全 文 ：Journal of Systematics and Evolution 46 (3): 239–257 (2008) doi: 10.3724/SP.J.1002.2008.08016
(formerly Acta Phytotaxonomica Sinica) http://www.plantsystematics.com
Taxon sampling and the accuracy of phylogenetic analyses
Tracy A. HEATH* Shannon M. HEDTKE* David M. HILLIS*, **
(Section of Integrative Biology and Center for Computational Biology and Bioinformatics, One University Station C0930,
The University of Texas, Austin, TX 78712, USA)
Abstract Appropriate and extensive taxon sampling is one of the most important determinants of accurate
phylogenetic estimation. In addition, accuracy of inferences about evolutionary processes obtained from phyloge-
netic analyses is improved significantly by thorough taxon sampling efforts. Many recent efforts to improve
phylogenetic estimates have focused instead on increasing sequence length or the number of overall characters in
the analysis, and this often does have a beneficial effect on the accuracy of phylogenetic analyses. However,
phylogenetic analyses of few taxa (but each represented by many characters) can be subject to strong systematic
biases, which in turn produce high measures of repeatability (such as bootstrap proportions) in support of incor-
rect or misleading phylogenetic results. Thus, it is important for phylogeneticists to consider both the sampling of
taxa, as well as the sampling of characters, in designing phylogenetic studies. Taxon sampling also improves
estimates of evolutionary parameters derived from phylogenetic trees, and is thus important for improved applica-
tions of phylogenetic analyses. Analysis of sensitivity to taxon inclusion, the possible effects of long-branch
attraction, and sensitivity of parameter estimation for model-based methods should be a part of any careful and
thorough phylogenetic analysis. Furthermore, recent improvements in phylogenetic algorithms and in computa-
tional power have removed many constraints on analyzing large, thoroughly sampled data sets. Thorough taxon
sampling is thus one of the most practical ways to improve the accuracy of phylogenetic estimates, as well as the
accuracy of biological inferences that are based on these phylogenetic trees.
Key words consistency, long-branch attraction, phylogenetic accuracy, phylogenomics, systematic error, taxon
sampling, Tree of Life.
The past two decades have seen great progress in
reconstructing the Tree of Life. The endeavor of
inferring the relationships among all living things is
not only of intrinsic interest to biologists, but also has
many practical applications throughout biology.
Phylogenetic trees allow biologists to make predic-
tions about biology, because we can infer when and
where various structures, molecules, or behaviors have
evolved in living organisms. Trees also provide
information about the expected distribution of these
features across taxonomic groups. Moreover, phylog-
enies facilitate the interpretation of comparative
observations by accounting for the historical
non-independence of organisms when analyzing
across various levels of biological organization (e.g.,
genes, genomes, individuals, populations, species, or
clades).
The number of practical applications of phy-
logenetics continues to grow each year. For example,
phylogenetics has become crucial for comparing and
interpreting the various genome sequencing projects.
The very reason that many such projects are under-
taken is to provide a broad evolutionary spectrum for
interpretation of genome function and evolution in the
framework of the Tree of Life (Eisen, 1998). Without
a phylogenetic framework, every genome would be a
new independent mystery, and the detection of gene
function would be greatly hindered without compara-
tive analyses. In the absence of phylogenetic com-
parisons, studying many gene functions in the human
genome would require experimentation and manipula-
tion that is not practical or ethical in humans. There-
fore, in a very real sense, the Tree of Life helps us to
understand how humans function and how we differ
from one another and from other species. The same is
also true of all the organisms that we eat, of all the
organisms that make us sick, of all the organisms that
maintain our ecosystems, and of all the organisms that
make the biological world interesting, entertaining,
and beautiful.
Applications of phylogenetics are by no means
limited to the functional and structural study of ge-
nomes, however. Phylogenetic applications span much
of biology, from human health (Bush et al., 1999) and
forensics (Hillis & Huelsenbeck, 1994; Metzker et al.,
2002) to conservation biology (Crandall et al., 2000)

———————————
Received: 2 February 2008 Accepted: 23 April 2008
* All authors contributed equally to this work.
** E-mail: dhillis@mail.utexas.edu; phone: 512-471-5792.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 240
and studies of behavior (Martins, 1996). Many of
these applications require accurate phylogenetic
estimates, not only in terms of tree topologies, but also
in branch lengths (for estimation of time and/or the
amount of change), ancestral character states (for
estimation of evolutionary transitions), and parameters
of evolutionary models (for study of evolutionary
processes). In general, accurate molecular phyloge-
netic estimates (estimates that represent true historical
relationships among species) are dependent on four
primary factors (Swofford et al., 1996): (1) appropri-
ate selection of target genes for analysis; (2) collection
of enough sequence data to obtain a robust and re-
peatable estimate; (3) use of accurate analytical
methods; and (4) sufficient taxon sampling for the
problem of interest. The first three of these factors
often receive the greatest attention from investigators,
but increased taxon sampling can be one of the most
practical and feasible approaches for improving
phylogenetic estimates (Zwickl & Hillis, 2002). Here
we explore and review the effects of taxon sampling
on phylogenetic analyses and their applications.
1 Dense taxon sampling improves phyloge-
netic accuracy
Phylogeneticists have long acknowledged that
data sets containing a large number of taxa create a
more complex computational problem for phyloge-
netic analysis. As more taxa are added to a phyloge-
netic data set, the number of possible tree topologies
increases very rapidly. In addition, the degree of
homoplasy (convergent changes or reversals) in-
creases with the number of taxa (Sanderson &
Donoghue, 1989). Regardless, numerous studies on
the importance of dense taxon sampling have indi-
cated that introducing additional taxa into a phyloge-
netic analysis results (on average) in more accurate
estimates of evolutionary relationships (Lecointre et
al., 1993; Philippe & Douzery, 1994; Hillis, 1996,
1998; Graybeal, 1998; Rannala et al., 1998; Zwickl &
Hillis, 2002; Pollock et al., 2002; Poe, 1998a, 2003;
DeBry, 2005; Hedtke et al., 2006). These studies
represent a broad range of approaches including
simulations, examinations of well-studied biological
groups, and comparisons to known phylogenies. Each
of these approaches has distinct advantages and
disadvantages (Hillis, 1995) and together they provide
a strong and consistent message about the importance
of dense taxon sampling. The benefits of denser taxon
sampling are especially evident in conjunction with
more thorough searches of solution space (Fig. 1).
Additionally, evaluations of phylogenetic analyses
often attribute problematic reconstruction and low
resolution to inadequate taxon sampling (e.g., Bremer
et al., 1999; Johnson, 2001; Lin et al., 2002; Braun &
Kimball, 2002; Chen et al., 2003; Freudenstein et al.,
2003; Sorenson et al., 2003; Albrecht et al., 2007).
Although the importance of taxonomic sampling
has been intensely investigated, many studies have
focused primarily on parsimony and distance methods.
Felsenstein (1978) demonstrated that under certain
circumstances, parsimony methods are inconsistent,
meaning they converge on an incorrect topology as
more and more characters are added for a limited
sample of taxa. When two non-adjacent taxa share
many homoplastic character states along long
branches, parsimony methods often interpret such
similarity as homology. The resulting tree depicts the
two taxa as sister to one another, attributing the shared
changes to a branch joining them; this effect is termed
long-branch attraction (LBA). Inconsistency is not
restricted to parsimony, however, as all phylogenetic
reconstruction methods can exhibit this behavior if
their assumptions are seriously violated or if there are
not enough taxa in the analysis to accurately estimate
the parameters of the evolutionary model (Felsenstein,
1978; Hendy & Penny, 1989; DeBry, 1992; Huelsen-
beck & Hillis, 1993; Yang, 1994; Huelsenbeck, 1995;




Fig. 1. Error in phylogenetic reconstruction typically decreases with
increased taxon sampling of a given taxonomic group. The benefits of
increased taxon sampling are particularly evident when searches of the
solution space are more thorough. In this graph (adapted from Zwickl
& Hillis, 2002, fig. 6), phylogenetic error decreases with increased
taxon sampling across all analyses. However, the benefits of adding
additional taxa are smaller if only the stepwise-addition algorithm (SA)
is used to find an approximate solution, compared to the more thorough
searches provided by stepwise-addition plus nearest-neighbor-
interchanges branch-swapping (SA+NNI) or tree bisection-reconnection
branch-swapping (SA+TBR). Analyses of larger data sets generally
require more thorough search algorithms (and thus more computational
effort), but result in greatly decreased phylogenetic error.
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

241
Lockhart et al., 1996; Gascuel et al., 2001; Huelsen-
beck & Lander, 2003; Susko et al., 2004; Philippe et
al., 2005). For example, maximum likelihood estima-
tion has been shown to be inconsistent in the presence
of severe branch-length heterogeneity (heterotachy, a
form of non-stationarity) if the substitution process is
assumed to be homogeneous across all lineages
(Kolaczkowski & Thornton, 2004; Spencer et al.,
2005; Philippe et al., 2005). This example emphasizes
the need for probabilistic models that incorporate
complex evolutionary processes (Yang & Roberts,
1995; Galtier & Gouy; 1998; Foster, 2004; Blanquart
& Lartillot, 2006; Gowri-Shankar & Rattray, 2007;
Blanquart & Lartillot, 2008; Kolaczkowski & Thorn-
ton, 2008).
Including additional taxa in a phylogenetic
analysis will increase the accuracy of the inferred
topology by dispersing homoplasy across the tree and
reducing the effect of long-branch attraction. Hillis
(1996) analyzed data simulated on a 228-taxon tree
and showed that simple parsimony and distance
methods accurately reconstruct the true topology when
provided with sequences 5,000 nucleotides in length.
At the time, this result was surprising because it
seemingly contradicted the common belief that accu-
rate phylogenetic reconstruction from very large data
sets was infeasible. Moreover, Hillis et al. (1994b) had
previously shown that analyses of much smaller data
sets, containing only 4 taxa, required considerably
longer sequences to attain the same level of accuracy.
The results of Hillis’s (1996) large-scale simulation
indicated that for phylogenies containing many taxa,
convergent substitutions or reversals (homoplasy) are
distributed among the many lineages in the tree and
therefore such misleading information is less likely to
overwhelm the true phylogenetic signal.
Because inadequate species sampling can result
in trees containing relatively long terminal branches,
sparsely sampled data sets are more likely to be
affected by LBA. Rannala et al. (1998) simulated
ultrametric trees under a simple model of cladogenesis
to investigate the impact of removing ingroup taxa on
the distribution of branch lengths. They demonstrated
that decreasing the proportion of sampled taxa leads to
an increase in the average length of terminal branches
and generates tree shapes that may be susceptible to
long-branch attraction (Fig. 2). Huelsenbeck and
Lander (2003) simulated sequences using simple
evolutionary models and determined that the probabil-
ity that parsimony is inconsistent becomes greater as
the proportion of taxa sampled decreases and substitu-
tion rates increase. Even under very simple models of
evolution, unweighted parsimony underestimated the
number of changes along branches and converged on
an incorrect topology (Huelsenbeck & Lander, 2003).
In general, many studies have shown that adding
taxa to bisect long branches can mitigate the effect of
LBA (Hendy & Penny, 1989; Graybeal, 1998; Poe &
Swofford, 1999; Poe, 2003). However, taxon addition
should be practiced judiciously to ensure that enough
taxa are added to sufficiently partition multiple long
branches (Graybeal, 1998; Poe, 2003) and that the
new taxa do not introduce additional long branches
(Kim, 1998). Prudent taxon addition is particularly



