ArticlePDF Available

Molecular systematics of the Hyaenidae: Relationships of a relictual lineage resolved by a molecular supermatrix


Abstract and Figures

The four extant species of hyenas (Hyaenidae; Carnivora) form a morphologically and ecologically heterogeneous group of feliform carnivorans that are remnants of a formerly diverse group of mammalian predators. They include the aardwolf (Proteles cristatus), a termite-feeding specialist, and three species with a craniodental morphology adapted to cracking the bones of prey and/or carcasses, the spotted hyena (Crocuta crocuta), brown hyena (Parahyaena brunnea), and striped hyena (Hyaena hyaena). Hyenas have been the subject of a number of systematic studies during the last two centuries, due in large part to the extensive fossil record of the group, with nearly 70 described fossil species. Morphological studies incorporating both fossil and living taxa have yielded different conclusions regarding the evolutionary relationships among living hyenas. We used a molecular supermatrix comprised of seven nuclear gene segments and the complete mitochondrial cytochrome b gene to evaluate phylogenetic relationships among the four extant hyaenid species. We also obtained sequence data from representative species of all the main families of the Feliformia (Felidae, Herpestidae, and Viverridae) to estimate the sister group of the Hyaenidae. Maximum parsimony and maximum likelihood analyses of the supermatrix recovered identical topologies. Furthermore, Bayesian phylogenetic analyses of the supermatrix, with among-site rate variation among data partitions parameterized in three different ways, also yielded the same topology. For each phylogeny reconstruction method, all but two nodes received 100% bootstrap or 1.00 posterior probability nodal support. Within the monophyletic Hyaenidae, Parahyaena and Hyaena were joined together, with Crocuta as the sister to this clade, and Proteles forming the most basal lineage. A clade containing two species of mongoose (core Herpestidae) plus Cryptoprocta ferox (currently classified in Viverridae) was resolved as the sister group of Hyaenidae. The pattern of relationships among the three bone-cracking hyaenids (Crocuta, Hyaena, and Parahyaena) is incongruent with recent cladistic assessments based on morphology and suggests the need to reevaluate some of the morphological characters that have been traditionally used to evaluate relationships among hyenas. Divergence time estimates based on a Bayesian relaxed molecular clock indicates that hyaenids diverged from their feliform sister group 29.2 MYA, in the Middle Oligocene. Molecular clock estimates also suggest that the origin of the aardwolf is much more recent (10.6 MYA) than that implied by a cladistic analysis of morphology ( approximately 20 MYA) and suggests that the aardwolf is possibly derived from a bone and meat eating lineage of hyaenids that were present in the Late Miocene. [Hyaenidae; phylogeny; cytochrome b; nuclear gene segments; Proteles; Crocuta; Hyaena; Parahyaena.].
Content may be subject to copyright.
Molecular Phylogenetics and Evolution 38 (2006) 603–620
1055-7903/$ - see front matter 2005 Elsevier Inc. All rights reserved.
Molecular systematics of the Hyaenidae: Relationships of a relictual
lineage resolved by a molecular supermatrix
Klaus-Peter KoepXia,¤, Susan M. Jenks b, Eduardo Eizirik c, Tannaz Zahirpour a,
Blaire Van Valkenburgh a, Robert K. Wayne a
a Department of Ecology and Evolutionary Biology, University of California, Los Angeles, CA 90095-1606, USA
b Departments of Biology and Psychology, The Sage Colleges, Troy, NY 12180, USA
c Faculdade de Biociências, PUCRS, Av. Ipiranga, 6681, Prédio 12, Porto Alegre, RS 90619-900, Brazil
Received 12 June 2005; accepted 31 October 2005
The four extant species of hyenas (Hyaenidae; Carnivora) form a morphologically and ecologically heterogeneous group of feliform
carnivorans that are remnants of a formerly diverse group of mammalian predators. They include the aardwolf (Proteles cristatus), a ter-
mite-feeding specialist, and three species with a craniodental morphology adapted to cracking the bones of prey and/or carcasses, the
spotted hyena (Crocuta crocuta), brown hyena (Parahyaena brunnea), and striped hyena (Hyaena hyaena). Hyenas have been the subject
of a number of systematic studies during the last two centuries, due in large part to the extensive fossil record of the group, with nearly 70
described fossil species. Morphological studies incorporating both fossil and living taxa have yielded diVerent conclusions regarding the
evolutionary relationships among living hyenas. We used a molecular supermatrix comprised of seven nuclear gene segments and the
complete mitochondrial cytochrome b gene to evaluate phylogenetic relationships among the four extant hyaenid species. We also
obtained sequence data from representative species of all the main families of the Feliformia (Felidae, Herpestidae, and Viverridae) to
estimate the sister group of the Hyaenidae. Maximum parsimony and maximum likelihood analyses of the supermatrix recovered identi-
cal topologies. Furthermore, Bayesian phylogenetic analyses of the supermatrix, with among-site rate variation among data partitions
parameterized in three diVerent ways, also yielded the same topology. For each phylogeny reconstruction method, all but two nodes
received 100% bootstrap or 1.00 posterior probability nodal support. Within the monophyletic Hyaenidae, Parahyaena and Hyaena were
joined together, with Crocuta as the sister to this clade, and Proteles forming the most basal lineage. A clade containing two species of
mongoose (core Herpestidae) plus Cryptoprocta ferox (currently classiWed in Viverridae) was resolved as the sister group of Hyaenidae.
The pattern of relationships among the three bone-cracking hyaenids (Crocuta, Hyaena, and Parahyaena) is incongruent with recent cla-
distic assessments based on morphology and suggests the need to reevaluate some of the morphological characters that have been tradi-
tionally used to evaluate relationships among hyenas. Divergence time estimates based on a Bayesian relaxed molecular clock indicates
that hyaenids diverged from their feliform sister group 29.2 MYA, in the Middle Oligocene. Molecular clock estimates also suggest that
the origin of the aardwolf is much more recent (10.6 MYA) than that implied by a cladistic analysis of morphology ( »20 MYA) and sug-
gests that the aardwolf is possibly derived from a bone and meat eating lineage of hyaenids that were present in the Late Miocene. [Hyae-
nidae; phylogeny; cytochrome b; nuclear gene segments; Proteles; Crocuta; Hyaena; Parahyaena.]
2005 Elsevier Inc. All rights reserved.
Keywords: Hyaenidae; Phylogeny; Cytochrome b; Nuclear gene segments; Proteles; Crocuta; Hyaena; Parahyaena
“There are few animals, whose history has passed under
the consideration of naturalists, that have given occasion
to so much confusion and equivocation as the Hyena
has done. It began very early among the ancients, and
*Corresponding author. Fax: +1 310 206 3987.
E-mail address: (K.-P. KoepXi).
604 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
the moderns have fully contributed their share.” (Bruce,
1790 cited in Kruuk, 1972)
1. Introduction
The Hyaenidae is one of the smallest families of the Car-
nivora, with a mere four extant species placed in four genera:
Crocuta (spotted hyena), Hyaena (striped hyena), Parahya-
ena (brown hyena) and Proteles (aardwolf); (Wozencraft,
1993). This meager extant diversity contrasts sharply with the
nearly 70 fossil species that have been described for the fam-
ily (Werdelin and Solounias, 1991). Biochemical and paleon-
tological data suggest that hyaenids Wrst arose from stem
feliforms in the Late Oligocene, about 25 million years ago
(MYA); (Werdelin and Solounias, 1991). Hyaenids reached
their peak diversity in the Late Miocene, with 20–30 species
present during this time, after which they precipitously
declined, eventually leaving four surviving species (Werdelin
and Solounias, 1991). The four extant species can therefore
be considered taxonomic relicts, the remnants of a once
diverse group (Brown and Lomolino, 1998; Simpson, 1953).
Extant hyenas form a morphologically and ecologically
heterogeneous group of carnivorans. Proteles cristatus, the
aardwolf, is a diminutive hyaenid (8–14kg) with greatly
reduced premolars and molars and very broad hard palate
and large tongue (Koehler and Richardson, 1990). These
features are associated with the aardwolf’s termite diet,
which it collects by licking from the soil surface, making the
aardwolf one of only two myrmecophagous species in the
extant Carnivora (the other being the sloth bear, Melursus
ursinus). Aardwolves are socially monogamous, but primar-
ily solitary foragers except when accompanied by cubs
(Richardson and Coetzee, 1988). At the other extreme is the
spotted hyena, Crocuta crocuta, a large (40–86 kg) cursorial
hypercarnivore that uses robust premolars to crush bones,
which are then ingested. Spotted hyenas live in large matri-
lineal social groups known as clans whose members hunt
cooperatively (Kruuk, 1972; Mills, 1990). Female spotted
hyenas are highly aggressive, tend to be larger than males,
and have male-like genitalia with a pseudophallus and
pseudoscrotum. The brown hyena (Parahyaena brunnea)
and the striped hyena (Hyaena hyaena) are also large in size
(28–48kg and 25–55 kg, respectively) and have dentitions
with robust premolars for crushing bones (Mills, 1982,
1990, 1999; Rieger, 1981). These two species are primarily
scavengers, although they will also hunt prey smaller or
larger than themselves (Mills, 1990; Rieger, 1981). Brown
and striped hyenas may form small clans, but both species
forage solitarily (Drea et al., 1999; Mills, 1990). Carnassials
of Crocuta, Hyaena, and Parahyaena are elongate and well
adapted for slicing meat. The similarities among these three
species (referred to hereafter as bone-cracking hyaenids) as
well as their Wrst occurrence in the fossil record suggests
that they evolved relatively recently from a common ances-
tor, whereas the divergent morphology of the aardwolf sug-
gests it evolved from a more distant ancestor early in
hyaenid evolution (Werdelin and Solounias, 1991).
The abundant fossil material available for hyaenids has
stimulated a number of paleontologists during the last two
centuries to examine evolutionary relationships among
taxa within the family, including relationships between fos-
sil taxa and the four living species. Prior to the emergence
of cladistics, taxa were grouped together according to over-
all morphological similarity. With regard to Crocuta,
Hyaena, and Parahyaena, two diVerent phylogenetic
hypotheses were proposed by paleontologists. The Wrst
explicit discussion of hyaenid phylogeny, presented in the
form of a key, was that of Gaudry (1862–1867). He consid-
ered H. hyaena and P. brunnea to be more closely related to
one another than either is to C. crocuta, a hypothesis that
was later supported by Pilgrim (1932), Ewer (1955), The-
nius (1966), and Hendey (1974). However, it should be
noted that Hendey (1974) regarded H. hyaena and P. brun-
nea as only distantly related to one another, having
diverged in the Miocene, and that P. brunnea shared a more
recent common ancestry with species of the extinct genus
Pachycrocuta. As a consequence of this, he named a new
subgenus, Parahyaena, for P. brunnea. Prior to Hendey
(1974), P. brunnea had been classiWed as Hyaena brunnea, as
most authors considered the brown and striped hyenas to
be closely related based on morphological similarity. Para-
hyaena was eventually elevated to full generic rank by Wer-
delin and Solounias (1991). The second hypothesis,
proposed by Schlosser (1890), suggested that P. brunnea
and C. crocuta were more closely related than either is to
H. hyaena. This second hypothesis was supported in the
Wrst cladistic analysis of hyaenid phylogeny (Galiano and
Frailey, 1977). Due to its highly reduced dentition and
other unique features associated with its specialized diet of
termites, the aardwolf was seen as a very distant relative of
the other hyenas and was often placed in a subfamily or
family of its own, the Protelinae or Protelidae.
More recently, Werdelin and Solounias (1991) have
addressed the issue of hyaenid interrelationships from a
paleontological and morphological perspective. One of the
most signiWcant results of this work is that, despite a thor-
ough survey of the skull, dentition and selected areas of the
postcranial skeleton, very few phylogenetically informative
characters were found. This is largely due to the majority of
characters with more than one character state being
uniquely derived in C. crocuta. Werdelin and Solounias
(1991) conducted two diVerent analyses in examining the
relationships among living taxa. In the Wrst, Proteles crista-
tus was excluded from the phylogenetic analysis (because
most dental characters are absent in this species due to its
reduced dentition) and an attempt was made to polarize
characters on an a priori basis using outgroup and ontoge-
netic information. The most parsimonious tree united
P. brunnea and C. crocuta as sister taxa. When P. cristatus
was included in the second analysis, trees with either P.
brunnea or H. hyaena as sister taxon to C. crocuta were
equally parsimonious. Werdelin and Solounias (1991)
found that a tree one step longer than the two equally
parsimonious trees was the traditional one that grouped
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 605
P. brunnea and H. hyaena as sister taxa. Based on these
analyses, the authors concluded that “ƒthe Recent Hyae-
nidae oVer no Wrm data in support of a resolved scheme of
interrelationships” (Werdelin and Solounias, 1991, p. 62).
Despite diYculties with inferring interrelationships among
living taxa by themselves, Werdelin and Solounias (1991)
conducted additional phylogenetic analyses that included the
four living taxa plus 57 “reasonably well known” fossil taxa.
