Syst. Biol. 52(4):439–459, 2003
c Society of Systematic Biologists
Copyright °
ISSN: 1063-5157 print / 1076-836X online
DOI: 10.1080/10635150390218132
Phylogenetics of Advanced Snakes (Caenophidia) Based on Four Mitochondrial Genes
CHRISTOPHER M. R. K ELLY,1,2 NIGEL P. B ARKER,3 AND M ARTIN H. VILLET 1
1
Department of Zoology and Entomology, Rhodes University, Grahamstown, South Africa
Department of Zoology, Oxford University, Oxford OX1 3PS, England; E-mail: [email protected]
3
Molecular Ecology and Systematics Facility, Department of Botany, Rhodes University, Grahamstown, South Africa
2
Abstract.—Phylogenetic relationships among advanced snakes (Acrochordus + Colubroidea = Caenophidia) and the position
of the genus Acrochordus relative to colubroid taxa are contentious. These concerns were investigated by phylogenetic analysis
of fragments from four mitochondrial genes representing 62 caenophidian genera and 5 noncaenophidian taxa. Four methods
of phylogeny reconstruction were applied: matrix representation with parsimony (MRP) supertree consensus, maximum
parsimony, maximum likelihood, and Bayesian analysis. Because of incomplete sampling, extensive missing data were
inherent in this study. Analyses of individual genes retrieved roughly the same clades, but branching order varied greatly
between gene trees, and nodal support was poor. Trees generated from combined data sets using maximum parsimony,
maximum likelihood, and Bayesian analysis had medium to low nodal support but were largely congruent with each
other and with MRP supertrees. Conclusions about caenophidian relationships were based on these combined analyses.
The Xenoderminae, Viperidae, Pareatinae, Psammophiinae, Pseudoxyrophiinae, Homalopsinae, Natricinae, Xenodontinae,
and Colubrinae (redefined) emerged as monophyletic, whereas Lamprophiinae, Atractaspididae, and Elapidae were not
in one or more topologies. A clade comprising Acrochordus and Xenoderminae branched closest to the root, and when
Acrochordus was assessed in relation to a colubroid subsample and all five noncaenophidians, it remained associated with the
Colubroidea. Thus, Acrochordus + Xenoderminae appears to be the sister group to the Colubroidea, and Xenoderminae should
be excluded from Colubroidea. Within Colubroidea, Viperidae was the most basal clade. Other relationships appearing in
all final topologies were (1) a clade comprising Psammophiinae, Lamprophiinae, Atractaspididae, Pseudoxyrophiinae, and
Elapidae, within which the latter four taxa formed a subclade, and (2) a clade comprising Colubrinae, Natricinae, and
Xenodontinae, within which the latter two taxa formed a subclade. Pareatinae and Homalopsinae were the most unstable
clades. [Advanced snakes; Bayesian analysis; Caenophidia; Colubroidea; data combination; maximum likelihood; maximum
parsimony; mitochondrial DNA; supertrees.]
Snakes have conventionally been divided into two snakes.” Dowling and Duellman (1978) and Dowling
major monophyletic groups, the Scolecophidia (blind et al. (1983, 1996) placed it within the Colubroidea,
snakes) and the Alethinophidia (true snakes) (Rieppel, associated with Homalopsinae or Natricinae. Recent
1979; Cadle, 1987; McDowell, 1987). Monophyly of both mtDNA-based studies (e.g., Kraus and Brown, 1998;
blind snakes and true snakes is supported by a variety Gravlund, 2001) have not resolved this issue, but
of osteological, soft anatomy, and ecological characters Slowinski and Lawson (2002), using both mtDNA and
(Lee and Scanlon, 2002). Various divisions of true snakes nuclear gene sequences, resolved Achrochordus as basal
have been attempted. Hoffstetter (1955) split them into to the Colubroidea with moderate to good bootstrap
the Henophidia (primitive snakes) and the Caenophidia support.
(advanced snakes), an arrangement followed by Romer
Colubroidea is substantially the most speciose snake
(1956), Underwood (1967), and Rieppel (1979). How- lineage, incorporating approximately 80% (2,500 species)
ever, there is little support for monophyly of primitive of recognized extant snake diversity (Dowling and
snakes (McDowell, 1987; Rieppel, 1988). In this article, Duellman, 1978). The proper subdivision of Colubroidea
taxa known to be nonmonophyletic will be distinguished and the relationships of its constituent clades are subby enclosure in single quotation marks. Currently, three jects of continued debate and conflict (Cadle, 1987;
successively less inclusive clades are recognized within McDowell, 1987). For instance, Underwood (1967) recogtrue snakes: the Macrostomata, the Caenophidia, and nized six colubroid families and 15 subfamilies based on
the Colubroidea. Lee and Scanlon (2002) listed the oste- a wide variety of morphological characters, Dowling and
ological, soft anatomy, and ecological synapomorphies Duellman (1978) split the group into three families and
of these clades. Immunological evidence (Cadle, 1988) 13 subfamilies using hemipenial morphology, maxillary
and mitochondrial DNA (mtDNA) sequence data (Heise dentition, and vertebral characters, and McDowell (1987)
et al., 1995; Kraus and Brown, 1998; Gravlund, 2001) identified four families and 18 subfamilies on the basis of
generally lend additional support to the monophyly of morphological and immunological data. Currently, four
these three subgroups of true snakes. Noncaenophidian families are formally recognized: Atractaspididae (one
macrostomatans are referred to here as “basal macros- or two subfamilies, 3% of colubroid diversity), Viperidae
tomatans.”
(two to four subfamilies, 10%), Elapidae (two subfamiAs currently defined, Caenophidia (advanced snakes) lies, 12%), and “Colubridae” (approximately 10 subfamicomprises a sister relationship between Colubroidea and lies, 75%) (Greene, 1997; Pough et al., 2001). In this study,
the genus Acrochordus (Acrochordidae) (Rieppel, 1988; we adopted a composite arrangement of the Colubroidea
Greene, 1997; Pough et al., 2001, Lee and Scanlon, 2002). (Table 1) adapted from Smith et al. (1977), Heymans
However, the placement of Acrochordus is contentious. (1982), Groombridge (1986), McDowell (1987), Cadle
Hoffstetter and Gayrard (1965), Underwood, (1967), and (1987, 1988), Underwood and Kochva (1993), Dowling
Heise et al. (1995) placed Acrochordus in the “primitive et al. (1996), Greene (1997), Slowinski et al. (1997), and
439
440
VOL. 52
SYSTEMATIC BIOLOGY
TABLE 1. Taxa used in this study, their gene representation, and sources of sequences (GenBank accession number or author). Sequences
are full length unless otherwise specified. The classification scheme employed is explained in the text. Multiple representatives of single genera
were assembled into supraspecific terminals at the generic level, and Typhlops lumbricalis and Rhamphotyphlops braminus were assembled into a
supraspecific terminal at the family level. Numbers in parentheses after taxon names indicate the data set (1, 2, or both) for each taxon.
Sources of sequences
Classification
Scolecophidia
Typhlopidae
Typhlops lumbricalis (2)
Rhamphotyphlops braminus (2)
Alethinophidia
Macrostomata
Tropidophiidae
Tropidophis haitianus (2)
Tropidophis wrighti (2)
Loxocemidae
Loxocemus bicolour (2)
Pythonidae
Python reticulatus (2)
Boidae
Boa constrictor (1, 2)
Caenophidia
Acrochordidae
Acrochordus granulatus (1, 2)
Acrochordus javanicus (1, 2)
Colubroidea
Atractaspididae
Aparallactinae
Aparallactus guentheria (1)
Aparallactus werneri (1)
Atractaspininae
Atractaspis bibronii (1)
Atractaspis corpulenta (1)
Atractaspis micropholis (1)
“Colubridae”
Calamariinae
Oreocalamus hanitschi (1)
Colubrinae
Boiga cynodon (1)
Boiga dendrophila (1)
Coluber constrictor (1)
Crotaphopeltis hotamboeia (1)
Dasypeltis medici (1)
Dasypeltis scabraa (1)
Dendrelaphis calligastra (1)
Dendrelaphis pictus (1)
Dinodon semicarinatum (1, 2)
Elaphe flavolineata (1)
Elaphe obsoleta (1)
Gastropyxis smaragdina (1)
Lycodon capucinus (1)
Lycodon laoensis (1)
Unplaced
Grayia ornata (1)
Homalopsinae
Cerberus rhynchops (1)
Enhydris enhydris (1)
Enhydris plumbea (1)
Natricinae
Macropisthodon rudis (1)
Natrix natrix (1)
Nerodia fasciata (1)
Nerodia rhombifera (1)
Nerodia taxispilota (1)
Thamnophis butleri (1)
Thamnophis elegans (1)
Pareatinae
Aplopeltura boa (1)
Pareas nuchalis (1)
Psammophiinae
Hemirhagerrhis nototaenia (1)
Hemirhagerrhis viperinusa (1)
Cytb
ND4
12Sr RNA
16S rRNA
Z46444 (0.32 kb)
Z46475
Z46445 (0.38 kb)
Z46476
U69845
Z46456 (0.38 kb)
Z46486
U69860
Z46448 (0.38 kb)
Z46478
Z46470
Z46495
Z46472
Z46502
Z46597
Z46499
Z46468
Z46525
L01765 (0.37 kb)
Gravlund, 2001 (0.28 kb)
Gravlund, 2001 (0.28 kb)
L01770
Gravlund, 2001
Gravlund, 2001 (0.36 kb)
U69865 (0.78 kb)
U69869 (0.76 kb)
U69746 (0.80 kb)
Forstner et al., 1995
AF217841 (0.80 kb)
U49296
AY235730 (0.77 kb)
U49315
U49314
AF039261 (0.65 kb)
U49306
AF217818 (0.80 kb)
U49303
U49300
AY235729 (0.79 kb)
AF139569
NC001945
U49304
NC001945
U49301
AF283643 (0.80 kb)
NC001945
NC001945
Z46469 (0.36 kb)
AF158435 (0.29 kb)
Z46493
AF158504 (0.34 kb)
Z46455
Z46485
AF158434 (0.29 kb)
AF158503 (0.34 kb)
Z46458
Z46492
AF158461 (0.28 kb)
AF158530 (0.34 kb)
Z46452
Z46481
Gravlund, 2001 (0.28 kb)
Gravlund, 2001
U49317
U49327
U49328
U49326
AF172392 (0.30 kb)
L33334 (0.30 kb)
U49322
U49324
AF305671 (0.31 kb)
U49312
U49311
AY235725 (0.75 kb)
(Continued on next page)
2003
441
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
TABLE 1. Continued
Sources of sequences
Classification
Malpolon monspessulanusa (1)
Psammophis condenarus (1)
Psammophis mossambicusa (1)
Psammophylax variabilisa (1)
Rhamphiophis acutusa (1)
Rhamphiophis oxyrhynchus (1)
Xenoderminae
Achalinus rufescens (1)
Xenodermus javanicus (1)
Xenodontinae
Alsophis portoricensis (1)
Dipsas catesbyi (1)
Farancia abacura (1, 2)
Helicops angulatus (1)
Helicops pictiventris (1)
Heterodon nasicus (1)
Heterodon simus (1)
Hydrodynastes bicinctus (1)
Liophis reginae (1)
Xenodon neuwiedi (1)
Xenodon servus (1)
Lamprophiinae
Lamprophis fuliginosus (1)
Lamprophis inornatusa (1)
Mehelya capensis (1)
Mehelya nyassaea (1)
Pseudoxyrophiinae
Leioheterodon madagascariensis
Madagascarophis colubrine (1)
Unplaced
Homoroselaps lacteus (1)
Elapidae
Elapinae
Aspidelaps scutatus (1)
Boulengerina annulata (1)
Boulengerina sp. (1)
Bungarus fasciatus (1)
Dendroaspis polylepis (1)
Dendroaspis sp. (1)
Elapsoidea nigra (1)
Maticora bivirgata (1)
Micruroides euryxanthus (1)
Micrurus diastema (1, 2)
Micrurus fulvius (1, 2)
Naja kaouthia (1)
Naja naja (1)
Ophiophagus hannah (1)
Hydrophiinae
Hydrophis semperi (1)
Laticauda colubrina (1)
Viperidae
Azemiophinae
Azemiops feae (1)
Causinae
Causus defilippi (1)
Causus rhombeatus (1)
Crotalinae
Agkistrodon contortrix (1)
Calloselasma rhodostoma (1)
Crotalus horridus (1, 2)
Crotalus viridis (1, 2)
Trimeresurus albolabris (1)
Trimeresurus tokarensis (1)
Viperinae
Bitis arietansa (1)
Vipera ammodytes (1)
Vipera ursinii (1)
a
Cytb
ND4
AY235721
AY235722
AY235724 (0.47 kb)
AY235723 (0.38 kb)
12Sr RNA
16S rRNA
Gravlund, 2001 (0.28 kb)
Z46450
Gravlund, 2001 (0.33 kb)
Z46479
Gravlund, 2001 (0.28 kb)
Gravlund, 2001
Z46443
Z46738
U49319
U49320
U49308
U69832 (0.80 kb)
U49307
U49310
Forstner et al., 1995
AF158448 (0.29 kb)
Z46459
Z46467
AF158408 (0.29 kb)
AF158517 (0.34 kb)
Z46496
Z46491
AF158478 (0.34 kb)
AF158428 (0.29 kb)
AF158494 (0.32 kb)
AF158430 (0.29 kb)
AF158433 (0.29 kb)
AF158479 (0.34 kb)
AF158501 (0.34 kb)
AF217840 (0.80 kb)
AF236814 (0.30 kb)
Z46449
Z46474
Z46457
Z46489
AY235727 (0.77 kb)
Gravlund, 2001 (0.28 kb)
Gravlund, 2001
AY235726
U49318
U49313
AF217833 (0.80 kb)
AF217828 (0.80 kb)
AF217829 (0.80 kb)
U96790
U96792
Z46466
Z46501
U96795
U96804
U96800
Z46433
Z46454
Z46483
Z46484
AF217842 (0.80 kb)
Z46453
Z46451
Z46482
Z46480
AF217822 (0.80 kb)
AF217834 (0.80 kb)
U96798
U96799
AF217830 (0.80 kb)
AF217832 (0.80 kb)
U49297
AF217820 (0.80 kb)
AF217812 (0.80 kb)
AF217823 (0.80 kb)
AF217839 (0.80 kb)
AF217835 (0.80 kb)
AF182539 (0.17 kb)
U49298
U41865
AF057187
AF057234
AF057186
AF057233
Z46473
AF057190
Z46460
Z46524
AF057237
Z46497
U41866
AF039268 (0.64 kb)
AF171918 (0.66 kb)
U41868
U41878
AF147860 (0.68 kb)
AF171909 (0.66 kb)
U41882
U41890
AY235728
AF182552 (0.17 kb)
Taxa for which Cytb data were generated in this study.