Fig. 2. Two simulations of a birth-death process to model cladogenesis. The speciation rate (λ) and extinction rate (μ) were fixed throughout the
simulation and arbitrarily set to λ /μ=2. A, Phylogenetic tree with complete (100%) taxon sampling (20 taxa total). B, Phylogenetic tree with 10%
taxon sampling (20 taxa sampled from 200 taxa total). When taxon sampling is low, terminal branch lengths are longer, indicating that sparsely
sampled data sets are susceptible to the effects of long-branch attraction. Adapted from Rannala et al. (1998; figs. 1 and 2).
Journal of Systematics and Evolution Vol. 46 No. 3 2008 242
important when conducting parsimony analyses since
this method is especially liable to inconsistency due to
long-branch attraction. Because parametric methods,
such as maximum likelihood, incorporate models that
account for unobserved substitutions, these methods
are less prone to the effects of long-branch attraction,
as long as the models of evolution are adequate.
However, enough taxa must be sampled to parameter-
ize these models effectively (Pollock et al., 2002). In
addition, longer branches require more accurate
models of evolution (because more unobserved
changes must be inferred), so increased taxon sam-
pling (which breaks up long branches) greatly benefits
parametric methods as well as nonparametric meth-
ods. We discuss methods for detecting and minimiz-
ing LBA more completely below.
Apart from its effect on topological accuracy, the
density of taxon sampling also has an impact on
branch-length estimation. Branch lengths provide
important information about the amount of change
that has occurred over the tree and are critical for
applications using phylogenies to make inferences
about evolution. Under the parsimony criterion,
branch lengths are often underestimated in sparsely
sampled regions of the tree because less information is
available to infer the history of unobserved substitu-
tions (Fitch & Bruschi, 1987; Fitch & Beintema,
1990). This artifact has been termed the node-density
effect (NDE) and may mislead studies that investigate
correlations between rates of molecular evolution and
biodiversity (Webster et al., 2003; Venditti et al.,
2006; Hugall & Lee, 2007). Maximum likelihood,
Bayesian, and distance methods are also susceptible to
node-density effects, particularly when the assumed
model of sequence evolution is overly simple and
substitution rates are high (Gojobori et al., 1982;
Bruno & Halpern, 1999; Hugall & Lee, 2007). If the
density of taxon sampling is increased, additional
internal nodes can reveal undetected substitutions and
improve estimates of branch lengths.
It has been shown that mis-estimation of branch
lengths can, in turn, lead to biased tree topologies
(Xia, 2006). Errors in estimates of genetic distance
become greater as the amount of divergence between
two sequences increases. Pairwise distance methods
for phylogenetic reconstruction typically use log-
transformed formulae to account for unobserved
substitutions (Swofford et al., 1996; Hoyle & Higgs,
2003). When using log-transformed formulae to
calculate genetic distances, particularly at high levels
of sequence divergence, there is a significant prob-
ability that the distance estimates will be undefined
even if the “true” model of sequence evolution is
assumed (Hoyle & Higgs, 2003). Therefore, when
conducting distance-based analyses, it is very impor-
tant to consider how taxa are sampled and avoid
inclusion of highly divergent sequences.
1.1 Increased taxon sampling versus increased
sequence length
Increasing the total number of characters in a
data set can increase resolution and support for a
phylogeny (Hillis et al., 1994a; Graybeal, 1994;
Rannala et al., 1998). In particular, increasing charac-
ter data such as the number of genes (or total number
of nucleotides) should reduce stochastic error or
character sampling bias (Phillips et al., 2004; Delsuc
et al., 2005). The rapidly increasing amount of se-
quence data available to researchers from whole
genomes, expressed sequence tags or cDNA libraries,
and individual gene-based studies means that many
analyses of these character-rich phylogenetic matrices
can greatly reduce stochastic error.
This plethora of sequence data has caused some
researchers to argue that large character data sets
alone are sufficient to estimate an accurate phylogeny,
notwithstanding the conclusions reached by numerous
studies showing the importance of taxon sampling.
For example, Rosenberg and Kumar (2001) conducted
a simulation study indicating that adding taxa to a
problematic phylogeny is less effective than adding
additional characters. This paper led to a debate in the
literature and reanalyses of the Rosenberg and Kumar
(2001) data (Zwickl & Hillis, 2002; Pollock et al.,
2002; Rosenberg & Kumar, 2003; Hillis et al., 2003).
Pollock et al. (2002) reanalyzed the Rosenberg and
Kumar (2001) data using a different approach to
summarizing results (measurement of error), and
Zwickl and Hillis (2002) re-conducted the Rosenberg
and Kumar (2001) study with a different approach to
study design that examined a fuller spectrum of taxon
sampling strategies. Both studies concluded that taxon
sampling has a very strong and positive effect on the
accuracy of phylogenetic reconstruction, and showed
in many cases that increasing the number of taxa had a
much greater beneficial effect than increasing the
number of characters. However, because of the
ever-increasing availability and generation of genomic
data and the difficulty of obtaining sequence data for
many taxa, the debate about the relative importance of
taxon sampling versus character sampling continues in
the literature (Hillis et al., 2003; Rosenberg & Kumar,
2003; Rokas et al., 2003; Cummings & Meyer, 2005;
Rokas et al., 2005; Hedtke et al., 2006; Gatesy et al.,
2007).
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

243
More recently, Rokas et al. (2003) argued that,
under parsimony, an accurate phylogeny for seven
yeast taxa could be obtained using twenty genes
randomly selected from a data set of 106, regardless of
taxonomic sampling. The authors based this claim on
the bootstrap support for each node in their tree: once
twenty genes were sampled, the bootstrap support for
each node rose above 95%, and the topology was
identical to that of the data set using all 106 genes.
Therefore, they concluded, accuracy was increased by
increasing the number of genes, with the implication
that many phylogenetic studies may be using an
insufficient number of genes to accurately reconstruct
topologies. The most obvious problem with their
study design was that the authors equated high boot-
strap support with accuracy (rather than more properly
with repeatability). Analyzing many characters may
result in convergence on a single answer, but small
systematic biases can result in convergence on the
incorrect answer (a point we explore in the next
section).
1.2 Increased character data can reduce stochas-
tic error, but can contribute to systematic error
Increasing the number of nucleotides will not
solve inaccurate reconstruction due to method incon-
sistency, or systematic error. All phylogenetic meth-
ods to date have conditions in which they will perform
inconsistently. Phylogenomic data sets, which tend to
have large numbers of characters with relatively few,
widely dispersed taxa, may be particularly prone to
problems with long-branch attraction. LBA has been
identified as causing spurious relationships in
large-scale studies of several taxonomic groups,
including mammals (Lin et al., 2002), metazoans
(Anderson & Swofford, 2004; Baurain et al., 2007),
arthropods (Delsuc et al., 2003) and angiosperms
(Stefanovic et al., 2004). When methods are inconsis-
tent, increasing character data can increase statistical
support for an inaccurate phylogeny (Huelsenbeck &
Hillis, 1993). This occurs because measures of support
such as the nonparametric bootstrap proportion or
Bayesian posterior probabilities are conditional on the
data and the method (Delsuc et al., 2005).
A reanalysis of the Rokas et al. (2003) data set by
Hedtke et al. (2006) showed that much of the conflict
between genes under parsimony is due to LBA. The
taxa sampled are unevenly dispersed across the tree of
yeast, and a long branch leading to the outgroup has a
tendency to pull particular taxa together across many
independent genes. Hedtke et al. (2006) additionally
used genes simulated on a more densely sampled
yeast phylogeny to demonstrate that for a given
bipartition affected by long-branch attraction, boot-
strap support for the wrong reconstruction increased
as genes were added to the analysis (Fig. 3). However,
when taxon sampling was increased, fewer genes were
needed to get acceptable support for the correct bipar-
tition. Although Hedtke et al. (2006) did not examine
increases in support or accuracy across the entire tree,
their results are consistent with the general finding
that increased taxon sampling improves accuracy
across the tree.
This discussion so far has assumed that all genes
share the same evolutionary history, and that by
increasing the number of nucleotides a researcher
increases the signal for that one history — i.e., that
topological conflicts between trees based on single
genes are stochastic. However, genes do not always
share an identical history, because of horizontal gene
transfer, introgression, incomplete lineage sorting, or
gene duplication/loss. In some cases, it may be more
appropriate to analyze genes separately rather than
concatenating all the data (e.g., Ane & Sanderson,
2005), particularly if a researcher is interested in
identifying processes that may have affected the
evolution of a species.
2 Strategies for effective taxon sampling
We have discussed how increased taxon sam-
pling can generally improve topological estimates.
Unfortunately, there is no “magic number” of taxa and
genes which ensures the accuracy of a phylogeny. The
percentage of taxa sampled within a taxonomic group
may be more important than the total number of taxa
(Yang & Goldman, 1997; Hillis, 1998). For small
clades of closely-related taxa, reduced taxonomic
sampling may not be problematic for phylogeny
reconstruction (Poe, 1998b). Critically, whether to
increase sampling within a taxonomic group of inter-
est should not depend on whether there is statistical
support for a topology, as strong support does not
indicate a lack of systematic error.
2.1 Information theory and taxon addition
Several techniques have been developed from
information theory to examine where taxa could be
added to an analysis to increase the precision of a
topological estimate (Goldman, 1998; Massingham &
Goldman, 2000; Geuten et al., 2007). The observed
information matrix, which is a measure of the sharp-
ness of the curve about the likelihood function, is
compared to the expected (or Fisher) information
matrix when a branch is added to different parts of the
Journal of Systematics and Evolution Vol. 46 No. 3 2008 244