The phylogenetic hypothesis they generated places Proteles
near the base of the hyaenid tree, allying this genus with the
extinct genus Plioviverrops, a small, mongoose-like insecti-
vore/omnivore whose fossil record extends from the Early to
Late Miocene, 16–5 MYA (Werdelin and Solounias, 1991,
1996). This hypothesis implies that Proteles has an early ori-
gin (»20 MYA) and that it is distantly related to the other
living taxa. However, the fossil record of Proteles only
extends to the Early Pleistocene (1.5 MYA) and suggests a
more recent origin for this genus. Resolving the discrepancy
between the cladistic age and the fossil age of Proteles is
important for understanding the ecological context that led
to the evolution of morphological specializations associated
with the myrmecophagous diet of the aardwolf. As for Para-
hyaena, Hyaena, and Crocuta, these taxa are part of the most
derived clade within the Hyaenidae (and hence, among the
most recently evolved), representing taxa with advanced
bone-cracking adaptations (such as greatly enlarged premo-
lars). Interestingly, in this phylogenetic scheme, Hyaena is
placed as more closely related to Crocuta than it is to Parahy-
aena. This working phylogenetic hypothesis has been used in
many subsequent studies of hyaenid morphological and eco-
logical evolution (e.g., Werdelin, 1996a,b; Werdelin and Sol-
ounias, 1996) and therefore forms the modern view of
hyaenid interrelationships based on morphology.
The phylogeny of extant hyaenids has been examined
using karyotypic and mitochondrial DNA data. Wurster and
Benirschke (1968) found that Proteles, Hyaena and Crocuta
were nearly identical in karyotype, suggesting a close rela-
tionship among these three genera. Jenks and Werdelin
(1998) used complete sequences of the mitochondrial cyto-
chrome b gene from the four living hyaenids plus two distant
outgroups (cat, Felis catus, and harbor seal, Phoca vitulina)
and found that Hyaena and Parahyaena were sister taxa, but
also that Crocuta and Proteles were united as sisters with
100% bootstrap support. This latter relationship had never
before been suggested in prior studies and the authors sus-
pected that long branch attraction (Felsenstein, 1978) had
drawn Proteles and Crocuta together. Consequently, they
conducted phylogenetic analyses using transversion parsi-
mony or third position transversions only. The resulting phy-
logenetic trees placed Proteles as the most basal hyaenid,
with Crocuta forming the sister group to Hyaena plus Para-
hyaena (Jenks and Werdelin, 1998). Taken altogether, previ-
ous morphological and molecular phylogenetic analyses have
supported every possible combination of relationships
between extant bone-cracking hyaenids.
Higher-order phylogenetic relationships between hyaenids
and other feliform families have also proven controversial.
Based on analyses of morphologic data, two phylogenetic
hypotheses have been supported. First, a sister relationship
of hyenas and cats (Felidae) has been suggested (Wozencraft,
1993; Wyss and Flynn, 1993). This hypothesis is also sup-
ported in a supertree analysis of the Carnivora (Bininda-
Emonds et al., 1999). Second, a sister relationship between
hyenas and mongooses (Herpestidae) has also been proposed
(Hunt, 1987, 1989; Veron, 1995). DNA–DNA hybridization
data have provided weak support for a relationship between
hyenas and civets (Viverridae); (Wayne et al., 1989). Most
recently, studies using mitochondrial and/or nuclear gene
sequences, but using only Crocuta as the sole representative
of Hyaenidae, have demonstrated a sister-group relationship
between herpestids and hyaenids (Flynn and Nedbal, 1998;
Flynn and Nedbal, 1998; Gaubert and Veron, 2003; Yoder
et al., 2003; Yu et al., 2004). Therefore, in addition to contro-
versial aspects regarding the relationships among hyaenid
species, the speciWc position of the family within the suborder
Feliformia remains uncertain.
Here, we use a molecular supermatrix comprised of
seven nuclear gene segments and the mitochondrial cyto-
chrome b gene, from all four living hyaenids and eight taxa
representing species from the other three feliform families,
to reconstruct phylogenetic relationships among these taxa.
The aims of our study are to: (1) estimate the relationships
among extant hyaenids; (2) assess the divergence time of
Proteles from the other three hyaenids to distinguish
whether the aardwolf evolved early or late in hyaenid evo-
lution; and (3) determine the sister group (Felidae, Herpes-
tidae, or Viverridae) of the Hyaenidae. We also conducted
phylogenetic analyses in which we combined our molecular
supermatrix with a modiWed version of the morphological
character matrix from Werdelin and Solounias (1991) to
evaluate the degree of congruence and incongruence
between molecular and morphological data sets.
2. Materials and methods
2.1. Taxon sampling
Tissue samples were obtained from two individuals of
each hyena species except H. hyaena (Table 1). Samples
from two or three representative species each of the Feli-
dae, Herpestidae, and Viverridae were included to deter-
mine the sister group of the Hyaenidae (Table 1). Our taxon
sampling includes Cryptoprocta ferox and Nandinia bino-
tata, which most classiWcations of the Carnivora assign to
the Viverridae (e.g., Wozencraft, 1993). Recent molecular
studies, however, indicate that Cryptoprocta is part of an
endemic radiation on Madagascar that is sister to the core
Herpestidae (Yoder et al., 2003) and that Nandinia is the
most basal extant feliform taxon (Flynn and Nedbal, 1998).
These and other studies (e.g., Gaubert and Veron, 2003)
suggest that the taxonomy of the Viverridae is in need of
substantial revision. Two species from the Canidae (Cani-
formia) were used as distant outgroups to root feliform
taxa (Table 1).
606 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
2.2. Laboratory methods
We extracted total genomic DNA from tissue samples
using phenol/chloroform methods followed by precipitation
with ethanol or with a QIAamp DNA Mini Kit (Qiagen,
Valencia, CA) according to the manufacturer’s protocol. We
ampliWed exon or exon/intron segments from the following
seven nuclear genes using the primers shown in Table 2:
WT1. For several of these loci, we designed internal primers
for use in both ampliWcation and sequencing. We also
obtained sequences of the complete mitochondrial cyto-
chrome b (cyt b) gene using a combination of published prim-
ers (Table 2). All gene segments were ampliWed by the
polymerase chain reaction (PCR) in an MWG-Biotech Pri-
mus 96 Plus thermal cycler with the following conditions: 28–
30 cycles of 94 °C for 30s, 52–54 °C for 30s, 72 °C for 45s, and
one cycle of 72 °C for 5min. Each 50 l reaction contained
35.7l sterile double-distilled water, 5l 10£ PCR buVer, 5l
of 25mM MgCl2, 1l of 10mM dNTP mix, 1l of both
25pM/l forward and reverse primers, 0.3l Taq polymerase
(Sigma–Aldrich, St. Louis, MO), and 1l of 0.1–1 g genomic
DNA. All PCRs included a negative control (no DNA).
AmpliWcation products were size-fractionated through 1%
agarose/Tris–acetic acid–EDTA gels and the appropriate
band was excised and then puriWed with an Ultra Clean Kit
(MoBio Laboratories, Solana Beach, CA). Products were
sequenced using the ampliWcation primers and the CEQ Dye
Terminator Cycle Sequencing Quick Start Kit (Beckman–
Coulter, Fullerton, CA). Sequencing reactions were precipi-
tated according to the manufacturer’s protocol and run on a
CEQ2000XL automated capillary sequencer (Beckman–
Coulter, Fullerton, CA). Chromatographs were checked for
accuracy and edited using Sequencher 3.1 (Gene Codes, Ann
Arbor, MI). Despite repeated eVorts at modifying thermal
cycling and/or sequencing conditions, we were unable to
obtain GNAT1 sequences from Nandina binotata and Para-
doxurus hermaphroditus or RAG1 sequences from Canis lupus
and Vulpes vulpes. These taxa were coded with question
marks (missing data) for the two loci in the combined phylo-
genetic analyses. APOB sequences from Crocuta crocuta, Pan-
thera leo, Nandinia binotata, and Vulpes vuples were from the
study by Amrine-Madsen et al. (2003).
2.3. Sequence alignment and treatment of gaps
Sequences for all loci were aligned by eye. Several
nuclear gene segments contained insertions and/or dele-
tions (indels) that necessitated introduction of gaps into
sequences. Alignment of sequences where indels occurred
was largely unambiguous. However, both GHR and WT1
had regions that contained microsatellite sequences that
varied in length among taxa making unambiguous align-
ment of these regions diYcult. Further, WT1 contained a
small region that was diYcult to align unambiguously.
Therefore, we excluded nine base pairs (bp) from the GHR
alignment and 107bp from the WT1 alignment (a total of
116bp) for all phylogenetic analyses. For maximum parsi-
mony analyses, potential phylogenetic signal contained in
indels was recovered by coding gaps according to the
method of Barriel (1994). This method treats indels as sin-
gle events regardless of length or complexity and uses ques-
tion marks associated with subsequent substitutions that
are contained within the indels, thereby minimizing tree
Species, common names and source information for taxa used in this study
Numbers in parentheses indicate number of individuals sequenced if >1. ClassiWcation according to Wozencraft, 1993.
Taxon Common name Source
Crocuta crocuta (2) Spotted hyena Field Station for Behavior Research Hyena Project, University of California, Berkeley
Hyaena hyaena Striped hyena Center for Reproduction on Endangered Species, Zoological Society of San Diego
Parahyaena brunnea (2) Brown hyena Center for Reproduction on Endangered Species, Zoological Society of San Diego
Proteles cristatus (2) Aardwolf Center for Reproduction on Endangered Species, Zoological Society of San Diego
Lynx canadensis Canadian lynx Kenai National Wildlife Refuge, Alaska
Panthera leo Lion Johannesburg Zoo, South Africa
Cryptoprocta ferox Fossa Center for Reproduction on Endangered Species, Zoological Society of San Diego
Genetta tigrina Large-spotted genet Kenya; collected by R.K. Wayne, University of California, Los Angeles
Nandinia binotata African palm civet Cameroon; collected by D.B. Pires, University of California, Los Angeles
Paradoxurus hermaphroditus Common palm civet Museum of Vertebrate Zoology 186574, University of California, Berkeley
Herpestes javanicus Javan mongoose Museum of Vertebrate Zoology 186570, University of California, Berkeley
Mungos mungo Banded mongoose Kenya; collected by R.K. Wayne, University of California, Los Angeles
Canis lupus Wolf Inuvik, Northwest Territories, Canada; collected by Peter Clarkson
Vulpes vulpes Red fox United Kingdom; collected by Eli GeVen, University of Tel Aviv
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 607
length. Gaps were treated as missing characters in maxi-
mum likelihood and Bayesian analyses.
2.4. Phylogeny estimation
2.4.1. Maximum parsimony
We reconstructed most parsimonious trees for the eight
individual loci, combined nuclear loci, and a combined
supermatrix (nuclear and mitochondrial sequences) using
PAUP* (SwoVord, 2002). All characters were equally
weighted in all analyses. For individual nuclear gene segments
and combined analyses of concatenated nuclear data and
supermatrix, we performed branch and bound searches, with
the initial upper bound computed via stepwise addition, the
‘furthest’ taxon sequence added to the search trees, and the
‘Multrees’ option (saving of all optimal trees) in eVect.
Because branch and bound searches are ineYcient for data
sets containing high amounts of homoplasy, a heuristic search
was employed for the cyt b data set, using 100 replicates of
stepwise addition and tree bisection–reconnection (TBR)
branch swapping. We evaluated branch support with 1000
pseudoreplicates of bootstrapping and the same branch and
bound (nuclear gene segments and combined analyses) or
heuristic (cyt b) search conditions used to reconstruct the
most parsimonious tree(s). We used the maximum parsimony
bootstrap analyses to detect if there was any signiWcant incon-
gruence among topologies derived from the eight individual
data sets, using the value of 770% as our criterion for signiW-
cant bootstrap support (e.g., Flynn and Nedbal, 1998).
2.4.2. Model selection
Modeltest 3.06 (Posada and Crandall, 1998) was used to
estimate the model and parameters of DNA evolution that
best Wt each of the following data partitions: (a) individual
loci; (b) combined nuclear DNA data; and (c) combined
supermatrix. Best-Wt models were chosen among a series of
nested models according to the Akaike information crite-
rion (AIC; Akaike, 1974) as implemented in Modeltest.
2.4.3. Maximum likelihood
Maximum likelihood (ML) analyses of the combined
supermatrix were performed using PAUP*, with the best-
Wtting model and associated parameters determined via
Modeltest speciWed in the tree search. We employed heuris-
tic searches with 100 replicates of random stepwise addition
and tree bisection–reconnection branch swapping. Three
hundred pseudoreplicates of bootstrapping with one repli-
cate of random stepwise addition and the ‘Multrees’ option
turned oV (DeBry and Olmstead, 2000) were employed to
measure support for the ML tree.
2.4.4. Bayesian analyses
We used MrBayes v2.01 (Huelsenbeck, 2000) and MrBa-
yes 3 (Ronquist and Huelsenbeck, 2003) to perform several
Gene abbreviation, type of sequence, and forward (F) and reverse (R) primer sequences for nuclear and mitochondrial gene segments ampliWed and
sequenced in this study
Gene names are APOB, apolipoprotein B; CHRNA1, cholinergic receptor, nicotinic, polypeptide 1 precursor; COL10A1, collagen type X I; GHR,
growth hormone receptor; GNAT1, guanine nucleotide binding protein, transducing polypeptide 1; RAG1, recombination activating protein 1; WT1,
Wilms tumor 1; Cyt b, cytochrome b.
