Group Title: BMC Evolutionary Biology
Title: A Higher-level MRP supertree of placental mammals
Full Citation
Permanent Link:
 Material Information
Title: A Higher-level MRP supertree of placental mammals
Physical Description: Book
Language: English
Creator: Beck, Robin
Bininda-Emonds, Olaf
Cardillo, Marcel
Liu, Fu-Guo
Purvis, Andy
Publisher: BMC Evolutionary Biology
Publication Date: 2006
Abstract: BACKGROUND:The higher-level phylogeny of placental mammals has long been a phylogenetic Gordian knot, with disagreement about both the precise contents of, and relationships between, the extant orders. A recent MRP supertree that favoured 'outdated' hypotheses (notably, monophyly of both Artiodactyla and Lipotyphla) has been heavily criticised for including low-quality and redundant data. We apply a stringent data selection protocol designed to minimise these problems to a much-expanded data set of morphological, molecular and combined source trees, to produce a supertree that includes every family of extant placental mammals.RESULTS:The supertree is well-resolved and supports both polyphyly of Lipotyphla and paraphyly of Artiodactyla with respect to Cetacea. The existence of four 'superorders' – Afrotheria, Xenarthra, Laurasiatheria and Euarchontoglires – is also supported. The topology is highly congruent with recent (molecular) phylogenetic analyses of placental mammals, but is considerably more comprehensive, being the first phylogeny to include all 113 extant families without making a priori assumptions of suprafamilial monophyly. Subsidiary analyses reveal that the data selection protocol played a key role in the major changes relative to a previously published higher-level supertree of placentals.CONCLUSION:The supertree should provide a useful framework for hypothesis testing in phylogenetic comparative biology, and supports the idea that biogeography has played a crucial role in the evolution of placental mammals. Our results demonstrate the importance of minimising poor and redundant data when constructing supertrees.
General Note: Start page 93
General Note: M3: 10.1186/1471-2148-6-93
 Record Information
Bibliographic ID: UF00100000
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: Open Access:
Resource Identifier: issn - 1471-2148


This item has the following downloads:


Full Text

BMC Evolutionary Biology Central

Research article

A higher-level MRP supertree of placental mammals
Robin MD Beck*1,2,3, Olaf RP Bininda-Emonds4,5, Marcel Cardillo', Fu-
Guo Robert Liu6 and Andy Purvis'

Address: 'Division of Biology, Imperial College London, Silwood Park campus, Ascot SL5 7PY, UK, 2Natural History Museum, Cromwell Road,
London SW7 5BD, UK, 3School of Biological, Earth and Environmental Sciences, University of New South Wales, NSW 2052, Australia, 4Lehrstuhl
fiur Tierzucht, Technical University of Munich, 85354 Freising-Weihenstephan, Germany, 5Institut fuir Spezielle Zoologie und Evolutionsbiologie
mit Phyletischem Museum, Friedrich-Schiller-Universitdit Jena, 07743 Jena, Germany and 6Department of Zoology, Box 118525, University of
Florida, Gainesville, Florida 32611-8552, USA
Email: Robin MD Beck*; Olaf RP Bininda-Emonds;
Marcel Cardillo; Fu-Guo Robert Liu; Andy Purvis
* Corresponding author

Published: 13 November 2006 Received: 23 June 2006
BMC Evolutionary Biology 2006, 6:93 doi: 10.1 186/1471-2148-6-93 Accepted: 13 November 2006
This article is available from:
2006 Beck et al; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Background: The higher-level phylogeny of placental mammals has long been a phylogenetic
Gordian knot, with disagreement about both the precise contents of, and relationships between,
the extant orders. A recent MRP supertree that favoured 'outdated' hypotheses (notably,
monophyly of both Artiodactyla and Lipotyphla) has been heavily criticised for including low-quality
and redundant data. We apply a stringent data selection protocol designed to minimise these
problems to a much-expanded data set of morphological, molecular and combined source trees,
to produce a supertree that includes every family of extant placental mammals.
Results: The supertree is well-resolved and supports both polyphyly of Lipotyphla and paraphyly
of Artiodactyla with respect to Cetacea. The existence of four 'superorders' Afrotheria,
Xenarthra, Laurasiatheria and Euarchontoglires is also supported. The topology is highly
congruent with recent (molecular) phylogenetic analyses of placental mammals, but is considerably
more comprehensive, being the first phylogeny to include all 113 extant families without making a
priori assumptions of suprafamilial monophyly. Subsidiary analyses reveal that the data selection
protocol played a key role in the major changes relative to a previously published higher-level
supertree of placentals.
Conclusion: The supertree should provide a useful framework for hypothesis testing in
phylogenetic comparative biology, and supports the idea that biogeography has played a crucial role
in the evolution of placental mammals. Our results demonstrate the importance of minimising poor
and redundant data when constructing supertrees.

Background biogeography. Until relatively recently, most comprehen-
The higher-level phylogeny of placental mammals has sive studies have relied purely on morphological data.
long been one of the most intensively studied problems in Such studies largely upheld the monophyly of all 18 [7]
systematics (e.g [1-6]), because a robust placental phylog- traditionally recognized orders but were rather less suc-
eny is crucial to understanding mammalian evolution and

Page 1 of 14
(page number not for citation purposes)


cessful in resolving the relationships between the orders
(e.g. [41).

Recent sophisticated analyses of molecular sequence data
have significantly revised and refined the view from mor-
phology, resulting in a well-resolved 'molecular consen-
sus' view of placental phylogeny [8,91 that is broadly
supported by many genes and gene combinations (see
Table 1). This consensus rejects the monophyly of two tra-
ditional placental orders: Artiodactyla (even-toed 'ungu-
lates') is paraphyletic with respect to Cetacea (whales;
[10,11]) and Lipotyphla (the insectivoress') is diphyletic
[12,13], being split into Afrosoricida and Eulipotyphla. At
the interordinal level, molecular data consistently resolve
extant placental groups into four 'superorders': Afrotheria,
Xenarthra, Laurasiatheria and Euarchontoglires (the latter
two comprising Boreoeutheria). Despite this recent
progress, regions of the topology are still uncertain, as dif-
ferent data types (e.g. nuclear genes, mitochondrial genes,
morphology) and methods of analysis (e.g. maximum
parsimony, maximum likelihood) often support conflict-
ing relationships. Notably, the location of the root of the
placental tree remains unresolved [8,14-161, and the pre-
cise relationships within each superorder are also some-
what unclear. Perhaps more importantly, taxonomic

Table I: Superorders and selected additional supraordinal clades
phylogeny (e.g. [8,9]).

coverage remains far from complete: the taxonomically
most inclusive higher-level single-matrix analysis of mam-
mals so far, that of Murphy et al. [17], included represent-
atives of only 54 of the 113 extant placental families
recognized by Wilson and Reeder [7]. Studies directly
combining molecular and morphological data have been
even more taxonomically limited, tending to focus on
specific areas of contention, such as afrotherian mono-
phyly [18], or relationships within Cetartiodactyla [11].
This is because comprehensiveness in such analyses is dif-
ficult to achieve, given the typically patchy taxonomic dis-
tribution of available data [19,20] that can be analysed in
a single matrix.

Supertree analysis provides an alternative route to com-
prehensive estimates of phylogeny [21]. This approach
combines existing phylogenetic tree topologies ('source
trees'), rather than their underlying data, by any of a
number of methods most commonly Matrix Represen-
tation with Parsimony (MRP; [22,23]). This procedure
produces a composite phylogeny, or 'supertree', that can
be taxonomically more comprehensive than any source
tree. Because supertree analyses sample at the level of tree
topologies [24], source trees based on any data (e.g. dis-
tances, which cannot be incorporated into ordinary phyl-

currently supported by the 'molecular consensus' view of placental

Supraordinal clades





Common names of orders

African insectivoress' tenrecss and golden moles)
Elephant shrews










Anteaters and sloths

Flying lemurs
Tree shrews

True insectivoress' (hedgehogs, shrews, true moles and Solenodon)
Even-toed ungulatess' and whales
Odd-toed ungulatess'

Orders follow [7] except that Xenarthra is divided into Cingulata and Pilosa, Lipotyphla insectivoreses') is split between Afrosoricida and
Eulipotyphla, and Cetacea (whales) and Artiodactyla (even-toed ungulatess) are combined as Cetartiodactyla. Higher-level names are taken from

Page 2 of 14
(page number not for citation purposes)



BMC Evolutionary Biology 2006, 6:93

genetic character matrices), can be used. As a result,
supertrees can be based on the broadest sampling of both
data and taxa, and so are often taxonomically more com-
prehensive than phylogenies of the same clades produced
by more direct approaches (e.g. [25-27]). Supertrees of
many clades have now been published, almost exclusively
using MRP (see [28] for a recent review).

