 50: 723–729.
Sullivan JP, Lavoue S, Hopkins CD. 2000. Molecular
systematics of the African electric fishes (Mormyroidea:
Teleostei) and a model for the evolution of their electric
organs. Journal of Experimental Biology 203: 665–683.
Vinh le S, Von Haeseler A. 2004. IQPNNI: moving fast through
tree space and stopping in time. Molecular Biology and
Evolution 21: 1565–1571.
Whelan S, Goldman N. 2001. A general empirical model of
protein evolution derived from multiple protein families
using a maximum-likelihood approach. Molecular Biology
and Evolution 18: 691–699.
Woese CR, Achenbach L, Rouviere P, Mandelco L. 1991.
Archaeal phylogeny: reexamination of the phylogenetic
position of Archaeoglobus fulgidus in light of certain
composition-induced artifacts. Systematic and Applied
Microbiology 14: 364–371.
Wolf YI, Rogozin IB, Koonin EV. 2004. Coelomata and not
Ecdysozoa: evidence from genome-wide phylogenetic
analysis. Genome Research 14: 29–36.
Yang Z. 1994. Maximum likelihood phylogenetic estimation
from DNA sequences with variable rates over sites:
approximate methods. Journal of Molecular Evolution 39:
306–314.
Yang Z. 1996. Among-site rate variation and its impact on
phylogenetic analyses. Trends in Ecology & Evolution 11:
367–370.
Yang Z. 1997. How often do wrong models produce better
phylogenies? Molecular Biology and Evolution 144:
105–108.
Yang Z, Roberts D. 1995. On the use of nucleic acid sequences
to infer early branchings in the tree of life. Molecular
Biology and Evolution 12: 451–458.
Zheng J, Rogozin IB, Koonin EV, Przytycka TM. 2007.
Support for the Coelomata clade of animals from a
rigorous analysis of the pattern of intron conservation.
Molecular Biology and Evolution 24: 2583–2592.
Zhou Y, Rodrigue N, Lartillot N, Philippe H. 2007. Evaluation
of the models handling heterotachy in phylogenetic
inference. BMC Evolutionary Biology 7: 206.
Zwickl DJ, Hillis DM. 2002. Increased taxon sampling greatly
reduces phylogenetic error. Systematic Biology 51:
588–598.