Gene Type of sequence Primer sequences (5!3) Reference
APOB Exon F: GTG CCA GGT TCA ATC AGT ATA AGT Amrine-Madsen et al., 2003
R: CCA GCA AAA TTT TCT TTT ACT TCA A Jiang et al., 1998
CHRNA1 Exon/intron F: GAC CAT GAA GTC AGA CCA GGA G Lyons et al., 1997
COL10A1 Exon F: ATT CTC TCC AAA GCT TAC CC Venta et al., 1996
F: GAT AAG ATT CTG TAT AAC AGG C internal primer; this study
R: TAG GAA TCC TGA GAA GGA GG internal primer; this study
GHR Exon/intron F: CCA GTT CCA GTT CCA AAG AT Venta et al., 1996
F: CTG TCC TAT GTT GAG AGC ATT TGC internal primer; this study
R: GAA ACA TTT TCC TCC AGA AGG G internal primer; this study
GNAT1 Exon/intron F: AGC ACC ATC GTC AAG CAG A Brouillette et al., 2000
RAG1 Exon F: GCT TTG ATG GAC ATG GAA GAA GAC AT Teeling et al., 2000
WT1 Exon/intron F: GAG AAA CCA TAC CAG TGT GA Venta et al., 1996
F: GGA AGC ATC CCA CAT TTC TCT TGC internal primer; this study
R: GAA TCA CAG GCT ACA AAC TGG GAC internal primer; this study
Cyt bMitochondrial coding F: CGA AGC TTG ATA TGA AAA ACC ATC GTT G L14724; Irwin et al., 1991
R: AAA CTG CAG CCC CTC AGA ATG ATA TTT GTC CTC A H15149; Kocher et al., 1989
F: GCA AGC TTC TAC CAT GAG GAC AAA TAT C L15162; Irwin et al., 1991
F: ATA GAC AAA ATC CCA TTC CA L15408; Irwin et al., 1991
R: TAG TTG TCA GGG TCT CCT AG H15494; KoepXi and Wayne, 1998
R: AAC TGC AGT CAT CTC CGG TTT ACA AGA C H15915; Irwin et al., 1991
608 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
diVerent Bayesian analyses in order to examine sensitivity
of the posterior probability and parameter estimates when
the supermatrix was analyzed under alternative models of
varying complexity. We used the best-Wtting model of evo-
lution selected with Modeltest for the likelihood function
component of Bayes’ rule, as applied to phylogenetic analy-
sis (Huelsenbeck et al., 2001).
Model parameters, along with tree topologies and
branch posterior probabilities, were estimated as part of the
analyses. The empirical nucleotide frequencies of the
supermatrix were speciWed as priors for all analyses.
We parameterized models in several diVerent ways. Three
diVerent Markov Chain Monte Carlo (MCMC) analyses
were conducted in which the eight data sets were assumed to
all evolve under the same model of evolution (i.e., GTR) but
diVered in the way the among-site rate variation, as approxi-
mated with the gamma distribution shape parameter (),
was partitioned among and within the eight data sets. Model
1 included a parameter for the proportion of invariable sites
(I) and applied a single gamma shape parameter for the
supermatrix (no partitioning of among-site rate variation).
Model 2 partitioned the gamma shape parameter among the
eight data sets that comprise the supermatrix by deWning the
eight partitions and using the “rates DsitespeciWcgamma” in
the MrBayes v2.01 command block. Model 3 partitioned the
gamma shape parameter among the seven nuclear gene data
sets and among the three codon positions of cyt b using the
same command for “rates” as in Model 2. Therefore, models
1–3 employed 2, 9, and 11 parameters, respectively, for the
among-site rate variation, along with the other parameters
of the best-Wt model. A single GTR substitution rate matrix
was applied across the supermatrix for all three models.
MCMC analyses used one cold and three heated chains that
ran for 1,000,000 generations, from which trees were sam-
pled every 100 generations. We conducted three independent
MCMC runs for each of the three models to conWrm that
the stationarity of likelihood values and parameter estimates
were consistent among runs. Burn-in plots were examined to
evaluate the number of generations elapsed before likeli-
hood scores had achieved stationarity. For each run, con-
sensus trees with node posterior probabilities were
generated based on 9000 or 7000 trees, depending on the
model, after the Wrst 1000 or 3000 trees (or 100,000 or
300,000 generations) were discarded as burn-in. Final con-
sensus trees derived from the three independent runs for
each model were based on a total of 27,003 (3£9001) or
21,003 (3 £7001) trees, depending on the model used (see
Section 3).
2.4.5. DNA and morphology
We also evaluated phylogenetic relationships among
extant hyenas by combining our molecular supermatrix
with morphological characters from the study by Werdelin
and Solounias (1991) in a total evidence analysis frame-
work using parsimony (Kluge, 1998). In their original anal-
yses, Werdelin and Solounias’ (1991) data matrix contained
26 coded morphological characters for the four hyena spe-
cies plus a ‘hypothetical ancestor’ that was used as the out-
group (Wve taxa total). Werdelin and Solounias conducted
phylogenetic analyses with and without P. cristatus because
some dental characters were absent from this species (and
therefore could not be coded) due to its highly reduced den-
tition relative to the three bone-cracking hyaenids. Their
data matrix without Proteles contained three characters
that were excluded from the matrix that included Proteles
(see Table 1 and Table 2 in Werdelin and Solounias, 1991).
We added these three characters (6: stage of reduction in
size of the upper Wrst molar relative to the upper fourth pre-
molar; 8: presence or absence of metaconid on the lower
Wrst molar; and 11: relative length of paracone and meta-
style of the upper fourth premolar) to the original 26 char-
acter matrix and simply coded them as missing for P.
cristatus. Furthermore, we discovered that two taxa had
been incorrectly coded for two characters, and we changed
their codings to make them consistent with original charac-
ter deWnitions given by Werdelin and Solounias (1991).
Character 39, size of mastoid crest, was originally coded as
0Dshort for P. brunnea by Werdelin and Solounias (1991).
However, these authors state:
“In H. hyaena, as well as the outgroups, the mastoid
crest ends at the postero-dorsal end of the external audi-
tory meatus. In P. brunnea and C. crocuta, on the other
hand, the mastoid crest continues beyond this point well
towards the ventral end of the external auditory meatus,
a condition that is probably derived” (Werdelin and Sol-
ounias, 1991, p. 56).
Therefore, we coded P. brunnea as 1Dlong for character
39, a condition shared with C. crocuta. We also changed the
coding for character 45, size of metacarpal I in Proteles cri-
status. This species was coded as 1 Dreduced in the original
analysis, but Werdelin and Solounias (1991) comment:
“This bone is vestigial in H. hyaena, P. brunnea and
C. crocuta. This is a derived state uniting these species
relative to Proteles, where the [metacarpal] I is much
larger” (p. 57).
Therefore, we coded P. cristatus as 0Dlarge for charac-
ter 45, a condition shared with the outgroup. Finally, we
note that the coding deWnitions for character 24, suture
between premaxillary and frontal on snout, were reversed
in Werdelin and Solounias (1991). Based on the authors’
deWnition for this character, 0Dpresent, and 1 Dabsent,
rather than the reverse. Nonetheless, taxa were all coded
correctly for this character.
For the total evidence analysis, we used two representa-
tive taxa of the feliform family that was found to be the sis-
ter group of the Hyaenidae based on the supermatrix
analyses that included all the taxa. To make the morphol-
ogy matrix congruent with the supermatrix, we added
another hypothetical ancestor with the same codings as the
Wrst hypothetical ancestor from the Werdelin and Solou-
nias (1991) study to the morphology matrix (six taxa in
total). This strategy assumes that the outgroup taxa used
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 609
for the supermatrix and the morphology matrix are equiva-
lent, even though Werdelin and Solounias (1991) could not
conWdently determine, based on cladistic analysis of mor-
phological characters, which feliform group was the sister
group to the hyaenids. Nonetheless, the primitive state
codings for the morphological characters in the hypotheti-
cal ancestor are based on a detailed evaluation of primitive
fossil hyaenids and other feliform groups by Werdelin and
Solounias (1991). Therefore, it is not unrealistic to expect
that some of these characters would be found in the true
feliform sister group of the hyaenids. Given these consider-
ations, we conducted total evidence parsimony analysis
using the exhaustive search option in PAUP*. Branch sup-
port for the shortest tree was estimated using 1000 pseu-
doreplicates of bootstrapping with branch and bound
search and the same search conditions as described for the
molecular supermatrix parsimony analyses. In addition, we
used TreeRot 2.0 (Sorenson, 1998) to quantify the Bremer
support for each node (Bremer, 1988) and to conduct parti-
tioned Bremer support analyses (PBS) in order to evaluate
the relative contribution of the nine data sets (eight gene
partitions and one morphology partition) to the total
Bremer support at each node.
2.5. Divergence dates
Divergence dates among the sampled taxa were esti-
mated using the Bayesian relaxed clock method developed
by Thorne et al. (1998) and Kishino et al. (2001). This
approach allows for heterogeneity of substitution rates
among branches of the tree (i.e., departures from a strict
molecular clock), while incorporating multiple fossil cali-
brations to constrain node ages. We used the ML topology
obtained with the molecular supermatrix and the program
estbranches to estimate branch lengths and a rate variance–
covariance matrix. The program divtime5b was then used to
estimate divergence dates among all sampled taxa, with
1,000,000 generations run after 100,000 generations of
burn-in, and nine fossil constraints obtained from McK-
enna and Bell (1997) and Turner (1987): (1) minimum of 4
MYA (i.e., Hyaena) and (2) maximum of 16.4 MYA for the
base of crown Hyaenidae (i.e., Proteles vs others); (3) mini-
mum of 16.4 MYA for Cryptoprocta vs. core Herpestidae;
(4) minimum of 3.58 MYA for Herpestes vs. Mungos; (5)
minimum of 16.4 MYA for Genetta vs. Paradoxurus; (6)
minimum of 5.3 MYA and (7) maximum of 23 MYA for
the base of Felidae; (8) minimum of 28.5 MYA for Felidae
vs. others; and (9) maximum of 55 MYA for the base of
Feliformia (Nandinia vs. others). The mean of the prior dis-
tribution for the ingroup root age was set at 50 MYA (con-
sistent with McKenna and Bell, 1997; Springer et al., 2003),
and its impact on the posterior distribution of node ages
was tested empirically. Eleven independent runs were per-
formed to evaluate convergence of point estimates and
credibility intervals for divergence dates, to test for consis-
tency among diVerent fossil constraints, and to assess the
eVect of modifying the prior for root-to-tip age.
3. Results
3.1. Sequence characteristics and incongruence among the
data partitions
The total length of the molecular supermatrix alignment,
including gaps, was 6218 characters. Five nuclear gene seg-
ments (APOB, CHRNA1, GHR, GNAT1, and WT1) con-
tained a total of 56 indels of various lengths. Forty (40)
indels were parsimony-informative for a particular feliform
family or for diVerentiating feliforms from the two canid
outgroup species. Interestingly, we discovered two large
insertions that occurred in diVerent locations in the GHR
gene segment. One insertion was 229 or 230bp in length
and found in the two herpestids, Herpestes javanicus and
Mungos mungo, respectively. The other insertion was 194 bp
long and occurred 3 to the Wrst insertion in the two canid
outgroup species. To utilize the potential phylogenetic sig-
nal contained by the indels, we coded them according to the
method proposed by Barriel (1994); (see Section 2). When
indels were coded in this way, the length of the supermatrix
alignment was 5730 characters. Sequences generated for
this study were deposited in GenBank under Accession
Nos. AY928668–AY928771 (Appendix A).
Among nuclear gene segments that we sequenced, WT1
contained the greatest number of variable and parsimony-
informative sites (after exclusion of ambiguously aligned
sequences, see Section 2), with APOB and GHR also con-
taining a large number of parsimony-informative sites
(Table 3). The cyt b gene, however, contained two to three
times more variable and parsimony-informative sites than
any of these three nuclear gene segments. Additional
sequence characteristics, tree statistics from the parsimony
analyses, and the best-Wtting models and their associated
parameters estimated using Modeltest for each of the data
partitions are presented in Table 3.
We detected one instance of signiWcant incongruence for
the APOB data partition in the separate parsimony analyses
of the eight gene segments, using our criterion of 770% boot-
strap support for strongly conXicting nodes. The most parsi-
monious tree for APOB places the felid clade as the sister
group to a clade comprised of the herpestids (including Cryp-
toprocta) and hyaenids, with a bootstrap support of 79%. This
conXicts signiWcantly with the most parsimonious trees for
CHRNA1, GNAT1, and WT1, which place the viverrid clade
as the sister group to herpestids and hyaenids, with a boot-
strap support of 77, 91, and 71%, respectively. This latter rela-
tionship is also found in the parsimony analyses of GHR,
RAG1, and cyt b, albeit with weak nodal support (<70%).
The COL10A1 gene tree was also incongruent with the other
gene trees in several respects, such as Wnding that Genetta and
Paradoxurus were not monophyletic and placed basally
among the feliforms, and the joining of Nandinia as the sister
to the herpestid clade (including Cryptoprocta). These rela-
tionships, however, received <50% bootstrap support. Despite
the signiWcant incongruence of a single node found in the
APOB gene tree, we do not think that this warrants exclusion
610 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
of the APOB data set from a combined analysis of other data
gene segments. Instead, the inXuence of such conXicting phy-
logenetic signals is best evaluated in the context of a combined
analysis of all data (Rokas et al., 2003).