Liu et al. ([29]; henceforth 'LEA') used a supertree
approach to infer the relationships among placental
mammal families from a combination of morphological
and molecular source trees. Their combined supertree,
based on 430 source trees from 315 references published
before March 1999, still remains by far the most compre-
hensive higher-level phylogeny of placentals published.

Overall, the LEA combined supertree ([29]; their Fig. 1)
seemed a reasonable compromise between the morpho-
logical and molecular phylogenies then available [29].
However, it conflicted with the majority of more recent
data in parts of its topology, supporting instead 'outdated'
views of placental phylogeny (see [11]). Most notably,
Artiodactyla was monophyletic, contradicting a wealth of
evidence already then available (and subsequently greatly
reinforced) for a Cetacea + Hippopotamidae (hippos)
clade (= Whippomorpha; summarised in [11]). Further-
more, interfamilial relationships within Artiodactyla
appeared anomalous [111. Monophyly of Lipotyphla was
also strongly supported, contradicting the association
between Afrosoricida and the other taxa (Paenungulata,
Macroscelidea and Tubulidentata) now considered to
comprise Afrotheria [30]. This was despite considerable
molecular evidence for both lipotyphlan polyphyly and
afrotherian monophyly prior to March 1999 (e.g.
[12,30]), both of which were actually reflected in the
molecular-only supertree of LEA ([29]; their Fig. 2A).

Gatesy et al. [11] argued in detail that the 'outdated' fea-
tures of the LEA supertree stemmed from any or all of 1)
uncritical selection of source trees that represent poor and
duplicated data; 2) assumptions of ordinal monophyly
without basis in the underlying data ('appeals to author-
ity'); and 3) inherent, methodological shortcomings in
the MRP method, if not the supertree approach as a whole
(see also [31]). Concentrating on the relationship
between Artiodactyla and Cetacea, Gatesy et al. [11]
claimed that all of the 33 MRP pseudocharacters support-
ing the monophyly of Artiodactyla in the combined
supertree derived from low quality source trees that repre-
sented 'appeals to authority, duplications of data, miscod-
ings, or derivatives of poorly justified trees' [11 ].

Motivated by concerns about source tree quality and
duplication in supertree analyses, Bininda-Emonds et al.
([32]; summarised in [33]) proposed a set of guidelines

for identifying suitable source trees, filtering out trees rep-
resenting duplicated or poor data, and minimizing
assumptions of higher-taxon monophyly. The underlying
principle of the guidelines is that only those source trees
that can be considered to represent 'independent phyloge-
netic hypotheses' should be included in a supertree.
Bininda-Emonds et al. [32] proposed that source trees
produced from independent character sets, such as differ-
ent genes or different morphological character sets, all
represent such independent phylogenetic hypotheses.
They also contended that different combinations of genes
and/or morphological characters likewise comprise inde-
pendent phylogenetic hypotheses, because of the possibil-
ity of signal enhancement (sensu [34]). To minimise data
duplication, they suggested that, where no clear cut choice
for a single tree presents itself for a given independent
character set (e.g. a particular gene), MRP 'mini-super-
trees' of all non-independent source trees based on that
character set should be created. Thus, each dataset
adjudged to be independent according the protocol is ulti-
mately represented (as far as possible) by a single, taxo-
nomically inclusive tree either an original source tree or
a 'mini-supertree' in the final supertree analysis. This
protocol has already been followed in the construction of
species-level supertrees of extant marsupials [25] and
cetartiodactyls [26].

Here, we apply the Bininda-Emonds et al. [32] guidelines
to a large set of source trees, including all those used by
LEA but also those from references published between
March 1999 (LEA's cut-off date) and September 2004, to
investigate higher-level placental phylogeny. We include
all 113 extant placental families, plus two recently extinct
and enigmatic groups Nesophontidae (West Indian
shrew-like insectivoress', currently included in Lipotyphla;
[7]) and Plesiorycteropodidae (a myrmecophagous form
recently assigned its own order, Bibymalagasia; [35]) the
relationships of which may be crucial to a better under-
standing of both the biogeographical history and patterns
of character evolution within placentals [36]. We use a
modified, 'semi-rooted' version of MRP that can compen-
sate for source trees that are not robustly rooted [37]. We
assess the degree of support for nodes in the supertree
using a supertree-specific support measure, reduced qual-
itative support (rQS; [26]); this varies from -1 (no sup-
port) to +1 (support from all relevant source trees), and is
described in Methods.

As a subsidiary analysis, we apply the guidelines of
Bininda-Emonds et al. [32] to the same 315 references
used by LEA in their combined supertree. By reproducing
their methodology as far as possible (e.g. exclusion of spe-
cific taxa, weighting of specific source trees, use of stand-
ard rooted MRP coding), except where these conflict with
the recommendations of the protocol, we aim to assess

Page 3 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

0022 2 eldae

^0 3^3~~+ -------------------------------- No lenldae PH
0 --0 0 40--atan-st--a--+0-00 Zlphlldae

Figure I

Supertree of extant placentals (50% majority rule consensus of 17 most parsimonious trees length =
8150.935), following application of the protocol of Bininda-Emonds et al. [32] to the complete set of refer-
ences. Asterisks indicate which branches collapse in the strict consensus. Numbers above branches represent reduced qualita-
tive support (rQS; [26,39]) values. The orders are indicated by brackets and the first three letters of their names following
Table I, with the additional fossil order Bibymalagasia indicated by BIB. Whippomorpha and the four superorders are also indi-

Page 4 of 14
(page number not for citation purposes)
Supe~~~~~~~~~~ree~~(pg ofetnnlcntlu5%mjriyrlbosnsso 7ms prioiu rer -lnoth fo=iainupss

BMC Evolutionary Biology 2006, 6:93

+0.043 Balaenidae
-- +1- Eschrichtiidae I
1 11 +0 032 Balaeno ptidae
SDelphinidae O CET
8 .036 +0.039 17 Phocoenidae 00
0 8Monodntida
+0.12C 0.098 +0.0 24 + 01 5 P.ata istidae
+0.782 +0.240 Hippo otamidae
S7+0.08 -Suidae
0Tayas uidae
+g ------------------------------------------Camelidae
,0 1 Equid
'00.1931 01 [- +0.019 Rhinocerotidae PER
S0 ---- Tapiridae
+7_______ 0 ________ i------------------- CARNIVORA
0.000 i Craseonycteridae
0--- --- aO* i ------- y t idae
+0.009 +4 Rhinopo ae

+7 -- +0.002 Megader--- lo tidae
+ 1I +5'! 04 | 0 00 N te i e
,0.227 -[ h-- stainolophidae

+0.4 + 0 +0Furi 0.0020e F r idae
+0. 7 -0 04 00 | 0 00 0 N t idae
------------------------------------ M y op o d id a e
+ 0 +l0.0 o idae
+0.141 00stac in-idae

+-- +0+0o6 +0.004 0Ldae
+6 I-----+10-+-.....---3-- Phllp ...idae

+0.028 +0 J041 SIR
0.0 +25. ~\ ___ +0.013 ____ |--------- le h n i a PRO
+6 + Noci idae 3 HYR

+0.89 +0.045 0.0 2109 C oce lidae MAC OER
+ -.t024 +3.5 T p idae SCA

+0009+0 rinace idae 0
+7 +OOO6 eno tide

+2 + 004ori idae
+9---- 0 0 3 -----+I-------------------------- Talyp oidae J
+0.013 +0101
+11 ~~~~ ~ ... I--------------------------------------- Dugmeon hgidae
0.02 +F-1 Elephantida PRO
0 .......00 0 0 +21. 1 1rcaidae 3HYR E

Figure 2ae MAC

Supertree of extant placentals (50% majority rule consensus of 5540 most parsimonious trees length =
4262.625), following application of the protocol of Bininda-Emonds et al. [32] to only those references used by

values, and numbers below represent decay indices. The orders and Whippomorpha are indicated and bracketed as in Figure I,
as are the two superorders (Xenarthra and Afrotheria) recovered as monophyletic in this analysis.
+Page 5 of 14

(page number not for citation purposes)
+5 +0.002 [TneiaChrscl1 ia AFR
+7 T-dpdia
+0.009 [1eaoncia
+0.004 N1
+0.013 +1 Daspo idae EE
+11 Myrmecphaidae

Figure 2
Supertree of extant placentals (50% majority rule consensus of 5540 most parsimonious trees length=
4262.625), following application of the protocol of Bininda-Emonds et al. [32] to only those references used by
Liu et al. [29]. Asterisks indicate which branches collapse in the strict consensus. Numbers above branches represent rQS
values, and numbers below represent decay indices. The orders and Whippomnorpha are indicated and bracketed as in Figure 1,
as are the two superorders (Xenarthra and Afrotheria) recovered as monophyletic in this analysis.