Z46446
Z46739
Z46471
L01768 (0.37 kb)
Z46498
L01769
442
SYSTEMATIC BIOLOGY
Pough et al. (2001). The authorship of names for the
higher categories used here is given in these papers. Unless otherwise specified, subsequent reference to higher
taxa will follow Table 1.
Underwood and Kochva (1993) found the Atractaspididae to be monophyletic based on a variety of morphological characters. This family has been poorly represented in most molecular studies, but its monophyly
is tentatively supported by mtDNA data (Kraus and
Brown, 1998). However, on the basis of serum albumin divergence estimates, Cadle (1994) questioned the
monophyly of Atractaspididae. Viperidae appears to
be monophyletic based on morphology (Liem et al.,
1971) and mtDNA (Heise et al., 1995; Kraus and Brown,
1998; Gravlund, 2001), as does Elapidae (morphology:
McDowell, 1968; McCarthy, 1985; mtDNA: Kraus and
Brown, 1998; Slowinski and Keogh, 2000). “Colubridae,”
by far the largest and most diverse colubroid family,
is demonstrably nonmonophyletic (Heise et al., 1995;
Kraus and Brown, 1998; Gravlund, 2001).
Two basic issues have plagued attempts at elucidation of relationships within advanced snakes. The first
is the difficulty of identifying monophyletic subdivisions. Such groupings have traditionally been based on
morphological data (Bogert, 1940; Underwood, 1967;
Bourgeois, 1968; Dowling and Duellman, 1978). However, because snakes lack limbs all of the morphological variation associated with such appendages is absent
in snakes, resulting in a restricted set of potential morphological characters (Dowling et al., 1996). The great
diversity of advanced snakes has led to difficulty in
assessment of these characters across the entire group
(Kraus and Brown, 1998), and high degrees of homoplasy are associated with many such characters (Kraus
and Brown, 1998). Because of these factors, few wellsupported clades have been recognized, leading to conflict in proposed classifications. Recent mtDNA-based
studies on select groups of advanced snakes have contributed substantially to identification of monophyletic
groups (Slowinski and Keogh, 2000; Vidal et al., 2000),
but much more such research is required.
Second, the generation of valid hypotheses of relationship also is problematic. Recent molecular studies
(Vidal et al., 2000; Lenk et al., 2001) have contributed
to resolution of relationships within putative monophyletic subsets of advanced snakes. However, all existing phylogenies of the group as a whole differ to
some extent in their topology, and little consensus has
been reached on the placement of even well-supported
clades. Many researchers agree that Acrochordidae is
sister to the Colubroidea (Cadle, 1988; Rieppel, 1988,
Lee and Scanlon, 2002), within which Viperidae is basal
(Cadle, 1988; Rieppel, 1988; Heise et al., 1995). However, McDowell (1979, 1987) and Heise et al. (1995)
considered Acrochordidae to be far distant from the
Colubroidea, and the basal position of the Viperidae
within Colubroidea was disputed by Kraus and Brown
(1998) and Gravlund (2001). Thus, even these general arrangements are questionable.
VOL. 52
The aims of this study were (1) to identify monophyletic higher taxa within advanced snakes, (2) to assess the support for currently recognized higher taxa
(Table 1), (3) to generate robust hypotheses of relationship within (where possible) and between the monophyletic taxa identified, and (4) to investigate the position of Acrochordus in relation to colubroids and “basal
macrostomatans.” We applied four different methods
of phylogenetic analysis: supertree consensus (Gordon,
1986; Sanderson et al., 1998), maximum parsimony
(MP), maximum likelihood (ML), and Bayesian inference
(Mau, 1996; Rannala and Yang, 1996; Mau and Newton,
1997; Yang and Rannala, 1997; Larget and Simon, 1999;
Mau et al., 1999). These analyses were based on DNA
sequence fragments from four (aims 1–3) or three (aim 4)
mitochondrial genes (investigated singly and in various
combinations). We included representatives from most
proposed families and subfamilies of advanced snakes,
from three of the “basal macrostomatan” families, and
from the blind snakes. Many morphological and molecular assessments of relationships among advanced snakes
have been hindered by limited character inclusion and by
taxon sampling problems. The effects of taxon sampling
have been reviewed by Lecointre et al. (1993) and, in the
current context, by Kraus and Brown (1998). Few new sequence data were generated in this study, but our analyses brought together for the first time published data
from many different reports. We have thus included significantly more taxa and characters than have been used
in previous studies.
M ATERIALS AND M ETHODS
Character and Taxon Sampling
It is well known that phylogenetic reconstruction benefits from increased character inclusion, provided all
characters share a common evolutionary history. This is
generally because of the increase in signal-to-homoplasy
ratio that occurs as more data are added, with a consequent increase in the likelihood of resolving hidden phylogenetic signal (e.g., de Queiroz et al., 1995; Nixon and
Carpenter, 1996). In the molecular paradigm, it is advantageous to include data from multiple genes because the
evolutionary rates of different genes often vary (Crozier,
1990). Each gene may thus help to resolve relationships
at different hierarchical levels. In this study, we included
as many data from as many genes as were available.
Mitochondrial DNA phylogenies of many vertebrate
groups are strongly influenced by taxon sampling
(Philippe and Douzery, 1994; Milinkovitch et al., 1996;
Kraus and Brown, 1998; Matthee et al., 2001). We thus
favored maximum taxon inclusion despite the consequent introduction of large quantities of missing data.
Elimination of incomplete taxa a priori removes potentially important character interactions from the analysis
(Novacek, 1992; Kearney, 1998) and falsely assumes that
all incomplete taxa will be equally detrimental to analyses (Kearney, 1998). Anderson (2001) noted that because
the stability of a taxon on a tree depends on topology,
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
elimination of the contribution of a taxon to topology
prevents it from determining a stable position. Consequently, the inclusion of incomplete taxa is required to
maintain a well-resolved tree.
Our complete data set comprised 98 species (91 advanced snakes, 5 “basal macrostomatans,” and 2 blind
snakes) represented by varying combinations of four mitochondrial gene fragments: cytochrome b (Cytb; maximum 804 bp), 12S ribosomal RNA (rRNA; maximum
421 bp and 31 gap characters), 16S rRNA (maximum
381 bp and 7 gap characters), and NAD dehydrogenase
subunit 4 (ND4; maximum 694 bp) (Table 1). We included representatives from four of the (approximately)
seven “basal macrostomatan” families and all families
and subfamilies of advanced snakes proposed to date
except Calliopheinae (Elapidae) and Pseudoxenodontinae (“Colubridae”) (McDowell, 1987), which contain one
and two genera, respectively.
Terminal Taxa
Whenever possible, we applied the exemplar approach (Yeates, 1995; Bininda-Emonds et al., 1998), in
which species are used as terminal entities. However,
to maximize gene representation per terminal in analyses involving gene combinations (and thus minimize
missing data), it was often necessary to combine data
from multiple species to form supraspecific terminals
(Bininda-Emonds et al., 1998) at the generic level and,
in one case (Typhlopidae), at the family level. The greatest potential fault with this approach (the assumption of
monophyly of supraspecific taxa; Yeates, 1995; Kron and
Judd, 1997; Bininda-Emonds et al., 1998) was not considered a problem in this study.
Splitting of the Data Set
The first three aims of this article involved assessment
of relationships within advanced snakes rather than relationships between advanced snakes and “basal macrostomatans.” To achieve these aims we created data set 1
with the following characteristics: (1) inclusion of data
from all four genes, (2) inclusion of all advanced snake
taxa, and (3) exclusion of all blind snakes and all “basal
macrostomatan” taxa except Boa constrictor, which was
used as the outgroup. We excluded these two groups of
snakes (1) because ND4 sequence data was lacking for
all “basal macrostomatans” except Boa and we wanted to
have an outgroup with no missing data and (2) because
we wanted to reduce the chances of long-branch attraction as far as possible and some “basal macrostomatans”
and blind snakes are only distantly related to advanced
snakes. In data set 1 with one outgroup and 62 ingroup
terminals, 27 taxa are supraspecific, 22 lack Cytb characters, 13 lack 12S characters, 19 lack 16S characters, 29
lack ND4 characters, and 16 (including the outgroup) are
represented by all four genes.
Our final aim was to investigate the position of Acrochordus in relation to colubroids and “basal macrostomatans.” We created data set 2 with the following characteristics: (1) exclusion of ND4 data because these data
443
were lacking for all “basal macrostomatans,” (2) inclusion of only a representative colubroid sample because
analysis of data set 1 showed that Acrochordus was not
nested within the Colubroidea, (3) inclusion of all “basal
macrostomatans,” and (4) inclusion of blind snakes as
the outgroup. In data set 2 with one outgroup and nine
ingroup terminals, five taxa are supraspecific, and all are
represented by the Cytb, 12S, and 16S genes.
