Samples analyzed

Tissues were obtained from 18 species in 17 genera representing the six major agamid groups of Moody (1980) (Groups I–VI), and one genus of the Chamaeleonidae (Table 1, see Appendix 1 for further details). We also incorporated into the analyses the published data (Honda et al., 1999a; Ota et al., 1999: Table 1). Although we examined only one chameleon (Bradypodion fischeri) in the present study, its designation as the representative of all chameleons should not lead to any substantial error in the results of the analyses, because the monophyly of this taxon, supported by a number of synapomorphs, deserves no doubt (e.g., Rieppel, 1981; Frost and Etheridge, 1989). We also examined samples of Anolis carolinensis of the Polychridae and Iguana iguana of the Iguanidae (sensu Frost and Etheridge, 1989), because these families are supposedly closest to the two acrodont families (Frost and Etheridge, 1989; Macey et al., 1997).

Table 1

Infrafamilial groups of the Agamidae defined in the previous studies and localities. Data sources are (A) present study; (B) Honda et al. (1999a); (C) Ota et al. (1999). See Appendix 1 for detailed localities and DDBJ accesion numbers. Asterisks denote Iguanidae (sensu lato).

DNA extraction, amplification and sequencing

Extraction, amplification and sequencing of DNA are described in detail elsewhere (Honda et al., 1999a, b). A part of mitochondrial 12S and 16S rRNA genes consisting of approximately 860 base pairs (bp) were amplified using the polymerase chain reaction (PCR; Saiki et al., 1988) with primers L1091, H1478, L2606, and H3056 (Kocher et al., 1989; Hedges et al., 1993).

Phylogenetic analyses

Alignments for DNA sequences were determined based on maximum nucleotide similarity using CLUSTAL W 1.4 (Thompson et al., 1994) with default gap penalties. The output was later adjusted by eye using manual alinger SeqApp 1.9 (Gilbert, 1993) according to secondary structures of rRNA genes. However, the resultant secondary structures (not given) were unstable because of lack of information on secondary structures in closely related taxa (Titus and Frost, 1996). Thus we use the data for 652 sites based on maximum similarity excluding insertions and deletions in the following analyses, although topologies derived from these two alignments are identical. This designation should involve any substantial error in the results of the analyses.

The neighbor-joining (NJ) method (Saitou and Nei, 1987) was applied to infer relationships among taxa on the basis of a pairwise matrix of the distance from Kimura's (1980) two-parameter model. Degrees of supports for internal branches in each tree were assessed by 1,000 bootstrap pseudoreplications (Felsenstein, 1985). The NJ analysis was performed by use of CLUSTAL W. Maximum-likelihood (ML) analysis was also conducted using fastDNAml 1.0.6 (Olsen et al., 1993). Jumble options were used to find a true ML tree. For the maximum-parsimony (MP) analysis, PAUP* 4.0b (Swofford, 1998) with heuristic option was used. The confidence was assessed by 1,000 bootstrap resamplings. In these three analyses, no bias was assumed between transition and transversion.

Morphological analysis, using PAUP* with heuristic option, was also conducted on the basis of 122 characters listed in Moody (1980). Morphological data were also analyzed in combination with DNA sequences. To combine two data sets, we adjusted 60 operational taxonomic units (OTUs) examined in Moody (1980) to 19 OTUs examined in the DNA analyses, and designated the chameleon as a presumptive outgroup on the basis of the results of analyses with molecular data alone (see below). The partition homogeneity test (Farris et al., 1994) was conducted to assess the homogeneity between DNA sequence and morphological data using PAUP* with heuristic 1,000 bootstrappings. Templeton's (1983) test, a two-tailed Wilcoxon signed ranks test (Felsenstein, 1985), was applied to examine statistical significance of the shortest tree generated from two data sets using MacClade 3.08a (Maddison and Maddison, 1992).

The interpretation of bootstrap proportions (BPs) is still in a state of uncertainty (see Felsenstein and Kishino, 1993; Hillis and Bull, 1993). We tentatively followed Shaffer et al. (1997), and considered BPs ≥90% as highly significant, 70≤BPs <90% as marginally significant, and BPs <70% as constituting limited evidence of monophyly. For the ML analysis and other statistics, significance level was set at P<1% to avoid type I errors.