Fig. 3. A, Well-sampled yeast tree used for simulating data. Dark lines represent four taxa known to be susceptible to long-branch attraction under
parsimony. B, Results of randomly sampling 1–25 genes for 4–40 taxa. As the number of genes in a parsimony analysis increases, the bootstrap for
the correct reconstruction of the four-taxon statement decreases, unless taxon sampling is sufficient to break the long branches. Adapted from Hedtke
et al. (2006; figs. 1 and 3).

topology (Geuten et al., 2007). In essence, this com-
parison indicates the increase in precision of the
topology estimate that is gained by adding a taxon
along a particular branch of the tree. Based on a data
set from angiosperms, Geuten et al. (2007) found that
information theory generally supports adding addi-
tional taxa close to where the long branch attaches to
the rest of the tree, in congruence with other studies
based on simulation (Graybeal, 1998; Poe & Swof-
ford, 1999). Unfortunately, these techniques, while
promising, have not been rigorously tested, and
generally assume that one has strong a priori expecta-
tions about where additional taxa might fall on the
phylogeny.
2.2 Detecting LBA
Most studies that explore specific recommenda-
tions for increased taxon sampling focus on detecting
and correcting LBA. LBA is often cited when a
topology fails to meet a priori expectations, but
caution must be applied before attributing unexpected
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

245
results to method inconsistency (Anderson & Swof-
ford, 2004) or method bias. Many claims of LBA in
published studies are a posteriori explanations of
unexpected results. Often, when taxa are added to an
analysis that originally generated an unexpected tree,
topological relationships change to match expectations
based on traditional taxonomy, and this change is
attributed to LBA (e.g. Lin et al., 2002; Stefanovic et
al., 2004; Philippe et al., 2005; Baurain et al., 2007).
Here we discuss strategies which could provide useful
heuristics for exploring whether method inconsistency
or bias, particularly LBA, is affecting the analysis.
Disagreement among independent data sets—
particularly those based on traditional morphological
taxonomy versus molecular analyses—may be the
first signal that LBA may be present in the data set
(Lin et al., 2002; Chen et al., 2003). If two long
branches are unexpectedly drawn together in one or
the other analysis, LBA may be the culprit. However,
a researcher must also consider the alternative that
traditional taxonomy may be wrong (e.g., Ammerman
& Hillis, 1992; Van Den Bussche et al., 1998). The
two long branches may in fact be sister taxa, or may
be brought together by LBA. To distinguish these
hypotheses, one technique is to prune each
long-branched taxon successively from the analysis,
and observe whether the topology changes (Bergsten,
2005). Presumably, if the position of a long-branched
taxon changes dependent upon the inclusion of other
long-branched taxa, LBA may be implicated. Unfor-
tunately, this test is not definitive, because other
characteristics of the excluded taxa could be affecting
the topology as a whole.
Simulated data generated under different hy-
potheses could be analyzed to compare possible
topologies and get a sense of potential error rates in
the analysis (Van Den Bussche et al., 1998; Sanderson
et al., 2000). In this parametric bootstrapping ap-
proach, data are simulated under each hypothesis
using model parameters estimated from the biological
sequence data. For example, one could identify the
alternative hypothesis by running a phylogenetic
analysis constrained to find the best tree which does
not place two suspect taxa as sister groups. If analyses
of replicate data sets simulated on this tree tend to
bring the two suspect taxa together more often than
expected by chance, then one cannot reject the possi-
bility that the two taxa are grouped due to systematic
error or bias. However, as with all parametric ap-
proaches, investigators should consider the adequacy
of the simulation model. Under-parameterized models
may underestimate the potential for LBA.
2.3 Outgroup sampling
Outgroup taxa tend to be on long branches, either
because of the processes of cladogenesis, because of
extinction events between the outgroup and the in-
group, or because of inappropriate selection of out-
group taxa. When the outgroup is distantly related to
the ingroup, long branches in the ingroup can be
drawn toward the base of the tree by LBA, affecting
ingroup relationships (Hillis, 1998; Rannala et al.,
1998) or misplacing the root of the tree (Holland et al.,
2003). Holland et al. (2003) used simulated data to
demonstrate this effect not just for parsimony, but for
maximum likelihood and distance analyses as well.
Graybeal (1998) suggested that error could be reduced
through the use of multiple outgroup taxa separated by
short internal branches.
One method to examine whether outgroup choice
is affecting the topology of the ingroup is to run the
analysis using the ingroup taxa only (Holland et al.,
2003; Bergsten, 2005). If the ingroup topology is
influenced by the long branch leading to the outgroup,
the unrooted topology with and without the outgroup
will change. This technique will not allow the re-
searcher to distinguish whether the rooting of the tree
(placement of the outgroup) is correct; only whether
the outgroup is influencing the ingroup topology. A
second method to evaluate outgroup choice is to
simulate a number of random sequences which are
each used to root the tree (Sullivan & Swofford,
1997). This can indicate the relative probabilities for
rooting positions when no historical signal is present
in the data for the outgroup. In both cases, if use of a
particular outgroup leads to LBA problems, sampling
more outgroup taxa may assist in detecting homoplasy
and reducing the effects of the long outgroup branch.
2.4 Ingroup sampling
If LBA is detected within the ingroup, long-
branch subdivision by addition of taxa could mitigate
this effect (Fig. 4; Hendy & Penny, 1989; Hillis,
1996). The strategy of long-branch subdivision has
been examined using both simulated (e.g., Graybeal,
1998; Poe & Swofford, 1999; Poe, 2003) and biologi-
cal (e.g., Poe, 1998a; Baurain et al., 2007) data. In
parsimony analyses, there are cases in which the
addition of one taxon can actually cause LBA; this has
been demonstrated using simulated data in the
four-taxon case when the new taxon introduces a new
long branch (Rannala et al., 1998; Poe & Swofford,
1999), or when there are three long branches, and
breaking one causes the remaining two to become
drawn together (Fig. 4; Poe & Swofford, 1999; Poe,
2003). In this case the new taxon introduces an
Journal of Systematics and Evolution Vol. 46 No. 3 2008 246


Fig. 4. Effects of long-branch subdivision on accuracy in four-taxon trees, based on simulated data. For each tree, branch lengths of long branches
are 0.5, of short terminal branches 0.1, and of internal branches 0.05. Dots indicate when a long branch was broken at length 0.1 from the internal
node. Poe (2003) evaluated 6 different reconstruction methods, 3 are summarized in this figure: unweighted maximum parsimony (MP), maximum
likelihood with an under-parameterized model of substitution (MLU), and maximum likelihood with the true model of substitution (MLT). The
numbers are the percentage of time the true, four-taxon tree was recovered in 100 replicate simulations. The arrows indicate increased ( ) or de-
creased ( ) accuracy as a result of the added taxa, and no arrow indicates that the accuracy was unaffected. Adapted from Poe (2003; fig. 3).


asymmetrical pattern of homoplasy and long-branch
attraction results in an incorrect reconstruction. In
addition, adding only one taxon may not be sufficient
to alleviate LBA if the branch is sufficiently long
(Poe, 2003; Hedtke et al., 2006) or if the added taxon
is placed towards the tip (Graybeal, 1998). However,
both these effects are diminished if enough taxa are
added along long branches (Poe, 1998a; Anderson &
Swofford, 2004; Hedtke et al., 2006). For example, in
the Hedtke et al. (2006) simulation study, adding taxa
increased accuracy for a particular bipartition only
when the added taxa divided a long branch that was
causing long-branch attraction.
2.5 Adding taxa with missing data
Adding taxa with incomplete character informa-
tion to a supermatrix has primarily been evaluated in
parsimony-based morphological analyses in reference
to fossil data. Incomplete fossil data are not always
beneficial, as they may reduce support for some nodes
(Wiens, 2003; Wiens, 2005; Cobbett et al., 2007).
However, it appears that well-chosen fossil data can
be helpful in breaking long branches, even if incom-
plete, as these morphological characters can be infor-
mative about character reconstruction at branching
nodes (Donoghue et al., 1989; Huelsenbeck, 1991).
The effect of missing character data on sequence
analysis is still being debated, with some researchers
arguing that adding taxonomic data is beneficial even
if the resulting supermatrix has a large proportion of
missing data (e.g., 25%: Philippe et al., 2004; 75%:
Wiens & Reeder, 1995; 95%: McMahon & Sanderson,
2006). However, Lemmon et al. (in press) used simu-
lated data to demonstrate that missing sequence data
can positively mislead model-based methods. This
depends in part on the relative rates of evolution for
sites with and without missing data, and the topologi-
cal position of those taxa with missing data. This is an
area of active research, and caution should be used
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