Molecular Techniques
Mitochondrial Cytb gene sequences from 10 species
(Appendix) were generated for the present study. The remaining data were obtained from GenBank or from the
literature (Table 1). Total genomic DNA was extracted
from liver, muscle, or blood samples using a modified
Chelex 100 protocol (Walsh et al., 1991) and was used as
a template in polymerase chain reaction (PCR) amplifications (Saiki et al., 1986). For each species, a fragment of
the Cytb gene a maximum of 804 bp long was amplified
with primers L14724 (50 -CGA AGC TTG ATA TGA AAA
ACC ATC GTT G-30 ) (Irwin et al., 1991) and H15547b
(50 -AAT AGG AAG TAT CAT TCT GGT TTA ATG-30 )
(Campbell, 1997, unpubl.). Reactions were conducted in
50- or 100-µl volumes typically with 35 cycles of 45 sec
at 95◦ C (denaturation), 45 sec at 52◦ C (annealing), and
3 min at 72◦ C (extension), followed by a final 10-min extension at 72◦ C. PCR products were purified using the
QIAquick PCR Purification Kit (Quiagen). Sequencing
was carried out with an ABI 3100 Genetic Analyzer, using the BigDye Terminator Cycle Sequencing Kit (Applied Biosystems) according to the manufacturer’s protocols. Both strands were sequenced for most taxa. Primers
used for sequencing included the amplification primers
and two internal primers: SNKP2 (50 -TGA (GA)GA CAA
ATA TCA TTC-30 ; this study) and SNKP3 (50 -GAA TGA
TAT TTG TC(C/T) TCA-30 ; this study). New sequences
were edited and contiguous sequences were assembled using the program Sequencher 3.0 (GeneCodes
Corporation).
Sequence Alignment
All sequences were aligned manually using the software DAPSA 4.0 (Harley, 1996). The arrangement of
Vidal et al. (2000), based on secondary structure, was
used as a guide for alignment of 12S sequences.
Sequence Characteristics
Unless otherwise specified, sequence characteristics
were investigated using data set 1 only, and analyses
were performed using PAUP∗ 4.0b8 (Swofford, 2000).
When using methods based on explicit models of DNA
evolution, we employed ModelTest 3.04 (Posada and
Crandall, 1998) to identify the model of DNA substitution that best fit the data. Investigations of mutational
saturation were conducted for each gene by plotting pdistance versus corrected estimates (based on the model
identified by ModelTest) of proportional sequence divergence. In all cases transitions (TIs) and transversions
444
VOL. 52
SYSTEMATIC BIOLOGY
(TVs) were considered separately, and protein-coding
genes were further divided into first, second, and third
codon positions. Phylogenetic signal in each single gene
and in data sets 1 and 2 was investigated using the g1
statistic for skewness of tree length distributions (Hillis
and Huelsenbeck, 1992), estimated from 100,000 random
trees, and the published critical values. However, a significant result may occur if phylogenetic signal is confined to limited portions of the topology. This possibility was tested in data set 1 by deleting all clades with
bootstrap support of ≥50% and then recalculating the
g1 statistic. The compliance of mtDNA with the molecular clock hypothesis was tested by applying the likelihood ratio test (Felsenstein, 1988) to a subset of data set 1
comprising all four genes and including only those taxa
represented by all genes.
Data Combination
Most phylogenetic analyses were conducted on combinations of the data from all four genes included in this
study. Data combination is justified here for the following
reasons. First, the entire animal mitochondrial genome
is inherited as a single unit, so different mitochondrial
genes are not independent estimates of organismal phylogeny (Moore, 1995; Page, 2000). Consequently, all four
genes should share the same evolutionary history and
should retrieve the same phylogenetic relationships. Second, no strong conflict in signal was revealed by separate
analyses of individual genes, thus providing no reason
to suspect that these genes differ in the phylogenetic relationships they retrieve.
Phylogenetic Analyses
Data set 1 was analyzed using supertree consensus
methods, MP, ML, and Bayesian inference. Analysis of
data set 2 involved all of the above methods except supertree consensus.
Subset Analysis and Supertree Consensus (Data Set 1)
Data set 1 included many missing data entries. In phylogeny reconstruction, missing characters can result in
almost random placement of affected taxa (Nixon and
Wheeler, 1992), leading to a proliferation of best trees
and a consequent reduction in the resolution of consensus trees (Huelsenbeck, 1991; Novacek, 1992). To avoid
this problem, we divided data set 1 into 15 subsets: each
single gene and all combinations of two, three, and four
genes. Each subset included only those terminal taxa represented by all of the subset’s constituent genes, thus
avoiding inclusion of missing data. All data subsets were
analyzed using MP (see Maximum parsimony section
below). The data subset combining all four genes was
analyzed using both MP and ML (see Maximum likelihood section below). The matrix representation with
parsimony (MRP) supertree consensus method (Baum,
1992; Ragan, 1992), implemented in the program RadCon (Thorley and Page, 2000), was then used to combine
the resultant most-parsimonious trees (MPTs). Supertree
consensus methods facilitate fusion of trees with differ-
ent but overlapping terminal sets. Nodes of the input
trees were weighted according to their bootstrap support (Ronquist, 1996), with 70–79% bootstrap support
weighted 3, 80–89% weighted 6, and 90–100% weighted
9 (base weight = 1). Following the suggestion of BinindaEmonds and Bryant (1998), Camin–Sokal parsimony
(prohibiting reversals) and Fitch parsimony (allowing
reversals) were assessed as alternatives. The most appropriate analysis was identified by tree comparison using the Shimodaira–Hasegawa (SH) test (see Hypothesis
Testing section below) (Goldman et al., 2000).
MP, ML, and Bayesian Analyses
Maximum parsimony.—Gaps in alignments were coded
using the complex indel coding system (Simmons and
Ochoterena, 2000), which allows all available information to be utilized. This method treats gaps that have
different 50 and/or 30 termini as presence/absence characters and is based on the application of six rules.
Three weighting schemes for nucleotide data were investigated for MP analysis of data sets 1 and 2 and of the
subsets of data set 1. Scheme 1 involved equal weighting,
in which all nucleotide positions were weighted equally
regardless of type or position. Scheme 2 involved differential position weighting, in which saturated substitution subsets were downweighted according to their
TI:TV ratio, as estimated by ML under the HKY85 + I + 0
(Hasegawa et al., 1985) model of nucleotide substitution.
Scheme 3 involved following the suggestion of Lenk et al.
(2000), in which the optimal DNA substitution model
identified by ModelTest was transferred to an MP framework by construction of an asymmetric step matrix based
on the inverse values of the instantaneous substitution
rate matrix (Q matrix) of the model, rounded to the nearest integer, and adjusted to satisfy the triangle inequality
(Sankoff, 1975). Each of these three schemes was applied
with (denoted “a”) and without subsequent weighting
based on the successive approximation method (Farris,
1969), using the rescaled consistency index of equally
weighted characters as a measure of homoplasy. Trees
from each scheme were compared using the SH test, and
the most suitable weighting scheme was thus identified.
Where there was no significant difference between the
trees from different schemes (P ≥ 0.05), the scheme producing the fewest MPTs was chosen.
All MP searches were performed with 1,000 random sequence-addition replicates and tree bisection–
reconnection (TBR) branch swapping using PAUP∗ 4.0b8
and including only informative characters. We used
heuristic searches in most cases, but data set 2 and
some of the subsets of data set 1 were small enough to
make exhaustive searches possible. Nonparametric bootstrap analyses (Felsenstein, 1985) with 1,000 pseudoreplicates and 10 random sequence-addition replicates were
conducted, applying character weights where relevant.
All characters were sampled with equal probability.
The stability of taxa across sets of bootstrap trees was
assessed using the leaf stability (LS) method (Thorley
and Wilkinson, 1999) implemented in RadCon.
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
Maximum likelihood.—ML analysis was restricted to
data sets 1 and 2 and to the subset of data set 1 comprising all four genes and only those taxa represented by
all genes. The general time reversible model with some
sites assumed to be invariable and with variable sites assumed to follow a discrete gamma distribution (GTR +
I + 0; Rodrı́guez et al., 1990; Yang, 1994) was selected by
ModelTest as the optimal model of nucleotide substitution for all datasets. PAUP∗ 4.0b8 was used to implement
heuristic ML searches with TBR branch swapping and
with starting trees generated by neighbor joining (NJ;
Saitou and Nei, 1987). Parameters under the GTR + I + 0
model were set according to the corresponding ModelTest outputs. Computational restrictions precluded bootstrap analysis within the ML paradigm.
Bayesian inference.—Bayesian phylogenetic analyses of
data sets 1 and 2 were conducted with MrBayes 3.0b3
(Huelsenbeck and Ronquist, 2001), using the GTR + I
+ 0 model of nucleotide substitution. Model parameter values were treated as unknown and were estimated
in each analysis. Random starting trees were used, and
analyses were run for 1 million generations, sampling
the Markov chains every 100 generations.
We used three methods to ensure that our analyses were not trapped in local optima: (1) each analysis was run twice, starting from different random trees,
and the log-likelihood (lnL) values at stationarity were
compared for convergence (Huelsenbeck and Bollback,
2001); (2) Metropolis-coupled Markov chain Monte Carlo
was applied, allowing a more extensive exploration
of parameter space (Huelsenbeck and Ronquist, 2001),
with one cold and three incrementally heated Markov
chains, applying the MrBayes default heating values; and
(3) topologies and posterior clade probabilities from different runs were compared for congruence (Huelsenbeck
and Imennov, 2002).
We plotted the log-likelihood scores of sample points
against generation time and considered stationarity
of Markov chains to have been reached when the
log-likelihood values reached a stable equilibrium
(Huelsenbeck and Ronquist, 2001). All sample points
prior to stationarity were discarded as burn-in values,
and remaining points were used to generate a 50% majority rule consensus tree, with each clade posterior probability value represented by the proportion of sample
points recovering that specific clade.
Hypothesis Testing
It was often necessary to statistically compare alternative phylogenetic hypotheses, e.g., to facilitate choice
of the best weighting scheme in MP analyses. For such
comparisons, we used the SH test with 10,000 bootstrap pseudoreplicates and the resampling estimated
log-likelihood approximation. This test was applied in
place of the Kishino–Hasegawa (KH) test (Kishino and
Hasegawa, 1989) because the KH test is invalid when
topologies being compared are specified a posteriori
(Shimodaira and Hasegawa, 1999). One potential problem with the SH test is that it may appear conservative
445
(Shimodaira, 2002), but in the current context this was
not considered a serious issue.
R ESULTS
Alignment and Sequence Characteristics
Sequences generated in this study were deposited in
GenBank under accession numbers AY235721–AY235730
(see Table 1). Alignment of protein coding sequences
(Cytb and ND4) was unambiguous. No gaps were introduced into the ND4 alignment, and the Cytb alignment
required a single triplet gap near the beginning of the
Cytb gene of Lamprophis inornatus. For the 16S sequences,
alignment was unambiguous except in two highly variable regions corresponding to loops (sites 2145–2170 and
2183–2189 in the mitochondrial genome of Dinodon semicarinatum), which were excluded from the alignment.
The 12S and 16S alignments required the introduction
of 31 and 7 gaps, respectively (data set 1), and 26 and 3
gaps, respectively (data set 2). Data set 1 comprised 2,338
characters (with approximately 35% missing data), 1,303
variable characters, and 1,071 parsimony-informative
characters. Data set 2 comprised 1,623 characters (with
insignificant missing data), 780 variable characters, and
503 parsimony-informative characters.
Sequence properties were determined based on data
set 1 only. Character information and nucleotide composition of sequences are summarized in Table 2. These
observations reflect patterns obtained by other authors for mitochondrial genes (Campbell, 1997, unpubl.;
Slowinski and Keogh, 2000; Lenk et al., 2001).
Table 3 includes minimum, maximum, and mean corrected (based on the scheme identified by ModelTest)
sequence divergences between advanced snake taxa after exclusion of terminals with extensive missing data.
These values suggest that Cytb and ND4 genes evolve at
comparable rates, as do 12S and 16S, but that the mitochondrial RNA genes evolve much more slowly than do
the protein-coding genes. This finding is in agreement
with other published findings (Crozier, 1990).