3.2. Combined nuclear data vs. cytochrome b gene analyses
Maximum parsimony and likelihood analyses recovered
identical topologies when the seven nuclear gene segments
were concatenated (Figs. 1A and B). Nandinia was strongly
supported as the most basal taxon and sister to the remaining
feliform clades. Felids formed the next successive clade, which
was sister to a group that included core viverrids (Genetta and
Paradoxurus), hepestids, and hyaenids. Within this group, we
found that the two herpestids plus Cryptoprocta (currently
classiWed in the Viverridae, Wozencraft, 1993) formed the sis-
ter group to the hyaenids with 100% bootstrap support in
both MP and ML analyses. Within the hyaenids, Proteles was
the most basal taxon and Crocuta was sister to the clade com-
prised of Hyaena and Parahyaena. Nine of the eleven resolved
nodes received 100% bootstrap support in both MP and ML
analyses. The two nodes that had bootstrap values <100%
had short internal branch lengths, especially in the ML analy-
sis (Fig. 1B). The node between felids and core viverrids had
bootstrap support of 99% (MP) and 68% (ML) while the
node between Proteles and Crocuta had bootstrap support of
96% (MP) and 84% (ML); (Figs. 1A and B).
Topologies of the MP and ML trees based on the cyt b
gene data diVered slightly from each other as well as from
those based on concatenated nuclear gene segments (Figs.
1C and D). The parsimony tree was largely congruent with
the nuclear gene tree, except that Crocuta was placed as the
sister taxon to Proteles, although with low bootstrap sup-
port (52%). This same grouping was recovered in the ML
tree using the TVM+ I + G model, except that the boot-
strap support was increased to 80%. The ML tree also
diVered from the combined nuclear gene analyses and the
cyt b MP tree by placing Cryptoprocta as the sister taxon to
a clade containing the herpestids and hyaenids. However,
this relationship had low bootstrap support (67%).
We examined whether base composition heterogeneity
in the cyt b data had somehow caused Proteles and Crocuta
to be joined together in the MP and ML analyses of these
data. Even though we cannot know the true phylogeny, the
combined nuclear data provide strong evidence for the rela-
tionships among the hyenas in both MP and ML analyses.
Therefore, we assessed the possibility that systematic error,
such as nonstationary base composition, had inXuenced the
cyt b topologies. Among the eight gene segments sequenced,
only the cyt b gene data was found to be signiWcantly heter-
ogeneous in base composition for both informative sites
and all sites, using the 2 test of homogeneity of base fre-
quencies as implemented in PAUP* (SwoVord, 2002). The
COL10A1 gene segment was the only other data partition
that was signiWcantly heterogeneous in base frequency for
the informative sites. The signiWcant heterogeneity in base
composition for the cyt b data may bias phylogenetic
reconstruction methods that fail to take nonstationary base
composition into account (Lockhart et al., 1994). Conse-
quently, we conducted minimum evolution analyses
(Rzhetsky and Nei, 1992) of cyt b data with bootstrapping
(1000 pseudoreplicates) using the LogDet/paralinear dis-
tance to correct for possible nonstationary base composi-
tion (Lake, 1994; Lockhart et al., 1994). To account for rate
Sequence characteristics, descriptive statistics from parsimony analyses, and models and substitution parameters derived from Modeltest for each of the
eight gene segment partitions and the two combined data partitions
No. of nucleotides, alignment length with gaps included in the alignment/alignment length with gaps coded according to Barriel (1994). A, C, G, and
TDempirical base frequencies; I, proportion of invariable sites; G, gamma shape parameter; TS/TV, transition/transversion ratio; rAC, rAG, rAT, rCG, rCT,
and rGT Drate of substitution for speciWed nucleotides.
APOB CHRNA1 COL10A1 GHR GNAT1 RAG1 WT1 Cyt bNuclear Supermatrix
No. of nucleotides 938/937 377/344 320/320 1113/675 463/445 1076/1076 791/686 1140 5078/4590 6218/5730
No. variable 203 155 68 215 89 130 217 503 1077 1580
No. parsimony-informative 135 99 49 115 57 71 140 390 666 1056
No. of trees 1 1 2 1 1 3 1 3 1 1
Tree length 24 202 100 270 101 185 281 1479 1390 2869
Retention index 0.911 0.882 0.779 0.897 0.936 0.688 0.841 0.381 0.852 0.623
Model TVM+G K81+G TIMef+I TIM+G TrN+I K80+I+G K81uf+G TVM+I+G GTR+G GTR+I+G
A0.325 0.280 0.263 0.299 0.192 0.265 0.245 0.296 0.275 0.280
C0.217 0.220 0.262 0.186 0.322 0.241 0.221 0.300 0.231 0.245
G0.176 0.250 0.224 0.193 0.290 0.267 0.270 0.135 0.233 0.212
T0.282 0.250 0.251 0.322 0.196 0.227 0.264 0.269 0.261 0.263
I 0.6846 0.5859 0.6556 0.4905 0.4461
G 0.7845 1.7452 0.6677 — 0.6739 0.6388 0.7782 0.4442 0.6396
TS/TV ratio 3.7689
rAC 1.4019 1.0000 1.0000 1.0000 1.0000 — 1.0000 40.7453 1.3155 2.6065
rAG 5.4906 3.0690 3.9056 2.4134 7.2681 — 4.1523 671.4096 4.4161 6.0604
rAT 0.6074 0.4896 0.4384 0.4436 1.0000 — 0.5375 38.5516 0.5582 0.8652
rCG 1.3371 0.4896 0.4384 0.4436 1.0000 — 0.5375 2.7985 0.8769 0.6699
rCT 5.4906 3.0690 6.0623 3.4420 8.2561 — 4.1523 671.4096 5.1955 11.1836
rGT 1.0000 1.0000 1.0000 1.0000 1.0000 — 1.0000 1.0000 1.0000 1.0000
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 611
heterogeneity in the LogDet distance calculations
(SwoVord et al., 1996), we used likelihood evaluation of the
three most parsimonious trees from the cyt b data to
estimate the proportion of invariable sites (I) under the
GTR model. We estimated I to be 0.5449, 0.5449, and
0.5448 for the three trees and used ID0.5449 in minimum
evolution analyses. The minimum evolution bootstrap con-
sensus tree using LogDet+I transformed distances showed
strong support for most feliform family clades, but the rela-
tionships among these clades and the position of Nanidina
and Cryptoprocta received <50% bootstrap support (results
not shown). Importantly, the relationships among the four
hyenas were identical to those found in nuclear gene trees
(Figs. 1A and B), except that the node between Proteles and
Crocuta only had a bootstrap support of 54%. These results
using LogDet+I distances therefore suggest that non-
stationary base composition may have aVected the MP and
ML analyses of the cyt b data because neither of these
methods alone corrects for such systematic bias.
In general, the cyt b gene trees had fewer nodes that were
well supported by bootstrap analyses than gene trees based
on the combined nuclear data (compare Figs. 1A and B with
Figs. 1C and D). The uncorrected pairwise sequence diver-
gence for the cyt b gene within the ingroup ranged from
7.6% (Hyaena vs. Parahyaena) to 19.5% (Panthera vs. Cro-
cuta). In contrast, for the combined nuclear gene data, the
sequence divergence ranged from 0.12% (Hyaena vs. Para-
hyaena) to 7.7% (Paradoxurus vs. Nandinia). Multiple substi-
tutions at the same sites in the cyt b gene most likely aVected
the phylogenetic signal at higher levels of sequence diver-
gence. Indeed, as measured by the retention index, cyt b data
contained a greater amount of homoplasy relative to the
individual or combined nuclear gene segment data (Table 3).
3.3. Supermatrix analyses
Parsimony analysis, maximum likelihood analysis, and
Bayesian inference of the molecular supermatrix all
resulted in identical topologies (Fig. 2). This topology was
identical to the one reconstructed based on the combined
nuclear data alone (Figs. 1A and B). The 5730 character
supermatrix used for the parsimony analyses had 1056 par-
simony-informative sites and a tree length of 2869 steps.
The retention index for the supermatrix parsimony tree was
lower than that for the concatenated nuclear gene tree
(0.623 vs. 0.852) as a result of the addition of the more
homoplastic cyt b data set. The best-Wtting model of substi-
tution for the supermatrix determined using Modeltest was
the GTR+ I+ G model (Table 3).
Fig. 1. (A) Maximum parsimony phylogram from analysis of concate-
nated nuclear gene data (LD1390). (B) Maximum likelihood phylogram
from analysis of concatenated nuclear gene data (¡ln LD14115.730). (C)
Maximum parsimony phylogram from analysis of the complete cyto-
chrome b gene (LD1479). (D) Maximum likelihood phylogram from
analysis of the complete cytochrome b gene (¡ln LD7205.903). Branch
lengths are drawn proportional to the number of changes (parsimony, A
and C) or the number of substitutions per site (likelihood, B and D); (see
scale bars). Family aYliation of major clades and lineages are shown at
the right of the tree. Bootstrap support values out of 1000 (A and C) or
300 (B and D) pseudoreplicates are shown above internodes.
612 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
In all phylogenetic analyses of the supermatrix, Nandi-
nia was the most basal taxon within the feliforms, fol-
lowed by a clade containing the two felids, which formed
the sister group to the viverrids (Genetta and Paradoxu-
rus), herpestids and hyaenids. The two herpestids and
Cryptoprocta were the sister group to the hyaenids. Based
on the distribution of the four viverrid taxa that we
included in our analyses, our results clearly suggest that
the Viverridae, as deWned in current classiWcations (e.g.,
Corbert and Hill, 1991; Wozencraft, 1993), is paraphy-
letic. Within hyaenids, Proteles was basal and sister to a
clade in which Crocuta formed the sister group to Parahy-
aena plus Hyaena (Fig. 2).
All but two of the nodes in the supermatrix tree were
supported by 100% MP and ML bootstrap values or 1.00
posterior probabilities in the Bayesian analyses (Fig. 2).
The two nodes that received less than maximal support
were the same ones as those in the trees based on the con-
catenated nuclear data (see Figs. 1A and B), speciWcally the
node between the felid clade and the clade of the remaining
feliforms except Nandinia (node 19) and the node between
Proteles and the remaining hyaenids (node 13). Bootstrap
values and posterior probability for these nodes varied
among the methods used to analyze the supermatrix. Nodes
19 and 13 received strong bootstrap support in the MP
analyses (99 and 80%, respectively), but relatively low sup-
port in the ML analyses (63% each). We used the “list of
apomorphies” output in PAUP* from the MP analyses to
examine the nature of nucleotide changes along the two
internal nodes within the Hyaenidae (i.e., nodes 13 and 12 in
Fig. 2). These nodes were supported by 10 and 12 unambig-
uous synapomorphies, respectively (Table 4). Of the 22
unambiguous synapomorphies, half of these were from the
cyt b gene, while the remainders were contributed by vari-
ous nuclear gene segments. The majority of changes were
transitions (A MG or C MT), two were transversions
(A MC) and a single indel in the GHR locus contributed
support to the clade that placed Crocuta as the sister to
Hyaena plus Parahyaena.
Burn-in plots (not shown) from MCMC chains using
models 1 and 2 showed that chain likelihood scores reached
stationarity within 100,000 generations and usually by
50,000 or 60,000 generations. For model 3, however, chain
likelihood scores did not achieve stationarity until about
250,000 generations. This result suggests that longer mixing
of MCMC chains was required for Model 3, most likely due
to the increased number of parameters that had to be esti-
mated. Therefore, consensus topologies, likelihood scores
and parameter estimates from the three independent
MCMC runs conducted for each model are based on 9000
post burn-in trees (after discarding the Wrst 1000 trees) for
Models 1 and 2, but only 7000 post burn-in trees for Model
3 (after discarding the Wrst 3000 trees). There was a clear
improvement in the chain likelihood scores from Model
1(GTR+ I+ G) to Model 3 (GTR + partition-codonSSG) in
which the among-site rate variation was partitioned among
the nuclear gene segments plus the three codon positions of
cyt b (Table 5). The post burn-in estimates of the likelihood
and nucleotide substitution model parameter values were
generally consistent among the three independent MCMC
runs for GTR+ I+G and GTR +partition SSG models, but
these varied more for the GTR + partition-codonSSG
Fig. 2. Maximum likelihood phylogram inferred from analysis of the
molecular supermatrix (¡ln LD22121.985). Internodes with asterisks (*)
indicate maximal bootstrap support (100%) from parsimony and likeli-
hood analyses (1000 and 300 pseudoreplicates, respectively) and 1.00 pos-
terior probability values from Bayesian analyses using Model 1, Model 2,
and Model 3 (see Section 2). Uncircled numbers shown at nodes 13 and 19
are, from top to bottom, maximum parsimony bootstrap, maximum likeli-
hood bootstrap, and node posterior probability values using Models 1, 2,
and 3, respectively. Circled numbers 12–22 indicate dated nodes referred
to in Table 6. Branch lengths are drawn proportional to the number of
substitutions per site (see scale bar). Family aYliation of major clades and
lineages are shown at the right of the tree.
Unambiguous synapomorphies for two clades within the Hyaenidae
The locus, the number corresponding to the base pair position in the gap-
coded alignment of the supermatrix, and the nucleotide transition are
listed below the Newick representation of the clade.
(Crocuta crocuta (Hyaena hyaena, Parahyaena brunnea))
APOB: 745, C !A
GHR: 1757, C !T; 1759, 1 !0; 2232, G !A
GNAT1: 2581, C !T
CYT b: 4728, A !G; 5113, C !T; 5166, A !G; 5577, A !G; 5619,
(Hyaena hyaena, Parahyaena brunnea)
APOB: 605, A !G
CHRNA1: 1096, T !C; 1212, A !C
COL10A1: 1571, G !A
RAG1: 2724, A !G; 2872, C !T
CYT b: 5025, C !T; 5310, A !G; 5415, A !G; 5470, C !T; 5497,
A!G; 5559, C !T
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 613
Parameter estimates (mean and 95% credibility intervals) for each of three MCMC runs using the three substitution models: GTR + I +G; GTR with
among site rate variation partitioned among the eight gene segment partitions (GTR + partitionSSG); and GTR with among site rate variation parti-
tioned among the seven nuclear gene segments and the three codon positions of the cyt b gene (GTR+ partition-codonSSG)
Parameter abbreviations are the same as in Table 3, except SS1, SS2, ƒ Dsite speciWc gamma shape parameter.
Run 1 Run 2 Run 3
GTR + I + G
¡ln Likelihood ¡22141.149 (¡22151.010 to ¡22133.310) ¡22141.052 (¡22150.880 to ¡22133.060) ¡22141.031 (¡22151.000 to ¡22133.560)
A0.271 (0.261–0.282) 0.272 (0.261–0.282) 0.272 (0.262–0.282)
C0.245 (0.235–0.255) 0.245 (0.235–0.255) 0.245 (0.236–0.255)
G0.221 (0.211–0.231) 0.221 (0.211–0.231) 0.221 (0.212–0.231)
T0.262 (0.251–0.273) 0.262 (0.252–0.273) 0.262 (0.252–0.273)
I 0.443 (0.372–0.505) 0.444 (0.375–0.507) 0.444 (0.373–0.505)
G 0.647 (0.492–0.843) 0.650 (0.498–0.853) 0.648 (0.494–0.844)
rAC 2.850 (2.160–3.803) 2.801 (2.127–3.625) 2.814 (2.133–3.618)
rAG 6.624 (5.150–8.584) 6.520 (5.099–8.278) 6.554 (5.092–8.168)
rAT 0.955 (0.675–1.320) 0.936 (0.662–1.294) 0.946 (0.674–1.261)
rCG 0.749 (0.481–1.093) 0.736 (0.500–1.072) 0.751 (0.493–1.055)
rCT 12.308 (9.696–16.015) 12.088 (9.469–15.207) 12.202 (9.519–15.202)
rGT 111
GTR + partitionSSG
¡ln Likelihood ¡21740.559 (¡21764.820 to ¡21725.700) ¡21734.731 (¡21750.000 to ¡21725.450) ¡21736.046 (¡21748.370 to ¡21727.340)
A0.279 (0.268–0.290) 0.281 (0.270–0.293) 0.280 (0.269–0.290)
C0.260 (0.250–0.270) 0.264 (0.253–0.274) 0.261 (0.251–0.271)
G0.208 (0.198–0.218) 0.204 (0.195–0.214) 0.207 (0.198–0.217)
T0.253 (0.243–0.262) 0.251 (0.242–0.261) 0.253 (0.242–0.263)
G 0.323 (0.286–0.364) 0.326 (0.291–0.363) 0.326 (0.289–0.367)
rAC 1.890 (1.503–2.355) 1.609 (1.269–2.034) 1.794 (1.355–2.305)
rAG 6.303 (5.060–7.837) 6.009 (4.863–7.424) 6.148 (4.796–7.725)
rAT 0.792 (0.571–1.073) 0.714 (0.499–0.973) 0.760 (0.553–1.019)
rCG 0.672 (0.473–0.953) 0.618 (0.436–0.864) 0.655 (0.453–0.891)
rCT 9.436 (7.482–11.592) 8.645 (6.952–10.777) 9.131 (7.153–11.325)
rGT 111
SS1 (APOB) 0.510 (0.350–0.750) 0.390 (0.338–0.551) 0.438 (0.410–0.499)
SS2 (CHRNA1) 0.962 (0.846–1.261) 0.892 (0.806–1.175) 1.069 (0.900–1.155)
SS3 (COL10A1) 0.565 (0.329–0.780) 0.594 (0.290–0.900) 0.697 (0.545–0.908)
SS4 (GHR) 0.637 (0.433–0.719) 0.538 (0.519–0.737) 0.549 (0.481–0.671)
SS5 (GNAT1) 0.413 (0.348–0.485) 0.430 (0.310–0.509) 0.443 (0.309–0.503)
SS6 (RAG1) 0.442 (0.276–0.571) 0.354 (0.263–0.465) 0.414 (0.335–0.597)
SS7 (WT1) 0.664 (0.467–0.856) 0.573 (0.544–0.707) 0.678 (0.569–0.720)
SS8 (Cyt b) 2.855 (2.733–3.069) 3.196 (2.950–3.306) 2.933 (2.752–3.123)
GTR + partition-codonSSG
¡ln Likelihood ¡21593.161 (¡21605.710 to ¡21585.130) ¡21604.171 (¡21613.650 to ¡21596.130) ¡21635.771 (¡21684.410 to ¡21602.510)
A0.284 (0.273–0.295) 0.281 (0.271–0.290) 0.280 (0.269–0.291)
C0.263 (0.254–0.274) 0.263 (0.252–0.273) 0.260 (0.251–0.271)
G0.203 (0.194–0.212) 0.207 (0.197–0.216) 0.208 (0.198–0.217)
T0.249 (0.240–0.259) 0.249 (0.239–0.260) 0.252 (0.242–0.261)
G 0.553 (0.478–0.653) 0.485 (0.423–0.555) 0.479 (0.360–0.612)
rAC 1.500 (1.138–1.987) 1.713 (1.295–2.295) 1.837 (1.436–2.262)
rAG 5.611 (4.446–6.996) 6.007 (4.823–7.594) 6.199 (5.039–7.700)
rAT 0.691 (0.497–0.928) 0.784 (0.569–1.095) 0.803 (0.585–1.066)
rCG 0.624 (0.422–0.866) 0.659 (0.459–0.893) 0.696 (0.489–0.959)
rCT 8.714 (6.847–10.868) 9.371 (7.493–11.860) 9.574 (7.595–11.963)
rGT 111
SS1 (APOB) 0.630 (0.628–0.630) 0.491 (0.477–0.570) 0.461 (0.323–0.672)
SS2 (CHRNA1) 0.645 (0.549–1.725) 0.628 (0.462–1.364) 2.419 (0.770–4.254)
SS3 (COL10A1) 1.783 (0.748–1.870) 0.854 (0.622–0.872) 0.731 (0.254–1.072)
SS4 (GHR) 0.596 (0.586-0.734) 0.728 (0.708–0.810) 0.651 (0.443–0.986)
SS5 (GNAT1) 0.194 (0.163–0.194) 0.301 (0.202–0.480) 0.473 (0.358–0.683)
SS6 (RAG1) 0.415 (0.412–0.470) 0.532 (0.402–0.627) 0.442 (0.300–0.751)
SS7 (WT1) 0.577 (0.557–0.800) 0.704 (0.654–0.919) 0.694 (0.419–0.958)
SS8 (Cyt b codon 1) 1.783 (1.169–1.803) 3.487 (2.073–3.776) 1.949 (1.525–6.802)
SS9 (Cyt b codon 2) 0.379 (0.355–0.751) 0.419 (0.360–0.727) 0.823 (0.364–1.019)
SS10 (Cyt b codon 3) 6.021 (5.142–6.118) 4.341 (3.741–4.452) 4.162 (3.207–5.082)
614 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
model (Table 5). The same topology (Fig. 2) was recovered
among the independent MCMC runs for each model as well
as among the three models. The posterior probability values
shown in Fig. 2 represent the Wnal consensus values derived
from the combination of results from the three separate
runs for each model (3 £9000 D 27,000 trees for Models 1
and 2 and 3£7000 D21,000 trees for Model 3). All but two
nodes received 1.00 posterior probability support. The two
nodes that had <1.00 posterior probability support diVered
greatly in their posterior probabilities according to the
model implemented. For node 19, the posterior probability
increased from 0.82 (Model 1) to 0.94 (Model 3). The pos-
terior probability of node 13 increased from 0.84 with
Model 1 to 0.94 with Model 2 but then decreased to 0.79
with Model 3 (Fig. 2). This suggests that, at least for these
two nodes, the estimated posterior probability was sensitive
to the way in which among-site rate variation was parti-
tioned among or within the individual gene segments. How-
ever, as we noted above with regard to the phylogenetic
results based on the concatenated nuclear data, these two
nodes were estimated as the shortest branches on the tree
(Fig. 2).
3.4. Analyses combining the molecular supermatrix and
morphologic characters
When we combined the molecular supermatrix with the
29 morphological characters from the study by Werdelin
and Solounias (1991), an exhaustive search using parsi-
mony found the tree shown in Fig. 3. Relationships among
the four hyaenids were identical to those based on the con-
catenated nuclear data and the supermatrix alone (Figs. 1A
and B, and 2). Nodes deWning hyaenid monophyly and
grouping Hyaena plus Parahyaena received 100% parsi-
mony bootstrap support, whereas the node between Pro-
teles and other hyaenids had a bootstrap support of 99%.
Bremer support was similarly high for these nodes, but the
PBS analyses revealed that the nine individual data sets
diVered in their relative contributions to total Bremer sup-
port at the nodes (Fig. 3). Each of the molecular data sets
provided a large contribution to hyaenid monophyly, but
among the 29 morphological characters, only 2 of these
provided support for the monophyly of hyenas. The node
between Proteles and other hyaenids had a Bremer support
value of 14, and this node received the most support from
the morphological data among the nine data partitions in
the PBS analyses. Further, among the molecular data sets,
CHRNA1 and COL10A1 provided conXicting information
for this relationship based on their negative values whereas
RAG1 and WT1 apparently contributed zero support. The
loci that did contribute positive support to this node are the
same ones that provided unambiguous synapomorphies for
this node in the parsimony analyses of the supermatrix
(Table 4). For the node that joins Hyaena and Parahyaena
together in a clade, all molecular data sets contributed posi-
tive support, with cyt b contributing the most support.
Interestingly, the phylogenetic signal from the morphologi-
cal data contributes conXicting signal to this node, as evi-
denced by the PBS value of ¡3 (Fig. 3). We note, however,
that the Bremer support and PBS values should be treated
with caution because the statistical meaning of these values
are unclear (DeBry, 2001). Nonetheless, Bremer support
and PBS analyses together provide a rough means of evalu-
ating the relative contributions of diVerent data sets to the
total nodal support of a tree in a combined analysis.
3.5. Divergence times
The relaxed clock estimates of divergence dates provided a
detailed estimate of the timing of diversiWcation of the four
Fig. 3. Maximum parsimony phylogram inferred from analysis of the combined molecular supermatrix and modiWed morphological character matrix from
Werdelin and Solounias (1991). For this tree, length (L)D918, consistency index (CI) excluding uninformative charactersD0.7960, and retention index
(RI) D0.7597. Numbers above internodes and to the left of boxes are bootstrap values (1000 pseudoreplicates). Numbers on top of boxes are Bremer support
(BS) index values. Numbers inside boxes are partitioned Bremer support (PBS) values from the nine data partitions used in the combined analysis (see key).
Branch lengths are drawn proportional to the number of changes (see scale bar). Morph Dmorphological data. Hypoth. ancestor Dhypothetical ancestor.
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 615
hyena species, as well as their separation from the other feli-
form clades (Table 6, Fig. 2). Hyenas diverged from the core
Herpestidae plus Cryptoprocta clade ca. 29.2 MYA, but the
diversiWcation of the four crown species is quite recent, at ca.
10.6 MYA (credibility interval [CI]: 7.3–15.0 MYA). This date
corresponds to the node uniting Proteles to the remaining
hyaenids, while the subsequent node (Crocuta united to
[Hyaena plus Parahyaena]) was estimated at 8.6 MYA (CI:
5.7–12.4 MYA), and the split between Hyaena and Parahya-
ena was estimated to have occurred ca. 4.2 MYA (CI: 2.6–6.4
MYA). Eleven diVerent runs of the dating analysis produced
consistent values, aYrming the convergence of the Bayesian
chains and congruence across varying combinations of fossil
constraints. Importantly, in one of the alternative runs we
released the upper limit (maximum) of 16.4 MYA for the base
of crown Hyaenidae, to test whether our young dates for this
clade could be biased due to use of this constraint. Results
from this run were almost identical to the ones shown here,
with point estimates and credible intervals for node ages
diVering by less than 1 million years from the values presented
in Table 6. ModiWcation of the mean of the prior distribution
for ingroup root age from 50 to 25 MYA did produce detect-
able changes, however the eVect was minor given the magni-
tude of the parameter change: point estimates for recent
nodes (e.g., all those within the Hyaenidae) were shifted by
0.3–0.8 million years, whereas those for basal nodes in the
Feliformia were moved by up to 4 million years.
4. Discussion
Phylogenetic analysis of the molecular supermatrix recov-
ered the same topology of relationships among the four liv-
ing hyaenids, regardless of the optimality criterion we used
(Fig. 2). Among the three bone-cracking hyaenids, Hyaena
and Parahyaena are joined together as sister taxa with maxi-
mum nodal support (100% MP and ML bootstrap and 1.00
Bayesian posterior probability), and Crocuta is the more
basal sister species to this clade. Proteles is the basal taxon in
the extant Hyaenidae and is a very close sister species to the
three bone-cracking species (Fig. 2). We Wnd this same pat-
tern of relationships in the phylogenetic analysis with the
concatenated nuclear gene data, also with high bootstrap
support (Figs. 1A and B). The cyt b data support a slightly
diVerent topology, however, in which the basal node divides
the hyaenids into two clades, one containing Hyaena plus
Parahyaena and the other containing Crocuta plus Proteles.
This same topology was recovered in a previous phylogenetic
study of hyaenids, also using the cyt b gene (Jenks and Wer-
delin, 1998). These authors attributed the pairing of Crocuta
with Proteles to a long-branch attraction artifact, possibly
caused by the use of distant outgroups, Felis catus, a feliform
carnivoran and Phoca vitulina, a caniform carnivoran, in
their analyses. However, a topology that places Proteles in its
generally accepted position at the base of the clade (as in our
Fig. 2) was obtained when all transition substitutions were
excluded or only third position transversions were analyzed
(Jenks and Werdelin, 1998).
Our results challenge two widely accepted ideas concern-
ing hyaenid relationships: Wrst, that Proteles is a distantly
related relict of an extinct clade of hyaenids, and second,
that the highly autapomorphic Crocuta is the most derived
hyaenid. In addition, our analysis rejects the recent place-
ment of Hyaena and Crocuta as sister taxa (Werdelin and
Solounias, 1991), and favors previous analyses that united
Hyaena and Parahyaena as sister taxa (Ewer, 1955; Gaudry,
1862–1867; Hendey, 1974; Pilgrim, 1932; Thenius, 1966).
Below we discuss each of these Wndings, as well as our anal-
ysis of the interrelationships among feliform carnivorans,
and the dating of these divergences.
4.1. The Position of Proteles
The fossil record and our data indicate that Crocuta,
Hyaena, Parahyaena, and Proteles are the living remnants of
a diverse clade of medium to large sized highly carnivorous
hyenas that were widespread in the Old World between 12
and 5 MYA (Werdelin and Solounias, 1991). The oldest
known deWnitive fossil hyaenids in Africa are from the Mid-
dle Miocene (ca. 14 MYA), about Wve million years after tec-
tonic activity established a land bridge (the Gomphothere
Bridge) that connected Eurasia and Africa, facilitating faunal
dispersal between the two continents for the Wrst time
(Agustí and Antón, 2002; Hunt, 1996; Turner and Antón,
2004). Based on the credibility interval for node 14 (Table 6),
the four extant species appear to have originated in Africa 7–
15 MYA, before the Early Pliocene (ca. 5 MYA), a time
when hyaenids were diverse in Africa (Hunt, 1996).
Our results suggest that the aardwolf diverged much
more recently (ca. 10.6 MYA) from its bone-cracking rela-
tives than the inferred divergence time of »18–20 MYA
based on morphology (Werdelin and Solounias, 1991) or a
molecular clock using third position transversions of the
cyt b gene (Jenks and Werdelin, 1998). This has two impor-
tant consequences for understanding the evolution of the
aardwolf, which Wrst appears in the African Pleistocene fos-
sil record, ca. 1.5 MYA (Richardson, 1987a,b; Werdelin and
Solounias, 1991). First, the younger age reduces the gap
Summary of divergence dating results for hyaenids and related carnivo-
rans obtained with a Bayesian relaxed clock approach using a molecular
supermatrix and nine fossil calibrations (see Section 2)
Node numbers refer to clades indicated in Fig. 2. CI, credibility interval.
Node/clade Included taxa Age (MYA) CI (MYA)
12 Hyaena +Parahyaena 4.2 2.6 – 6.4
13 Crocuta +Hyaena +Parahyaena 8.6 5.7–12.4
14 Hyaenidae 10.6 7.3–15.0
15 Core Herpestidae 11.8 8.1–16.7
16 Cryptoprocta + Core Herpestidae 24.4 18.3–32.2
17 Hyaenidae +Clade 16 29.2 22.5–37.9
18 Core Viverridae 25.2 18.9–33.2
19 Clades 17 +18 35.2 27.7–45.1
20 Felidae 14.7 10.2–20.5
21 Felidae +Clade 19 36.5 28.9–46.5
22 Nandinia + Clade 21 43.3 33.4–54.1
616 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
between the estimated time of origin of the aardwolf and its
earliest appearance in the fossil record to approximately
nine million years, as opposed to 16.5–18.5 million years as
implied by previous morphological studies (Werdelin and
Solounias, 1991). Second, this result suggests that the aard-
wolf is derived from a more recently evolved group of hyae-
nids and thus challenges ideas about the morphological
transition that led to the evolution of the aardwolf’s unique
ecology and associated craniodental morphology. Based on
their morphological analyses of fossil and extant hyaenids,
Werdelin and Solounias (1991) placed Proteles near the
base of the hyaenid tree, in an unresolved position between
Plioviverrops and other primitive hyaenids (e.g., Tungurictis,
Tongxinictis, and Thalassictis). This position was favored
because Plioviverrops species tended to be small insecti-
vores and omnivores, and thus a transition to myrmeco-
phagy seemed relatively plausible (Thenius, 1966; Werdelin
and Solounias, 1991). In this view, the morphological adap-
tations associated with myrmecophagy are ancestral fea-
tures retained from Middle to Late Miocene hyaenids
(Werdelin and Solounias, 1991).
In contrast, our estimated divergence time suggests that
the aardwolf may instead have a shared ancestry with the
Lycyaena-Chasmaporthetes lineage, which were cursorial
meat and bone eating hyaenids that Wrst arose in the late
Middle Miocene, ca. 11.8 MYA (Werdelin and Solounias,
1991, 1996). Fossils of Lycyaena Wrst appear in north Afri-
can deposits at around this time and then become more
dominant in the Late Miocene and Pliocene of Africa (Wer-
delin and Solounias, 1991). Derivation of aardwolves from
Lycyaena ancestors was originally proposed by Ewer and
Cooke (1964) and later reiterated by Ewer (1973). In sup-
port of this relationship, aardwolves share the more derived
dental formula with the three extant bone-cracking hyenas
(Gingerich, 1974), which is characterized by fewer molars
and premolars. However, Werdelin and Solounias (1991)
considered the ancestry of Proteles from Lycyaena to be
implausible because of the divergent craniodental morphol-
ogy of Lycyaena compared to Proteles. If Proteles was a
descendant of a taxon similar to Plioviverrops that had 6–7,
rather than 4–5 cheek teeth as in the extant bone-cracking
hyenas, more cheek teeth would be expected to be retained
by Proteles, despite their reduction to pegs. In addition, the
short branch length between the split of Proteles and Cro-
cuta suggests that Proteles shares a close common ancestry
(ca. 2 million years) with the more derived bone-cracking
hyaenids (Table 6).
Although our data suggest that Proteles may have had a
more recent ancestry than previously thought, it does not
indicate when Proteles acquired its unique dietary special-
ization. We suggest that the remarkable transition from
meat eating carnivore to termite specialist may have
occurred at a time when grassland ecosystems were becom-
ing more widespread worldwide, including the sub-Saharan
African savannas, from the Late Miocene to the Early
Pleistocene (Jacobs et al., 1999; Potts and Behrensmeyer,
1992). Notably, >90% of the diet of aardwolves consists of
nasute harvester termites of the genus Trinervitermes, which
consume grasses (Kruuk and Sands, 1972; Richardson,
1987a). These termites forage on the soil surface in dense
concentrations, and consequently the aardwolf may ingest
as many as 300,000 termites in a single night (Richardson,
1987b). Trinervitermes is thought to have evolved during
the Late Miocene (Weesner, 1960) and is one of the most
derived genera within the isopteran family Termitidae (B.
Thorne, pers. Comm.). Therefore, the aardwolf may have
evolved from carnivorous and/or bone-eating ancestors
into a niche that was largely unexploited by other carnivo-
rans, possibly as a result of increased competition with
other large carnivorans, which reached their greatest diver-
sity during the Pliocene (Hunt, 1996; Turner and Antón,
2004; Werdelin, 2003). A test of the hypothesis of coevolu-
tion between the aardwolf and Trinervitermes might be pro-
vided by reconstructing the phylogenetic history of the
Termitidae as well as using this phylogeny to understand
the evolution of foraging mode in this family of termites.
4.2. The evolution of Crocuta
The spotted hyena appears to be the most derived of
the living species based on unique features of its morphol-
ogy and behavior (Werdelin and Solounias, 1991). Rela-
tive to the other two extant, bone-cracking species,
Crocuta is more specialized as a hypercarnivore, hunting
for prey rather than scavenging, foraging in groups rather
than alone, and regularly capturing prey larger than itself.
This is reXected in its robust skull and teeth that easily
crack very large bones and rapidly slice through tough
skin (Werdelin, 1996a,b; Van Valkenburgh et al., 2004).
Relative to striped and brown hyenas, spotted hyenas
have a broader muzzle, larger carnassials, and enhanced
jaw muscle mechanics (Van Valkenburgh et al., 2004).
Beyond these feeding adaptations, spotted hyenas are
unique in their extreme sexual monomorphism, in which
females have masculinized genitalia associated with high
levels of aggression. This collection of autapomorphies
has led to the assumption that Crocuta is the most special-
ized of the three bone-cracking hyaenids (Werdelin and
Solounias, 1991). However, our phylogenetic analysis
indicates otherwise; the divergence between Crocuta and
Hyaena plus Parahyaena occurred ca. 8.6 MYA (Table 6).
Based on the fossil record, their common ancestor was
likely an African bone-cracking species that split to form
two lineages, one leading to the dominantly hunting Cro-
cuta, and the other to the dominantly scavenging Hyaena
plus Parahyaena clade.
This evolutionary scenario presents some intriguing
questions concerning the evolution of the unique genital
monomorphism of Crocuta. It has been argued that female
masculinization evolved as a byproduct of selection for
aggression when feeding at carcasses (Frank, 1986).
Female spotted hyenas produce relatively large young that
must be nourished with copious quantities of maternal
milk. High levels of competitive aggression occurs during
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 617
group carcass feeding, and dominant females gain greater
access to carcasses and raise more young successfully
(Frank et al., 1995; Holekamp et al., 1996; Holekamp and
Smale, 2000). Because there is no known osteological indi-
cator of hyena masculinization at present, the origin of this
trait cannot be determined. Moreover, there is evidence
that both androgenic and nonandrogenic mechanisms are
involved in the development of the penile clitoris of female
Crocuta (Drea et al., 1998, 1999). However, Frank’s sce-
nario would suggest that masculinization should have
evolved concomitantly with the ability to cooperatively
hunt large prey. It is also possible that the evolution of
cooperative hunting and consequent feeding competition
may have accelerated the evolution of an emerging suite of
variations in the genitalia of female Crocuta due to nonan-
drogenic mechanisms. The craniodental morphology of
the earliest Crocuta from the Late Pliocene of Africa and
Asia (McKenna and Bell, 1997) is consistent with a group
hunter of large prey and thus the genital monomorphism
may have evolved very near to the origin of this species. It
is also possible that genital monomorphism was present in
the common ancestor of all three bone-crackers, but this
would require a loss of multiple characters within the scav-
enging lineage. Surprisingly, no fossil or living intermedi-
ates have been found between either Crocuta and other
hyaenids or between Proteles and other hyaenids. The
absence of intermediate forms for these unusual species
suggests that the transitions were rapid (but see Werdelin
and Solounias, 1991, who found morphological evolution
within hyaenids to be generally gradual in nature). Indeed,
given the short internal branch lengths and long terminal
branch lengths for the Hyaenidae in our tree (Fig. 2), Pro-
teles and Crocuta may have acquired their autapomorphic
morphologies well after their divergence. As noted above
in the discussion of Proteles, the guild of large African car-
nivores was very diverse in the Pliocene. Intense competi-
tion for food could have driven the evolution of both the
aardwolf and spotted hyena, with the former exiting the
guild and the other evolving unique behaviors and mor-
phology to succeed.
4.3. Hyaena and Parahyaena as sister taxa
With regard to relationships among Crocuta, Hyaena
and Parahyaena, our phylogeny contradicts the hypothesis
of relationships based on cladistic analyses of morphologi-
cal characters that either place Hyaena or Parahyaena as
the sister taxon to Crocuta (Galiano and Frailey, 1977;
Werdelin and Solounias’, 1991). In Werdelin and Solounias
(1991) cladistic analysis of the four extant genera, trees in
which either Hyaena or Parahyaena was paired with Cro-
cuta were equally parsimonious. However, these authors
suggested that the tree in which Hyaena was paired with
Crocuta was better supported because these taxa shared
three synapomorphies, relative to the two synapomorphies
joining Parahyaena and Crocuta together. In either case,
these relationships were weakly supported. Nonetheless,
our tree is congruent with earlier views based on overall
morphological similarity that suggested that Parahyaena
and Hyaena are more closely related than either of these are
to Crocuta (Ewer, 1955; Gaudry, 1862–1867; Hendey, 1974;
Pilgrim, 1932; Thenius, 1966).
Given the small number of extant taxa involved and
research eVort spanning more than 100 years, why have
morphological studies of hyaenid phylogeny not been able
to reach a consensus? Part of the answer lies with changing
philosophies associated with methods of phylogeny recon-
struction and how morphological character similarity
among taxa are thus evaluated (e.g., phenetic similarity vs.
shared ancestral and shared derived traits). Moreover, a
thorough survey of craniodental and postcranial skeletal
morphology revealed few phylogenetically informative
characters available for inferring relationships among
extant taxa (Werdelin and Solounias, 1991). The short time
intervals between branching events among hyaenids as evi-
denced in our phylogeny and dating analyses suggests that
there was little time for informative morphological charac-
ters to become Wxed. Even though mitochondrial and
nuclear sequence data unambiguously support Hyaena and
Parahyaena as sister taxa, the PBS analysis of the combined
molecular and morphological character matrices showed
that the morphological data contributes conXicting signal
to this relationship (Fig. 3).
Our phylogeny suggests that some morphological char-
acters that have been traditionally used in hyaenid system-
atics should be reevaluated. The craniodental skeleton has
been the primary focus of almost all morphologically based
inferences of hyaenid phylogeny. This is because cranioden-
tal remains comprise the vast majority of evidence when it
comes to fossil hyaenids and therefore aspects of the cra-
niodental skeleton allow extinct and extant taxa to be ana-
lyzed simultaneously. The availability of a rich fossil record
for the Hyaenidae as a whole has been critical to deWning
character states and accurately establishing polarities, par-
ticularly with regard to the four extant taxa (Werdelin and
Solounias, 1991). Nevertheless, our phylogeny clearly indi-
cates that the morphological characters that join either
Hyaena or Parahyaena with Crocuta in the analyses using
only extant taxa are either homoplasies or shared ancestral
character states (symplesiomorphies). For example, Hyaena
and Crocuta are united by the following character states:
(1) anterior position of infraorbital foramina relative to the
middle of the upper third premolar; (2) presence of a pre-
maxillary-frontal suture; (3) large processes for the nuchal
ligament; and (4) a straight scapular spine (Werdelin and
Solounias, 1991; present study). Beyond the implication
that the polarities of these characters should be reinter-
preted, there is the more important question of whether
these (and perhaps other) characters actually constitute
homologies or homoiologies (Reidl, 1978). Homoiologies
are non-heritable morphological similarities that result
from similar epigenetic responses to environmental stimuli
(Lieberman, 2000). Because bone has the ability to remodel
in response to changes in mechanical loading, many
618 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
osseous features of the cranial and postcranial skeleton are
potentially homoiologies (Gibbs et al., 2000; Lieberman,
2000). With regard to hyenas, for example, the size of the
nuchal processes (coded as either small or large) that serve
for the attachment of the neck musculature may reXect how
these processes are mechanically loaded in diVerent species
of hyenas and which may itself be related to the behavioral
function of these muscles. Similarities in nuchal processes
(as well as scapular spine shape) between diVerent species of
hyena may be due to similar loading regimes rather than
common ancestry (i.e., homology). Further, because of inte-
grated development among individual skull bones, change
in shape or structure of one bone is usually inXuenced by
changes in neighboring bones (Lieberman, 2000). As a con-
sequence, osseous characters that have been used in
appraisals of hyaenid phylogeny, such as relative position
of the infraorbital foramen in the maxilla bone and the
presence or absence of a premaxillary-frontal suture, can-
not be considered independent in the phylogenetic sense.
This also makes it diYcult to understand whether such
characters represent true homologies.
Comparative studies of craniofacial development and
the eVects of mechanical loading in diVerent species of
hyena oVer a potential approach to evaluate the homolo-
gous nature of skull characters and thus their usefulness in
phylogenetic inference. Such integrated studies have proven
especially useful in testing homologies in the cranial skele-
ton of hominids (Lieberman, 2000 and references therein).
The study by Binder and Van Valkenburgh (2000) examin-
ing correlated changes in skull morphology and bite
strength in captive juvenile and adult spotted hyenas is
especially pertinent in this regard. If the ontogenetic
approach employed by these researchers could be extended
to brown and striped hyenas and perhaps be integrated
with landmark analyses (e.g., O’Higgins, 2000), this would
perhaps better illuminate the phylogenetic signal in the cra-
nial skeleton of hyenas.
As noted above, cyt b data support a topology which
conXicts with that produced by the supermatrix, in that the
basal node divides the hyaenids into two clades, one con-
taining Hyaena plus Parahyaena and the other containing
Crocuta plus Proteles. Given that the topology based on the
supermatrix is the phylogenetic hypothesis with the greatest
support based on molecular data, then the present study
and the one by Jenks and Werdelin (1998) indicate that the
cyt b sequences by themselves contain misleading phyloge-
netic signal with regard to hyaenid relationships. This mis-
leading signal may stem from a combination of homoplasy
in transitions at Wrst and third codon positions (Jenks and
Werdelin, 1998) and/or nonstationary base composition
(this study). While model-based methods like ML are usu-
ally expected to mitigate problems associated with homo-
plasy, we found that this was not the case here (Fig. 1D)
and showed that nonstationary base composition was a
more likely cause of the misleading signal (see Section 3). In
the parsimony analysis of the supermatrix, we found that
cyt b sequences contain unambiguous synapomorphies for
the two internal nodes within the hyaenid phylogeny (Table
5). Furthermore, PBS analyses of a data matrix in which we
combined molecular and morphological characters showed
that the cyt b data contributed substantial and positive sig-
nal to these same two nodes (Fig. 3). Therefore, the cyt b
data are consistent with the topology derived from the
nuclear gene data when the former data set is combined
with the latter but is inconsistent (at least in ML and
equally weighted MP analyses) when analyzed alone. These
Wndings support the idea that combining data from multi-
ple genomes (mitochondrial and nuclear in animals, along
with the chloroplast genome in plants) can overcome the
location-dependent idiosyncrasies that may be particular to
certain genomes or linkage groups (Cummings et al., 1995).
Additionally, our study reinforces the notion that a single
gene or linkage group is unlikely to contain robust phyloge-
netic signal for all levels of a phylogeny and underscores
the importance of sampling multiple loci (Cummings et al.,
1995; Rokas et al., 2003).
4.4. Interrelationships among the feliform carnivorans
Our results show that the clade containing Herpestes,
Mungos and Cryptoprocta is placed as the sister group to the
Hyaenidae with 100% nodal support (Fig. 2). Furthermore,
we estimate that these clades diverged around 29 MYA, in
the Middle Oligocene (Table 6). This age is somewhat older
than the 25 MYA age suggested by the hyaenid fossil record
(Werdelin and Solounias, 1991), but this latter age falls
within our estimated credibility interval (Table 6). The join-
ing of hyaenids and herpestids as sister taxa supports earlier
phylogenetic estimates based on analyses of the auditory
bulla (Hunt, 1989). Our results are also congruent with previ-
ous molecular studies using mitochondrial gene sequences
(cyt b and ND2) and nuclear gene sequences (intron 1 of the
transthyretin gene and exon 1 of the interphotoreceptor rete-
noid-binding protein) that sampled only Crocuta as the sole
representative of the Hyaenidae (Flynn and Nedbal, 1998;
Gaubert and Veron, 2003; Yoder et al., 2003; Yu et al., 2004).
The two genera of mongooses represented in our study (Her-
pestes and Mungos) are classiWed in the Herpestidae while the
Malagasy-endemic carnivoran Cryptoprocta is traditionally
classiWed as a member of the Viverridae. Our Wndings, how-
ever, conWrm the results of a recent study of Malagasy Car-
nivora that showed that carnivorans endemic to Madagascar
(traditionally classiWed in both the Herpestidae and Viverri-
dae) are descended from a single common ancestor and share
ancestry with the Herpestidae (Yoder et al., 2003). Further-
more, among the four viverrid taxa we sequenced in our
study, Nandinia was the most basal lineage within the feli-
forms (Fig. 2). Collectively, our results lend additional sup-
port to the conclusions of several recent studies that have
shown that the Viverridae, as traditionally circumscribed, is
not monophyletic (Flynn and Nedbal, 1998; Gaubert and
Veron, 2003; Yoder et al., 2003; Yu et al., 2004). Overall, the
pattern of interrelationships among the feliform families
inferred in this study are concordant with these other molec-
K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620 619
ular-based studies, suggesting that diVerent regions of the
feliform carnivoran genome are tracking the same phyloge-
netic history. Although a larger sample of feliform taxa is
obviously required to further validate these Wndings, such
concordance across diVerent studies nonetheless provides
conWdence that a stable phylogenetic hypothesis for the
primary families of the Feliformia is emerging.
We are grateful to the individuals and institutions listed
in Table 1 for providing the tissue and/or DNA samples
used in this study. We thank P. Adam and C. Bardeleben
for comments and editorial suggestions that improved the
manuscript. This study was supported by research funds
from the U.S. National Science Foundation.
Appendix A. Supplementary material
Supplementary data associated with this article can be
found, in the online version, at doi:10.1016/j.ympev.
Agustí, J., Antón, M., 2002. Mammoths, Sabertooths, and Hominids: 65
Million Years of Mammalian Evolution in Europe. Columbia Univ.
Press, New York, NY.
Akaike, H., 1974. A new look at the statistical model identiWcation. IEEE
Trans. Autom. Contr. 19, 716–723.
Amrine-Madsen, H., KoepXi, K.-P., Wayne, R.K., Springer, M.S., 2003. A
new phylogenetic marker, apolipoprotein B, provides compelling evi-
dence for eutherian relationships. Mol. Phylogenet. Evol. 28, 225–240.
Barriel, V., 1994. Molecular phylogenies and how to code insertion/dele-
tion events. Life Sci. 317, 693–701.
Binder, W.J., Van Valkenburgh, B., 2000. Development of bite strength
and feeding behaviour in juvenile spotted hyenas (Crocuta crocuta). J.
Zool. Lond. 252, 273–283.
Bininda-Emonds, O.R.P., Gittleman, J.L., Purvis, A., 1999. Building large
trees by combining phylogenetic information: a complete phylogeny of
the extant Carnivora (Mammalia). Biol. Rev. 74, 143–175.
Bremer, K., 1988. The limits of amino acid sequence data in angiosperm
phylogenetic reconstruction. Evolution 42, 795–803.
Brouillette, J.A., Andrew, J.R., Venta, P.J., 2000. Estimate of nucleotide
diversity in dogs with a pool-and-sequence method. Mammal. Genome
11, 1079–1086.
Brown, J.H., Lomolino, M.V., 1998. Biogeography. Sinauer Associates,
Sunderland, MA.
Corbert, G.B., Hill, J.E., 1991. A World List of Mammalian Species.
Oxford University Press, London.
Cummings, M.P., Otto, S.P., Wakeley, J., 1995. Sampling properties of DNA
sequence data in phylogenetic analysis. Mol. Biol. Evol. 12, 814–822.
DeBry, R.W., 2001. Improving the interpretation of the decay index for
DNA sequence data. Syst. Biol. 50, 742–752.
DeBry, R.W., Olmstead, R.G., 2000. A simulation study of reduced tree-
search eVort in bootstrap resampling analysis. Syst. Biol. 49, 171–179.
Drea, C.M, Weldele, M.L., Forger, N.G., Coscia, E.M., Frank, L.G., Glick-
man, S.E., 1998. Androgens and masculinization of genitalia in the
spotted hyaena (Crocuta crocuta). 2. EVects of prenatal anti-androgens.
J. Reprod. Fert. 113, 117–127.
Drea, C.M., Corsica, E.M., Glickman, S.E., 1999. Hyenas. In: Knobil, E.,
Neill, J.D. (Eds.), Encyclopedia of Reproduction Vol. 2. Academic
Press, San Diego, CA, pp. 718–725.
Ewer, R.F., 1955. The fossil carnivores of the Transvaal caves. The lycyaenas
of Sterkfontein and Swartkrans, together with some general considerations
of the Transvaal fossil hyaenids. Proc. Zool. Soc. Lond. 124, 839–857.
Ewer, R.F., 1973. The Carnivores. Cornell Univ. Press, Ithaca, NY.
Ewer, R.F., Cooke, H.B.S.., 1964. The Pleistocene mammals of southern
Africa. In: Davis, D.H.S. (Ed.), Ecological Studies in Southern Africa.
W. Junk, The Hague, pp. 35–48.
Felsenstein, J., 1978. Cases in which parsimony and compatibility methods
will be positively misleading. Syst. Zool. 27, 401–410.
Flynn, J.J., Nedbal, M.A., 1998. Phylogeny of the Carnivora (Mammalia):
congruence vs incompatibility among multiple data sets. Mol. Phyloge-
net. Evol. 9, 414–426.
Frank, L.G., 1986. Social organization of the spotted hyaena (Crocuta cro-
cuta) II. Dominance and reproduction. Anim. Behav. 35, 1510–1527.
Frank, L.G., Holekamp, K.E., Smale, L., 1995. Dominance, demography
and reproductive success of female spotted hyenas. In: Sinclair, A.R.E.,
Arcese, P. (Eds.), Serengeti II: Dynamics, Management, and Conserva-
tion of an Ecosystem. Univ. Chicago Press, Chicago, IL, pp. 364–384.
Galiano, H., Frailey, D., 1977. Chasmaporthetes kani, new species from
China, with remarks on the phylogenetic relationships of genera within
the Hyaenidae (Mammalia, Carnivora). Am. Mus. Nov. 2632, 1–16.
Gaubert, P., Veron, G., 2003. Exhaustive sample set among Viverridae
reveals the sister-group of felids: the linsangs as a case of extreme mor-
phological convergence within Feliformia. Proc. R. Soc. Lond. B 270,
Gaudry, A., 1862–1867. Animaux fossiles du Mont Leberon. F. Savy,
Paris, France.
Gingerich, P.D., 1974. Proteles cristatus Sparrman from the Plesitocene of
South Africa, with a note on tooth replacement in the aardwolf (Mam-
malia: Hyaenidae). Ann. Transvaal Mus. 29, 49–55.
Gibbs, S., Collard, M., Wood, B., 2000. Soft-tissue characters in higher pri-
mate phylogenetics. Proc. Natl. Acad. Sci. USA 97, 11130–11132.
Hendey, Q.B., 1974. The late Cenozoic Carnivora of the southwestern
Cape Province. Ann. South Afr. Mus. 63, 1–369.
Holekamp, K.E., Smale, L., Szykman, M., 1996. Rank and reproduction in
the female spotted hyaena. J. Reprod. Fert. 108, 229–237.
Holekamp, K.F., Smale, L., 2000. Feisty females and meek males: repro-
ductive strategies in the spotted hyena. In: Wallen, K., Schneider, K.,
Schneider, J. (Eds.), Reproduction in Context. The MIT Press, Cam-
bridge, MA, pp. 257–285.
Huelsenbeck, J.P., 2000. MrBayes: Bayesian inference of phylogeny. Bio-
metrics 17, 754–755.
Huelsenback, J.P., Ronquist, F., Nielsen, R., Bollbeck, J.P., 2001. Beyesian
inference of phylogeny and its impact on evolutionary biology. Science
294, 2310–2314.
Hunt Jr., R.M, 1987. Evolution of the Aeluroid Carnivora: signiWcance of
auditory structure in the nimravid cat Dinictis. Am. Mus. Novitates
2886, 1–74.
Hunt Jr., R.M, 1989. Evolution of the Aeluroid Carnivora: SigniWcance of
the ventral promontorial process of the petrosal, and the origin of basi-
cranial patterns in the living families. Am. Mus. Novitates 2930, 1–32.
Hunt Jr., R.M, 1996. Biogeography of the Order Carnivora. In: Gittleman,
J.L. (Ed.), Carnivore Behavior, Ecology and Evolution, vol. 2. Cornell
Univ. Press, Ithaca, NY, pp. 485–541.
Irwin, D.M., Kocher, T.D., Wilson, A.C., 1991. Evolution of the cyto-
chrome b gene of mammals. J. Mol. Evol. 32, 128–144.
Jacobs, B.F., Kingston, J.D., Jacobs, L.L., 1999. The origin of grass-domi-
nated ecosystems. Ann. Missouri Bot. Gard. 86, 590–643.
Jenks, S.M., Werdelin, L. 1998. Taxonomy and systematics of living hyae-
nas (Family Hyaenidae). In: Mills, M.G.L., Hofer, H. (Compilers),
Hyaenas: Status Survey and Conservation Action Plan. IUCN/SSC
Hyaena Specialist Group, IUCN, Gland, Switzerland, pp. 8–17.
Jiang, Z., Priat, C., Galibert, F., 1998. Traced orthologous ampliWed
sequence tags (TOASTs) and mammalian comparative maps. Mam-
mal. Genome 9, 577–587.
Kishino, H., Thorne, J.L., Bruno, W.J., 2001. Performance of a divergence
time estimation method under a probabilistic model of rate evolution.
Mol. Biol. Evol. 18, 352–361.
620 K.-P. KoepXi et al. / Molecular Phylogenetics and Evolution 38 (2006) 603–620
Kluge, A.G., 1998. Total evidence or taxonomic congruence: cladistics or
consensus classiWcation. Cladistics 14, 151–158.
Kocher, T.D., Thomas, W.K., Meyer, A., Edwards, S.V., Paabo, S., Villa-
blanca, F.X., Wilson, A., 1989. Dynamics of mitochondrial DNA evo-
lution in animals: ampliWcation and sequencing with conserved
primers. Proc. Natl. Acad. Sci. USA 86, 6196–6200.
Koehler, C.E., Richardson, P.R.K., 1990. Proteles cristatus. Mammal. Spe-
cies 363, 1–6.
KoepXi, K.-P., Wayne, R.K., 1998. Phylogenetic relationships of otters
(Carnivora: Mustelidae) based on mitochondrial cytochrome b
sequences. J. Zool. 246, 401–416.
Kruuk, H., 1972. The Spotted Hyena. Univ. of Chicago Press, Chicago, IL.
Kruuk, H., Sands, W.A., 1972. The aardwolf (Proteles cristatus Sparrman)
1783 as a predator of termites. E. Afr. Wildl. J. 10, 211–227.
Lake, J.A., 1994. Reconstructing evolutionary trees from DNA and pro-
tein sequences: paralinear Distances. Proc. Natl. Acad. Sci. USA 91,
Lieberman, D.E., 2000. Ontogeny, homology and phylogeny in the homi-
nid craniofacial skeleton: the problem of the browridge. In: O’Higgins,
P., Cohn, M.J. (Eds.), Development, Growth and Evolution. Academic
Press, San Diego, CA, pp. 85–122.
Lockhart, P.J., Steel, M.A., Hendy, M.D., Penny, D., 1994. Recovering evo-
lutionary distances under a more realistic model of sequence evolution.
Mol. Biol. Evol. 11, 605–612.
Lyons, L.A., Laughlin, T.F., Copeland, N.G., Jenkins, N.A., Womack,
J.E., O’Brien, S.J., 1997. Comparative anchor tagged sequences
(CATS) for integrative mapping of mammalian genomes. Nat.
Genet. 15, 47–56.
McKenna, M.C., Bell, S.K., 1997. ClassiWcation of Mammals Above the
Species Level. Columbia Univ. Press, New York, NY.
Mills, M.G.L., 1982. Hyaena brunnea. Mammalian Species 194, 1–5.
Mills, M.G.L., 1990. Kalahari Hyenas: comparative behavior and ecology
of two species. Unwin-Hymen, London.
Mills, M.G.L., 1999. The Brown Hyena. Africa Environ. Wildl. 7, 80–88.
O’Higgins, P., 2000. Quantitative approaches to the study of craniofacial
growth and evolution: advances in morphometric techniques. In:
O’Higgins, P., Cohn, M.J. (Eds.), Development, Growth and Evolution.
Academic Press, San Diego, CA, pp. 163–185.
Pilgrim, G.E., 1932. The fossil Carnivora of India. Paleont. Indica 18,
Posada, D., Crandall, K.A., 1998. MODELTEST: testing the model of
DNA substitution. Bioinformatics 14, 817–818.
Potts, R., Behrensmeyer, A.K., 1992. Late cenozoic terrestrial ecosystems.
In: Behrensmeyer, A.K., Damuth, J.D., DiMichele, W.A., Potts, R.,
Sues, H.-D., Wing, S.L. (Eds.), Terrestrial Ecosystems Through Time:
Evolutionary Paleoecology of Terrestrial Plants and Animals. Univ. of
Chicago Press, Chicago, IL, pp. 419–541.
Reidl, R.J., 1978. Order in Living Organisms. Wiley, New York, NY.
Richardson, P.R.K., 1987a. Food consumption and seasonal variation in
the diet of the aardwolf Proteles cristatus in southern Africa. Z. Säuge-
tierk. 52, 307–325.
Richardson, P.R.K., 1987b. Aardwolf: the most highly specialized myrme-
cophagous mammal? South Afr. J. Sci. 83, 405–410.
Richardson, P.R.K., Coetzee, E.M., 1988. Mate desertion in response to
female promiscuity in the socially monogamous aardwolf. S. Afr. J.
Zool. 23, 306–308.
Rieger, I., 1981. Hyaena hyaena. Mammalian Species 150, 1–5.
Rokas, A., Williams, B.L., King, N., Carroll, S.B., 2003. Genome-scale
approaches to resolving incongruence in molecular phylogenies.
Nature 425, 798–804.
Ronquist, F., Huelsenbeck, J.P., 2003. MrBayes 3: Bayesian phylogenetic
inference under mixed models. Bioinformatics 19, 1572–1574.
Rzhetsky, A., Nei, M., 1992. A simple method for estimating and testing
minimum-evolution trees. Mol. Biol. Evol. 9, 945–967.
Schlosser, M., 1890. Die AVen, Lemuren, Chiropteren, Insectivoren, Mar-
supialier, Creodonten und Carnivoren des europäischen Tertiärs. III.
Beiträge zur Paläontologie und Geologie Österreich-Ungarns und des
Orients 8, 1–107.
Simpson, G.G., 1953. The Major Features of Evolution. Simon and
Schuster, New York, NY.
Sorenson, M.D. 1998. TreeRot, version 2. Univ. of Michigan, Ann Arbor,
Springer, M.S., Murphy, W.J., Eizirik, E., O’Brien, S.J., 2003. Placental
mammal diversiWcation and the Cretaceous-Tertiary boundary. Proc.
Natl. Acad. Sci. USA 100, 1056–1061.
SwoVord, D.L., Olsen, G.J., Waddell, P.J., Hillis, D.M., 1996. Phylogenetic
inference. In: Hillis, D.M., Moritz, C., Mable, B.K. (Eds.), Molecular
Systematics. Sinauer Associates, Sunderland, MA, pp. 407–514.
SwoVord, D.L. 2002. PAUP*. Phylogenetic Analysis Using Parsimony
(* and Other Methods), Version 4.0. Sinauer Associates, Sunderland,
Teeling, E.C., Scally, M., Kao, D.J., Romagnoli, M.L., Springer, M.S..,
2000. Molecular evidence regarding the origin of echolocation and
Xight in bats. Nature 403, 188–192.
Thenius, E., 1966. Zur Stammegeschichte der Hyänen (Carnivora, Mam-
malia). Z. Saugetier. 31, 293–300.
Thorne, J.L., Kishino, H., Painter, I.S., 1998. Estimating the rate of evolu-
tion of the rate of molecular evolution. Mol. Biol. Evol. 15, 1647–1657.
Turner, A., 1987. New fossil carnivore remains from Plio-Pleistocene
deposits in the Sterkfontein Valley (Mammalia: Carnivora). Ann.
Transvaal Mus. 34, 203–226.
Turner, A., Antón, M., 2004. Evolving Eden: An Illustrated Guide to the
Evolution of the African Large-Mammal Fauna. Columbia Univ.
Press, New York, NY.
Van Valkenburgh, B., Wang, X., Damuth, J., 2004. Cope’s rule, hyper-
carnivory, and extinction in North American canids. Science 306,
Venta, P.J., Brouillette, J.A., Yuzbasiyan-Gurkan, V., Brewer, G.J., 1996.
Gene-speciWc universal mammalian sequence-tagged sites: application
to the canine genome. Biochem. Genet. 34, 321–341.
Wayne, R.K., Benveniste, R.E., Janczewski, D.N., O’Brien, S.J., 1989.
Molecular and biochemical evolution of the Carnivora. In: Gittleman,
J.L. (Ed.), Carnivore Behavior, Ecology and Evolution. Cornell Univ.
Press, Ithaca, NY, pp. 465–494.
Weesner, F.M., 1960. Evolution and biology of termites. Ann. Rev. Ento-
mol. 5, 153–170.
Werdelin, L., 1996a. Carnivoran ecomorphology: a phylogenetic perspec-
tive. In: Gittleman, J.L. (Ed.), Carnivore Behavior, Ecology and Evolu-
tion, vol. 2.. Cornell Univ. Press, Ithaca, NY, pp. 582–624.
Werdelin, L., 1996b. Community-wide character displacement in Miocene
hyaenas. Lethaia 29, 97–106.
Werdelin, L., Solounias, N., 1991. The Hyaenidae: Taxonomy, systematics
and evolution. Fossils and Strata 30, 1–104.
Werdelin, L., Solounias, N., 1996. The evolutionary history of hyaenas in
Europe and western Asia during the early Miocene. In: Bernor, R.L.,
Fahlbusch, V., Mittmann, H.-W. (Eds.), The Evolution of Western
Euarasian Neogene Mammal Faunas. Columbia Univ. Press, New
York, NY, pp. 290–306.
Werdelin, L., 2003. Mio-Pliocene Carnivora from Lothagam, Kenya. In:
Leakey, M.G., Harris, J.M. (Eds.), Lothagam: The Dawn of Humanity
in Eastern Africa. Columbia Univ. Press, New York, NY, pp. 261–678.
Wozencraft, W.C., 1993. Carnivora. In: Wilson, D.E., Reeder, D.M. (Eds.),
Mammal Species of the World: A Taxonomic and Geographic Refer-
ence. Smithsonian Institution Press, Washington, DC, pp. 279–348.
Wurster, D.H., Benirschke, K., 1968. Comparative cytogenetic studies in
the Order Carnivora. Chromosoma 24, 336–382.
Wyss, A.R., Flynn, J.J., 1993. A phylogenetic analysis and deWnition of the
Carnivora. In: Szalay, F.S., Novacek, M.J., McKenna, M.C. (Eds.),
Mammal Phylogeny: Placcentals. Springer-Verlag, Newyork, Ny, pp.
Yoder, A.D., Burns, M.M., Zehr, S., Delefosse, T., Veron, G., Goodman,
S.M., Flynn, J.J., 2003. Single origin of Malagasy Carnivora from an
African ancestor. Nature 421, 734–737.
Yu, L., Li, Q.-w., Ryder, O.A., Zhang, Y.-p., 2004. Phylogenetic relation-
ships within the mammalian order Carnivora indicated by sequences of
two nuclear genes. Mol. Phylogenet. Evol. 33, 694–705.

Supplementary resources (100)