Page 5 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

the specific impact of the protocol on the overall supertree
topology. Specifically, we focus on whether monophyly of
both Artiodactyla and Lipotyphla are affected in terms of
changes in topology, or in support values as measured by
the decay index (DI; [38]). We also examine whether
other changes in topology and support are in better or
worse agreement with contemporary evidence. This will
help determine whether the criticized aspects of the origi-
nal LEA topology reflect an inherent, unavoidable, weak-
ness of supertree analysis per se, or avoidable weaknesses
in the source dataset that was originally used and that can
be remedied using a suitable protocol for source tree col-

Results and discussion
The search of the final MRP supertree matrix from the full
analysis recovered 17 most parsimonious trees of length
8150.935 (the tree length is not a whole number because
of the downweighting procedure used to account for the
presence of nonmonophyletic families in some source
trees; see Methods). The strict consensus of the 17 trees is
highly resolved, with the only conflict occurring within
the hystricognath rodents. The 50% majority rule consen-
sus is illustrated here (Figure 1), with those branches that
collapse in the strict consensus identified by asterisks.
There are no unsupported novel clades (sensu [39]).
Repeating the analysis with the extinct Nesophontidae
and Plesiorycteropodidae deleted from the original source
trees has no effect on the higher-level relationships among
the extant taxa.

The supertree presented here is highly congruent with
most recent estimates of placental phylogeny at both the
inter- and intraordinal levels [8,9]. However, because the
primary goal of our analyses was to investigate ordinal
composition and interordinal relationships, we did not
include single order source trees in our dataset. As such,
the intraordinal relationships presented here are not
based on the maximum amount of data available.
Although even this amount of data has yielded relation-
ships that are largely congruent with current phylogenetic
opinion, we would recommend the use of relevant super-
trees (e.g. [26,27]) or other similarly comprehensive phy-
logenies for intraordinal relationships.

Four principal clades, or 'superorders', are present: Afroth-
eria (rQS = +0.112), Xenarthra (rQS = +0.052), Laurasia-
theria (rQS = +0.186) and Euarchontoglires (rQS = -
0.085). In upholding the monophyly of these super-
orders, this supertree supports the hypothesis that plate
tectonics have been crucial in the early evolution of mod-
em placentals [40]. The superorders may have undergone
their initial divergences in biogeographical areas that were
separate throughout much of the Cretaceous and Ceno-
zoic: Afrotheria in Afro-Arabia, Xenarthra in South Amer-

ica, and both Euarchontoglires and Laurasiatheria in
Laurasia [40]. However, recent studies suggest that a
number of fossil 'condylarths' from Laurasia are afrothe-
rian [18,41], conflicting with a strict tectonic-based inter-
pretation of placental phylogeny. Regardless, these four
superorders indicate that morphological convergence has
been more pervasive than previously thought [8,36,42].

In agreement with most recent phylogenetic analyses of
placental mammals, the supertree upholds the mono-
phyly of 16 of the 18 extant orders recognized in Wilson
and Reeder [7]. Although a 'seed tree' that assumes mono-
phyly of all 18 of these orders (including Artiodactyla and
Lipotyphla) was included as a source tree (see Methods),
the 16 that are monophyletic in the supertree are sup-
ported by between 17 (Sirenia) and 156 (Primates) other
source trees. Lipotyphla is polyphyletic, with Afrosoricida
in Afrotheria and Eulipotyphla (here including the extinct
Nesophontidae) in Laurasiatheria, and Artiodactyla is par-
aphyletic with respect to Cetacea.

The supertree supports Afrotheria as the sister to the
remaining superorders, in agreement with most nuclear
and nuclear + mitochondrial trees (e.g. [43-45]). A recent
analysis of retroposon integration [16] supported a
xenarthran root, congruent with morphological evidence
for a split between Xenarthra and all other extant placen-
tals (Epitheria) [46], although alternatives could not be
statistically rejected. Within Afrotheria, the first split is
between Paenungulata (rQS = +0.121) and Afroinsec-
tiphilia (rQS = -0.019; here including the extinct Ple-
siorycteropodidae). Within Paenungulata, Procaviidae
(Hyracoidea) and Dugongidae + Trichechidae (Sirenia)
are sister taxa (rQS = -0.014), in agreement with some
sequence data (e.g. [13,43]) and retroposons [47]. Within
Afroinsectiphilia, the supertree recovers both Afrosoricida
(rQS = +0.023) and Afroinsectivora (Afrosoricida + Mac-
roscelididae; rQS = -0.029), with Orycteropodidae
(Tubulidentata) as the sister to Afroinsectivora; this is
again congruent with most sequence data (e.g. [8]),
although chromosome-painting supports monophyly of
(Macroscelididae + Orycteropodidae) [48] and retro-
posons support monophyly of (Afrosoricida + Oryctero-
podidae) [47]. Based on source trees from MacPhee [35]
and Asher et al. [18], Plesiorycteropodidae is recovered as
the sister to Orycteropodidae, indicating that the extinct
bibymalagasy is afrotherian, as might be suspected from
its known distribution (the Holocene of Madagascar;
[35]) and from features of its astragalus that are shared
with a number of extant afrotherians [35,36]. Relation-
ships within Xenarthra, the only superorder that is cur-
rently also supported by morphology, are congruent with
both morphological [49] and molecular [14] evidence.

Page 6 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

Euarchonta (rQS = +0.058) and Glires (rQS = -0.112) are
both monophyletic, together forming the clade Euarchon-
toglires. The low rQS value for Glires probably reflects the
inclusion of source trees that support rodent polyphyly or
paraphyly (e.g. [50,511), although morphological [18]
and most recent molecular phylogenies [17,43,52] sup-
port rodent monophyly, as recovered here. Tupaiidae
(Scandentia) form the sister group to a Cynocephalidae
(Dermoptera) + Primates clade. The supertree topology
within Euarchontoglires, at both the inter- and intra-ordi-
nal levels, is highly congruent with most recent, mainly
molecular evidence [17,43,44,52].

Within Laurasiatheria, a monophyletic Eulipotyphla (rQS
= +0.018) is the sister to the remaining taxa. This contra-
dicts the hypothesis that Erinaceidae (hedgehogs) are
basal placentals, as has been suggested by mitochondrial
trees (e.g. [50]),. A Solenodontidae + Nesophontidae (rQS
= -0.001) clade is congruent with biogeographic evidence,
as both taxa are known only from the West Indies, but
compelling evidence for the true affinities of Nesophonti-
dae is still lacking [36]; the position advocated for it here
is based on only three source trees. A sister-group relation-
ship between Erinaceidae and Soricidae (shrews) to the
exclusion of Talpidae (true moles) agrees with most
molecular estimates (e.g. [53]), but is only relatively
weakly supported here (rQS = -0.004). Within the remain-
ing taxa, Chiroptera (including a paraphyletic Microchi-
roptera with respect to Megachiroptera) are the sister
group to Fereuungulata (rQS = +0.176). Carnivora and
Manidae (Pholidota) together form Ferae (rQS = +0.039),
with Cetartiodactyla and Perissodactyla as sister taxa (=
Cetungulata; rQS = +0.081). Different molecular data
continue to yield incompatible topologies within Fereu-
ungulata (see [44]); the topology favoured here is argua-
bly more congruent with morphological data because the
sister relationship between cetartiodactyls and perissodac-
tyls requires only a single origin of 'ungulate' features
within Laurasiatheria. However, a recent transposon anal-
ysis [54] recovered a clade comprising carnivorans, peris-
sodactyls and bats (pholidotans were not sampled, but are
also probably members of this group), which has been
named Pegasoferae. Artiodactyla is paraphyletic, with
Whippomorpha (rQS = +0.139) as the sister to the rumi-
nants, forming Cetruminantia (rQS = +0.127); Suidae +
Tayassuidae (pigs + peccaries; rQS = +0.073) and Cameli-
dae (camels) comprise successive sister groups. The
cetartiodactylan topology is congruent with both molecu-
lar [10] and combined morphological and molecular [11]

Overall, our supertree topology is in much better agree-
ment with the current consensus view of placental phylog-
eny than is that of LEA. Why? The three main possibilities
are (a) that the LEA topology resulted from either poor

and/or duplicated data, or assumptions of monophyly,
which the Bininda-Emonds et al. [32] guidelines have
largely removed; (b) that other, minor, differences in the
technical details of the two studies are responsible, or (c)
that phylogenies published after March 1999 (the cut-off
point of LEA) are more in agreement with the molecular
consensus, and that these studies are now in a majority.
Our subsidiary analysis in effect, repeating the LEA anal-
ysis using the Bininda-Emonds et al. [32] guidelines can
help discriminate between these three possibilities.