Plots of uncorrected versus corrected sequence divergence (not shown) were linear for both substitution types
in the 12S and 16S sequences and at first and second coding positions in the Cytb and ND4 sequences. In both of
the protein genes, third-position TVs deviated from linearity but maintained a positive gradient. However, for
third-position TIs the relationship had a final gradient
of effectively zero, suggesting that this class of substitution is saturated. The TI:TV ratios for third positions
were estimated at 64:1 (Cytb) and 16:1 (ND4) (data set 1)
and at 36:1 (Cytb) (data set 2), so third-position TIs were
downweighted in the relevant MP weighting schemes
by factors of 64, 16, and 36, respectively.
Significant phylogenetic signal (P < 0.01) was present
in each single gene (g1 = −0.213 for Cytb, −0.376 for
ND4, −0.430 for 12S, −0.440 for 16S) in data set 1 both
before (63 terminals, g1 = −0.375) and after (36 terminals,
g1 = −0.302) removal of clades with bootstrap support
>50% and in data set 2 (10 terminals, g1 = −0.600). The
likelihood-ratio test strongly rejected the hypothesis that
446
VOL. 52
SYSTEMATIC BIOLOGY
TABLE 2. Nucleotide composition and character information for each gene fragment, based on data set 1. For third positions in protein-coding
genes, character information is given for transitions (TI) and transversions (TV) separately. There was no significant change in numbers of variable
and parsimony-informative characters when calculations were repeated without the outgroup. Tests of base homogeneity across taxa were based
on variable sites only.
Base frequencies
Gene
Cytb
ND4
12S
16S
a
Codon position
or data type
1
2
3
1
2
3
Bases
Gaps
Bases
Gaps
No. (%) characters
A
C
G
T
Probability of base
homogeneity across taxaa
0.33
0.22
0.38
0.36
0.16
0.46
0.38
0.23
0.27
0.38
0.27
0.31
0.32
0.24
0.18
0.12
0.05
0.18
0.12
0.05
0.18
0.26
0.39
0.19
0.20
0.41
0.17
0.20
1.000
1.000
0.000
1.000
1.000
0.000
0.999
0.35
0.21
0.19
0.25
1.000
Total
Variable
Parsimony informative
268
268
87 TI, 181 TV
231
231
58 TI, 174 TV
421
31
381
7
128 (48%)
74 (28%)
86 TI, 177 TV (98%)
131 (57%)
68 (29%)
58 TI, 172 TV (99%)
229 (54%)
31 (100%)
142 (37%)
7 (100%)
98
46
83 TI, 172 TV
110
47
52 TI, 169 TV
173
21
99
1
Based on a chi-square test.
mtDNA conforms to molecular clock assumptions (n =
16, χ 2 = 36.12, P < 0.001).
Phylogenetic Analyses of Data Set 1
Subset analysis.—Relevant information on each data
subset is given in Table 4. Optimal MP analysis of
individual genes produced a single MPT in each case
(not shown). These analyses retrieved roughly comparable clades, but branching order varied considerably
between gene trees. However, none of the differences between these trees were strongly supported by bootstrap
values, and one of the most striking features of all four
phylogenies was extremely poor bootstrap support,
especially for backbone nodes. This result parallels
trends in bootstrap or decay index support obtained in
similar mtDNA sequence–based studies involving both
snake phylogenetics (Kraus and Brown, 1998; Slowinski
and Keogh, 2000; Vidal et al., 2000; Gravlund, 2001)
and the phylogenetics of other groups (e.g., artiodactyl
mammals: Matthee et al., 2001). However, there was a
marked increase in bootstrap support (Table 4) and tree
congruence in analyses involving gene combinations.
Predictably, for analyses of data subsets the proportion
of nodes with bootstrap support of ≥50% was significantly correlated positively with the number of variable
and parsimony-informative characters and negatively
with the number of terminal taxa (P < 0.05). Also, it
is significant to note that the MPT from analysis of the
Cytb + ND4 data subset (the only genes representing
TABLE 3. Sequence divergences between ingroup taxa for each gene
(after exclusion of terminals with extensive missing data), corrected
using the results of ModelTest analyses.
Corrected pairwise distances
Gene
Maximum
Minimum
Mean
Cytb
ND4
12S
16S
2.27
2.95
0.70
0.39
0.33
0.16
0.05
0.04
1.26
1.26
0.31
0.15
Aparallactus) contained a monophyletic Atractaspididae,
with 99% bootstrap support.
Figure 1 illustrates the single trees from MP and ML
analysis of the data subset comprising all four genes and
including only those taxa represented by all genes. Almost all nodes in the MP tree have support values significantly >50%. In the retrieval of major clades and the
relationships between these clades, the ML tree is entirely
consistent with the optimal topologies from all analyses
of data set 1 (Figs. 2–5), and the MP tree differs only
in the placement of Atractaspididae, despite extensive
taxon sampling problems.
Supertree consensus.—In the MRP supertree consensus analysis, Fitch parsimony performed better than
Camin–Sokal parsimony, although this difference was
not significant (P > 0.05). Figure 2 illustrates the tree
with the highest log-likelihood from the set of 13 mostparsimonious supertrees retrieved.
Maximum parsimony.—Weighting Scheme 1a (equally
weighted data with successive approximation) was identified as the optimal scheme for MP analysis of data set 1.
This scheme was superior to all other weighting schemes,
significantly so for all except Scheme 1 (without successive approximation) (P < 0.05). Figure 3 illustrates
the tree with the highest log-likelihood from the set of
11 MPTs retrieved. Eighty percent of nodes have ≥50%
bootstrap support, and 58% of nodes have ≥70% bootstrap support. LS values for the MP bootstrap tree profile
are given in Table 5. Pareatines (Aplopeltura boa and Pareas nuchalis) and homalopsines (Enhydris and Cerberus
rhynchops) were considerably less stable than other taxa.
Maximum likelihood.—ML analysis of data set 1 produced a single tree (Fig. 4).
Bayesian inference.—The two independent Bayesian
analyses of data set 1 converged on similar log-likelihood
scores and reached stationarity at no later than 200,000
generations (plots not shown). The first 2,000 sample
points were thus discarded as burn-in, leaving 8,000 samples for construction of a 50% majority rule consensus
tree. The consensus topologies of both runs were almost
identical, and the posterior probability values supporting congruent nodes were highly correlated (not shown),
2003
447
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
TABLE 4. Subsets of data set 1 (single genes and all combinations of two, three, and four genes) for supertree consensus. Weighting schemes
are equal weighting (1), differential codon position weighting in which saturated substitution subsets were downweighted according to their
TI:TV ratio (2), and differential weighting based on the output of ModelTest (3). Those schemes incorporating further weighting by successive
approximation are marked with an “a.” Those marked with an asterisk were significantly better than competing schemes (α = 0.05), as assessed
using the SH test. Where differences between weighting schemes were not significant, the scheme producing the fewest MPTs was chosen.
Bootstrap analyses applied character weights where relevant.
No. characters
Data subset
Cytb
ND4
12S
16S
Cytb, ND4
Cytb, 12S
Cytb, 16S
ND4, 12S
ND4, 16S
12S, 16S
Cytb, ND4, 12S
Cytb, ND4, 16S
Cytb, 12S, 16S
ND4, 12S, 16S
All genes
No. terminals
Variable
Parsimony
informative
41
34
50
44
18
38
31
22
23
43
16
16
31
22
16
465
429
260
149
797
707
574
613
524
398
981
890
808
736
1086
399
378
194
100
636
570
468
479
418
288
742
681
631
559
809
Best weighting scheme
Scheme
No. MPTs
3a
2
1
1
1a∗
1a
1a
1a
1a∗
1
1a
1a
1a
1a∗
1a
1
2
102
64
1
1
1
1
1
2
1
1
1
1
1
providing further evidence that the analyses converged.
Figure 5 illustrates the 50% majority rule consensus tree
from one of the runs. Forty-six percent of nodes have
posterior probability values ≥95%.
Phylogenetic relationships.—These results are based on
Figures 2–5 (MRP, MP, ML, and Bayesian analyses of
data set 1). All analyses supported monophyly of advanced snakes with respect to Boa constrictor, and clades
1–9 (Figs. 2–5) were always retrieved (ignoring Homoroselaps, the placement of which appeared to be relatively
random).
In clade 2 (Viperidae), there was much support for
monophyly of the subfamily Crotalinae and of the Old
World (Trimeresurus and Calloselasma) and New World
(Crotalus and Agkistrodon) crotalinae taxa. Viperinae was
not monophyletic.
In clade 6, MP and ML analyses provided moderate
support for monophyly of the Pseudoxyrophiinae, but
Lamprophiinae and Atractaspididae were only monophyletic in the MRP tree. The MP, ML, and Bayesian
topologies placed Aparallactinae in the Lamprophiinae,
as sister to Mehelya, and Atractaspininae was variously
associated with the Elapidae, Lamprophiinae, and
Homoroselaps. Elapidae was monophyletic in the MP
and ML topologies. Two elapid clades were present
throughout, Maticora + Micrurus + Micruroides and Naja
+ Boulengerina, and a Dendroaspis + Hydrophis clade
was retrieved in the MRP, MP, and ML analyses. The
elapid subfamilies Elapinae and Hydrophiinae were not
monophyletic.
In clade 7 (Natricinae excluding Macropisthodon), the
African and European genus Natrix was basal to a New
World (Nerodia and Thamnophis) clade (Thamnophiini;
Rossman and Eberle, 1977). The Southeast Asian species
Macropisthodon rudis was consistently placed within
Colubrinae.
Chosen scheme (if
different from best scheme)
Scheme
No. MPTs
2a
1a
1a
1
1
1
1a
1
Proportion of nodes
with bootstrap
support ≥50%
0.47
0.61
0.36
0.32
0.93
0.71
0.75
0.89
0.90
0.60
0.92
0.85
0.86
0.95
1.00
In clade 8 (Xenodontinae), there was much support
for monophyly of the Xenodontini (e.g., Jenner, 1981, unpubl.), here represented by Xenodon, and Liophis. In the
MP and ML topologies, the South American taxa (Hydrodynastes, Helicops, Alsophis, Xenodon and Liophis) and
Central American taxa (Dipsas) formed a clade distinct
from Farancia and Heterodon (North American representatives). In the ML and Bayesian topologies, the South,
Central, and North American taxa separated into distinct
clades. Homoroselaps emerged within the Xenodontinae
in the MRP and MP topologies.
In clade 9, the Colubrinae was monophyletic (excluding Dendrelaphis, with Gastropyxis in the Bayesian
topology), and Macropisthodon, Dendrelaphis, and Calamariinae (Oreocalamus) were always closely associated.
Within Colubrinae, all trees contained the clades Lycodon
+ Dinodon and Crotaphopeltis (Dasypeltis, Boiga), and in
the MP, ML, and Bayesian topologies these two clades
were collectively monophyletic. Grayia usually emerged
in a relatively basal position in clade 9.
There was widespread agreement between the four
analyses regarding relationships between these major
clades, and the basal portion of the tree was identical for
all reconstructions: Boa((Acrochordidae, Xenoderminae)
(Viperidae (remaining Colubroidea))). In the MRP, ML,
and Bayesian topologies, Pareatinae was the next to
branch off. Clade 4 and clade 6 were always sister groups,
clade 7 and clade 8 were sister groups in the MRP, MP,
and ML topologies, and clades 7–9 were always collectively monophyletic.
Phylogenetic Analyses of Data Set 2
Maximum parsimony.—Weighting Scheme 1 (equally
weighted data without successive approximation) was
identified as the optimal scheme for MP analysis of
448
SYSTEMATIC BIOLOGY
VOL. 52
FIGURE 1. The single best phylograms from MP and ML analysis of all four genes combined, including only those terminal taxa with no
missing genes. (a) MP, applying the optimal weighting scheme (Scheme 1a): consistency index = 0.60, retention index = 0.68. (b) ML, applying
the GTR + I + 0 model of DNA substitution with parameter values estimated by ModelTest: −ln L = 16585.7. Numbers are bootstrap (MP)
support values for the associated nodes. Supraspecific terminals are labeled with generic names only. Labels next to the tree are subfamily and
family names (Table 1).
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
449
FIGURE 2. The most likely of 13 best phylograms (generated using Fitch parsimony) from the MRP supertree consensus method, using as
input the MPTs resulting from optimal weighted analyses of all subsets of data set 1: consistency index (CI) = 0.64, retention index (RI) = 0.86.
Nodes on the input trees were weighted according to their bootstrap support. When assessed in relation to the original unweighted sequence data,
CI = 0.25, RI = 0.31. Open circles mark nodes that collapse in the strict consensus, and asterisks mark nodes shared with the MP hypothesis (Fig. 3).
Branch lengths are proportional to the number of unambiguous changes in the original sequence data. Supraspecific terminals are labeled with
generic names only. Labels next to the tree are subfamily names (Table 1), and the associated numbers correspond to clade numbers in the text.
data set 2. Figure 6a illustrates the single MPT retrieved.
Seventy-one percent of nodes have ≥50% bootstrap support, and 29% of nodes have ≥70% bootstrap support.
Maximum likelihood.—ML analysis of data set 2 produced a single tree (Fig. 6b).
Bayesian inference.—The two independent Bayesian
analyses of data set 2 converged on similar loglikelihood scores and reached stationarity at no later
than 50,000 generations (plots not shown). The first
500 sample points were thus discarded as burn-in,
450
SYSTEMATIC BIOLOGY
VOL. 52
FIGURE 3. The most likely of 11 best phylograms from MP analysis of data set 1, applying the optimal weighting scheme (Scheme 1a):
consistency index = 0.36, retention index = 0.54. Open circles mark nodes that collapse in the strict consensus. Asterisks mark nodes shared
with the ML hypothesis (Fig. 4) (ignoring Homoroselaps). Numbers on the tree are bootstrap support values for the associated nodes, and branch
lengths are proportional to the number of unambiguous changes. Supraspecific terminals are labeled with generic names only. Labels next to the
tree are subfamily names (Table 1), and the associated numbers correspond to clade numbers in the text.
leaving 9,500 samples for construction of a 50% majority rule consensus tree. The consensus topologies
of both runs were identical, and the posterior probability values supporting congruent nodes were highly
correlated (not shown). Figure 6c illustrates the 50%
majority rule consensus tree from one of the runs. Sixty-
seven percent of nodes have posterior probability values
≥95%.
Phylogenetic relationships.—In their placement of Acrochordus as sister to the Colubroidea, the three topologies
were identical. Support for this relationship was poor in
the MP reconstruction but high in the Bayesian analysis.
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
TABLE 5. The stability of terminal taxa (LS values) across the MP
bootstrap tree profile of data set 1 (1,635 rooted trees) returned by the
program RadCon.
Genus (leaf)
Boa
Acrochordus
Achalinus
Xenodermus
Crotalus
Agkistrodon
Trimeresurus
Calloselasma
Vipera
Azemiops
Bitis
Causus
Micrurus
Micruroides
Naja
Boulengerina
Maticora
Dendroaspis
Aspidelaps
Bungarus
Dasypeltis
Boiga
Dinodon
Lycodon
Crotaphopeltis
Ophiophagus
Coluber
Elapsoidea
Elaphe
Hydrophis
Laticauda
Aparallactus
Mehelya
Helicops
Atractaspis
Farancia
Alsophis
Heterodon
Madagascarophis
Gastropyxis
Dipsas
Leioheterodon
Grayia
Lamprophis
Dendrelaphis
Nerodia
Thamnophis
Macropisthodon
Natrix
Hydrodynastes
Xenodon
Liophis
Homoroselaps
Psammophylax
Hemirhagerrhis
Rhamphiophis
Malpolon
Psammophis
Oreocalamus
Enhydris
Cerberus
Aplopeltura
Pareas
LS (difference)
1.000
0.997
0.996
0.996
0.982
0.982
0.982
0.982
0.982
0.982
0.982
0.982
0.884
0.884
0.884
0.884
0.884
0.882
0.882
0.881
0.881
0.881
0.881
0.881
0.880
0.880
0.879
0.879
0.879
0.878
0.877
0.877
0.876
0.874
0.874
0.874
0.874
0.873
0.871
0.870
0.869
0.867
0.864
0.858
0.858
0.854
0.854
0.853
0.852
0.850
0.848
0.848
0.817
0.810
0.810
0.810
0.810
0.809
0.801
0.717
0.717
0.634
0.634
451
D ISCUSSION
Utility of mtDNA
In this study, individual genes were often incongruent
in their retrieval of phylogenetic relationships (although
none of this incongruence was strongly supported by
bootstrap values), and support overall was extremely
poor, especially for internal nodes. In contrast, different gene combinations retrieved congruent relationships
and nodal support was generally high. The significant
positive and negative correlations of bootstrap support
with the number of variable and parsimony-informative
characters and the number of terminal taxa, respectively,
suggests that support (and by implication accuracy) is
largely dependent on the character: terminal taxon ratio (a conclusion reached by Bremer et al., 1999, for the
Rubiaceae). This is likely to apply especially to data with
high levels of homoplasy, so we infer that all of the genes
used here are substantially homoplastic at the hierarchical levels investigated.
The pattern of nodal support was similar in many of
the trees; support was moderate or high up to the level of
the Viperidae. Beyond this similarity, support was generally restricted to terminal and subterminal nodes, with
little support for many backbone nodes. This pattern was
consistent for both the rapidly evolving (Ctyb and ND4)
and the more slowly evolving (12S and 16S) genes, suggesting that the relatively rapid overall rate of mtDNA
evolution is not a factor hampering retrieval of higher
relationships. Some researchers have suggested that the
Miocene radiation of Colubroidea was extremely rapid
(e.g., Cadle, 1994; Greene, 1997). An explosive colubroid
radiation above the level of the Viperidae would account
in part for the patterns and high levels of homoplasy
observed here. Any rapid radiation will result in short
branch lengths linking radiated taxa if the rate of evolution of characters used in tree construction is not speeded
by the radiation. There is no reason to expect the rate of
evolution of mtDNA to increase during rapid radiation.
Additionally, if the radiation were relatively ancient (as
postulated here), there would be loss of phylogenetic signal along the short branches due to multiple hits (substitutional saturation), resulting in homoplasy.
Single mitochondrial genes are not optimal for investigations of higher relationships among advanced snakes,
but the effectiveness of mtDNA increases with increasing character:taxon ratios. Despite extensive taxon sampling problems, the trees in Figure 1 are almost entirely
consistent with all of the optimal topologies from analyses of data set 1. Thus, the failure of previous molecular
studies to accurately retrieve higher relationships among
advanced snakes may have been due to insufficient character inclusion rather than taxon sampling problems.
Caenophidian Phylogenetic Relationships
Many of the groups for which we have found evidence of monophyly are well supported by morphological characters and have been retrieved in other molecular
452
SYSTEMATIC BIOLOGY
VOL. 52
FIGURE 4. The single ML phylogram from analysis of data set 1 under the GTR + I + 0 model of DNA substitution, with parameter values
estimated by ModelTest: −ln L = 34275.4. Asterisks mark nodes shared with the MP hypothesis (Fig. 3) (ignoring Homoroselaps). Branch lengths
are proportional to the number of substitutions per site. Supraspecific terminals are labeled with generic names only. Labels next to the tree are
subfamily names (Table 1), and the associated numbers correspond to clade numbers in the text.
studies. The Viperidae, Xenoderminae, Pareatinae,
Psammophiinae, Pseudoxyrophiinae, Homalopsinae,
Natricinae, Xenodontinae, and Colubrinae (with redefinition) fall into this category (e.g., Rossman and Eberle,
1977; McDowell, 1987; Cadle, 1988, 1994; Heise et al.,
1995; Dowling et al., 1996; Kraus and Brown, 1998; Vidal
et al., 2000; Gravlund, 2001). Our results serve to reinforce these clades.
Clade 1.—Consequent upon the presence of a wellsupported basal (Acrochordus + Xenoderminae) clade
in all of our topologies, we suggest removal of
Xenoderminae from the Colubroidea. Kraus and Brown
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
453
FIGURE 5. The 50% majority rule consensus tree from Bayesian analysis of data set 1 under the GTR + I + 0 model of DNA substitution:
mean −ln L = 34339.7. Asterisks mark nodes shared with the ML hypothesis (Fig. 4) (ignoring Homoroselaps). Numbers on the tree are posterior
probability values for the associated nodes, and branch lengths are averaged over all input trees. Supraspecific terminals are labeled with generic
names only. Labels next to the tree are subfamily names (Table 1), and the associated numbers correspond to clade numbers in the text.
(1998) also reached this conclusion (using the ND4 data
included here), and Dowling (1989, unpubl.) placed
Xenoderminae with Homalopsinae in his superfamily
“Acrochordoidea.”
Clade 2.—Within Viperidae, Azemiops feae has been
considered either basal (Liem et al., 1971) or related to
Asian representatives of the Crotalinae (e.g., Parkinson,
1999). However, all analyses of data set 1 strongly support placement of Azemiops in a clade with Viperinae
and Causinae. Causinae was suggested by Groombridge
(1984) to be sister to the remaining viperids, and Dowling
et al. (1996) found Causinae to lie outside a Viperinae +
Crotalinae clade. However, Causinae and Viperinae
are united by several synapomorphic morphological
454
SYSTEMATIC BIOLOGY
VOL. 52
FIGURE 6. The best phylograms from MP, ML, and Bayesian analyses of data set 2. (a) MP (single tree), applying the optimal weighting
scheme (scheme 1, without subsequent successive approximation): length = 1,514, consistency index = 0.53, retention index = 0.32. (b) ML
(single tree), applying the optimal model (GTR + I + 0) and parameters identified by ModelTest: −ln L = 8969.2. (c) Bayesian analysis (50%
majority rule consensus tree), applying the GTR + I + 0 model of DNA substitution: mean −ln L = 8979.4. Numbers are bootstrap (MP) and
posterior probability (Bayesian analysis) values for the associated nodes. Supraspecific terminals are labeled with generic or family names only.
Labels next to the tree are higher taxon names (Table 1).
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
characters (Underwood, 1999), and our results suggest
a link between these taxa, although not the sister relationship assumed by Lenk et al. (2001). Despite much
support for our groupings, we consider our taxon sampling within the Viperidae to be insufficient to clarify
subfamily relationships. The basal position of Viperidae
within the Colubroidea, as obtained here, is supported
by the immunological results of Cadle (1988). The preferred topology of Kraus and Brown (1998, generated
from ND4 sequences using TVs only) places Viperidae
in a terminal position within Colubroidea, whereas their
NJ tree based on all variable sites is consistent with the results of the present study. Our saturation analysis of ND4
sequences indicates that only third-position TIs are substantially saturated. We thus consider Kraus and Brown’s
exclusion of other transitions to be in error.
Clade 3.—Pareatinae is a morphologically well-defined
group (McDowell, 1987), but its affinities are obscure.
In our MRP, ML, and Bayesian topologies, Pareatinae
branched immediately after the Viperidae and was sister
to the remaining Colubroidea. However, support for this
relationship was generally poor and pareatines were relatively unstable in the MP bootstrap tree profile (Table 5).
Similarly, Slowinski and Lawson (2002) were unable to
confidently resolve the relationships of this group other
than to ascertain that it tends to occupy a basal position in the Colubroidea, sometimes associating with the
Viperidae. We can thus make no conclusions about placement of the Pareatinae.
Clade 4.—Psammophiinae is also morphologically well
defined (Bogert, 1940; Underwood, 1967; Bourgeois,
1968; Dowling and Duellman, 1978); however, the relationship of Psammophiinae to other taxa is disputed. On
the basis of retinal characters, hemipenial morphology
(Underwood, 1967), and minimal molecular evidence
(George and Dessauer, 1970; Minton and Salanitro, 1972),
Rasmussen (1985) included them in the Colubrinae.
However, our results support no such association, and
Cadle (1994) found significantly greater albumin divergence among psammophiine genera than among
colubrines, suggesting a much earlier divergence of
the former group compared with the latter (assuming
a molecular clock). McDowell (1987) included psammophiines in his Boaedontinae (roughly comparable to
the Lamprophiinae of this study), but this position is also
incongruent with the present results. All of our analyses
placed Psammophiinae around the middle of the tree, as
sister to clade 6.
Clade 5.—Homalopsinae has been associated with the
Acrochordidae and the Natricinae (e.g., Dowling and
Duellman, 1978), with Dasypeltis (now considered a colubrine) (Underwood, 1967; Rasmussen, 1985), with the
Boiginae (now part of the Colubrinae) (Underwood,
1967), and with Viperidae (Rasmussen, 1985; Gravlund,
2001). The present results support none of these relationships but offer no solid alternatives. The retrieval of Homoroselaps as sister to the Homalopsinae in our Bayesian
analysis is almost certainly erroneous. The affinities of
Homalopsinae remain obscure.
455
Clade 6.—Several key issues relate to clade 6. The
first issue is the monophyly of the Atractaspididae.
The Atractaspididae as here constituted was originally
proposed by Bourgeois (1968) on the basis of skull
morphology (her subfamily Aparallactinae, within the
“Colubridae”). Heymans (1975) identified the same
grouping on the basis of jaw musculature and erected
the family Atractaspididae (Heymans, 1982), within
which Atractaspis was assigned to the subfamily Atractaspidinae (our Atractaspininae). The remaining genera were placed in the subfamily Aparallactinae. Since
then, the monophyly of Atractaspididae (with respect
to Atractaspis) has been debated, and the placement
of Aparallactus within Atractaspididae has been questioned. Underwood and Kochva (1993) identified two
unique morphological synapomorphies for the Atractaspididae: wedge-shaped dorsal process of the premaxilla and lateral position of the rectal glands. In our
MRP topology, the Atractaspididae was monophyletic
and was sister to the remainder of clade 6. The MPT
from analysis of the Cytb + ND4 data subset (the only
genes representing Aparallactus) also contained a monophyletic Atractaspididae, with 99% bootstrap support
(in the absence of lamprophiine representatives). However the MP, ML, and Bayesian trees placed Aparallactus
sister to Mehelya (Lamprophiinae). This placement is in
agreement with the findings of McDowell (1987), who
placed Aparallactus in his Boaedontinae (roughly comparable to our Lamprophiinae), and with the immunological data of Cadle (1994), which revealed no close
relationship between Aparallactus and Atractaspis. We
consider our sample of atractaspid taxa insufficient to
resolve this issue, and solid conclusions on the status of
the Atractaspididae await inclusion of additional related
taxa.
The second issue is the position of Homoroselaps. Prior
to McDowell’s (1968) study, this southern African genus
was considered an elapid on the basis of its proteroglyphous maxilla. McDowell (1968) argued for its transfer to
the Aparallactinae, but Underwood and Kochva (1993)
returned it to the Elapidae based on a variety of morphological characters. However, Underwood and Kochva
stated that the majority of evidence indicated a primary
dichotomy between Homoroselaps and the Elapidae. In a
comprehensive mtCytb study of the Elapidae, Slowinski
and Keogh (2000) found Homoroselaps to lie outside a
clade containing the remaining elapids and found no
association betweenHomoroselaps and their xenodontine
representatives (Farancia and Heterodon). Our placements
of Homoroselaps within the Xenodontinae (MRP and MP
topologies) and as sister to the Homalopsinae (Bayesian
topology) are almost certainly erroneous, but our ML tree
suggests that the affinities of Homoroselaps may lie with
the Atractaspididae.
The third issue is the monophyly of and relationships
between the Lamprophiinae and Pseudoxyrophiinae.
As constituted here, Lamprophiinae + Pseudoxyrophiinae would be roughly equivalent to the Boaedontinae
of McDowell (1987) and would also be similar to the
456
SYSTEMATIC BIOLOGY
former Lycodontinae of Underwood (1967), Dowling
and Duellman (1978), Dowling et al. (1983), and Cadle
(1987). The type genus of Lycodontinae (Lycodon) is a colubrine, so this name has been relegated to the synonymy
of Colubrinae (Cadle, 1994; Kraus and Brown, 1998).
Lamprophiinae and Pseudoxyrophiinae contain most
noncolubrine, nonnatricine, nonpsammophiine African
snakes (Lamprophiinae) and Madagascan snakes (Pseudoxyrophiinae) belonging to the “Colubridae.” We know
of no unique morphological synapomorphies supporting either of these groupings. McDowell (1987) tentatively included Mehelya in his Boaedontinae but noted
that the salivary glands, hemipenes, and vertebrae were
atypical. He also tentatively included Aparallactus on the
basis of skull musculature. Our results certainly suggest that Mehelya is a lamprophiine and that it is closely
related to Aparallactus. Based on immunological comparisons, Cadle (1994) found evidence of monophyly
of a sample of three African lamprophiines (Lamprophis,
Mehelya, and Lycodonomorphus) but retrieved no association between this clade and Leioheterodon (a pseudoxyrophiine). Our analyses recovered no especially close
relationship between the Lamprophiinae and Pseudoxyrophiinae, and the MRP and Bayesian analyses placed
Pseudoxyrophiinae within the Elapidae. Thus, inclusion
of additional taxa from the Lamprophiinae, Pseudoxyrophiinae, Atractaspididae, and Elapidae will be necessary to resolve the status and composition of the Lamprophiinae and Pseudoxyrophiinae.
The fourth issue is the composition of and relationships within the Elapidae. Our failure to demonstrate
monophyly of the Elapinae and Hydrophiinae may result from taxon sampling problems rather than from real
disagreements. We consider our sample of elapid taxa
too small for solid conclusions about intraelapid relationships to be made. However, the strongly supported association of Dendroaspis and Hydrophis is interesting in its
possible implications and requires further investigation.
At the least it suggests that the African mambas (Dendroaspis) are not typical elapines, an inference supported
by hemipenial morphology (Branch, pers. comm.).
The fifth issue is the age of the elapid, atractaspid,
and lamprophiine radiations. Each of these three taxa
has been considered by various authors to be an
ancient colubroid lineage: Elapidae (McDowell, 1987;
Gravlund, 2001), Atractaspididae (Rasmussen, 1985;
McDowell, 1987; Underwood and Kochva, 1993; Cadle,
1994; Gravlund, 2001), and Lamprophiinae (Cadle, 1994;
Gravlund, 2001). All of our analyses, however, suggest
that these taxa group together in a relatively terminal position on the colubroid tree and are not extremely early
radiations.
Clade 7.—Natricinae is characterized mainly by
hemipenial synapomorphies; the entire hemipenis is
spinose, usually with large asymmetrically placed spines
at the base. Calyces are usually absent, and the sulcus spermaticus is forked and centripetal or is simple
(McDowell, 1987). Most studies have found the Natricinae to be monophyletic, with the New World taxa
(Thamnophiini) and Old World taxa grouping separately
VOL. 52
(Rossman and Eberle, 1977; Dowling et al., 1996; Kraus
and Brown, 1998; Fig. 6). All of our topologies contained
a monophyletic core natricine group but placed the one
Asian taxon (Macropisthodon rudis) with the Colubrinae.
However, in a suboptimal ML tree (not significantly different from the best topology) Macropisthodon emerged
as sister to the remaining Natricinae. The Natricinae are
usually found to be closest to the Colubrinae. Dowling
et al. (1996) and Cadle (1994) retrieved this relationship
from protein comparisons and immunological data, respectively. However, our analyses suggest that Natricinae is most closely allied to the Xenodontinae. This result was also obtained by Dowling et al. (1983), although
the natricine–colubrine link was retrieved as an equally
plausible alternative. However, because Macropisthodon
grouped with the colubrines in our best trees, addition of
further Asian taxa to our analysis may shift the Natricinae towards a sister relationship with the Colubrinae.
Clade 8.—Xenodontinae is an extremely large (>600
species) and heterogeneous New World assemblage
(Greene, 1997) characterized mainly by a forked sulcus spermaticus or, if the sulcus is single, a unicapitate hemipenis (Jenner, 1981, unpubl.). Based on albumin immunological data, Cadle (1984, 1985) recognized
two major lineages: the South American xenodontines
and the Central American xenodontines. However, he
found no clear evidence of relationship between the
two, and the North American representatives showed
no affinities with either group. However, using mtDNA
12S and 16S gene fragments and a comprehensive sample of taxa, Vidal et al. (2000) found Xenodontinae to
be monophyletic and divided into North, Central, and
South American clades, with the latter two being collectively monophyletic. Within the South American group,
the tribes Xenodontini, Hydropsini, and Pseudoboini
were monophyletic. Most xenodontines possess derived
type IIIb retinas, whereas pseudoboine retinas are of the
ancestral type I, which suggests that retinal composition
is an extremely labile character (possibly strongly correlated with lifestyle) and is thus of little use in snake
phylogenetics. Our results supported monophyly of the
Xenodontini and of Xenodontinae as a whole. The MP,
ML, and Bayesian hypotheses agreed that the South
and Central, American forms are collectively monophyletic with respect to North American taxa, and in
the ML and Bayesian topologies the South, Central, and
North American taxa separated into distinct clades. The
present results and those of Vidal et al. (2000) suggest
that xenodontines are derived colubroids. This finding
is in discord with claims by numerous other authors that
the Xenodontinae represents a cluster of extremely old
New World colubroids (e.g., Cadle, 1984, 1985; Dowling
et al., 1983; Dowling et al., 1996). However, Kraus and
Brown (1998) suggest some inconsistencies associated
with these studies that may explain the observed discrepancies. Our retrieval of Homoroselaps within the Xenodontinae is almost certainly erroneous.
Clade 9.—Colubrinae is one of the most often retrieved
clades in phylogenetic analyses of snakes. It is generally
diagnosed by the apparently synapomorphic presence of
2003
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
asymmetric hemipenes, although not all colubrine taxa
exhibit this condition (e.g., McDowell, 1987; Kraus and
Brown, 1998). Most studies agree that Colubrinae occupies a terminal position in the Colubroidea (e.g., Cadle,
1988, 1994; Heise et al., 1995; Dowling et al., 1996), a
result also obtained by us. We agree with McDowell
(1987), Cadle (1994), and Kraus and Brown (1998) that
Lycodon (and probably other members of Dowling and
Duellman’s, 1978, tribe Lycodontini) belongs in the Colubrinae. Further, our analyses consistently retrieved a
close and well-supported association between Dinodon
(also part of the tribe Lycodontini) and Lycodon, a result
mirrored by Vidal et al. (2000). We thus suggest that a
tribe Lycodontini be recognized within the Colubrinae,
comprising (until more data are brought to bear on the
issue) the taxa included in the Lycodontini of Dowling
and Duellman (1978). Another grouping retrieved in all
of our analyses of data set 1 was Crotaphopeltis(Dasypeltis,
Boiga). These taxa (excluding Dasypeltis) have been included in the subfamily Boiginae of Bourgeois (1968) and
the tribe Boigini of Dowling and Duellman (1978) and
were found by Gravlund (2001) to form a clade. Dasypeltis
possesses a suite of highly specialized adaptations linked
to its exclusive diet of birds’ eggs, and its placement on
the basis of morphological characters has been problematic. Dowling and Duellman (1978) recognized its colubrine affinities and placed it in their tribe Colubrini
of the Colubrinae. However, Dasypeltis has often been
assigned its own family (Dasypeltinae; Bourgeois,
1968) and has even been associated with the Homalopsinae (Underwood, 1967; Rasmussen, 1985). Our
results and those of Gravlund (2001, data included here)
clearly support inclusion of Dasypeltis in the Colubrinae,
and we suggest recognition of Dowling and Duellman’s
tribe Boigini, expanded to include Dasypeltis. Our MP,
ML, and Bayesian trees showed a sister relationship between the Lycodontini and Boigini of this study, a result
also obtained by Heise et al. (1995), Vidal et al. (2000), and
Gravlund (2001). Based on only two taxa, Dendrelaphis
(Asian) and Gastropyxis (African), we find no support for
the tribe Philothamnini (Dowling and Duellman, 1978).