247
when combining taxa with missing data until this
issue has been more completely explored using both
simulated and biological data sets.
2.6 Alternatives to increased taxon sampling
Several other techniques of combating method
inconsistency have been suggested. Because inconsis-
tency or bias results from violation of model assump-
tions (e.g., not accurately modeling multiple substitu-
tions), finding a better-fitting model could solve the
problem (Olsen, 1987; Whelan & Goldman, 2001;
Lartillot & Philippe, 2004; Anderson & Swofford,
2004; Delsuc et al., 2005; Baurain et al., 2007; Lartil-
lot et al., 2007). To reduce inconsistency due to LBA,
it has been suggested that either eliminating
fast-evolving sites from the analysis (Delsuc et al.,
2002; Delsuc et al., 2005; Rodriguez-Ezpeleta et al.,
2007) or coding all the data to represent only less
frequent transversions (“RY” coding: Phillips et al.,
2004; Delsuc et al., 2005) would reduce saturation and
compositional bias in the data set, and thus reduce
LBA. Using amino acid data (with models which take
into account site heterogeneity, e.g., Lartillot &
Philippe, 2004) may be another alternative to using
raw sequence data. However, the actual effect of
saturated sequences on phylogenetic analyses has been
incompletely explored. Hillis (1998) simulated se-
quence data on a tree with such long branch lengths
that the sequences would not be recognizable as
homologous, but phylogenetic methods were still able
to reconstruct the correct tree when taxon sampling
was sufficient. Exclusion of data (whether by exclud-
ing sites or reducing their information) may be useful
in eliminating the problem of a particular set of long
branches, but this may be at the expense of resolution
in other regions of the tree.
3 Taxon sampling affects parameter esti-
mation
Many advances in phylogenetic analysis over the
past two decades have involved model-based ap-
proaches, such as maximum likelihood and Bayesian
analyses (Swofford et al., 1996; Ronquist & Huelsen-
beck, 2003; Felsenstein, 2004). In general, these
parametric methods outperform nonparametric meth-
ods in both simulations and experimental studies
(Hillis et al., 1994a; Huelsenbeck, 1995; Cunningham
et al., 1997). However, accurate phylogenetic results
from model-based studies depend, at least in part, on
reasonably accurate parameter estimates for the
models of evolution (Goldman, 1993; Hillis et al.,
1994b; Cunningham et al., 1998; Lemmon &
Moriarty, 2004; Brown & Lemmon, 2007). One of the
reasons that increased taxon sampling results in more
accurate phylogenetic estimation for these model-
based methods is that sampling additional taxa also
improves parameter estimation (Pollock et al., 1999;
Sullivan et al., 1999; Pollock & Bruno, 2000; Pollock
et al., 2002). In addition, as branch lengths are short-
ened, there are fewer unobserved changes that need to
be inferred, so the accuracy of the inference becomes
less dependent on the model of evolution.
In addition to their effect on phylogenetic analy-
ses, the parameters of evolutionary models are them-
selves of interest to biologists. These parameters are
often gene-specific, so collecting genomic-scale data
from many genes across only a few taxa does little to
improve our estimates of the details of evolutionary
models. Instead, a thorough taxon-sampling approach
is needed for each gene. Of course, the evolutionary
processes may not be static across the Tree of Life for
any given gene, so models that account for non-
stationarity in these processes can provide better
descriptions of evolutionary history (Yang & Roberts,
1995; Galtier & Gouy, 1998; Foster, 2004; Blanquart
& Lartillot, 2006; Boussau & Gouy, 2006; Gowri-
Shankar & Rattray, 2007; Blanquart & Lartillot, 2008;
Kolaczkowski & Thornton, 2008). These models relax
the assumption of time-homogeneity and can be used
to detect signatures of complex evolutionary proc-
esses, such as compositional heterogeneity or hetero-
tachy (branch-length heterogeneity), that are known to
exist in biological data (Lockhart et al., 1992; Foster
et al., 1997; Mooers & Holmes, 2000; Lopez et al.,
2002; Jermiin et al., 2004; Ane et al., 2005). Non-
stationary models can greatly increase the need for
even more thorough taxon sampling, because the
model parameters may need to be estimated multiple
times across the tree, rather than once for all taxa. It is
important to note, however, that under non-stationary
models, the number of parameters can increase as
more sequences are added, thus increasing the com-
putational difficulty of phylogenetic reconstruction
from large data sets. Nonetheless, this obstacle may be
mitigated by the use of carefully constructed priors in
a Bayesian MCMC framework (Yang, 2006) and with
the development of computational methods for calcu-
lating likelihoods from non-reversible models (Bous-
sau & Gouy, 2006).
Some of the parameters that have been shown to
be important for phylogenetic estimation include
site-specific rates of evolutionary change; rates of
change across first, second, and third positions of
Journal of Systematics and Evolution Vol. 46 No. 3 2008 248
codons; rates of change relative to changes in func-
tional groups of amino-acid residues; relative rates of
the various classes of transitions and transversions
between nucleotide states; branch-specific rates of
evolutionary change; and taxon-specific differences in
base composition (Olsen, 1987; Steel et al., 1993;
Hasegawa & Hashimoto, 1993; Hillis et al., 1993;
Leipe et al., 1993; Goldman & Yang, 1994; Steel,
1994; Swofford et al., 1996). The number of taxa that
are needed to effectively estimate these parameters
differ greatly across the parameters, but all of the
estimates are improved by more thorough taxon
sampling. For instance, Pollock and Bruno (2000)
noted significant improvement in parameter estima-
tion (and in turn, phylogenetic estimation) as their
taxon samples increased from 4 to 8 to 16 to 24 taxa.
They concluded that both phylogenetic reconstruction
and estimation of unknown evolutionary processes
show greater improvement through increasing taxon
sampling than by increasing sequence length. In some
cases, reasonable parameter estimates may be ob-
tained from external data sources, such as the HIV
database, and then applied to a more limited set of
taxa in the phylogenetic analysis (Hillis, 1999). How-
ever, for most taxa, the appropriate comparative data
must be obtained by the investigator for a specific
group of species under study.
4 Dense taxon sampling improves infer-
ences of evolutionary processes
Beyond simply broadening our understanding of
species relationships, phylogenetic trees are essential
tools used in many areas of biology. Phylogenies are
often used to explain broad evolutionary patterns and
processes such as the evolution of adaptive traits,
ancestral character states, the timing of species diver-
gences, and variation in evolutionary rates. Many of
the applications developed for these types of analyses
require robust and accurate estimates of phylogeny
(topology, branch lengths, and root position). This is
an important consideration in and of itself; however,
post-tree reconstruction applications are sensitive to
reduced levels of data sampling, even when provided
with an accurate phylogenetic tree.
4.1 Comparative methods
Comparative analyses are a fundamental compo-
nent in the fields of evolutionary biology, behavior,
and ecology. The development of statistical methods
that incorporate phylogenetic trees (Felsenstein, 1985)
have allowed for robust and reliable tests of the
evolution of adaptive traits and the processes that
might drive diversification. For example, these meth-
ods have been used to reveal patterns in the biodiver-
sity of marine teleost fishes (Alfaro et al., 2007) and
to show that independent origins of dietary specializa-
tion have been a major factor in the evolution of
defensive mechanisms in neotropical poison frogs
(Darst et al., 2005). Comparative analyses of character
evolution using phylogenetic comparative methods
require attention to adequate sampling at many levels.
At the intraspecific level, poor sampling of organismal
attributes can lead to measurement error, which may
result in an underestimation of the variance of con-
trasts between sister taxa (Ricklefs & Starck, 1996).
Generation of a robust phylogeny is extremely impor-
tant since different comparative methods have differ-
ent ways of dealing with topological uncertainty
(Purvis et al., 1994). In addition, fewer taxa (and thus
fewer internal nodes for calculating contrasts) can lead
to increased variance and uncertainty in the results.
Ackerly (2000) used simulated data to show that the
statistical power of several comparative tests de-
creased as the sample size of taxa decreased, and that
careful attention should be paid to how species are
sampled for these analyses. Biased taxon sampling,
particularly with respect to the characters of interest,
can lead to systematic biases in the calculation of
statistical correlations between characters. The results
presented by Ackerly (2000) indicate that uniform,
random sampling of taxa does not introduce error in
phylogenetic comparative methods.
4.2 Ancestral character states
An integral component of phylogenetic compara-
tive analyses and other evolutionary applications is the
reconstruction of ancestral character states. These
methods use phylogenetic trees and branch lengths to
infer the states of discrete or continuous characters at
ancestral nodes, and have been used to reconstruct
such diverse ancestral characters as the advertisement
calls of frogs in the genus Physalaemus (Ryan &
Rand, 1995, 1998), the fruiting-body forms of homo-
basidiomycetes (Hibbett, 2004), and ancient bacterial
protein sequences (Gaucher et al., 2003). Dense taxon
sampling is also an important consideration for ances-
tral-state reconstruction methods. Salisbury and Kim’s
(2001) analyses of simulated data and trees indicated
that the accuracy of parsimony ancestral-state estima-
tion decreases with reduced taxon sampling and
increased rates of character evolution (Fig. 5). Be-
cause parsimony methods do not account for unob-
served changes, they usually underestimate the num-
ber of changes along a branch (Fitch & Bruschi, 1987;
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

249



Fig. 5. The mean probabilities, Pr (Correct), of correctly estimating
the root state of a binary character evolving at 3 different rates (r) on
subsamples of 512-taxon, pure-birth model tree topologies. Each
point is the mean for a sample of 100 trees and the error bars represent
the ± 1 standard deviation. Adapted from Salisbury and Kim (2001; fig.
1).