The subsidiary analysis found 5535 trees of length
4262.625, using the same 4:1 weighting scheme of LEA
(see Methods). A strict consensus is fully resolved at the
interordinal level (the only conflicts are within rodents
and bats), and there are no novel unsupported clades.
Again, although a 'seed tree' that assumes monophyly of
the orders recognized by Wilson and Reeder [7] was
included as a source tree (see Methods), those that are
monophyletic in the supertree are supported by between
six (Sirenia) and 60 (Rodentia) other source trees. Figure
2 is a 50% majority rule consensus, with branches that
collapse in the strict consensus indicated by asterisks. The
equally weighted analysis (not shown) recovers largely
identical unrooted relationships, with neither Artiodac-
tyla nor Lipotyphla recovered as monophyletic.

Artiodactyla is paraphyletic, with Cetacea and Hippopot-
amidae forming Whippomorpha. Support values (DI = 6;
rQS = +0.099) indicate that this clade is relatively well-
supported, and are similar to those for Ruminantia (DI =
7; rQS = +0.124). Whippomorpha and Ruminantia are
sisters, forming Cetruminantia, which also has reasonable
support (DI = 6; rQS = +0.111).

Lipotyphla is polyphyletic, with separate eulipotyphlan
(DI = 7, rQS = +0.009) and afrosoricid (DI = 7, rQS =
+0.002) clades. Significantly, Afrosoricida is part of a
monophyletic Afrotheria (DI = 5, rQS = +0.045), the exist-
ence of which was controversial in 1999. Afrotheria was
not recovered in the combined supertree of LEA, although
it was present in their molecular-only supertree (their Fig-
ure 2A).

The two major changes from the LEA topology Cetacea
nesting within a paraphyletic Artiodactyla, and diphyly of
Lipotyphla (both of which were recovered in the LEA
molecular-only supertree) seen in this reanalysis are
both in better accord with the state of phylogenetic
knowledge in 1999, and are in agreement with our full
supertree (Figure 1). They indicate that the potential prob-
lem of 'time-lag' in supertrees, where inclusion of older
studies biases the supertree topology towards outdated
views of relationships, is not an inherent limitation of the

Page 7 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

Notably, the DI support values in this reanalysis are
almost always lower, and in many cases greatly so, than
their equivalents in the original combined LEA supertree.
For example, DI support for the monophyly of the order
Chiroptera, drops from 74 to 6, and similarly large drops
are seen for Lagomorpha (69 to 5), Perissodactyla (139 to
10) and Rodentia (37 to 3). Some interordinal groupings
also show reduced DI values (e.g. Glires, 26 to 5; Ferae, 21
to 7; Paenungulata, 108 to 25.5). These declines probably
reflect the exclusion of some duplicate trees and, particu-
larly, the avoidance of a priori assumptions of monophyly.
As such, the DI values in this analysis are probably a more
accurate indication of the actual support for each group.

Table 2 lists the relative similarity of different topologies
as measured by the normalised partition metric [55,56]
and 'explicitly agree' triplets. It indicates that both the
application of the source tree selection protocol of
Bininda-Emonds et al. [32] and the inclusion of more
recent source trees are important in explaining the differ-
ences between our updated supertree topology and the
original LEA supertree. For instance, the 4:1 upweighted
supertree from the subsidiary analysis ('LEA+P 4:1') is
~18% more similar to the large molecular tree of Murphy
et al. ('MEA'; [17]) than is the original LEA supertree,
according to the normalised partition metric. This effect is
attributable solely to the application of the protocol,
which was sufficient to bring the LEA dataset in line with
the molecular consensus in a number of key areas. The full
supertree, however, was ~15% more similar again to the
Murphy et al. [17] tree. It is also only ~9% different from
the large molecular and morphological analysis of Gatesy
et al. ('GEA'; [111). These latter results reflect the inclusion
of the more recent source trees published since the study
by LEA. The 'explicitly agree' triplet scores confirm these

The supertree from our main analysis is a well-resolved,
comprehensive, and reasonably robust higher-level phyl-

ogeny of placental mammals. It agrees strongly with the
weight of current data (e.g. [11,17,18,42,43,45]), suggest-
ing that MRP supertrees can accurately reflect available
phylogenetic evidence (contra [11]). To our knowledge, it
is the first placental phylogeny of any kind to include all
extant families, and has over two times the taxonomic
coverage of the most comprehensive non-supertree analy-
sis so far [17].

The supertree is based on a large set of stringently-selected
source trees derived from analyses of a very wide range of
characters and character types (including morphology,
mitochondrial genes and nuclear genes) analysed using
improved coding [32,37], searching [57] and robustness-
checking [26] methods from those used in the previous
supertree assessment of placental phylogeny by LEA. It
appears from our subsidiary analysis that at least some of
the key differences between our supertree and the original
LEA study lie with the selection of independent source
trees and in the avoidance of a priori assumptions of
monophyly. This finding confirms that the inclusion of
poor or duplicated data is not inherent to supertree con-
struction (as implied by [11]; see [33]), although, as in all
areas of science, it remains an issue of which researchers
need to be mindful.

The supertree hopefully provides a valuable, comprehen-
sive framework for research into the evolution and bioge-
ography of placental mammals. We suggest that this
topology is suitable for use in comparative studies that
require a higher-level phylogeny of placentals. Supertrees,
if carefully constructed, can combine apparent accuracy
(as judged by available character evidence) with compre-
hensiveness, suggesting that they may play an important
role in phylogenetics for some time to come.

Finding and Filtering Source Trees
The 315 references used by LEA are listed online as supple-
mentary information to their paper [58]. To identify addi-

Table 2: Normalised partition metric [55,56] and 'explicitly agree' triplet scores of supertrees and supermatrices.

'Explicitly agree triplets'

LEA+P 1:1
LEA+P 4:1
Full ST


0.241 I
0.02 I

LEA+P 1:1



Normalised partition metric
LEA+P 4:1 Full ST









'LEA' = Combined morphological and molecular supertree of Liu et al. ([29]; their Figure I); 'LEA+P 1:1' = 1:1 equally weighted supertree following
application of protocol to the LEA references (50% majority rule consensus; not shown); 'LEA+P 4:1' = 4: I upweighted supertree following
application of protocol to the LEA references (50% majority rule consensus; Figure 2); 'Full ST' = extended analysis supertree, based on an updated
set of references (50% majority rule consensus; Figure I); 'MEA' = molecular topology of Murphy et al. ([ 17] their Figure I); 'GEA' = morphological
and molecular topology of Gatesy et al. ([I I]; their Figure 4).