The choice of Oreocalamus as our sole representative
of the Calamariinae was unfortunate because the placement of this genus is uncertain (Coborn, 1991). McDowell
(1987) considered Calamariinae to be a morphologically well-defined group of unknown affinities. Our results consistently placed it in close association with the
Colubrinae, normally grouped with Dendrelaphis and
Macropisthodon. Oreocalamus and Dendrelaphis are both restricted to East Asia, so their proposed relationship may
be worth further pursuit (Kraus and Brown, 1998). Colubrine affinities of the Calamariinae were also suggested
by Dowling and Duellman (1978), who recognized the
taxon as a tribe of their Colubrinae.
The enigmatic genus Grayia has previously been associated with the Colubrinae based on immunological data
(Cadle, 1994). This is surprising given the large differences in hemipenial morphology between the two taxa:
Grayia has a symmetrical hemipenis with a forked sulcus spermaticus, whereas Colubrinae is characterized
457
by an asymmetric organ. Our data are insufficient to
resolve the exact relationship between Grayia and the
Colubrinae, but in all our trees Grayia occupied a relatively basal position. We found no association of Grayia
with the Natricinae (McDowell, 1987; Cadle, 1994) or the
Xenodontinae (McDowell, 1987).
Position of Acrochordus
Acrochordidae and the Colubroidea share a variety
of morphological synapomorphies, but multiple morphological synapomorphies of the Colubroidea act to
separate these two taxa (Lee and Scanlon, 2002), and
Schwaner and Dessauer (1982) found substantial differences in the transferrins of these two groups. Consequently, Acrochordidae and Colubroidea have generally
been considered sister taxa (Rieppel, 1988; Greene, 1997;
Pough et al., 2001; Lee and Scanlon, 2002), a relationship
also retrieved by Slowinski and Lawson (2002). Our analyses strongly support a sister relationship between Acrochordus + Xenoderminae and the Colubroidea, and based
on Figure 6 we reject claims by other authors (McDowell,
1987; Heise et al., 1995) that Acrochordidae may occupy
a position closer to the root of the Alethinophidia. In this
study, we found no association between Acrochordidae
and either Natricinae or Homalopsinae (contra Dowling
and Duellman, 1978; Dowling et al., 1983, 1996).
ACKNOWLEDGMENTS
We thank Bill Branch, Don Broadley, and Ulrich Joger for provision
of tissue samples. Peter Gravlund allowed us to use some of his 12S
and 16S sequence data not yet available in GenBank. Peter Lenk explained the method for conversion of ModelTest results to MP weighting schemes, the PAUP∗ support group explained some of the intricacies of that program, and Bill Branch and Don Broadley provided useful
biological information. Chris Simon, L. Lacey Knowles, and two anonymous reviewers provided constructive criticism. C.M.R.K. thanks TLJC
and Jan Legge for continued support and encouragement. This study
was funded by National Research Foundation (South Africa) grants to
N.P.B. and M.H.V. and by the Rhodes Trust (C.M.R.K.).
R EFERENCES
ANDERSON, J. S. 2001. The phylogenetic trunk: Maximal inclusion of
taxa with missing data in an analysis of the Lepospondyli (Vertebrata,
Tetrapoda). Syst. Biol. 50:170–193.
B AUM , B. R. 1992. Combining trees as a way of combining data sets for
phylogenetic inference, and the desirability of combining gene trees.
Taxon 41:3–10.
BININDA-EMONDS , O. R. P., AND H. N. B RYANT . 1998. Properties of
matrix representation with parsimony analyses. Syst. Biol. 47:497–
508.
BININDA-EMONDS , O. R. P., H. N. B RYANT , AND A. P. R USSELL. 1998.
Supraspecific taxa as terminals in cladistic analysis: Implicit assumptions of monophyly and a comparison of methods. Biol. J. Linn. Soc.
64:101–133.
B OGERT , C. M. 1940. Herpetological results of the Vernay Angola
Expedition. Part 1. Snakes, including an arrangement of African
Colubridae. Bull. Am. Mus. Nat. Hist. 77:1–107.
B OURGEOIS , M. 1968. Contribution à la morphologie comparée du
crâne des Ophidiens de l’Afrique Centrale. Publ. Univ. Off. Congo
13:1–239.
B REMER, B., R. K. J ANSEN, B. O XELMAN, M. B ACKLUND , H. LANTZ, AND
K. K IM . 1999. More characters or more taxa for a robust phylogeny—
Case study from the coffee family (Rubiaceae). Syst. Biol. 48:413–435.
CADLE, J. E. 1984. Molecular systematics of xenodontine colubrid
snakes. III. An overview of xenodontine phylogeny and the history
of New World snakes. Copeia 1984:641–652.
458
SYSTEMATIC BIOLOGY
CADLE, J. E. 1985. The Neotropical colubrid snake fauna: Lineage components and biogeography. Syst. Zool. 34:1–20.
CADLE, J. E. 1987. Geographic distributions: Problems in phylogeny
and zoogeography. Pages 77–105 in Snakes: Ecology and evolutionary biology (R. A. Seigel, J. T. Collins, and S. S. Novak, eds.).
Macmillan, New York.
CADLE, J. E. 1988. Phylogenetic relationships among advanced snakes:
A molecular perspective. Univ. Calif. Publ. Zool. 119:1–77.
CADLE, J. E. 1994. The colubrid radiation in Africa (Serpentes:
Colubridae): Phylogenetic relationships and evolutionary patterns
based on immunological data. Zool. J. Linn. Soc. 110:103–140.
COBORN, J. 1991. The atlas of snakes of the world. T.F.H. Publications,
Neptune City, New Jersey.
CROZIER , R. H. 1990. From population genetics to phylogeny: Uses and
limits of mitochondrial DNA. Aust. Syst. Bot. 3:111–124.
DE Q UEIROZ, A., M. J. D ONOGHUE, AND J. K IM . 1995. Separate versus
combined analysis of phylogenetic evidence. Annu. Rev. Ecol. Syst.
26:657–681.
D OWLING , H. G., AND W. E. D UELLMAN. 1978. Systematic herpetology: A synopsis of families and higher categories. HISS Publications,
New York.
DOWLING , H. G., C. A. HAAS , S. B. HEDGES , AND R. HIGHTON. 1996.
Snake relationships revealed by slow-evolving proteins: A preliminary survey. J. Zool. 241:1–28.
DOWLING , H. G., R. HIGHTON, G. C. M AHA, AND L. R. M AXSON.
1983. Biochemical evaluation of colubrid snake phylogeny. J. Zool.
201:309–329.
FARRIS , J. S. 1969. A successive approximations approach to character
weighting. Syst. Zool. 18:374–385.
FELSENSTEIN, J. 1985. Confidence limits on phylogenies: An approach
using the bootstrap. Evolution 39:783–791.
FELSENSTEIN, J. 1988. Phylogenies from molecular sequences: Inferences and reliability. Annu. Rev. Genet. 22:521–565.
G EORGE, D. W., AND H. C. D ESSAUER. 1970. Immunological correspondence of transferrins and the relationships of colubrid snakes. Comp.
Biochem. Physiol. 33:617–627.
G OLDMAN, N., J. P. ANDERSON, AND A. G. R ODRIGO . 2000. Likelihoodbased tests of topologies in phylogenetics. Syst. Biol. 49:652–
670.
GORDON, A. 1986. Consensus supertrees: The synthesis of rooted trees
containing overlapping sets of leaves. J. Classif. 3:335–348.
GRAVLUND , P. 2001. Radiation within the advanced snakes
(Caenophidia) with special emphasis on African opistoglyph colubrids, based on mitochondrial sequence data. Zool. J. Linn. Soc.
72:99–114.
GREENE, H. W. 1997. Snakes: The evolution of mystery in nature. Univ.
California Press, Berkeley.
GROOMBRIDGE, B. C. 1984. The facial carotid artery in snakes (Reptilia:
Serpentes): Variations and possible cladistic significance. Amphib.Reptilia 5:145–155.
GROOMBRIDGE, B. C. 1986. Phyletic relationships among viperine
snakes. Pages 219–222 in Studies in herpetology; Proceedings of the
3rd Ordinary General Meeting of the Societas European Herpetologica (Z. Rocek, ed.). Charles Univ., Prague.
HARLEY, E. H. 1996. DAPSA, version 4.0. A program for DNA and
protein sequence alignment. Univ. Cape Town, Cape Town.
HASEGAWA, M., H. K ISHINO , AND T. YANO . 1985. Dating of the human–
ape splitting by a molecular clock of mitochondrial DNA. J. Mol.
Evol. 22:160–174.
HEISE, P. J., L. R. M AXSON, H. G. D OWLING , AND S. B. HEDGES . 1995.
Higher-level snake phylogeny inferred from mitochondrial DNA sequences of 12S rRNA and 16S rRNA genes. Mol. Biol. Evol. 12:259–
265.
HEYMANS , J. C. 1975. La musculature mandibulaire et le groupe parotidien des Aparallactinae et Atractaspinae (Serpentes Colubridae) a
majorite fouisseurs. Rev. Zool. Afr. 89:889.
HEYMANS , J. C. 1982. Contribution a la phylogenèse des ophidians de
l’Afrique centrale. Ann. Soc. R. Zool. Belg. 112:79–87.
HILLIS , D. M., AND J. P. HUELSENBECK . 1992. Signal, noise, and reliability in molecular phylogenetic analyses. J. Hered. 83:189–195.
HOFFSTETTER, R. 1955. Squamates de type moderne. Pages 606–662 in
Traı̂té de paléontology, Volume 5 (J. Piveteau, ed.). Masson, Paris.
VOL.
52
HOFFSTETTER, R., AND Y. G AYRARD . 1965. Observations sur l’ostéologie
et la classification des Acrochordidae (Serpentes). Bull. Mus. Natl.
Hist. Nat. 36:677–696.
HUELSENBECK , J. P. 1991. When are fossils better than extant taxa in
phylogenetic analysis? Syst. Zool. 40:458–469.
HUELSENBECK , J. P., AND J. P. BOLLBACK . 2001. Empirical and hierarchical Bayesian estimation of ancestral states. Syst. Biol. 50:351–366.
HUELSENBECK , J. P., AND N. S. IMENNOV . 2002. Geographic origin of
human mitochondrial DNA: Accommodating phylogenetic uncertainty and model comparison. Syst. Biol. 51:155–165.
HUELSENBECK , J. P., AND F. RONQUIST . 2001. MrBayes: Bayesian inference of phylogeny. Department of Biology, Univ. Rochester,
Rochester, New York.
I RWIN, D. M., T. D. K OCHER, AND A. C. WILSON. 1991. Evolution of the
cytochrome b gene of mammals. J. Mol. Evol. 32:128–144.
J ENNER, J. V. 1981. A zoogeographic study and the taxonomy of
the xenodontine colubrid snakes. Unpublished Ph.D. dissertation,
New York University.
KEARNEY, M. 1998. Systematics of the amphisbaenian family
Rhineuridae: Missing data and resolution. J. Vertebr. Paleontol.
18(suppl.):55A.
K ISHINO , H., AND M. HASEGAWA. 1989. Evaluation of the maximum
likelihood estimate from the evolutionary tree topologies from DNA
sequence data, and the branching order in Hominoidea. J. Mol. Evol.
29:170–179.
K RAUS , F., AND W. M. B ROWN. 1998. Phylogenetic relationships of colubroid snakes based on mitochondrial DNA sequences. Zool. J. Linn.
Soc. 122:455–487.
KRON, K. A., AND W. S. J UDD . 1997. Systematics of the Lyonia group
(Andromedeae, Ericaceae) and the use of species as terminals in
higher-level cladistic analyses. Syst. Bot. 22:479–492.
LARGET , B., AND D. L. S IMON. 1999. Markov chain Monte Carlo algorithms in the Bayesian analysis of phylogenetic trees. Mol. Biol. Evol.