Mitochondrial sequence variation

Aligned sequences from 12S and 16S rRNA genes are presented in Appendix 2. All sequences showed strong bias against guanine on the light strand (A=33.6–37.7%, C=21.3–27.2%, G=17.9–20.0%, T=18.7–25.3%). Several observations demonstrated that such a bias represents that in mitochondrial genome, not in the nuclear integrated copies of mitochondrial genes (e.g., Zhang and Hewitt, 1996; Macey et al., 1997). We thus interpreted these sequences as those of authentic mitochondrial DNA.

The percentage of transitions with total sequence is plotted against that of transversions in Fig. 2A. Transitions exceeded transversions at low levels of sequence divergence. This agrees with previous studies on animal mitochondrial DNA that reported an initial high (>50%) transition bias which gradually decreases over time (Brown et al., 1982; Hedges et al., 1991; Fuller et al., 1998). The percent of transition is plotted against the total sequence divergence in Fig. 2B. The scatter plots did not exhibit a transition plateau (usually corresponding to 40–50% transitions), which is the point where multiple substitutions are occurring at the same site (Brown et al., 1982; Thomas et al., 1989; Hedges et al., 1991). Therefore, these data are considered to be useful for phylogenetic inference (e.g., Brown et al., 1982; Fuller et al., 1998).

Fig. 2

(A) Percentage of transitions with total sequence versus that of transversions for 12S and 16S rRNA genes in the 21 taxa examined. Closed circles denote comparisons within Acrodont (i.e., agamines, leiolepidines and chameleons). Open circles denote comparisons between Acrodont and outgroups. (B) Percentage of transitions with the variation versus that of sequnece divergence.

The 12S rRNA fragment consisted of 423 total sites, 326 of which were variable. For the 16S rRNA fragment, there were 437 total aligned sites, 272 of which were variable. Inter-generic nucleotide replacements within Agamidae varied from 70 bp (Physignathus lesueurii vs. Pogona) to 236 bp (Aphaniotis vs. Leiolepis).

Phylogenetic relationships

The NJ dendrogram derived from aligned sequences is shown in Fig. 3A. Ten nodes (nodes 1–10) were supported with significant BPs. The monophyly of Acrodonta (i.e., agamines, leiolepidines and chameleons) was supported in all bootstrap iterations (node 1: BP=100%). The ingroup portion of this dendrogram was divided into two major lineages, of which one was monotypic with Bradypodion of the Chamaeleonidae. The other (node 2: 100%) contained all genera of the Agamidae examined. These agamid lizards showed a dichotomous relationship. One of the major clusters (node 3: 85%) further split into two subcluster (nodes 5, 6). Of these, node 5 (93%) consisted of Leiolepis and Uromastyx (Group I [Moody, 1980] or Leiolepidinae [Frost and Etheridge, 1989]), whereas, node 6 (87%) accommodated several Australian genera. In the latter, Lophognathus (Group III) was first diverged, followed by the Hypsilurus (Group IV), leaving Physignathus (Group II), Chlamydosaurus (Group III) and Pogona (Group III) as monophyletic (node 10: 96%). The other major cluster (node 4: 84%) split into two subclusters (nodes 7, 8). Of these, node 7 (100%) consisted of Phrynocephalus and Agama (Group VI), whereas node 8 (100%) contained all Group V genera (i.e., Acanthosaura, Aphaniotis, Calotes, Draco, Gonocephalus, Japalura, Phoxophrys and Ptyctolaemus).

Fig. 3

(A) Neighbor-joining (NJ) dendrogram deriving from distance matrix from 12S and 16S rRNA sequence data. Numbers beneath branches are bootstrap proportions (BPs) at least 50% of the 1,000 bootstrap pseudoreplications. Nodes with bold numbers are identical with ML and MP analyses. Bar equals 0.1 unit of Kimura's (1980) two-parameter distance. (B) Maximum-likelihood (ML) dendrogram (ln likelihood =–14358.4). All branches were supported in significantly positive (P<1%). Bar equals 0.1 unit. (C) Maximum parsimony (MP) cladogram using heuristic option (2,202 steps, 443 bp informative under the condition of parsimony, consistency index =0.41). Branches without BP values were not supported in ⩾50% of the replicates.

Relationships depicted as a result of ML (Fig. 3B) and MP analyses (Fig. 3C) showed no inconsistency with those expressed in the NJ dendrogram in terms of topology of nodes 1–10, except for the absence of node 3 in MP. In the ML dendrogram, all branches were supported with significant P-values. Likewise, the MP cladogram, though giving no supports to node 3, showed no conflicts with other two analyses at the level of BPs ≥50%.