Page 8 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

tional relevant references that might contain further
source trees, we searched BioAbstracts, Web of Knowl-
edge, Zoological Record and BIOSIS online literature
databases using the following search terms: mammal*,
euther* or placental* together with any of phylogen*, sys-
tematic*, cladistic*, classif*, taxonom*, cladogram*, phe-
nogram* or fossil*. We examined the online abstracts
(where available) of the ~3000 initial references identi-
fied, and excluded those that did not appear to contain
relevant phylogenetic information. The remaining ~1000
(including supplementary information such as electronic
appendices) were examined in full, as were all of the LEA

We rejected potential source trees for any of several possi-
ble reasons. Trees that did not provide unequivocal evi-
dence that actual datasets underlie their topologies (e.g.,
many reviews, taxonomies and informal composites of
existing phylogenies) were rejected; we considered une-
quivocal evidence to include character lists, apomorphy
lists, sequence alignments, character matrices or distance
matrices. Trees reproduced from earlier references (and
thus dissociated from their underlying datasets) were also
excluded, although we examined the original references
where possible. Source trees in which characters were
mapped onto an independent topology were rejected,
unless the authors demonstrated that the distribution of
the mapped characters was congruent with the assumed
tree. References lacking phylogeny depictions and not
providing sufficient information in the text to infer a rea-
sonably well-resolved source tree were not used, nor were
those that included only unrooted trees, unless the pres-
ence of non-placental taxa or clearly identified paralogous
genes made rooting uncontroversial. References contain-
ing only source trees whose terminal taxa could not be
identified to the family-level or below for example mor-
phological studies where taxa are not identified beyond
the ordinal level, or molecular studies that employ inter-
familial chimeric sequences (see [59]) were also not
used. LEA coded such trees with each order replaced by an
unresolved polytomy comprising its constituent families,
but because the composition of the currently recognized
placental orders (Lipotyphla and Artiodactyla, in particu-
lar) is in question, in addition to their interrelationships,
we considered it necessary to exclude trees that would
have forced us to assume ordinal monophyly a priori.
Source trees that included some terminal taxa that were
above the family-level, but that were otherwise suitable
for inclusion in the final supertree, were coded with the
suprafamilial taxa deleted. Because our focus is on inter-
ordinal relationships, in general we only coded additional
source trees for the full analysis that included representa-
tives of at least three placental orders recognized by Wil-
son and Reeder [7]. Exceptions to this were artiodactyl-
only, lipotyphlan-only and rodent-only trees (with repre-
sentatives from at least three families), all of which were

coded because the monophyly of each of these orders has
been seriously challenged in recent years ([11,18,50]
respectively). The number of taxa present in each source
tree varied between three and 55.

This initial filtering rejected 93 of the references originally
used by LEA, leaving 222 for reanalysis. These comprised
the complete set of initial source trees for our subsidiary
analysis (see below). Trees from 208 further publications
also met the filtering criteria. The topologies of all suitable
trees presented in these 430 references [see Additional file
1 ] such as multiple most parsimonious cladograms and/
or trees produced under different phylogenetic methods
(e.g. parsimony, distance and likelihood) and weighting
schemes [see Additional file 2] constitute the data set for
our full analysis. They were reproduced by importing an
appropriate taxon list into TreeView 1.6.6 [60], changing
the resultant 'bush' to the appropriate topology based on
all relevant information present as diagrams, tables and
accompanying text (where sufficient to imply an informa-
tive phylogeny), and saving it as a NEXUS-formatted tree-
file [61]. We always chose the optimal trees (or consensus
thereof), where indicated, over constrained or sub optimal
trees preferred by the authors based on a priori assump-
tions as to correct phylogeny of placentals (which repre-
sent 'appeals to authority' sensu [11]). However, if
multiple optimal trees were presented, and the authors
explicitly preferred one or a subset of these, we followed
this preference [32]. In cases of gene paralogy in molecu-
lar analyses, where the same species may be represented in
multiple different positions within the same tree, all pos-
sible permutations of the positions of each placental
taxon were entered.

To standardise terminal taxa among source trees, all taxa
in all source trees were initially synonymised by hand to
species using the taxonomy presented in Wilson and
Reeder [7]. In the absence of specific information, sub-
families and families were synonymised with the type spe-
cies of the genus giving them their names (following
[32]). For example, Bos, Bovinae and Bovidae were all
coded as Bos taurus. Terminal taxa that could not be iden-
tified to the family-level or below were pruned from the
source trees, and source trees with fewer than three taxa
remaining were not used. Taxa represented only by com-
mon names that did not unequivocally identify families
(e.g. 'monkey') were likewise deleted.

Species-level terminal taxa were then synonymised to
higher-level terminals using the Perl script synonoTree
[32], following Wilson and Reeder's [7] taxonomy. For
those source trees where synonymisation resulted in non-
monophyletic terminals (i.e. members of the same higher
taxon did not form a monophyletic group in the original
source tree), synonoTree outputs multiple trees with the

Page 9 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

non-monophyletic terminal taxa in each of their possible

For the subsidiary analysis, species-level terminals were
likewise synonymised to family-level, except that carnivo-
rans and primates were synonymised to order-level, as in
LEA. Non-placental terminals were deleted, as were the
families Ctenodactylidae, Ctenomyidae, Moschidae, Neo-
balaenidae and Petromuridae, which were excluded by
LEA because their inclusion led to a considerable loss of
resolution in their original analysis.

Establishing Independent Source Trees
Bininda-Emonds et al. [32] advocated that only 'inde-
pendent evolutionary hypotheses' should be included in
a supertree analysis (but see [31] for a critique of Bininda-
Emonds et al.'s definition of independence). Source trees
that represent the same character and taxon sets (e.g., mul-
tiple most-parsimonious trees, or maximum parsimony
and maximum likelihood trees of the same dataset) are
clearly non-independent. We combined each set of such
non-independent source trees into a single 'mini-super-
tree' [321, for both our full and subsidiary data sets in turn.
To identify non-independent source trees (sensu [32]), all
source trees were initially sorted into groups representing
the same character set (e.g. all MTCO1 trees, all 12S + 16S
rRNA trees or all DNA-hybridisation trees), with gene
names synonymised where possible according to the tax-
onomy proposed by the Human Genome Organisation
Gene Nomenclature Committee [62] and the GeneCards
database [63]. We have assumed that different introns,
exons or domains of the same gene represent the same
non-independent character set in this study, unless there
was strong evidence to the contrary. Within each group of
non-independent source trees, if multiple trees from the
same reference and representing the exact same data set
were present (e.g. multiple most parsimonious cladog-
rams), these were combined into a mini-supertree, which
could then be used to represent that dataset in the final
supertree analysis.

If, after this procedure, any of a group of non-independent
source trees or mini-supertrees was a strict taxonomic sub-
set of any other, the taxonomically less inclusive source
tree (or trees) was excluded from the final analysis as
being redundant. If there was only partial taxonomic over-
lap between source trees representing the same character
set, we did not create a mini-supertree of these, as any lack
of resolution in the mini-supertree may be because of
insufficient taxonomic overlap, rather than genuine
incongruence between the source trees. Instead, these par-
tially overlapping source trees were included separately.

MRP analyses
We used matrix representation with parsimony (MRP;
[22,23]) for both the mini- and overall supertree analyses:
each source tree is encoded using additive binary coding,
with each taxon coded as '1' if it descends from a particu-
lar node in the source tree, '0' if it does not, and '?' if it is
not present in that source tree. This procedure is per-
formed for all informative nodes in the source tree. A sin-
gle matrix containing the combined 'matrix
representations' of every source tree is then subjected to
parsimony analysis; the resultant most parsimonious tree
(or trees) is the supertree, and contains every taxon
present in any source tree [22,23]. All MRP matrices were
generated using the Perl script SuperMRP.

Within our full dataset, all MRP matrices were produced
using 'semi-rooted' MRP coding [37]. This modification
of standard MRP coding does not use an all-zero 'MRP
outgroup' to root every source tree, but only those where
the position of the root is held to have been determined
robustly. As such, the method does not enforce question-
able rooting decisions present in the source trees, such as
rooting based on a priori assumptions about the relation-
ships of the in-group. This modification may be particu-
larly advisable for groups where the position of the root
remains unclear, such as placentals (see [15]). Here, we
consider the presence of non-placental outgroup taxa
(such as marsupials and non-mammals) or paralogous
genes to represent robust rooting information. We syno-
nymised all such 'real outgroups' to the name 'Real_OG',
and used this taxon to root the MRP supertrees. For our
subsidiary analysis, we instead followed LEA and used
standard MRP coding with the hypothetical, all-zero MRP
outgroup common to all source trees to root the supertree

The resultant MRP matrices were analysed using PAUP*
4.0b10 [64]. We used reversible parsimony with all char-
acters weighted equally, unless some of the source trees
contained non-monophyletic families, in which case we
downweighted the associated MRP characters appropri-
ately. For example, a single non-monophyletic family in
two distinct positions in a single initial source tree would
be included in two, non-independent source trees (in a
different position in each), and the MRP characters corre-
sponding to those trees would each be given a weight of
0.5. Although weighting of MRP characters in proportion
to the degree of support for their corresponding nodes has
been shown to improve performance [65], we could not
implement this in our study due to the non-comparable
indices used (e.g. bootstrapping, jackknifing, decay indi-
ces, Bayesian posterior probabilities) in different source
trees, and the absence of support values of any kind for
many source trees.