16:750–759.
LECOINTRE, G., H. PHILIPPE, H. L. V ÂN LÊ, AND H. LE G UYADER . 1993.
Species sampling has a major impact on phylogenetic inference. Mol.
Phylogenet. Evol. 2:205–224.
LEE, M. S. Y., AND J. D. S CANLON. 2002. Snake phylogeny based on
osteology, soft anatomy and ecology. Biol. Rev. 77:333–401.
LENK , P., S. K ALYABINA, M. WINK , AND U. J OGER. 2001. Evolutionary
relationships among the true vipers (Reptilia: Viperidae) inferred
from mitochondrial DNA sequences. Mol. Phylogenet. Evol. 19:94–
104.
LIEM , K. L., H. M ARX, AND G. R. R ABB . 1971. The viperid snake
Azemiops: Its comparative cephalic anatomy and phylogenetic position in relation to Viperinae and Crotalinae. Fieldiana Zool. 59:65–
126.
M ATTHEE, C. A., J. D. B URZLAFF, J. F. TAYLOR, AND S. K. D AVIS . 2001.
Mining the mammalian genome for artiodactyl systematics. Syst.
Biol. 50:367–390.
MAU, B. 1996. Bayesian phylogenetic inference via Markov chain Monte
Carlo methods. Ph.D. Dissertation, Univ. Wisconsin, Madison.
MAU, B., AND M. NEWTON. 1997. Phylogenetic inference for binary
data on dendrograms using Markov chain Monte Carlo. J. Comput.
Graph. Stat. 6:122–131.
MAU, B., M. NEWTON, AND B. LARGET . 1999. Bayesian phylogenetic
inference via Markov chain Monte Carlo methods. Biometrics 55:1–
12.
MCCARTHY, C. J. 1985. Monophyly of elapid snakes (Serpentes: Elapidae): An assessment of the evidence. Zool. J. Linn. Soc. 83:79–93.
MCD OWELL, S. B. 1968. Affinities of the snakes usually called Elaps
lacteus and E. dorsalis. Zool. J. Linn. Soc. 47:561–578.
MCD OWELL, S. B. 1979. A catalogue of the snakes of New Guinea and
the Solomons, with special reference to those in the Bernice P. Bishop
Museum. Part III. Boinae and Acrochordoidea. J. Herpetol. 13:1–92.
MCD OWELL, S. B. 1987. Systematics. Pages 3–50 in Snakes: Ecology
and evolutionary biology (R. A. Seigel, J. T. Collins, and S. S. Novak,
eds.). Macmillan, New York.
MILINKOVITCH, M. C., R. G. LED UC, J. ADACHI , F. FARNIR, M.
G EORGES , AND M. HASEGAWA. 1996. Effects of character weighting
and species sampling on phylogeny reconstruction: A case study
based on DNA sequence data in cetaceans. Genetics 144:1817–1833.
2003
459
KELLY ET AL.—PHYLOGENETICS OF ADVANCED SNAKES
M INTON, S. A., AND S. K. S ALANITRO . 1972. Serological relationships
among some colubrid snakes. Copeia 1972:246–252.
MOORE, W. S. 1995. Inferring phylogenies from mtDNA variation:
Mitochondrial gene trees versus nuclear gene trees. Evolution
49:718–726.
NIXON, K. C., AND J. M. CARPENTER. 1996. On simultaneous analysis.
Cladistics 12:221–241.
NIXON, K. C., AND Q. D. WHEELER. 1992. Extinction and the origin of
species. Pages 119–143 in Extinction and phylogeny (M. J. Novacek
and Q. D. Wheeler, eds.). Columbia Univ. Press, New York.
NOVACEK , M. J. 1992. Fossils, topologies, missing data and the higher
level phylogeny of eutherian mammals. Syst. Biol. 41:58–73.
PAGE, R. D. M. 2000. Extracting species trees from complex gene trees:
Reconciled trees and vertebrate phylogeny. Mol. Phylogenet. Evol.
14:89–106.
PARKINSON, C. L. 1999. Molecular systematics and biogeographical
history of pitvipers as determined by mitochondrial ribosomal sequences. Copeia 1999:576–586.
PHILIPPE, H., AND E. D OUZERY. 1994. The pitfalls of molecular phylogeny based on four species, as illustrated by the
Cetacea/Artiodactyla relationships. J. Mammal. Evol. 2:133–152.
POSADA, D., AND K. A. CRANDALL. 1998. ModelTest: Testing the model
of DNA substitution. Bioinformatics 14:817–818.
POUGH, F. H., R. M. ANDREWS , J. E. CADLE, M. L. CRUMP, A.
H. S AVITZKY, AND K. D. WELLS . 2001. Herpetology, 2nd edition.
Prentice-Hall, Englewood Cliffs, New Jersey.
R AGAN, M. A. 1992. Phylogenetic inference based on matrix representation of trees. Mol. Phylogenet. Evol. 1:53–58.
RANNALA, B., AND Z. YANG . 1996. Probability distribution of molecular
evolutionary trees: A new method of phylogenetic inference. J. Mol.
Evol. 43:304–311.
RASMUSSEN, J. B. 1985. A re-evaluation of the systematics of the African
rear-fanged snakes of Bogert’s groups XIII–XVI, including a discussion of some evolutionary trends within Caenophidia. Pages 531–
548 in Proceedings of the International Symposium on African Vertebrates (K.-L. Schuchmann, ed.). Zoologisches Forschungsinstitut
und Museum Alexander Koenig, Bonn.
RIEPPEL, O. 1979. A cladistic classification of primitive snakes based on
skull structure. Z. Zool. Syst. Evolutionsforsch. 17:140–150.
RIEPPEL, O. 1988. A review of the origin of snakes. Evol. Biol. 22:37–130.
RODR ı́ GUEZ, F., J. F. OLIVER, A. M AR ı́ N, AND J. R. M EDINA. 1990. The
general stochastic model of nucleotide substitution. J. Theor. Biol.
142:485–501.
R OMER, A. S. 1956. Osteology of the reptiles. Univ. Chicago Press,
Chicago.
R ONQUIST , F. 1996. Matrix representation of trees, redundancy and
weighting. Syst. Biol. 45:247–253.
R OSSMAN, D. A., AND W. G. EBERLE. 1977. Partition of the genus Natrix,
with preliminary observations on evolutionary trends in natricine
snakes. Herpetologica 33:34–43.
SAIKI , R. K., T. L. B UGAWAN, G. T. HORN, K. B. M ULLIS , AND H.
A. ERLICH. 1986. Analyses of enzymatically amplified β-globin and
HLADQα DNA with allele-specific oligonucleotide probes. Nature
324:163–166.
S AITOU, N., AND M. NEI . 1987. The neighbor-joining method: A new
method for reconstructing phylogenetic trees. Mol. Biol. Evol. 4:406–
425.
S ANDERSON, M. J., A. PURVIS , AND C. HENZE. 1998. Phylogenetic supertrees: Assembling the trees of life. Trends Ecol. Evol. 13:105–109.
SANKOFF, D. 1975. Minimum mutation trees of sequences. J. Appl.
Math. 21:35–42.
S CHWANER, T. D., AND H. C. D ESSAUER. 1982. Comparative immunodiffusion survey of snake transferrins focused on the relationships of
the natricines. Copeia 1982:541–549.
SHIMODAIRA, H. 2002. An approximately unbiased test of phylogenetic
tree selection. Syst. Biol. 51:492–508.
SHIMODAIRA, H., AND M. HASEGAWA. 1999. Multiple comparisons
of log-likelihoods with applications to phylogenetic inference. Mol.
Biol. Evol. 16:1114–1116.
SIMMONS , M. P., AND H. O CHOTERENA. 2000. Gaps as characters in
sequence-based phylogenetic analyses. Syst. Biol. 49:369–381.
SLOWINSKI , J. B., AND J. S. K EOGH. 2000. Phylogenetic relationships
of elapid snakes based on cytochrome b mtDNA sequences. Mol.
Phylogenet. Evol. 15:157–164.
SLOWINSKI , J. B., A. K NIGHT , AND A. R. R OONEY. 1997. Inferring species
trees from gene trees: A phylogenetic analysis of the Elapidae (Serpentes) based on the amino acid sequences of venom proteins. Mol.
Phylogenet. Evol. 8:349–362.
SLOWINSKI , J. B., AND R. LAWSON. 2002. Snake phylogeny: Evidence from nuclear and mitochondrial genes. Mol. Phylogenet. Evol.
24:194–202.
SMITH, H. M., R. B. S MITH, AND H. L. S WAIN. 1977. A summary of
snake classification (Reptilia: Serpentes). J. Herpetol. 11:115–121.
S WOFFORD , D. L. 2000. PAUP∗ : Phylogenetic analysis using parsimony (∗ and other methods), version 4. Sinauer, Sunderland,
Massachusetts.
T HORLEY, J. L., AND R. D. M. PAGE. 2000. RadCon: Phylogenetic tree
comparison and consensus. Bioinformatics 16:486–487.
T HORLEY, J. L., AND M. WILKINSON. 1999. Testing the phylogenetic
stability of early tetrapods. J. Theor. Biol. 200:343–344.
UNDERWOOD , G. 1967. A contribution to the classification of snakes.
British Museum of Natural History, London.
UNDERWOOD , G. 1999. Morphological evidence on the affinities of
vipers. Kaupia 8:3–8.
UNDERWOOD , G., AND E. K OCHVA. 1993. On the affinities of the burrowing asps Atractaspis (Serpentes: Atractaspididae). Zool. J. Linn.
Soc. 107:3–64.
VIDAL, N., S. G. K INDL, A. WONG , AND S. B. HEDGES . 2000. Phylogenetic relationships of xenodontine snakes inferred from 12S and 16S
ribosomal RNA sequences. Mol. Phylogenet. Evol. 14:389–402.
°
R
WALSH, P. S., D. A. M ETZGER, AND R. HIGUCHI . 1991. Chelex 100 as
a medium for simple extraction of DNA for PCR-based typing from
forensic material. BioTechniques 10:506–513.
YANG , Z. 1994. Estimating the pattern of nucleotide substitution. J. Mol.
Evol. 39:105–111.
YANG , Z., AND B. RANNALA. 1997. Bayesian phylogenetic inference using DNA sequences: A Markov chain Monte Carlo method. Mol.
Biol. Evol. 14:717–724.
YEATES , D. K. 1995. Groundplans and exemplars: Paths to the tree of
life. Cladistics 11:343–357.
First submitted 3 January 2002; reviews returned 13 September 2002;
final acceptance 29 March 2003
Associate Editor: L. Lacey Knowles
APPENDIX 1:
LOCALITY D ATA AND VOUCHER NUMBERS OF TAXA
FOR WHICH CYTB S EQUENCES WERE G ENERATED
Accession numbers refer to snake specimens rather than tissue samples. Collection acronyms are as follows: CAS, California
Academy of Sciences; CMRK, Christopher M. R. Kelly; HLMD, Hessisches Landesmuseum Darmstadt; NMZB, Natural History Museum,
Bulawayo, Zimbabwe; PEM, Port Elizabeth Museum, South Africa.
Aparallactus guentheri, East Usambara Mountains, Tanzania (PEM
R5678); Dasypeltis scabra, Moebase, northern Mozambique (PEM
R13479); Hemirhagerrhis viperinus, Kaokoveld, Namibia (CAS (AMB5989)); Malpolon monspessulanus, Telmit, Morocco (HLMD-RA1180);
Psammophis mossambicus, Namagure, northern Mozambique (PEM
R13217); Psammophylax variabilis, Mulanje Massif, Malawi (CMRK
M01); Rhamphiophis acutus, Luzamba, Angola (PEM R13485); Lamprophis inornatus, Kasuga, Eastern Cape, South Africa (PEM R5676);
Mehelya nyassae, Kwekwe, Zimbabwe (NMZB 16570); Bitis arietans, Grahamstown, South Africa (PEM R5661).