Fitch & Beintema, 1990; Huelsenbeck & Lander,
2003). Dense taxon sampling can reduce this effect
and improve the accuracy of parsimony ancestral-state
estimates. Maximum likelihood and Bayesian meth-
ods for reconstructing ancestral states have also been
developed (Pagel, 1994; Schluter et al., 1997; Pagel,
1999; Huelsenbeck & Bollback, 2001; Pagel et al.,
2004). These parametric ancestral-state reconstruction
methods are also sensitive to high rates of character
evolution. However, Schluter et al. (1997) showed
that parsimony ancestral-state reconstruction methods
often fail to identify ambiguous-node state estimates.
Conversely, maximum likelihood and Bayesian
methods are less likely to provide misleading results
because these methods incorporate branch-length
information and explicit models of character evolution
and quantify uncertainty in ancestral-state estimates
(provided that the model assumptions are adequate).
Bayesian approaches, in particular, use Markov chain
Monte Carlo sampling to accommodate and quantify
uncertainty in the tree topology, branch lengths,
ancestral states, and model parameters (Huelsenbeck
& Bollback, 2001; Pagel et al., 2004). Denser taxon
sampling reduces the number of unobserved evolu-
tionary events, and so is also expected to simplify and
improve the reconstruction of ancestral states in
model-based analyses.
4.3 Divergence time estimation
A primary field of research in evolutionary biol-
ogy involves estimation of the timing and rate of
evolutionary processes. In these applications, phy-
logenetic trees are used to date speciation events and
infer lineage-specific substitution rates. Reliable
estimates of species divergence times are fundamental
components for understanding historical biogeogra-
phy, testing hypotheses of adaptive character evolu-
tion, and estimating speciation and extinction rates.
However, divergence time estimation is hindered by
the fact that the rate of evolution and time are intrin-
sically linked when inferring genetic distances be-
tween lineages. Several methods test for variation in
the rates of molecular evolution or tease apart the rate
of substitution and time by applying models for
estimating lineage-specific substitution rates. These
methods include strict molecular clock models (Zuck-
erkandl & Pauling, 1962; Langley & Fitch, 1974),
local molecular clocks (Kishino & Hasegawa, 1990;
Rambaut & Bromham, 1998; Yoder & Yang, 2000;
Yang & Yoder, 2003), non-parametric and semi-
parametric methods for estimating autocorrelated
substitution rates (Sanderson, 1997, 2002), and
Bayesian methods for estimating autocorrelated and
uncorrelated rates (Thorne et al., 1998; Huelsenbeck
et al., 2000; Kishino et al., 2001; Thorne & Kishino,
2002; Drummond et al., 2006; Lepage et al., 2006).
These various approaches have been applied to a
number of biological data sets (e.g., Yang & Yoder,
2003; Smith et al., 2006; Bell, 2007; Hugall et al.,
2007; Roelants et al., 2007; Zhou & Holmes, 2007).
Current implementations of most of these methods
require a fixed tree topology and sometimes fixed
branch lengths (Thorne & Kishino, 2002; Sanderson,
2003; Lepage et al., 2007; for exceptions see Drum-
mond et al., 2006). Because of their reliance on phy-
logenetic data, these methods can be sensitive to taxon
sampling density. Robinson et al. (1998) evaluated the
effect of reduced taxon sampling on the performance
of the relative-rates test. The relative-rates test (Sarich
& Wilson, 1973; Wu & Li, 1985) is used to compare
the substitution rates between two species and has
been extended for analyzing larger phylogenetic trees
to detect rate variation (Li & Bousquet, 1992;
Takezaki et al., 1995). The simulation study of Rob-
inson et al. (1998) showed that increased proportions
of taxon sampling improved the accuracy of the
relative-rates test.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 250
Most of the work exploring the accuracy of mo-
lecular dating methods has revealed that these meth-
ods are very sensitive to the fossil calibrations used
and little is known about the impact of taxon sampling
on divergence time estimates (Yang & Rannala, 2006;
Rutschmann et al., 2007; Hugall et al., 2007). A recent
study by Hug and Roger (2007) used two biological
data sets with low levels of taxon sampling (30 meta-
zoan taxa with two outgroup species, and a 36 taxon
data set that spanned all eukaryotes) and concluded
that, for these data sets, reduced taxon sampling was
not an important factor in the estimation of node
times. However, their analyses showed that the choice
and application of fossil calibration points resulted in
a significant impact on the estimates of node ages.
From their results, Hug and Roger (2007) recom-
mended that biologists should focus on improving the
number and quality of their fossil calibrations and not
on increasing taxon sampling, provided there are
enough taxa to obtain a reliable estimate of phylog-
eny. However, because of the sparsely sampled data
sets used in this study and the demonstrated extreme
sensitivity of these data to fossil constraints, Hug and
Roger’s (2007) results may not apply to a more gen-
eral set of conditions and the importance of dense
taxon sampling for estimating species divergence
times is still an open question.
Node-density effects, as a result of uneven taxon
sampling, may adversely affect molecular dating
analyses (Hugall & Lee, 2007). Based on the studies
demonstrating the sensitivity of divergence time
estimation methods to fossil calibration choice (Near
& Sanderson, 2004; Near et al., 2005; Roger & Hug,
2006; Yang & Rannala, 2006; Ho, 2007; Hugall et al.,
2007; Rutschmann et al., 2007), together with studies
emphasizing the importance of increased taxon sam-
pling on phylogenetic reconstruction methods and the
estimation of evolutionary parameters (Lecointre et
al., 1993; Hillis, 1996, 1998; Graybeal, 1998; Rannala
et al., 1998; Pollock & Bruno, 2000; Zwickl & Hillis,
2002; Pollock et al., 2002; Poe, 2003; DeBry, 2005;
Hedtke et al., 2006), we recommend increased collec-
tion of fossils and improved taxon sampling density
for these types of analyses, whenever possible. Maxi-
mizing the number of fossil calibration points goes
hand-in-hand with increasing taxon sampling because
densely sampled trees provide a greater number of
internal nodes on which an investigator can place a
fossil calibration. Moreover, investigators are far more
restricted by the availability of fossils and other types
of information for calibrating divergences than by the
availability of extant taxa. Further investigation using
simulations and well-sampled data sets of living and
fossil taxa should help shed light on this issue. Be-
cause extensive taxon sampling (especially of fossil
taxa) is sometimes impractical, Bayesian methods for
divergence time estimation present promising oppor-
tunities to account for uncertainty in phylogenies by
simultaneously estimating the tree topology and
branching times (Drummond et al., 2006). These
methods can also incorporate information on taxon
sampling density in the form of priors on the distribu-
tion of divergence times (Yang & Rannala, 1997,
2006).
4.4 Evaluating diversification rates
Phylogenetic trees are fundamental for under-
standing variation in species diversity. Methods for
elucidating patterns of speciation and extinction
measure the shape of phylogenies to detect shifts in
diversification rates or to estimate global net diversi-
fication rates. Phylogenetic tree shape can be meas-
ured by quantifying how node ages are distributed
over time or by calculating the degree of asymmetry
among lineages in the tree. Measures of tree shape can
be compared to a null model that assumes all lineages
have experienced the same rate of diversification
(Shao & Sokal, 1990; Kirkpatrick & Slatkin, 1993;
Nee et al., 1994b; Pybus & Harvey, 2000; Agapow &
Purvis, 2002). Analyses of the temporal distribution of
diversification events use branch lengths obtained
from time-adjusted phylogenies to estimate and detect
large shifts in speciation and extinction rates (Nee et
al., 1994b; Pybus & Harvey, 2000). For example,
Becerra (2005) applied these methods to investigate
temporal and biogeographic processes that may have
shaped the diversity of the plant genus Bursera. The
results of this study indicate that the radiation of this
group is associated with the establishment of tropical
dry forest habitat in Mexico. However, inadequate
taxon sampling has a significant impact on these
methods (Nee et al., 1994a). Nee et al. (1994a) used
lineages-by-time plots to show that incomplete taxon
sampling can result in an apparent reduction in the
rate of diversification over time, even when the tree
evolved under constant rates of speciation and extinc-
tion.
Analyses based on topology measure asymmetry
in the distribution of lineages over a tree to test for
changes in diversification rates. These methods evalu-
ate the balance either at a single node or over the
entire tree (Shao & Sokal, 1990; Kirkpatrick & Slat-
kin, 1993; Agapow & Purvis, 2002) and are often used
to detect patterns characteristic of rapid radiations in
phylogenetic trees (Guyer & Slowinski, 1993; Chan &
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

251
Moore, 1999, 2002). The degree of taxon sampling is
an important consideration when conducting these
analyses. Several studies have shown that published
phylogenies are (on average) much more imbalanced
than expected under a model assuming constant
diversification rates (Guyer & Slowinski, 1991;
Heard, 1992; Mooers, 1995; Purvis & Agapow, 2002;
Holman, 2005; Blum & François, 2006; Heath et al.,
2008). Mooers (1995) compared the level of tree
imbalance in a collection of published phylogenies
and found that incomplete trees are more imbalanced
than completely sampled phylogenies. Another study
by Heath et al. (2008) examined the effect of random
taxon sampling on empirical trees and on phylogenies
simulated under different models of cladogenesis.
They found that reduced taxon sampling of empirical
trees and trees simulated under variable and autocor-
related speciation and extinction rates causes an
increase in node imbalance. These results suggest that
poor taxon sampling leads to an increase in the ap-
parent rate variation because of the overrepresentation
of older nodes. The bias caused by incomplete species
sampling must be considered when using phylogenies
to test hypotheses about species diversity.
5 Innovations in reconstruction algorithms
and the analysis of large data sets
Until recently, computational constraints on
phylogenetic analyses made inclusion of large num-
bers of taxa impractical for many biologists. However,
developments in computational power, parallel com-
putation, and phylogenetic algorithms have greatly
decreased computational constraints for phylogenetic
analyses of many taxa, even for the most computa-
tionally demanding parametric approaches (e.g.,
Brauer at al., 2002; Lemmon & Milinkovitch, 2002;
Guindon & Gascuel, 2003; Ronquist & Huelsenbeck,
2003; Stamatakis, 2006; Minh et al., 2005; Zwickl,
2006). Quick but imprecise clustering techniques,
such as the widely used neighbor-joining algorithm
(Saitou & Nei, 1987), are rapidly being replaced by
methods that more thoroughly explore solution space
using a clearly defined model of evolution (imple-
mented in such programs as RAXML, GARLI,
PhyML, IQPNNI, MRBAYES, and PAUP*; see http:
//evolution.genetics.washington.edu/phylip/software.
html for more information). Analyses of hundreds to
thousands of taxa have become routine for parsimony,
maximum likelihood, and Bayesian approaches, and
analyses of tens-of-thousands of taxa are now feasible,
even for parametric methods. Therefore, computa-
tional constraints can no longer be viewed as a serious
impediment to thorough taxon sampling. Instead, the
limitations of taxon sampling have now shifted to
problems of taxon availability and the constraints of
specimen and data collection. Given the many benefits
of thorough taxon sampling summarized in this paper,
we advise biologists to carefully consider taxon
sampling design in planning, conducting, and inter-
preting phylogenetic analyses. In many cases, in-
creasing taxon sampling is one of the most practical
and beneficial approaches to increasing the accuracy
of phylogenetic estimates and the biological infer-
ences that are derived from phylogenetic trees.
Acknowledgements We thank professors Deyuan
Hong, Zhiduan Chen, Chengxin Fu, Michael J.
Donoghue, and Yin-Long Qiu for hosting the sympo-
sium on “Evolutionary Biology in the 21st Cen-
tury—Tracing Patterns of Evolution through the Tree
of Life,” and for the invitation to D.M.H. to present
this paper. The symposium was supported by a grant
from the National Natural Science Foundation of
China. A.J. Abrams, J. Brown, M. Morgan, G. Pauly,
R. Springman, M. Swenson, R. Symula, D. Zwickl,
Hervé Philippe, and an anonymous reviewer provided
helpful comments on this manuscript. D.M.H. grate-
fully acknowledges grant support from the United
States National Science Foundation (NSF). T.A.H.
was funded by a graduate research traineeship pro-
vided by an NSF IGERT grant in Computational
Phylogenetics and Applications to Biology awarded to
the University of Texas, Austin. S.M.H. was sup-
ported by an NSF pre-doctoral fellowship.
References
Ackerly DD. 2000. Taxon sampling, correlated evolution, and
independent contrasts. Evolution 54: 1480–1492.
Agapow PM, Purvis A. 2002. Power of eight tree shape
statistics to detect nonrandom diversification: a
comparison by simulation of two models of cladogenesis.
Systematic Biology 51: 866–872.
Albrecht C, Kuhn K, Streit B. 2007. A molecular phylogeny of
Planorboidea (Gastropoda, Pulmonata): insights from
enhanced taxon sampling. Zoologica Scripta 36: 27–39.
Alfaro ME, Santini F, Brock CD. 2007. Do reefs drive
diversification in marine teleosts? Evidence from the
pufferfish and their allies (order Tetraodontiformes).
Evolution 61: 2104–2126.
Ammerman LK, Hillis DM. 1992. A molecular test of bat
relationships: Monophyly or diphyly? Systematic Biology
41: 222–232.
Anderson FE, Swofford DL. 2004. Should we be worried about
long-branch attraction in real data sets? Investigations
using metazoan 18S rDNA. Molecular Phylogenetics and
Journal of Systematics and Evolution Vol. 46 No. 3 2008 252
Evolution 33: 440–451.
Ane C, Burleigh JG, McMahon MM, Sanderson MJ. 2005.
Covarion structure in plastid genome evolution: A new
statistical test. Molecular Biology and Evolution 22:
914–924.
Ane C, Sanderson MJ. 2005. Missing the forest for the trees:
Phylogenetic compression and its implications for
inferring complex evolutionary histories. Systematic
Biology 54: 146–157.
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.
Becerra JX. 2005. Timing the origin and expansion of the
Mexican tropical dry forest. Proceedings of the National
Academy of Sciences USA 102: 10919–10923.
Bell CD. 2007. Phylogenetic placement and biogeography of
the North American species of Valerianella (Valerianaceae:
Dipsacales) based on chloroplast and nuclear DNA.
Molecular Phylogenetics and Evolution 44: 929–941.
Bergsten J. 2005. A review of long-branch attraction. Cladistics
21: 163–193.
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.
Blum MGB, Françios O. 2006. Which random processes
describe the Tree of Life? A large-scale study of
phylogenetic tree imbalance. Systematic Biology 55:
685–691.
Boussau B, Gouy M. 2006. Efficient likelihood computations
with nonreversible models of evolution. Systematic
Biology 55: 756–768.
Brauer MJ, Holder MT, Dries LA, Zwickl DJ, Lewis PO, Hillis
DM. 2002. Genetic algorithms and parallel processing in
maximum-likelihood phylogeny inference. Molecular
Biology and Evolution 19: 1717–1726.
Braun EL, Kimball RT. 2002. Examining basal avian
divergences with mitochondrial sequences: Model
complexity, taxon sampling, and sequence length.
Systematic Biology 51: 614–625.
Bremer B, Jansen RK, Oxelman B, Backlund M, Lantz H, Kim
KJ. 1999. More characters or more taxa for a robust
phylogeny—case study from the coffee family
(Rubiaceae). Systematic Biology 48: 413–435.
Brown JM, Lemmon AR. 2007. The importance of data
partitioning and the utility of bayes factors in Bayesian
phylogenetics. Systematic Biology 56: 643–655.
Bruno WJ, Halpern AL. 1999. Topological bias and
inconsistency of maximum likelihood using wrong
models. Molecular Biology and Evolution 16: 564–566.
Bush RM, Fitch WM, Bender CA, Cox NJ. 1999. Positive
selection on the H3 hemagglutinin gene of human
influenza virus A. Molecular Biology and Evolution 16:
1457–1465.
Chan KMA, Moore BR. 1999. Accounting for mode of
speciation increases power and realism of tests of
phylogenetic asymmetry. American Naturalist 153:
332–346.
Chan KMA, Moore BR. 2002. Whole-tree methods for
detecting differential diversification rates. Systematic
Biology 51: 855–865.
Chen WJ, Bonillo C, Lecointre G. 2003. Repeatability of clades
as a criterion of reliability: A case study for molecular
phylogeny of Acanthomorpha (Teleostei) with larger
number of taxa. Molecular Phylogenetics and Evolution
26: 262–288.
Cobbett A, Wilkinson M, Wills MA. 2007. Fossils impact as
hard as living taxa in parsimony analyses of morphology.
Systematic Biology 56: 753–766.
Crandall KA, Bininda-Emonds ORP, Mace GM, Wayne RK.
2000. Considering evolutionary processes in conservation
biology. Trends in Ecology and Evolution 15: 290–295.
Cummings MP, Meyer A. 2005. Magic bullets and golden
rules: Data sampling in molecular phylogenetics. Zoology
108: 329–336.
Cunningham CW, Jeng K, Husti J, Badgett M, Molineux IJ,
Hillis DM, Bull JJ. 1997. Parallel molecular evolution of
deletions and nonsense mutations in bacteriophage T7.
Molecular Biology and Evolution 14: 113–116.
Cunningham CW, Zhu H, Hillis DM. 1998. Best-fit
maximum-likelihood models for phylogenetic inference:
Empirical tests with known phylogenies. Evolution 52:
978–987.
Darst CR, Menendez-Guerrero PA, Coloma LA, Cannatella
DC. 2005. Evolution of dietary specialization and
chemical defense in poison frogs (Dendrobatidae): A
comparative analysis. American Naturalist 165: 56–69.
DeBry RW. 1992. The consistency of several phylogeny-
inference methods under varying evolutionary rates.
Molecular Biology and Evolution 9: 537–551.
DeBry RW. 2005. The systematic component of phylogenetic
error as a function of taxonomic sampling under
parsimony. Systematic Biology 54: 432–440.
Delsuc F, Brinkmann H, Philippe H. 2005. Phylogenomics and
the reconstruction of the Tree of Life. Nature Reviews
Genetics 6: 361–375.
Delsuc F, Phillips MJ, Penny D. 2003. Comment on “Hexapod
origins: Monophyletic or paraphyletic?” Science 301:
1482.
Delsuc F, Scally M, Madsen O, Stanhope MJ, de Jong WW,
Catzeflis FM, Springer MS, Douzery EJP. 2002.
Molecular phylogeny of living xenarthrans and the impact
of character and taxon sampling on the placental tree
rooting. Molecular Biology and Evolution 19: 1656–1671.
Donoghue MJ, Doyle JA, Gauthier J, Kluge AG, Rowe T. 1989.
The importance of fossils in phylogeny reconstruction.
Annual Review of Ecology and Systematics 20: 431–460.
Drummond AJ, Ho SY, Phillips MJ, Rambaut A. 2006. Relaxed
phylogenetics and dating with confidence. PLoS Biology
4: e88.
Eisen JA. 1998. Phylogenomics: Improving functional
predictions for uncharacterized genes by evolutionary
analysis. Genome Research 8: 163–167.
Felsenstein J. 1978. Cases in which parsimony or compatibility
methods will be positively misleading. Systematic
Zoology 27: 401–410.
Felsenstein J. 1985. Phylogenies and the comparative method.
American Naturalist 125: 1–15.
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

253
Felsenstein J. 2004. Inferring phylogenies. Sunderland, MA:
Sinauer Associates.
Fitch WM, Beintema JJ. 1990. Correcting parsimonious trees
for unseen nucleotide substitutions—the effect of dense
branching as exemplified by ribonuclease. Molecular
Biology and Evolution 7: 438–443.
Fitch WM, Bruschi M. 1987. The evolution of prokaryotic
ferredoxins—with a general method correcting for
unobserved substitutions in less branched lineages.
Molecular Biology and Evolution 4: 381–394.
Foster PG. 2004. Modeling compositional heterogeneity.
Systematic Biology 53: 485–495.
Foster PG, Jermiin LS, Hickey DA. 1997. Nucleotide
composition bias affects amino acid content in proteins
coded by animal mitochondria. Journal of Molecular
Evolution 44: 282–288.
Freudenstein JV, Pickett KM, Simmons MP, Wenzel JW. 2003.
From basepairs to birdsongs: phylogenetic data in the age
of genomics. Cladistics 19: 333–347.
Galtier N, Gouy M. 1998. Inferring pattern and process:
maximum-likelihood implementation of a nonhomo-
geneous model of DNA sequence evolution for
phylogenetic analysis. Molecular Biology and Evolution
15: 871–879.
Gascuel O, Bryant D, Denis F. 2001. Strengths and limitations
of the minimum evolution principle. Systematic Biology
50: 621–627.
Gatesy J, DeSalle R, Wahlberg N. 2007. How many genes
should a systematist sample? Conflicting insights from a
phylogenomic matrix characterized by replicated
incongruence. Systematic Biology 56: 355–363.
Gaucher EA, Thomson JM, Burgan MF, Benner SA. 2003.
Inferring the palaeoenvironment of ancient bacteria on the
basis of resurrected proteins. Nature 425: 285–288.
Geuten K, Massingham T, Darius P, Smets E, Goldman N.
2007. Experimental design criteria in phylogenetics:
Where to add taxa. Systematic Biology 56: 609–622.
Gojobori T, Ishii K, Nei M. 1982. Estimation of average
number of nucleotide substitutions when the rate of
substitution varies with nucleotide. Journal of Molecular
Evolution 18: 414–423.
Goldman N. 1993. Simple diagnostic statistical tests of models
for DNA substitution. Journal of Molecular Evolution 37:
650–661.
Goldman N. 1998. Phylogenetic information and experimental
design in molecular systematics. Proceedings of the Royal
Society of London Series B-Biological Sciences 265:
1779–1786.
Goldman N, Yang Z. 1994. Codon-based model of nucleotide
substitution for protein-coding DNA-sequences. Mole-
cular Biology and Evolution 11: 725–736.
Gowri-Shankar V, Rattray M. 2007. A reversible jump method
for Bayesian phylogenetic inference with a nonhomo-
geneous substitution model. Molecular Biology and
Evolution 24: 1286–1299.
Graybeal A. 1994. Evaluating the phylogenetic utility of
genes—a search for genes informative about deep
divergences among vertebrates. Systematic Biology 43:
174–193.
Graybeal A. 1998. Is it better to add taxa or characters to a
difficult phylogenetic problem? Systematic Biology 47:
9–17.
Guindon S, Gascuel O. 2003. A simple, fast, and accurate
algorithm to estimate large phylogenies by maximum
likelihood. Systematic Biology 52: 696–704.
Guyer C, Slowinski JB. 1991. Comparisons of observed
phylogenetic topologies with null expectations among 3
monophyletic lineages. Evolution 45: 340–350.
Guyer C, Slowinski JB. 1993. Adaptive radiation and the
topology of large phylogenies. Evolution 47: 253–263.
Hasegawa M, Hashimoto T. 1993. Ribosomal-RNA trees
misleading? Nature 361: 23.
Heard SB. 1992. Patterns in tree balance among cladistic,
phenetic, and randomly generated phylogenetic trees.
Evolution 46: 1818–1826.
Heath TA, Zwickl DJ, Kim J, Hillis DM. 2008. Taxon sampling
affects inferences of macroevolutionary processes from
phylogenetic trees. Systematic Biology 57: 160–166.
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.
Hibbett D. 2004. Trends in morphological evolution in
homobasidiomycetes inferred using maximum likelihood:
a comparison of binary and multistate approaches.
Systematic Biology 53: 889–903.
Hillis DM. 1995. Approaches for assessing phylogenetic
accuracy. Systematic Biology 44: 3–16.
Hillis DM. 1996. Inferring complex phylogenies. Nature 383:
130–131.
Hillis DM. 1998. Taxonomic sampling, phylogenetic accuracy,
and investigator bias. Systematic Biology 47: 3–8.
Hillis DM. 1999. Phylogenetics and the study of HIV. In:
Crandall KA ed. The Evolution of HIV. Baltimore: The
Johns Hopkins University Press. 105–121.
Hillis DM, Allard MW, Miyamoto MM. 1993. Analysis of
DNA sequence data: phylogenetic inference. Methods in
Enzymology 224: 456–487.
Hillis DM, Huelsenbeck JP. 1994. Support for dental HIV
transmission. Nature 369: 24–25.
Hillis DM, Huelsenbeck JP, Cunningham CW. 1994a.
Application and accuracy of molecular phylogenies.
Science 264: 671–677.
Hillis DM, Huelsenbeck JP, Swofford DL. 1994b. Hobgoblin of
phylogenetics. Nature 369: 363–364.
Hillis DM, Pollock DD, McGuire JA, Zwickl DJ. 2003. Is
sparse taxon sampling a problem for phylogenetic
inference? Systematic Biology 52: 124–126.
Ho SYW. 2007. Calibrating molecular estimates of substitution
rates and divergence times in birds. Journal of Avian
Biology 38: 409–414.
Holland BR, Penny D, Hendy MD. 2003. Outgroup
misplacement and phylogenetic inaccuracy under a
molecular clock—A simulation study. Systematic Biology
52: 229–238.
Holman EW. 2005. Nodes in phylogenetic trees: The relation
between imbalance and number of descendent species.
Systematic Biology 54: 895–899.
Hoyle DC, Higgs PG. 2003. Factors affecting the errors in the
estimation of evolutionary distances between sequences.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 254
Molecular Biology and Evolution 20: 1–9.
Huelsenbeck JP. 1991. When are fossils better than extant taxa
in phylogenetic analysis. Systematic Zoology 40:
458–469.
Huelsenbeck JP. 1995. Performance of phylogenetic methods in
simulation. Systematic Biology 44: 17–48.
Huelsenbeck JP, Bollback JP. 2001. Empirical and hierarchical
Bayesian estimation of ancestral states. Systematic
Biology 50: 351–366.
Huelsenbeck JP, Hillis DM. 1993. Success of phylogenetic
methods in the 4-taxon case. Systematic Biology 42:
247–264.
Huelsenbeck JP, Lander KM. 2003. Frequent inconsistency of
parsimony under a simple model of cladogenesis.
Systematic Biology 52: 641–648.
Huelsenbeck JP, Larget B, Swofford D. 2000. A compound
Poisson process for relaxing the molecular clock. Genetics
154: 1879–1892.
Hug LA, Roger AJ. 2007. The impact of fossils and taxon
sampling on ancient molecular dating analyses. Molecular
Biology and Evolution 24: 1889–1897.
Hugall AF, Foster R, Lee MS. 2007. Calibration choice, rate
smoothing, and the pattern of tetrapod diversification
according to the long nuclear gene RAG-1. Systematic
Biology 56: 543–563.
Hugall AF, Lee MSY. 2007. The likelihood node density effect
and consequences for evolutionary studies of molecular
rates. Evolution 61: 2293–2307.
Jermiin LS, Ho SYW, Ababneh F, Robinson J, Larkum AWD.
2004. The biasing effect of compositional heterogeneity
on phylogenetic estimates may be underestimated.
Systematic Biology 53: 638–643.
Johnson KP. 2001. Taxon sampling and the phylogenetic
position of passeriformes: Evidence from 916 avian
cytochrome b sequences. Systematic Biology 50:
128–136.
Kim J. 1998. Large-scale phylogenies and measuring the
performance of phylogenetic estimators. Systematic
Biology 47: 43–60.
Kirkpatrick M, Slatkin M. 1993. Searching for evolutionary
patterns in the shape of a phylogenetic tree. Evolution 47:
1171–1181.
Kishino H, Hasegawa M. 1990. Converting distance to time:
application to human evolution. Methods in Enzymology
183: 550–570.
Kishino H, Thorne JL, Bruno WJ. 2001. Performance of a
divergence time estimation method under a probabilistic
model of rate evolution. Molecular Biology and Evolution
18: 352–361.
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
[online]. Molecular Biology and Evolution. doi:10.1093/
molbev/msn042. Available from mbe.oxfordjournals.org
[accessed 3 March 2008].
Langley CH, Fitch WM. 1974. An examination of the
constancy of the rate of molecular evolution. Journal of
Molecular Evolution 3: 161–177.
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.
Lecointre G, Philippe H, Van Le HL, Le Guyader H. 1993.
Species sampling has a major impact on phylogenetic
inference. Molecular Phylogenetics and Evolution 2:
205–224.
Leipe DD, Gunderson JH, Nerad TA, Sogin ML. 1993. Small
subunit ribosomal RNA+ of Hexamita inflata and the
quest for the first branch in the eukaryotic tree. Molecular
and Biochemical Parasitology 59: 41–48.
Lemmon AR, Brown JM, Stanger-Hall K, Lemmon EM. The
effect of missing data in likelihood-based phylogenetics.
Systematic Biology. (in press)
Lemmon AR, Milinkovitch MC. 2002. The metapopulation
genetic algorithm: An efficient solution for the problem of
phylogeny estimation. Proceedings of the National
Academy of Sciences USA 99: 10516–10521.
Lemmon AR, Moriarty EC. 2004. The importance of proper
model assumption in Bayesian phylogenetics. Systematic
Biology 53: 265–277.
Lepage T, Bryant D, Philippe H, Lartillot N. 2007. A general
comparison of relaxed molecular clock models. Molecular
Biology and Evolution 24: 2669–2680.
Lepage T, Lawi S, Tupper P, Bryant D. 2006. Continuous and
tractable models for the variation of evolutionary rates.
Mathematical Biosciences 199: 216–233.
Li P, Bousquet J. 1992. Relative-rate test for nucleotide
substitutions between two lineages. Molecular Biology
and Evolution 9: 1185–1189.
Lin YH, McLenachan PA, Gore AR, Phillips MJ, Ota R, Hendy
MD, Penny D. 2002. Four new mitochondrial genomes
and the increased stability of evolutionary trees of
mammals from improved taxon sampling. Molecular
Biology and Evolution 19: 2060–2070.
Lockhart PJ, Howe CJ, Bryant DA, Beanland TJ, Larkum
AWD. 1992. Substitutional bias confounds inference of
cyanelle origins from sequence data. Journal of Molecular
Evolution 34: 153–162.
Lockhart PJ, Larkum AWD, Steel MA, Waddell PJ, Penny D.
1996. Evolution of chlorophyll and bacteriochlorophyll:
The problem of invariant sites in sequence analysis.
Proceedings of the National Academy of Sciences USA
93: 1930–1934.
Lopez P, Casane D, Philippe H. 2002. Heterotachy, an
important process of protein evolution. Molecular Biology
and Evolution 19: 1–7.
Martins EP. 1996. Phylogenies and the comparative method in
animal behavior. Oxford: Oxford University Press.
Massingham T, Goldman N. 2000. EDIBLE: experimental
design and information calculations in phylogenetics.
Bioinformatics 16: 294–295.
McMahon MM, Sanderson MJ. 2006. Phylogenetic
supermatrix analysis of GenBank sequences from 2228
papilionoid legumes. Systematic Biology 55: 818–836.
Metzker ML, Mindell DP, Liu XM, Ptak RG, Gibbs RA, Hillis
DM. 2002. Molecular evidence of HIV-1 transmission in a
criminal case. Proceedings of the National Academy of
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

255
Sciences USA 99: 14292–14297.
Minh BQ, Vinh LS, von Haeseler A, Schmidt HA. 2005.
pIQPNNI: parallel reconstruction of large maximum
likelihood phylogenies. Bioinformatics 21: 3794–3796.
Mooers AO. 1995. Tree balance and tree completeness.
Evolution 49: 379–384.
Mooers AO, Holmes EC. 2000. The evolution of base
composition and phylogenetic inference. Trends in
Ecology and Evolution 15: 365–369.
Near TJ, Meylan PA, Shaffer HB. 2005. Assessing concordance
of fossil calibration points in molecular clock studies: An
example using turtles. American Naturalist 165: 137–146.
Near TJ, Sanderson MJ. 2004. Assessing the quality of
molecular divergence time estimates by fossil calibrations
and fossil-based model selection. Philosophical
Transactions of the Royal Society of London Series
B-Biological Sciences 359: 1477–1483.
Nee S, Holmes EC, May RM, Harvey PH. 1994a. Extinction
rates can be estimated from molecular phylogenies.
Philosophical Transactions of the Royal Society of
London Series B-Biological Sciences 344: 77–82.
Nee S, May RM, Harvey PH. 1994b. The reconstructed
evolutionary process. Philosophical Transactions of the
Royal Society of London Series B-Biological Sciences
344: 305–411.
Olsen GJ. 1987. Earliest phylogenetic branchings: comparing
rRNA-based evolutionary trees inferred with various
techniques. Cold Spring Harbor Symposia on Quantitative
Biology 52: 825–837.
Pagel M. 1994. Detecting correlated evolution on
phylogenies—a general-method for the comparative-
analysis of discrete characters. Proceedings of the Royal
Society of London Series B-Biological Sciences 255:
37–45.
Pagel M. 1999. The maximum likelihood approach to
reconstructing ancestral character states of discrete
characters on phylogenies. Systematic Biology 48:
612–622.
Pagel M, Meade A, Barker D. 2004. Bayesian estimation of
ancestral character states on phylogenies. Systematic
Biology 53: 673–684.
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, Lartillot N, Brinkmann H. 2005. 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 PWH,
Casane D. 2004. Phylogenomics of eukaryotes: Impact of
missing data on large alignments. Molecular Biology and
Evolution 21: 1740–1752.
Philippe H, Zhou Y, Brinkmann H, Rodrigue N, Delsuc F.
2005. Heterotachy and long-branch attraction in
phylogenetics. BMC Evolutionary Biology 5: 50.
Phillips MJ, Delsuc F, Penny D. 2004. Genome-scale
phylogeny and the detection of systematic biases.
Molecular Biology and Evolution 21: 1455–1458.
Poe S. 1998a. The effect of taxonomic sampling on accuracy of
phylogeny estimation: Test case of a known phylogeny.
Molecular Biology and Evolution 15: 1086–1090.
Poe S. 1998b. Sensitivity of phylogeny estimation to taxonomic
sampling. Systematic Biology 47: 18–31.
Poe S. 2003. Evaluation of the strategy of long-branch
subdivision to improve the accuracy of phylogenetic
methods. Systematic Biology 52: 423–428.
Poe S, Swofford DL. 1999. Taxon sampling revisited. Nature
398: 299–300.
Pollock DD, Bruno WJ. 2000. Assessing an unknown
evolutionary process: Effect of increasing site-specific
knowledge through taxon addition. Molecular Biology and
Evolution 17: 1854–1858.
Pollock DD, Taylor WR, Goldman N. 1999. Coevolving
protein residues: Maximum likelihood identification and
relationship to structure. Journal of Molecular Biology
287: 187–198.
Pollock DD, Zwickl DJ, McGuire JA, Hillis DM. 2002.
Increased taxon sampling is advantageous for
phylogenetic inference. Systematic Biology 51: 664–671.
Purvis A, Agapow PM. 2002. Phylogeny imbalance: taxonomic
level matters. Systematic Biology 51: 844–854.
Purvis A, Gittleman JL, Luh HK. 1994. Truth or consequences
—effects of phylogenetic accuracy on 2 comparative
methods. Journal of Theoretical Biology 167: 293–300.
Pybus OG, Harvey PH. 2000. Testing macro-evolutionary
models using incomplete molecular phylogenies.
Proceedings of the Royal Society of London Series
B-Biological Sciences 267: 2267–2272.
Rambaut A, Bromham L. 1998. Estimating divergence dates
from molecular sequences. Molecular Biology and
Evolution 15: 442–448.
Rannala B, Huelsenbeck JP, Yang Z, Nielsen R. 1998. Taxon
sampling and the accuracy of large phylogenies.
Systematic Biology 47: 702–710.
Ricklefs RE, Starck JM. 1996. Applications of phylogenetically
independent contrasts: A mixed progress report. Oikos 77:
167–172.
Robinson M, Gouy M, Gautier C, Mouchiroud D. 1998.
Sensitivity of the relative-rate test to taxonomic sampling.
Molecular Biology and Evolution 15: 1091–1098.
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.
Roelants K, Gower DJ, Wilkinson M, Loader SP, Biju SD,
Guillaume K, Moriau L, Bossuyt F. 2007. Global patterns
of diversification in the history of modern amphibians.
Proceedings of the National Academy of Sciences USA
104: 887–892.
Roger AJ, Hug LA. 2006. The origin and diversification of
eukaryotes: problems with molecular phylogenetics and
molecular clock estimation. Philosophical Transactions of
the Royal Society B-Biological Sciences 361: 1039–1054.
Rokas A, Kruger D, Carroll SB. 2005. Animal evolution and
the molecular signature of radiations compressed in time.
Science 310: 1933–1938.
Rokas A, Williams BL, King N, Carroll SB. 2003.
Genome-scale approaches to resolving incongruence in
molecular phylogenies. Nature 425: 798–804.
Ronquist F, Huelsenbeck JP. 2003. MrBayes 3: Bayesian
phylogenetic inference under mixed models.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 256
Bioinformatics 19: 1572–1574.
Rosenberg MS, Kumar S. 2001. Incomplete taxon sampling is
not a problem for phylogenetic inference. Proceedings of
the National Academy of Sciences USA 98: 10751–10756.
Rosenberg MS, Kumar S. 2003. Taxon sampling,
bioinformatics, and phylogenomics. Systematic Biology
52: 119–124.
Rutschmann F, Eriksson T, Abu Salim K, Conti E. 2007.
Assessing calibration uncertainty in molecular dating: The
assignment of fossils to alternative calibration points.
Systematic Biology 56: 591–608.
Ryan MJ, Rand AS. 1995. Female responses to ancestral
advertisement calls in túngara frogs. Science 269:
390–392.
Ryan MJ, Rand AS. 1998. Evoked vocal response in male
túngara frogs: pre-existing biases in male responses?
Animal Behavior 56: 1509–1516.
Saitou N, Nei M. 1987. The neighbor-joining method—a new
method for reconstructing phylogenetic trees. Molecular
Biology and Evolution 4: 406–425.
Salisbury BA, Kim JH. 2001. Ancestral state estimation and
taxon sampling density. Systematic Biology 50: 557–564.
Sanderson MJ. 1997. A nonparametric approach to estimating
divergence times in the absence of rate constancy.
Molecular Biology and Evolution 14: 1218–1231.
Sanderson MJ. 2002. Estimating absolute rates of molecular
evolution and divergence times: a penalized likelihood
approach. Molecular Biology and Evolution 19: 101–109.
Sanderson MJ. 2003. r8s: inferring absolute rates of molecular
evolution and divergence times in the absence of a
molecular clock. Bioinformatics 19: 301–302.
Sanderson MJ, Donoghue MJ. 1989. Patterns of variation in
levels of homoplasy. Evolution 43: 1781–1795.
Sanderson MJ, Wojciechowski MF, Hu JM, Khan TS, Brady
SG. 2000. Error, bias, and long-branch attraction in data
for two chloroplast photosystem genes in seed plants.
Molecular Biology and Evolution 17: 782–797.
Sarich VM, Wilson AC. 1973. Generation time and genomic
evolution in primates. Science 179: 1144–1147.
Schluter D, Price T, Mooers AO, Ludwig D. 1997. Likelihood
of ancestor states in adaptive radiation. Evolution 51:
1699–1711.
Shao KT, Sokal RR. 1990. Tree balance. Systematic Zoology
39: 266–276.
Smith AB, Pisani D, Mackenzie-Dodds JA, Stockley B,
Webster BL, Littlewood DT. 2006. Testing the molecular
clock: molecular and paleontological estimates of
divergence times in the Echinoidea (Echinodermata).
Molecular Biology and Evolution 23: 1832–1851.
Sorenson MD, Oneal E, Garcia-Moreno J, Mindell DP. 2003.
More taxa, more characters: the hoatzin problem is still
unresolved. Molecular Biology and Evolution 20:
1484–1498.
Spencer M, Susko E, Roger AJ. 2005. Likelihood, parsimony,
and heterogeneous evolution. Molecular Biology and
Evolution 22: 1161–1164.
Stamatakis A. 2006. RAxML-VI-HPC: maximum likelihood-
based phylogenetic analyses with thousands of taxa and
mixed models. Bioinformatics 22: 2688–2690.
Steel M. 1994. Recovering a tree from the leaf colourations it
generates under a Markov model. Applied Mathematics
Letters 7: 19–23.
Steel M, Szekely L, Erdos PL, Waddell P. 1993. A complete
family of phylogenetic invariants for any number of taxa
under Kimura 3ST model. New Zealand Journal of Botany
31: 289–296.
Stefanovic S, Rice DW, Palmer JD. 2004. Long branch
attraction, taxon sampling, and the earliest angiosperms:
Amborella or monocots? BMC Evolutionary Biology 4:
35.
Sullivan J, Swofford DL. 1997. Are guinea pigs rodents? The
utility of models in molecular phylogenetics. Journal of
Mammalian Evolution 4: 77–86.
Sullivan J, Swofford DL, Naylor GJP. 1999. The effect of taxon
sampling on estimating rate heterogeneity parameters of
maximum-likelihood models. Molecular Biology and
Evolution 16: 1347–1356.
Susko E, Inagaki Y, Roger AJ. 2004. On inconsistency of the
neighbor-joining, least squares, and minimum evolution
estimation when substitution processes are incorrectly
modeled. Molecular Biology and Evolution 21:
1629–1642.
Swofford DL, Olsen JL, Waddell PJ, Hillis DM. 1996.
Phylogenetic inference. In: Hillis DM, Moritz C, Mable
BK eds. Molecular systematics. Sunderland, MA: Sinauer
Associates. 407–514.
Takezaki N, Rzhetsky A, Nei M. 1995. Phylogenetic test of the
molecular clock and linearized trees. Molecular Biology
and Evolution 12: 823–833.
Thorne JL, Kishino H. 2002. Divergence time and evolutionary
rate estimation with multilocus data. Systematic Biology
51: 689–702.
Thorne JL, Kishino H, Painter IS. 1998. Estimating the rate of
evolution of the rate of molecular evolution. Molecular
Biology and Evolution 15: 1647–1657.
Van Den Bussche RA, Baker RJ, Huelsenbeck JP, Hillis DM.
1998. Base compositional bias and phylogenetic analyses:
A test of the “flying DNA” hypothesis. Molecular
Phylogenetics and Evolution 10: 408–416.
Venditti C, Meade A, Pagel M. 2006. Detecting the
node-density artifact in phylogeny reconstruction.
Systematic Biology 55: 637–643.
Webster AJ, Payne RJH, Pagel M. 2003. Molecular phylogenies
link rates of evolution and speciation. Science 301:
478–478.
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.
Wiens JJ. 2003. Missing data, incomplete taxa, and
phylogenetic accuracy. Systematic Biology 52: 528–538.
Wiens JJ. 2005. Can incomplete taxa rescue phylogenetic
analyses from long-branch attraction? Systematic Biology
54: 731–742.
Wiens JJ, Reeder TW. 1995. Combining data sets with different
numbers of taxa for phylogenetic analysis. Systematic
Biology 44: 548–558.
Wu CI, Li WH. 1985. Evidence for higher rates of nucleotide
substitution in rodents than in man. Proceedings of the
National Academy of Sciences USA 82: 1741–1745.
Xia X. 2006. Topological bias in distance-based phylogenetic
methods: problems with over- and under-estimated genetic
HEATH et al.: Taxon sampling and the accuracy of phylogenetic analyses

257
distances. Evolutionary Bioinformatics 2: 375–387.
Yang Z. 1994. Statistical properties of the maximum-likelihood
method of phylogenetic estimation and comparison with
distance matrix-methods. Systematic Biology 43:
329–342.
Yang Z. 2006. Computational Molecular Evolution. New York:
Oxford University Press.
Yang Z, Goldman N. 1997. Are big trees indeed easy? Trends
in Ecology and Evolution 12: 357–357.
Yang Z, Rannala B. 1997. Bayesian phylogenetic inference
using DNA sequences: a Markov chain Monte Carlo
method. Molecular Biology and Evolution 14: 717–724.
Yang Z, Rannala B. 2006. Bayesian estimation of species
divergence times under a molecular clock using multiple
fossil calibrations with soft bounds. Molecular Biology
and Evolution 23: 212–226.
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.
Yang Z, Yoder AD. 2003. Comparison of likelihood and
Bayesian methods for estimating divergence times using
multiple gene loci and calibration points, with application
to a radiation of cute-looking mouse lemur species.
Systematic Biology 52: 705–716.
Yoder AD, Yang Z. 2000. Estimation of primate speciation
dates using local molecular clocks. Molecular Biology and
Evolution 17: 1081–1090.
Zhou Y, Holmes EC. 2007. Bayesian estimates of the
evolutionary rate and age of hepatitis B virus. Journal of
Molecular Evolution 65: 197–205.
Zuckerkandl E, Pauling L. 1962. Molecular disease, evolution,
and genetic heterogeneity. In: Kasha M, Pullman B eds.
Horizons in Biochemistry. New York: Academic Press.
189–225.
Zwickl DJ. 2006. Genetic algorithm approaches for the
phylogenetic analyses of large biological sequence
datasets under the maximum likelihood criterion. Ph.D.
Dissertation. Austin, TX: The University of Texas at
Austin.
Zwickl DJ, Hillis DM. 2002. Increased taxon sampling greatly
reduces phylogenetic error. Systematic Biology 51:
588–598.