Page 10 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

Branch-and-bound tree searches were used for all our
mini-supertree analyses, and the mini-supertree was taken
to be the strict consensus of all equally most parsimoni-
ous solutions. The final MRP matrices of both full and
subsidiary data sets were analysed using the parsimony
ratchet [57], with the PAUP* instruction block produced
using the Perl script PerlRat. For the full analysis, 20
batches of 500 replicates were carried out, with 25% of the
characters randomly chosen to be upweighted by a factor
of two in each ratchet replicate, followed by a brute force
heuristic search starting from the set of shortest trees
found among all 20 batches. The subsidiary matrix was
considerably smaller, so 50 batches of 500 replicates were
carried out, again followed by a brute force search. TBR
branch swapping was employed in all ratchet searches.
For the iterative reweighting steps, a maximum of one tree
was held at each step, whereas the maximum number of
trees for final brute force searches was equal to the prod-
uct of the number of batches and 1 + the number of repli-

Final datasets
The full dataset included 725 trees [see Additional file 3],
of which 109 were MRP mini-supertrees, and 54 were due
to nonmonophyletic taxa in some source trees. 652 were
based on molecular data, 58 on morphology and 15 on
combined molecular and morphological data. Following
Bininda-Emonds and Sanderson [66], a 'seed tree' was
added to ensure sufficient overlap among source trees.
This assigned all 115 terminal taxa to their respective
orders without specifying any further relationships. Ordi-
nal membership came from Wilson and Reeder [7], except
Plesiorycteropodidae (not listed by [7]), which was
treated as an additional order, Bibymalagasia, following
MacPhee [35]. These tree descriptions were converted into
a 'semi-rooted' MRP matrix of 6715 pseudocharacters [see
Additional file 4]. We did not differentially weight MRP
characters from different source trees, apart from the
downweighting of multiple non-independent trees aris-
ing because nonmonophyletic families.

The subsidiary data set comprised 466 trees [see Addi-
tional file 5], of which 48 were MRP mini-supertrees, and
24 resulted because of nonmonophyletic taxa in some
source trees. 408 were based on molecular data, 43 on
morphology and ten on combined molecular and mor-
phological data. We again included a 'seed tree', as above.
Tree descriptions were converted into a standard MRP
matrix of 1857 pseudocharacters [see Additional file 6].
We followed LEA in performing two analyses, one with
equal weightings and one in which larger trees were
upweighted by a factor of four (this upweighted analysis
was the basis for their Figure 1), on the assumption that
such trees tend to be of higher quality. In the latter analy-
sis, we upweighted all source trees that were originally

upweighted by LEA and that we retained after application
of the protocol (53 in total).

The seed trees used in both analyses derive from the tax-
onomy presented in Wilson and Reeder [7], and therefore
violate the source tree collection guidelines (because the
taxonomy is not based on an explicit dataset). They were
chosen because the taxonomy is a widely used standard
for mammals, is fully comprehensive, and has a relatively
low information content ensuring that it will be easily
overruled by any robust source trees. However, because
the taxonomy supports ordinal monophyly, it will bias
both analyses slightly in this direction. Nevertheless, as we
discuss below and in Results and Discussion, the degree of
support for the orders whose monophyly is upheld is
much too great to be attributed to the seed trees alone.

Support values
We calculated the supertree-specific support measure,
reduced qualitative support (rQS; [26]) to assess the sup-
port for nodes in both the full and subsidiary analyses.
This measure is a modified version of qualitative support
(QS), as developed by Bininda-Emonds [39], in which
support for each supertree clade is calculated by compar-
ing the supertree with each of its source trees in turn. As
such, it avoids problems associated with the inherent
non-independence of MRP pseudocharacters that renders
the use of the more familiar support measures, such as the
bootstrap or decay index (DI), invalid [39,67]. Fortu-
nately, QS values are roughly correlated with bootstrap
values [39].

For rQS, each supertree clade is supported ('Hard Match')
contradicted ('Hard Mismatch'), or is neither supported
nor contradicted ('Equivocal) by each source tree. rQS val-
ues range from -1 to +1, indicating a greater proportion of
hard mismatches and hard matches among the set of
source trees, respectively. An rQS value of -1 indicates an
unsupported novel clade, the presence of which has been
argued by some to be a negative feature of MRP supertrees
(e.g. [68]). rQS avoids the problems that affect QS identi-
fied by Wilkinson et al. [69]. Other supertree-specific met-
rics for assessing support, such as V [69], triplet-based
methods [70] and modified bootstrap methods [71,72],
have also been recently proposed, but are not used here.
All rQS values were determined using the Perl script Qua-
liTree [39].

Results from the rQS analyses also confirmed that the
inclusion of seed trees had a minimal effect on the topol-
ogies of the resultant supertrees. For the full analysis, the
seed tree was informative for only 19 of the 113 nodes on
the supertree. It directly conflicted with 10 of these 19
nodes, indicating that it was being overruled about half
the time, and its removal did not affect rQS values signif-

Page 11 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

icantly (mean difference between values with and without
seed tree = -2.179 x 10-4, df= 18, t = -0.641, one-tailed P-
value = 0.74). These findings indicate that the supertree is
reflecting the signal from the 725 other source trees, rather
than the seed tree. Similar results were apparent for the
LEA+P analysis, where the seed tree conflicted with five of
the 13 informative nodes on this supertree and its
removal also did not alter rQS values significantly (mean
difference between values with and without seed tree =
7.352 x 10-5, df= 12, t = 0.113, one-tailed P-value = 0.46).

For the differentially-weighted subsidiary analysis, we
additionally computed DI values for each node, but solely
for comparison with the values reported by LEA, given
that the measure is not strictly valid in a supertree context.
Analyses used the program AutoDecay [73] to specify con-
straint trees and PerlRat to specify the ratchet search
parameters for PAUP*. Because of the large number of
nodes to be examined, the ratchet searches were more lim-
ited (two runs comprising 20 batches of 100 replicates
and one run comprising 5 batches of 200 replicates, for
each node) than that used to derive the entire supertree,
and the concluding brute force search was omitted. The
more limited nature of the searches means that the DI for
each node may overestimate the real value in some cases.

We used the normalised partition metric (also known as
the Robinson-Foulds topological distance [55,56]) and
'explicitly agree' triplets to quantify the topological differ-
ences between: 1) the full supertree (Figure 1), 2) the sub-
sidiary analysis of the LEA references alone, using the 4:1
weighting scheme (Figure 2), 3) the subsidiary analysis,
using 1:1 equal weighting (topology not shown), 4) the
original LEA combined supertree, 5) the topology of Mur-
phy et al. ([171; their Figure 1; this is the taxonomically
most comprehensive molecular phylogeny of placental
mammals currently available), and 6) the combined
molecular and morphological topology of Gatesy et al.
([11]; their Figure 4). The normalised partition metric
scores were calculated using the perl script partitionMet-
ric, whilst the 'explicitly agree' triplet scores were calcu-
lated using COMPONENT [74]; for both metrics, trees
pruned to have identical taxon sets for each pairwise com-

Authors' contributions
RMDB collected most of the source trees, carried out all of
the analyses and wrote most of the manuscript, as part of
an MSc in Advanced Methods in Taxonomy and Biodiver-
sity at Imperial College and the Natural History Museum,
London. ORPBE co-supervised RMDB, wrote the Perl
scripts, advised on the analyses, and wrote significant por-
tions of the manuscript. MC collected some of the source
trees, prepared parts of the supplementary file, and helped
write the manuscript. FGRL collected some of the source

trees and helped write the manuscript. AP conceived of
and developed the research project, supervised RMDB,
and wrote significant portions of the manuscript. All
authors read and approved the final manuscript.