Comparisons of phylogenetic hypotheses

The MP cladogram derived from morphological data is shown in Fig. 4. The ingroup portion was divided into two major clusters, of which one, consisting of Leiolepis and Uromastyx (Group I), was supported in 100% BP. The other major cluster, supported in 97% BP, contained Groups II–VI. Within the latter, Groups III–VI constituted a cluster (75%). Monophyly of Group V was also supported (74%). By contrast, monophyly was not supported with significant BP values for each of the Groups II, III, and IV. As to Group VI, monophyly was rather weakly supported (66%).

Fig. 4

Strict consensus tree of 48 equally most-parsimonious tress generated from analysis of morphological data provided by Moody (1980) (762 steps, 119 informative under the condition of parsimony, consistency index=0.23). Branches without BP values were not supported in⩾50% of 1,000 replicates. Species names and intra-group BP values are omitted.

Independent MP analyses of DNA sequence and morphological data sets yielded different phylogenetic hypotheses (compare Fig. 3C and Fig. 4). When the Templeton's (1983) test was applied to morphological data of 19 OTUs, topology in the MP cladogram was significantly more parsimonious than that from molecular data (Ts=5, n=75). When this test was applied to molecular data, the MP cladogram was also significantly shorter than that from morphological data (Ts=44, n= 712). The partition homogeneity test revealed a significant heterogeneity between two data sets. These results imply that the molecular and morphological data sets are conflicting. We thus separately examined data sets phylogenetically in order to avoid a decrease in OTUs in the combined analysis, although the relationships derived from the combined analysis (not given) was largely consistent with the MP analysis of DNA sequence data.

Monophyly of the family Agamidae

Acrodonta (i.e., agamines, leiolepidines and chameleons) are known to exclusively share a number of morphological features, such as maxillaries in broad contact behind the premaxilla (Moody, 1980; Estes et al., 1988; Frost and Etheridge, 1989). A support to the monophyly of Acrodonta has also been provided from the analyses of data for sequences and the secondary structures in mitochondrial tRNA genes as well (Macey et al., 1997). Our results further confirmed its monophyly and strongly support the validity of those morphological characters as synapomorphs of Acrodonta.

Moody (1980), in his phylogenetic analysis on the basis of morphological data, hypothesized the primary dichotomy of Agamidae (sensu lato) into the Leiolepis–Uromastyx clade (Group I) and the remainder (Groups II–VI) (Fig. 1). Thus, Moody (1980, 1983) and Böhme (1982) separated Leiolepis and Uromastyx from the Agamidae (sensu lato), reviving the family Uromastycidae (=subfamily Leiolepidinae: Frost and Etheridge, 1989). Borsuk-Bialynicka and Moody (1984) reduced this group to the subfamilial status. All these authors took an a priori assumption for monophyly of the Agamidae (sensu lato). However, Frost and Etheridge (1989) lumped Agamidae with Chamaeleonidae, and recognized three sub-families, Agaminae, Leiolepidinae and Chamaeleoninae, in the latter, because their cladistic analysis of morphological data yielded two alternative equally parsimonious relationships: ((Agaminae, Chamealeoninae), Leiolepidinae), and ((Agaminae, Leiolepidinae), Chamealeoninae). Macey et al. (1997), based on the analysis of DNA sequence data of some portions of mitochondrial genes (tRNAs, ND1, ND2 and COI) and those combined with morphological data listed in Frost and Etheridge (1989), indicated the closest relationship between Physignathus and Phrynocephalus (i.e., representatives of Agamidae sensu stricto) within Acrodonta with significant BP values. They went so far as to even argue that both the Agamidae (sensu lato) and the Leiolepidinae are metataxa, because they failed to support the monophyly of the Agamidae or of leiolepidines (Uromastyx and Leiolepis). By contrast, our analysis of other mitochondrial gene (12S and 16S rRNAs) sequences strongly supported the monophyly of the family Agamidae (against Chamaeleonidae as representative by Bradypodion fischeri) and the sister relationship of Uromastyx and Leiolepis (see further discussion below). With respect to the intergeneric phylogeny, the sequence variation in 12S and 16S rRNA genes may possibly be regarded as being more informative than those in other domains, judging from results of recent studies on other taxa (e.g., Heise et al., 1995; Georges et al., 1999). We thus consider that our results strongly support the validity of the family Agamidae (contra Frost and Etheridge, 1989).

Diversification of Agamidae

Among the results of analyses of morphological data by Moody (1980) using unweighted and weighted Wagner tree algorithms, and compatibility methods, phylogenetic relationships were rather unstable, and only five branching topologies were consistently supported: (1) the dichotomy between the Group I and the Groups II–VI; (2) the monophyly of the members of the Groups III–VI; (3) the monophyly of the members of the Groups V–VI; (4) the monophyly of the Group III; and (5) the monophyly of the Group VI. Our analysis of his data with bootstrap resamplings (Felsenstein, 1985) resolved only (1), (2) and the monophyly of the Group V with significant BPs (Fig. 4). This may suggest that the agamid lizards have few phylogenetically informative morphological characters. By contrast, our approach using the molecular data seems to be much more useful in resolving the relationships of agamids, indicating a large dichotomy between a relatively primitive group including Australian and Melanesian members (Groups I–IV), and a more advanced group (Groups V–VI).

Joger (1991), on the basis of immunological data, argued for the monophyly of the Group VI. His results, however, failed to support the monophyly of the Group I, the Groups II–IV complex, or the Group V. Our results strongly suggest the monophyly of each of the latter three groups as well. We suspect that Joger's (1991) analysis suffered partially because of the limited resolving power of immuno-distance data as a result of more or less subjective estimate of the intensity of precipitin arcs (Greer, 1986), and also because of the insufficient number of samples from the “non-Group VI” members.

Although Uromastyx has been occasionally referred to as a typical agamid (Camp, 1923; Jollie, 1960), it also has a few highly specialized morphological features (Moody, 1980). Some authors (e.g., Moody, 1980, 1983; Borsuk-Bialynicka and Moody; Frost and Etheridge, 1989) thus classified this genus, together with its putative closest relative Leiolepis, to an independent family or subfamily. Nevertheless, analysis of sequence data for mitochondrial tRNAs, ND1, ND2 and COI genes did not support the dichotomy of the Leiolepis–Uromastyx clade and the remainder (Macey et al., 1997). Our results of NJ and ML analyses of data for mitochondrial 12S and 16S rRNA gene sequences did not support the validity of the subfamily Leiolepidinae or family Uromastycidae, either, because, although the two genera constituted a well supported clade (contra Macey et al., 1997), they did not show a sister relationship with all remaining agamids. MP analysis yielded no substantial account with respect to this problem. We thus consider any taxonomic separation of Leiolepis and Uromastyx from the remainder as inappropriate.

Members of the Group II are morphologically relatively primitive and are characterized by herbivory (e.g., Moody, 1980). In a different group of Iguania, primitive members are reported to constitute basal branches in a phylogenetic tree (Frost and Etheridge, 1989). Contrary to such a pattern, phylogenetic relationships inferred from DNA sequence in the present study do not support the early divergence of the primitive Group II, and suggest its possible non-monophyly. This is surprising especially when considering that both of the two taxa representing the Group II in our analyses are currently assigned to a single genus (Physignathus). It is thus probable that the morphological and ecological similarities among the Group II members actually represent symplesiomorphy or convergence. Further analyses are strongly desired to revise the generic arrangement of the two species examined here.

Lophognathus (Group III), occasionally regarded as synonymous with Physignathus (Wermuth, 1967; Matsui, 1992), was distantly located from either of the two Physignathus species within the Australian radiation (Fig. 3). This seems to support the validity of Lophognathus (e.g., Moody, 1980; Cogger, 1994). The morphological similarities between this genus and the two species of Physignathus may reflect symplesiomorphy or convergence, too.

Moody (1980) assumed the monophyly of Group III, and attributed its diversity to the Australian in situ radiation. Our results, while supporting the Australian origin of its diversity (see below), negate the monophyly of the Group III.

In our analysis of Moody's (1980) morphological data, the monophyly of the Groups V and VI was not supported at all (Fig. 4). By contrast, our molecular analyses strongly suggested the sister relationship of these two monophyletic groups. Considering the closest associations of the arboreal Group V with the likewise arboreal Group IV, and of the terrestrial Group VI with the largely terrestrial Group III in Fig. 4, we suspect that the results of the former analysis is influenced by the convergent characters independently evolved in response to similar ecological requirements. Our molecular analyses, on the other hand, failed to elucidate relationships within the Group V in detail. This may reflect the almost concurrent radiation of all lineages of this group examined here.

Biogeography of Agamidae

It is noteworthy that node 6 in our phylogenetic relationships (Fig. 3) exclusively consists of Australian agamids except for the Southeast Asian Physignathus cocincinus. Some authors argued that the ancestors of at least a part of the agamid fauna of Australian region (including Papua New Guinea and adjacent islands) have entered this continent from Eurasia through the Sunda Islands, a fringing archipelago between Malay Peninsula and New Guinea (e.g., Hecht, 1975; Tyler, 1979; Cogger and Heatwole, 1981; Witten, 1982, 1983). Others claimed that almost all agamids were derived from the Australian endemic radiation, and that they have no direct relationships with the tropical Asian agamids (e.g., Baverstock and Donnellan, 1990; King, 1990). Witten (1982, 1983), on the basis of morphological and karyological features, divided the Australian agamids into two groups. He assumed members of the smaller group, including Physignathus (Group II) and Hypsilurus (Group IV, referred to as Gonocephalus at that date: see below), as recent derivatives from the Asian stock, and the others as originating from an old endemic radiation in Australia. Moody (1980), while postulating the Australian endemic radiation for the Group III, supposed that the agamid fauna of this region increased through multiple colonizations from Southeast Asia. However, according to the phylogenetic relationships inferred above, the Australian endemic members (Groups II–IV) seem to have been derived entirely through an in situ radiation. Distribution of Physignathus cocincinus is thus considered as a consequence of the secondary dispersal. Occurrence of Hypsilurus in Melanesia also seems to represent a colonization from Australia. Such a process contrasts with that for the diversification of Australian varanids, because the current diversity of this gigantic lizard family in the Australian region is considered to have increased through multiple colonizations from Asia on the basis of DNA sequence data (Fuller et al., 1998).

Darlington (1957) highlighted the zoogeographically characteristic distribution of the genus Gonocephalus (sensu lato), which occurs across the Wallacia, a border of the Oriental and Australian faunal realms. However, Moody (1980), on the basis of morphological data, argued for the distant affinity between the species occurring in the west and east of Wallacia, and insisted on the validity of Hypsilurus, a nominate genus synonymized of Gonocephalus at that date, to accommodate the latter species. Witten (1983), however, pointed out that the number of micro-chromosomes in Gonocephalus (or Hypsilurus) spinipes from the Australian region is equivalent to those in several Asian species, and is greater than those in most other Australian agamids. He regarded such a similarity pattern as indicative of a closer phylogenetic affinity of this species with Asian agamids. Based on the immunological data, Baverstock and Donnellan (1990), and King (1990) supported Moody's (1980) view and considered “Gonocephalus” east of Wallacia as a part of the post Gondwanaland endemic Australian radiation. Furthermore, Ota et al. (1992) also indicated distinct chromosomal differences between Gonocephalus grandis and G. miotypanum from Southeast Asia, and “G”. spinipes. Our results, demonstrating a much closer phylogenetic affinity of Hypsilurus with other Australian agamids than with the Southeast Asian Gonocephalus, further support the latter view.

As contrasted with the agamid fauna of the Australian region, that of the Asian and African region consists of two distinct components—the Leiolepis–Uromastyx clade (node 5), and the other Asian–African agamids (node 4). Leiolepis and Uromastyx, although phylogenetically closest to each other, are geographically greatly isolated in Southeast Asia, and West Asia and Africa, respectively. This strongly suggests the relict nature of these genera.

The Groups V (node 8) and VI (node 7), while being monophyletic to each other (node 4), differ from each other geographically (distributed in South–Southeast Asia, and central–West Asia and Africa, respectively: Moody, 1980; Matsui, 1992). Judging from the fact that the current diversity of the Group VI is centered in West Asia and Africa (Moody, 1980; Matsui, 1992), it is likely that this group and the Group V diverged trough the vicariance between West Asia and South Asia, and that the central Asian representatives of the Group VI (a few species of Phrynocephalus) were derived from the secondary dispersal.

Cracraft (1974) divided the all modern groups of lizards into northern (Laurasian) and southern (Gondwanan) elements, and placed the Agamidae in the latter. However, present results do not support the Gondwanaland origin of the Agamidae, because they failed to demonstrate secondary derivations of the Laurasian (i.e., Asian) agamids from the Gondwanan (i.e., African and Australian) relatives. On the other hand, fossil evidence suggests the East Asian origin of the Agamidae (Borsuk-Bialynicka and Moody, 1984). Thus, it is more likely that the common ancestor of the family emerged in Asia, and that African and Australian agamids were originated through the secondary dispersals from Asia. This view is circumstantially supported by the highest species diversity in Southeast Asia, and by the absence of the agamid lizards in other Gondwanan areas such as Madagascar and South America.