Additional material

Additional file 1
RTF file listing all 315 original references used by Liu et al. [29] and rea-
sons for excluding source trees from some of these references, plus an addi-
tional 204 references identified as containing valid source trees.
Click here for file

Additional file 2
XLS file giving details of all source trees coded from the references listed
in Additional file 1.
Click here for file

Additional file 3
NEXUS file (can be viewed using a text editor) containing all source trees
used in the complete analysis, in parenthetical notation. Weights given to
each source tree are also listed.
Click here for file

Additional file 4
NEXUS file (can be viewed using a text editor) containing semi-rooted
MRP pseudocharacter matrix used in the complete analysis, the weighting
scheme used and 50% majority-rule consensus (illustrated in Figure 1)
...,, 111 parsimony analysis of this matrix.
Click here for file

Additional file 5
NEXUS file (can be viewed using a text editor) containing all source trees
used in the subsidiary LEA analysis, in parenthetical notation. Weights
given to each source tree are also listed.
Click here for file

Additional file 6
NEXUS file (can be viewed using a text editor) containing standard MRP
pseudocharacter matrix used in the complete analysis, 4:1 and 1:1 weight-
ing schemes and 50% majority-rule consensuses (for both weighting
schemes; 4:1 consensus illustrated in Figure 2) following parsimony anal-
ysis of this matrix.
Click here for file
2148-6-93-S6 .nex]

Page 12 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

We thank Rachel Tomlins, Lalitha Sundaram, Meredith Murphy Thomas,
Kate Jones and the librarians at Cambridge University, Imperial College and
the Natural History Museum, London, for their help in locating the many
references examined in this study, and Christopher Phennah for access to
phylogenetic software. All trees mentioned in this study plus the annotated
MRP matrices have been deposited in TreeBASE (Study Accession # SI 1620;
Matrix Accession # M29 19 and 2920). All perl scripts used in this paper can
be downloaded from the homepage of ORPBE, under "Programs" [75].
Financial support was provided by the BMBF (Germany) through the "Bio-
informatics for the Functional Analysis of Mammalian Genomes" project
(ORPBE), and by NERC (UK) through research grant N ER/A/S/2001/00581
(RMDB, AP, MC). Further financial support during the write-up of this work
was provided by the Leverhulme Trust through Study Abroad Studentship
SAS/301 10 (RMDB).

I. Gregory WK: The orders of mammals. Bulletin of the American
Museum of Natural History 1910, 27:1-524.
2. Simpson GG: The principles of classification and a classifica-
tion of mammals. Bulletin of the American Museum of Natural History
1945, 85:1-350.
3. McKenna MC: Toward a phylogenetic classification of the
Mammalia. In Phylogeny of the Primates Edited by: Luckett WP and
Szalay FS. New York, Plenum; 1975:21-46.
4. Novacek MJ: Mammalian phylogeny shaking the tree. Nature
1992, 356:121-125.
5. Szalay FS, Novacek MJ, McKenna MC: Mammal phylogeny. Vol-
ume 2. Placentals. New York, Springer-Verlag; 1993.
6. Rose KD, Archibald JD: The Rise of Placental Mammals: Origins
and Relationships of the Major Extant Clades. Baltimore, John
Hopkins University Press; 2005.
7. Wilson DE, Reeder DM: Mammal species of the world. Wash-
ington, D.C., Smithsonian Institution Press; 1993.
8. Springer MS, Stanhope MJ, Madsen 0, deJong WW: Molecules con-
solidate the placental mammal tree. Trends in Ecology & Evolu-
tion 2004, 1 9:430-438.
9. Springer MS, Murphy WJ, Eizirik E, O'Brien SJ: Molecular evidence
for major placental clades. In The Rise of Placental Mammals: Ori-
gin, Timing, and Relationships of the Major Extant Clades Edited by: Rose
KD and Archibald JD. Baltimore, The John Hopkins University Press;
10. Gatesy J, Milinkovitch M, Waddell V, Stanhope M: Stability of cla-
distic relationships between Cetacea and higher-level artio-
dactyl taxa. Systematic Biology 1999, 48:6-20.
I I. Gatesy J, Matthee C, DeSalle R, Hayashi C: Resolution of a super-
tree/supermatrix paradox. Systematic Biology 2002, 51:652-664.
12. Stanhope MJ, Waddell VG, Madsen 0, de Jong WW, Hedges SB,
Cleven CC, Kao DJ, Springer M: Molecular evidence for multiple
origins of Insectivora and for a new order of endemic African
insectivore mammals. Proceedings of the National Academy of Sci-
ences of the United States of America 1998, 95:9967-9972.
13. Roca AL, Bar-Gal GK, Eizirik E, Helgen KM, Maria R, Springer MS,
O'Brien SJ, Murphy WJ: Mesozoic origin for West Indian insec-
tivores. Nature 2004, 429:649-651.
14. Delsuc F, Scally M, Madsen 0, Stanhope MJ, de Jong WW, Catzeflis
FM, Springer MS, Douzery EJP: Molecular phylogeny of living
xenarthrans and the impact of character and taxon sampling
on the placental tree rooting. Mol Biol Evol 2002, 19:1656-1671.
15. de Jong WW, van Dijk MAM, Poux C, Kappe G, van Rheede T, Mad-
sen 0: Indels in protein-coding sequences of Euarchontog-
lires constrain the rooting of the eutherian tree. Molecular
Phylogenetics and Evolution 2003, 28:328-340.
16. Kriegs JO, Churakov G, Kiefmann M, Jordan U, Brosius J, Schmitz J:
Retroposed elements as archives for the evolutionary his-
tory of placental mammals. PLoS Biol 2006, 4:e9 1.
17. Murphy WJ, Eizirik E, Johnson WE, Zhang YP, Ryder OA, O'Brien SJ:
Molecular phylogenetics and the origins of placental mam-
mals. Nature 2001, 409:614-618.
18. Asher RJ, Novacek MJ, GeislerJH: Relationships of endemic Afri-
can mammals and their fossil relatives based on morpholog

ical and molecular evidence. journal of Mammalian Evolution 2003,
19. Sanderson MJ, Driskell AC: The challenge of constructing large
phylogenetic trees. Trends in Plant Science 2003, 8:374-379.
20. Bininda-Emonds ORP: Supertree construction in the genomic
age. In Molecular Evolution: Producing the Biochemical Data Edited by:
Zimmer EA and Roalson E. Elsevier; 2005:745-757.
21. Sanderson MJ, Purvis A, Henze C: Phylogenetic supertrees:
Assembling the trees of life. Trends in Ecology & Evolution 1998,
22. Baum BR: Combining trees as a way of combining datasets for
phylogenetic inference, and the desirability of combining
gene trees. Taxon 1992, 41:3-10.
23. Ragan MA: Phylogenetic inference based on matrix represen-
tation of trees. Molecular Phylogenetics and Evolution 1992, 1:53-58.
24. Bininda-Emonds ORP: Trees versus characters and the super-
tree / supermatrix "paradox". Systematic Biology 2004,
25. Cardillo M, Bininda-Emonds ORP, Boakes E, Purvis A: A species-
level phylogenetic supertree of marsupials. journal of Zoology
2004, 264:1 1-31.
26. Price SA, Bininda-Emonds ORP, Gittleman JL: A complete phylog-
eny of the whales, dolphins and even-toed hoofed mammals
(Cetartiodactyla). Biological Reviews 2005, 80:445-473.
27. Jones KE, Purvis A, MacLarnon A, Bininda-Emonds ORP, Simmons
NB: A phylogenetic supertree of the bats (Mammalia: Chi-
roptera). Biological Reviews 2002, 77:223-259.
28. Bininda-Emonds ORP: The evolution of supertrees. Trends in Ecol-
ogy & Evolution 2004, 19:315-322.
29. Liu FGR, Miyamoto MM, Freire NP, Ong PQ, Tennant MR, Young TS,
Gugel KF: Molecular and morphological supertrees for euthe-
rian (placental) mammals. Science 2001, 291:1786-1789.
30. Stanhope MJ, Madsen 0, Waddell VG, Cleven GC, DeJWW, Springer
MS: Highly congruent molecular support for a diverse super-
ordinal clade of endemic African mammals. Molecular Phyloge-
netics and Evolution 1998, 9:501-508.
31. GatesyJ, Baker RH, Hayashi C: Inconsistencies in arguments for
the supertree approach: supermatrices versus supertrees of
Crocodylia. Systematic Biology 2004, 53:342-355.
32. Bininda-Emonds ORP,Jones KE, Price SA, Cardillo M, Grenyer R, Pur-
vis A: Garbage in, garbage out: Data issues in supertree con-
struction. In Phylogenetic supertrees: Combining information to reveal
the tree of life Edited by: Bininda-Emonds ORP. Dordrecht, the Neth-
erlands, Kluwer Academic; 2004:267-280.
33. Bininda-Emonds ORP, Jones KE, Price SA, Grenyer R, Cardillo M,
Habib M, Purvis A, Gittleman JL: Supertrees are a necessary not-
so-evil: A comment on Gatesy et al. Systematic Biology 2003,
34. De Queiroz A, Donoghue MJ, Kim J: Separate versus combined
analysis of phylogenetic evidence. Annual Review of Ecology and
Systematics 1995, 26:657-681.
35. MacPhee RDE: Morphology, adaptations, and relationships of
Plesiorycteropus, and a diagnosis of a new order of eutherian
mammals. Bull Amer Mus Nat Hist 1994:1-214.
36. Helgen KM: Major mammalian clades: a review under consid-
eration of molecular and palaeontological evidence. Mamm
Biol 2003, 68:1-15.
37. Bininda-Emonds ORP, Beck RMD, Purvis A: Getting to the roots
of matrix representation. Systematic Biology 2005, 54:668-672.
38. Bremer K: The limits of amino acid sequence data in
angiosperm phylogenetic reconstruction. Evolution 1988,
39. Bininda-Emonds ORP: Novel versus unsupported clades:
Assessing the qualitative support for clades in MRP super-
trees. Systematic Biology 2003, 52:839-848.
40. Hedges SB: Afrotheria: Plate tectonics meets genomics. Pro-
ceedings of the National Academy of Sciences of the United States ofAmer-
ica 2001, 98:1-2.
41. Zack SP, Penkrot TA, Bloch JI, Rose KD: Affinities of'hyopsodon-
tids' to elephant shrews and a Holarctic origin of Afrotheria.
Nature 2005, 434:497-501.
42. Madsen 0, Scally M, Douady CJ, Kao DJ, DeBry RW, Adkins R,
Amrine HM, Stanhope MJ, de Jong WW, Springer MS: Parallel adap-
tive radiations in two major clades of placental mammals.
Nature 2001, 409:610-614.

Page 13 of 14
(page number not for citation purposes)

BMC Evolutionary Biology 2006, 6:93

43. Murphy WJ, Eizirik E, O'Brien SJ, Madsen 0, Scally M, Douady CJ,
Teeling E, Ryder OA, Stanhope MJ, de Jong WW, Springer MS: Res-
olution of the early placental mammal radiation using Baye-
sian phylogenetics. Science 2001, 294:2348-2351.
44. Waddell PJ, Shelley S: Evaluating placental inter-ordinal phylog-
enies with novel sequences including RAG I, gamma-fibrino-
gen, ND6, and mt-tRNA, plus MCMC-driven nucleotide,
amino acid, and codon models. Molecular Phylogenetics and Evolu-
tion 2003, 28:197-224.
45. Amrine-Madsen H, Koepfli KP, Wayne RK, Springer MS: A new phy-
logenetic marker, apolipoprotein B, provides compelling evi-
dence for eutherian relationships. Molecular Phylogenetics and
Evolution 2003, 28:225-240.
46. Shoshani J, McKenna MC: Higher taxonomic relationships
among extant mammals based on morphology, with
selected comparisons of results from molecular data. Molec-
ular Phylogenetics and Evolution 1998, 9:572-584.
47. Nishihara H, Satta Y, Nikaido M, Thewissen JG, Stanhope MJ, Okada
N: A retroposon analysis of Afrotherian phylogeny. Mol Biol
Evol 2005, 22:1823-1833.
48. Robinson TJ, Fu B, Ferguson-Smith MA, Yang F: Cross-species
chromosome painting in the golden mole and elephant-
shrew: support for the mammalian clades Afrotheria and
Afroinsectiphillia but not Afroinsectivora. Proc Biol Sci 2004,
49. Gaudin TJ: Phylogenetic relationships among sloths (Mamma-
lia, Xenarthra, Tardigrada): the craniodental evidence. Zoo-
logical journal of the Linnean Society 2004, 140:255-305.
50. Arnason U, Adegoke JA, Bodin K, Born EW, Esa YB, Gullberg A, Nils-
son M, Short RV, Xu XF, Janke A: Mammalian mitogenomic rela-
tionships and the root of the eutherian tree. Proc NatlAcad Sci
USA 2002, 99:8151-8156.
51. Misawa K, Janke A: Revisiting the Glires concept -phyloge-
netic analysis of nuclear sequences. Molecular Phylogenetics and
Evolution 2003, 28:320-327.
52. Huchon D, Madsen 0, Sibbald M, Ament K, Stanhope MJ, Catzeflis F,
de Jong WW, Douzery EJP: Rodent phylogeny and a timescale
for the evolution of Glires: Evidence from an extensive taxon
sampling using three nuclear genes. Mol Biol Evol 2002,
53. Douady CJ, Chatelier PI, Madsen 0, de Jong WW, Catzeflis F,
Springer MS, Stanhope MJ: Molecular phylogenetic evidence
confirming the Eulipotyphla concept and in support of
hedgehogs as the sister group to shrews. Mol Phylogenet Evol
2002, 25:200-209.
54. Nishihara H, Hasegawa M, Okada N: Pegasoferae, an unexpected
mammalian clade revealed by tracking ancient retroposon
insertions. Proceedings of the National Academy of Sciences of the
United States of America 2006, 103:9929-9934.
55. Robinson DR, Foulds LR: Comparison of phylogenetic trees.
Mathematical Biosciences 1981, 53:13 1-147.
56. Steel MA, Penny D: Distributions of tree comparison metrics-
some new results. Systematic Biology 1993, 42:126-141.
57. Nixon KC: The parsimony ratchet, a new method for rapid
parsimony analysis. Cladistics 1999, 15:407-414.
58. Science Magazine supplementary data [ 1/5509/1786/DC I]
59. Malia MJ, Lipscomb DL, Allard MW: The misleading effects of
composite taxa in supermatrices. Molecular Phylogenetics and
Evolution 2003, 27:522-527.
60. Page RDM: TREEVIEW: An application to display phyloge-
netic trees on personal computers. Computer Applications in the
Biosciences 1996, 12:357-358.
61. Maddison DR, Swofford DL, Maddison WP: NEXUS: An extensi-
ble file format for systematic information. Systematic Biology
1997, 46:590-621.
62. Wain HM, Lush M, Ducluzeau F, Povey S: Genew: the human gene
nomenclature database. Nucleic Acids Research 2002, 30:169-171.
63. Rebhan M, Chalifa-Caspi V, PriluskyJ, Lancet D: GeneCards: ency-
clopedia for genes, proteins and diseases. .
64. Swofford DL: PAUP*: Phylogenetic Analyses Using Parsimony
(*and other methods). 4.0b 10 edition. Sunderland, Massachusetts,
Sinauer; 2002.
65. Bininda-Emonds ORP, Bryant HN: Properties of matrix repre-
sentation with parsimony analyses. Systematic Biology 1998,

66. Bininda-Emonds ORP, Sanderson MJ: An assessment of the accu-
racy of MRP supertree construction. Systematic Biology 2001,
67. Purvis A: A modification to Baum and Ragan's method for
combining phylogenetic trees. Systematic Biology 1995,
68. Pisani D, Wilkinson M: Matrix representation with parsimony,
taxonomic congruence, and total evidence. Systematic Biology
2002, 51:151-155.
69. Wilkinson M, Pisani D, Cotton JA, Corfe I: Measuring support and
finding unsupported relationships in supertrees. Systematic
Biology 2005, 54:823-83 I.
70. Cotton JA, Slater CSC, Wilkinson M: Discriminating supported
and unsupported relationships in supertrees using triplets.
Systematic Biology 2006, 55:345-350.
71. BurleighJG, Driskell AC, Sanderson MJ: Supertree bootstrapping
methods for assessing phylogenetic variation among genes
in genome-scale data sets. Systematic Biology 2006, 55:426-440.
72. Moore BR, Smith SA, Donoghue MJ: Increasing data transparency
and estimating phylogenetic uncertainty in supertrees:
approaches using nonparametric bootstrapping. Systematic
Biology 2006, 55:662-676.
73. Eriksson T: AutoDecay. 5.0th edition. Bergius Foundation, Royal
Swedish Academy of Sciences, Stockholm, Program distributed by the
author; 2001.
74. Page RDM: COMPONENT. 2.0th edition. London, Natural History
Museum; 1993.
75. Homepage of Olaf Bininda-Emonds [

Page 14 of 14
(page number not for citation purposes)

Publish with BioMed Central and every
scientist can read your work free of charge
"BioMed Central will be the most significant development for
disseminating the results of biomedical research in our lifetime."
Sir Paul Nurse, Cancer Research UK
Your research papers will be:
available free of charge to the entire biomedical community
peer reviewed and published immediately upon acceptance
cited in PubMed and archived on PubMed Central
yours you keep the copyright

Submit your manuscript here: BioMedcentral adv.asp

BMC Evolutionary Biology 2006, 6:93

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs