| Edited by Yoko Satta. Norihiro Okada: Corresponding author. E-mail: nokada@bio.titech.ac.jp |
Phylogenetic relationships among marsupials have been studied by analyzing complete mitochondrial genomes and concatenated nuclear genes (Amrine-Madsen et al., 2003; Nilsson et al., 2003; Nilsson et al., 2004). Marsupials are separated into two cohorts: the Ameridelphia and Australidelphia (Szalay, 1982). Szalay (1982) originally recognized the Australidelphia cohort, which contains the genus Dromiciops and Australian marsupials, based on morphological features that both groups share, such as a continuous lower ankle joint pattern. Identification of this cohort was later supported by mitochondrial genome and concatenated nuclear gene sequence analyses (Amrine-Madsen et al., 2003; Nilsson et al. 2003; Nilsson et al. 2004).
Paleontological evidence suggests that all Australian marsupials immigrated to Australia from South America via Antarctica (Woodburne and Case, 1996). Szalay (1982) proposed two hypotheses concerning Dromiciops. One hypothesis states that all Australian marsupials are monophyletic and Dromiciops is a sister group of all Australian marsupials, and that the single ancestor of all extant Australian marsupials migrated from South America to Australia. The other hypothesis states that Australian marsupials are not monophyletic and that Dromiciops is nested in Australian marsupials. In this hypothesis, the ancestor of Dromiciops once migrated to Australia then returned to South America, or multiple ancestors of Australian marsupials migrated independently to Australia (Nilsson et al., 2004). An analysis with concatenated 6.4-kb nuclear genes did not resolve this issue (Amrine-Madsen et al., 2003). A mitochondrial genome sequence analysis supports the monophyly of Dromiciops and Australian omnivorous marsupials (Isoodon macrourus) and, thus, the latter hypothesis (Nilsson et al., 2004). Recent morphological analysis also supports the latter hypothesis (Horovitz and Sanchez-Villagra, 2003).
However, despite the results of these mitochondrial genome and nuclear gene analyses, the familial relationships among Diprotodontia remain unresolved. (Amrine-Madsen et al. 2003; Nilsson et al., 2004). The order Diprotodontia contains Australian marsupials distributed in New Guinea and Australia, including koala, wombat, cuscus and kangaroo. Diprotodontia are morphologically characterized by syndactylus (the second and third digits of the hind limbs are completely fused) and teeth morphology (a pair of incisors in the lower jaw is procumbent) (Marshall et al., 1990). Recent classification recognizes 19 Diprotodontian families and, of these, the extant families are Phascolarctidae (koala), Vombatidae (wombats), Macropodidae (kangaroos and wallabies), Potoroidae (bettongs, potoroos and rat kangaroos), Phalangeridae (brush-tailed possums), Petauridae (gliders and striped possums), Pseudocheiridae (ringtail possums), Acrobatidae (feather-tailed possum and pygmy glider), Tarsipedidae (honey possum) and Burramyidae (pygmy possums) (Table 1; Aplin and Archer, 1987). The suborder Vombatiformes includes two families, namely Phascolarctidae and Vombatidae, and other families are grouped as the suborder Phalangerida. In Phalangerida, there are eight extant families; six of the eight, excluding Macropodidae and Potoroidae, are grouped as possums.
 View Details | Table 1. Diprotodontian families |
Each family is well recognized by morphological characteristics. Molecular studies with the sequence of the mitochondrial gene, ND2, supports monophyly for each Diprotodontian family (Osborne et al., 2002). The monophyly of suborder Vombatiformes is well supported by several morphological and molecular studies (Kirsch, 1977; Aplin and Archer, 1987; Flannery, 1987; Marshall et al., 1990; Springer et al., 1994; Osborne et al., 2002; Amrine-Madsen et al., 2003; Asher et al. 2004; Nilsson et al., 2004). Phylogenetic relationships among families in Phalangerida, however, are not well resolved. Flannery (1987) suggested two hypotheses for phylogenetic relationships among these families from morphological evidence:
1. Macropodoidea (Macropodidae + Potoroidae) is a sister group of Phalangeridae, and other possums (Petauridae, Pseudocheiridae, Tarsipedidae, Acrobatidae and Burramyidae) are a sister group of Macropodoidea and Phalangeridae (Tree 1 in Fig. 1).
 View Details | Fig. 1. Two hypotheses by Flannery for the phylogenetic relationships among families in Phalangerida. Tree 1 and 2 indicate Flannery’s hypothesis trees predicted by morphological data. Tree3 is indicated by mitochondrial genome ND2 sequence (Osborene et al., 2002) |
2. Macropodoidea (Macropodidae + Potoroidae) is a sister group of a monophyletic group composed of all possums (Phalangeridae, Petauridae, Pseudocheiridae, Tarsipedidae, Acrobatidae and Burramyidae) (Tree 2 in Fig. 1).
Hypothesis 2 proposes that Macropodoidea is a group independent of other possums. Flannery showed only one synapomorphy supportive of hypothesis 2. In contrast, he recognized four synapomorphies for hypothesis 1. Kirsch’s serological data showed close relatedness between Macropus (Macropodidae) and Trichosurus (Phalangeridae), supporting hypothesis 1 (Kirsch, 1977). The analysis of the ND2 amino acid sequence data also supports a close relationship between Phalangeridae and Macropodoidea (Macropodidae + Potoroidae) (Osborne et al., 2002), also supporting hypothesis 1.
Another difficulty in determining phylogenetic relationships among families in Phalangerida is that the position of Vombatiformes differs in different analyses. According to the two hypotheses of Flannery, Vombatiformes is considered to be monophyletic as an outgroup taxon. In Osborn’s ND2 analysis, however, Vombatiformes is close to Phalangeridae + Macropodoidea (Macropodidae + Potoroidae) (Tree 3 in Fig. 1), implying that suborder Phalangerida is paraphyletic. The same result is seen in the complete mitochondrial genome data analysis (Nilsson et al., 2004). However, these results (Fig. 1, Tree 3) contradict Flannery’s monophyletic premise. On the other hand, Maximum Likelihood (ML) analyses of 6.4-kb concatenated nuclear genes suggested the monophyly of Phalangerida with a high bootstrap support (Amrine-Madsen et al., 2003).
In the present study, we sequenced the complete mitochondrial genomes of six Diprotodontian species. Then we performed ML analysis at the amino acid (aa) and nucleotide (nt) levels to elucidate the phylogenetic relationships in Diprotodontia at the family level and to revaluate the phylogenetic position of Dromiciops, currently considered to be a member of Australidelphia.
We isolated DNA from ethanol-preserved liver samples from Phascolarctos cinensis (koala), Lagorchestes hirsutus (western hare wallaby), Phalanger interpositus (Stein’s cuscus), Dactylopsila trivirgata (common striped possum) and Distoechurus pennatus (feather-tail possum) provided from South Australian Museum. Petaurus breviceps (sugar glider) was purchased from pet shop. Sample information is shown in Table 1. Tissue samples were digested with proteinase K. Total genomic DNA was then isolated by phenol/chloroform extraction and ethanol precipitation (Blin and Stafford, 1976) and stored at 4°C.
We designed universal primers based on the complete mitochondrial sequence of Macropus robustus (accession number Y10524) and Trichosurus vulpecula (AF357238). The forward and reverse PCR primers are shown in Table 2. We PCR amplified 3~4-kb mitochondrial DNA (mtDNA) fragments of each samples using 30 cycles of denaturation at 94°C for 30 s, annealing at 50°C for 1 min and extension at 72°C for 3 min in thin-walled reaction tubes. The reaction mixture contained 2.5 U Taq polymerase (TaKaRa, Siga, Japan), Taq buffer (Takara), 0.4 mM dNTPs, 2.5 mM MgCl2, 10 pM primers and 100 ng of genomic DNA. The PCR products completely covered the mitochondrial genomes and were confirmed by electrophoresis in a 2.0% SeaKem GTG agarose gel (TaKaRa, Siga, Japan). The control region of Phalanger interpositus and Petaurus breviceps were not amplified in this study.
 View Details | Table 2. Primers to amplify complete mitochondrial DNA |
PCR products of about 2~5 kb fragments were amplified and used as templates for direct sequencing and primer walking. Cycle sequencing using BigDye terminator (ABI, Foster City, CA) was performed with 25 cycles of denaturation at 96°C for 30 s, annealing at 47°C for 15 s, and extension at 60°C for 2 min in a total volume of 10 μl.
We used the 23 mtDNA sequences. The sequences of 12 mitochondrial proteins (mt-proteins) encoded in the same strand of mtDNA were used for phylogenetic analyses. Alignments of sequences were carefully checked by eye, and all positions with gaps or ambiguous alignments, in addition to overlapping regions between ATP6 and ATP8 and between ND4 and ND4L, were excluded. The total number of remaining codons was 3,453.
The phylogenetic analyses were carried out at both the aa and nt levels with the ML method (Felsenstein, 1981; Kishino et al., 1990). We used the ProtML program in the MOLPHY package (ver. 2.3) (Adachi and Hasegawa, 1996a), the TREE-PUZZLE program for quartet-puzzling (QP) analysis (Strimmer and von Haeseler, 1996), and the CodeML program in the PAML package (ver. 3.14) (Yang, 1997) for analysis of the aa sequences with the mtREV-F model (Adachi and Hasegawa, 1996b) and analysis of the nucleotide sequences of the protein-encoding genes with the codon-substitution (CS) model (Goldman and Yang, 1994; Yang et al., 1998). For the CS model, we used Miyata et al.’s (1979) distance with geometric formulae (Yang et al., 1998). We also used the BaseML program in the PAML package with the GTR + Γ model (Rodriguez et al., 1990) to analyze nt sequences, in which only the first and second codon positions were used.
In using the CodeML, BaseML, and TREE-PUZZLE programs, the discrete Γ distribution model (with eight categories, except for the CS model for which four categories were used because of the computational burden) for the site-heterogeneity (Yang, 1996) was adopted, and the shape parameter (α) of the Γ distribution was optimized. Bootstrap probabilities (BPs) were estimated by the RELL (resampling of estimated log-likelihoods) method (Kishino et al., 1990) with 10,000 bootstrap resamplings. The RELL method has been shown to be efficient in estimating BPs without performing ML estimation for each resampled data (Hasegawa and Kishino, 1994).
From preliminary analyses of the concatenated aa sequences of the 12 mt-proteins using the NJ, MP and TREE-PUZZLE methods, we were able to fix some clades that were supported with high BPs or QP supports and had no biological controversy surrounding them. This reduced the number of candidate trees provided to more sophisticated analyses. Even so, the number of candidate trees was still too large to allow exhaustive analysis with a computationally intensive method, and therefore we carried out an approximate likelihood analysis with the ProtML program for all the candidate trees. The most serious problem of the ML method when applied to data from many species is the explosively increasing number of possible trees, but most of these trees are very poor and unpromising. By quickly eliminating these trees by an approximate method, the ML method can be applied to a large phylogeny. In estimating the branch lengths for each tree topology by ML, we usually use the time-consuming Newton-Raphson method. The approximate likelihood option implemented in ProtML avoides this process and estimates an “approximate likelihood” from the initial values for the Newton-Raphson method given by the ordinary least squares. We could examine all the possible trees with the approximate likelihood method, and, by excluding unpromising trees by this approximate criterion, we could select the best 20,000 trees for the full likelihood analysis. There is a strong correlation between the approximate likelihood and the maximum likelihood, and, in a practical sense, this is a good method to reduce the computational burden (Adachi and Hasegawa, 1996a). However, even 20,000 trees may be too many for most sophisticated models. Therefore, we further reduced the number of candidate trees by selecting the best trees among the 20,000 that had log-likelihood scores differing by less than 2 SEs from that of the highest likelihood tree derived from a simpler model.
The mitochondrial genomes of six marsupials determined in our analysis contain 13 proteins, 2 ribosomal RNA and 22 transfer RNA genes. The gene orders are same as typical marsupials (Janke et al., 1994, 1997). The tRNA-Lys is highly mutated, and the anticodons of tRNA-Asp are GCC. These phenomena are also typical to marsupial mitochondria where it is reported that tRNA-Lys from nuclear DNA is imported to mitochondria and help translation (Döner et al., 2001) and anticodon of tRNA-Asp GCC which usually code for Gly is modified to GUC after transcription (Börner et al., 1996).
In placental mitochondrial control region, the presence of three conserved sequence block (CSB I, CSB II and CSB III) were reported (Walberg and Clayton, 1981). And two of CSBs (CSB II and CSB III) were found in Didelphis and Macropus (Janke et al., 1994, 1997). In our analysis, all CSBs are found in Isoodon, Distoechrirus, Lagorchestes, Phascolarctos, Macropus and Vombatus.
Fig. 2 shows a QP tree of the concatenated aa sequences of 12 mt-proteins. This tree indicates five stable monophyletic clades. These are Macropus + Lagorchestes clade (Macropodidae) (1), the Trichosurus + Phalanger clade (Phalangeridae) (2), the Vombatus + Phascolarcotos clade (Vombatiformes) (6), the Caenolestes + Rhyncholestes clade (9), and the Didelphis + Thylamys clade (10), and were confirmed with 95, 85, 89, 98, and 100% QP support, respectively. These clades were also supported with high BPs from the NJ and MP analyses (data not shown); therefore, we fixed these clades in the subsequent ML analyses. We used some eutherians and monotremes as outgroups (the relationships among the outgroup species were also fixed, as shown in Fig. 2). The number of possible trees among (1) Macropus + Lagorchestes, (2) Trichosurus + Phalanger, (3) Distoechurus, (4) Petaurus, (5) Dactylopsila, (6) Vombatus + Phascolarcotos, (7) Isoodon, (8) Dromiciops, (9) Caenolestes + Rhyncholestes, (10) Didelphis + Thylamys is 34,459,425. These trees were examined using the approximate likelihood option of ProtML for the concatenated mt-proteins, and the best 20,000 trees were selected for full likelihood analyses by the ProtML program with the mtREV-F model. The best 1,155 trees with log-likelihood scores within 2 SEs compared to the ML tree by the ProtML analysis were selected for more sophisticated analyses using the CodeML program with the mtREV-F + Γ model and by the BaseML with the GTR + Γ model applied to the nucleotides at the first and second positions in each codon.
 View Details | Fig. 2. A quartet-puzzling tree of concatenated amino acid sequences of 12 mt-proteins using the mtREV-F + Γ model. This tree was used for making the conclusive tree shown in Fig. 3. The horizontal length of each branch is proportional to the estimated number of amino acid substitutions. Numbers indicate QP support values. Eutherians and monotremes were used as outgroups. The numbers (1)–(10) under the branch indicate monophyletic groups and single taxa. The five stable monophyletic groups are shown in bold. Dasypus novemcinctus, nine-banded armadillo (accession number Y11832) Tamandua tetradactyla, southern tamandua (AJ421450), Chrysochloris asiatica, Cape golden mole (AB096866), Dugong dugon, dugong (AY075116), Procavia capensis, cape hyrax (AB096865), Orycteropus afer, aardvark (Y18475), Didelphis virginiana, Virginia opossum (Z29573), Trichosurus vulpecula, common brush-tailed possum (AF357238), Isoodon macrourus, brindled bandicoot (AF358864), Macropus robustus, wallaroo euro (Y10524), Tachyglossus aculeatus, short-beaked echidna (AJ303116), Ornithorhynchus anatinus, platypus (X83427), Caenolestes fuliginosus, Ecuadorian caenolestid (AJ508400), Dromiciops gliroides, colocolos (AJ508402), Rhyncholestes raphanurus, Chilean caenolestid (AJ50839), Thylamys elegans, elegant fat-tailed opossum (AJ508401), Vombatus ursinus, wombat (AJ304826), Phascolarctos cinensis koala (this work ; AB241053), Lagorchestes hirsutus western hare wallaby (this work; AB241056), Phalanger interpositus Stein’s cuscus (this work; AB241057), Petaurus breviceps suger glider (this work; AB241055), Dactylopsila trivirgata common striped possum (this work; AB241054) and Distoechrus pennatus feathertail possum (this work; AB241052). |
Fig. 3 shows the ML tree given by the CodeML analysis. Strong support was given to the Petaurus + Dactylopsila clade (Petauridae), which had 99% BP (100% BP by the BaseML analysis of the first and second codon positions), and to the Diprotodontia clade that consists of groups (or a taxa) 1–6 with 98% (94%) BP. The grouping of the Australian (Au) marsupials with the South American (Am) Dromiciops (the Australidelphia hypothesis by Szalay) was supported with 92% (87%) BP. The ML tree suggests that Dromiciops is nested within the Australian marsupials. The BP for the monophyly of Australian marsupials, excluding Dromiciops as an outgroup, was only 2% (4%), and thus this relationship was not supported (Tree 5 in Table 3).
 View Details | Fig. 3. An ML tree of concatenated amino acid sequences of 12 mt-proteins using the mtREV-F + Γ model. The monophyly of Isoodon and Dromiciops, the monophyly of Phalangerida and the sister relationship between Macropodidae and Phalangeridae were supported in this tree. The horizontal length of each branch is proportional to the estimated number of amino acid substitutions. Numbers indicate percent BPs estimated by the RELL method for amino acid analysis, and numbers in parentheses indicate percent BPs estimated by the RELL method for nt analysis of the first and second positions in codons. Numbers (1)–(10) under the branch indicate the monophyletic groups shown in Fig. 2. |
 View Details | Table 3. Comparison of trees among marsupials |
In the ML tree, the Didelphis + Thylamys clade (Didelphimorpha) represents the most basal lineage in the marsupial tree, and the Caenolestes + Rhyncholestes clade (Paucituberculata) is the sister group of the Australian marsupials + Dromiciops. The BP for the monophyly of Didelphis + Thylamys + Caenolestes + Rhyncholestes was 2.5% (8.9%); therefore, monophyly of the two major South American families was unlikely by our analysis. The monophyly of South American marsupials, including Dromiciops, was 0.07% (0.00%). Therefore, from our analyses, these groupings seem unlikely, consistent with previous analyses of mitochondrial genomes and nuclear genes (Springer et al., 1998; Amrine-Madsen et al., 2003, Nilsson et al. 2004). However, the South American clade including Dromiciops cannot be rejected by the conservative test of the Weighted Shimodaira-Hasegawa (WSH) test (Tree 6 in Table 3).
The monophyly of Phalangerida, including (1) Macropodidae, (2) Phalangeridae, (3) Distoechurus, (4) Petaurus and (5) Dactylopsila, was supported by 83% BP in the aa analysis and by 84% BP in the nt analysis of first and second positions within codons. The Petaurus + Dactylopsila + Distoechurus clade received 80% BP in the aa analysis but only 32% in the nt analysis of first and second positions within codons. We carried out further ML analysis of nt sequences using the CS model for a limited number of species (Fig. 4). This is the ML tree of 105 unrooted trees of the following six groups: (1) Macropus + Lagorchestes, (2) Trichosurus + Phalanger, (3) Distoechurus, (4) Petaurus, (5) Dactylopsila and (6) Vombatus + Phascolarcotos. This tree differs from that in Fig. 3 in that Distoechurus is not a sister group of the Petaurus + Dactylopsila clade, and the grouping of these three species received only 46% BP. All the BPs were low except for the Petaurus + Dactylopsila grouping; the relationships among Macropus + Lagorchestes, Distoechurus, Petaurus + Dactylopsila and Trichosuru + Phalanger remain ambiguous by our ML analyses.
 View Details | Fig. 4. An ML tree of concatenated codon sequences of 12 mt-protein genes using the CS model. Numbers indicate percent BPs estimated by the RELL method for amino acid analysis (above the branch) and CS analysis (below the branch). Numbers (1)–(10) under the branch indicate the monophyletic groups shown in Fig. 2. |
Although the hypothesis of the South American order Microbiotheria (Dromiciops) being nested among Australian marsupials (Fig. 3 and Table3, Trees-1, 2, 3 and 4) was the most likely by our analysis consistently with the previous mitochondrial genome analysis (Nilsson et al., 2004), the support was not conclusively strong. Tree-5, where Dromiciops is sister to all Australian marsupials (Marshall et al., 1990), could not be rejected by the AU, KH and WSH tests (Table 3). Indeed, as also was obtained in the mitochondria genome analysis of Nilsson et al. (2004), the most likely tree was that of Dromiciops nested within Australian marsupials. However, Tree 4 of Australidelphia monophyly had a log-likelihood score lower than that of the ML tree by only 15.0 ± 10.3 (± 1 SE) and could not be rejected with the SH test (p = 0.63).
Szalay (1982) proposed the cohorts Ameridelphia and Australidelphia (including Dromiciops in South America), and the monophyly of Australidelphia has been supported by previous molecular studies (e.g., Springer et al., 1998; Amrine-Madsen et al., 2003; Nilsson et al., 2004). Although the conservative WSH test could not reject an alternative possibility (Tree 6 in Table 3), the Australidelphia clade including Dromiciops was also very strongly supported by our analyses.
The best tree differed between the aa analysis and the nt analysis of the first and second position in codons. The former analysis preferred Tree 1, in which Australidelphia is paraphyletic, whereas the latter preferred Tree 2, in which Australidelphia is monophyletic. However, the difference was not statistically significant. The log-likelihood differences between the two trees were only 21.5 ± 17.8 and 3.9 ± 15.9 for the aa and nt analyses, respectively, and the AU, KH and WSH tests could not reject the alternative tree. However, the Bayesian posterior probability (PP; Huelsenbeck et al., 2001) rejected Tree 2 with a p-value of 5 × 10, but this might reflect the tendency of Bayesian PPs to predict with overconfidence when the model assumed in the analysis differs from reality (which is always the case to some degree) (Waddell et al., 2001). Therefore, this molecular analysis should be regarded as having only limited resolution concerning Australidelphia monophyly.
Diprododontia is separated into suborders, Vombatiformes and Phalangerida (Aplin and Archer, 1987). The monophyly of Phalangerida is ambiguous because the synapomorphy of Phalangerida is poorer than that of Vombatiformes (Aplin and Archer, 1987). While monophyly of Phalangerida is supported by serological analysis, not enough samples were analyzed in that study to conclusively determine monophyly (Kirsch, 1977). Only the concatenated nuclear gene sequence analysis strongly supports the monophyly of Phalangerida (Amrine-Madsen, 2003); all other analyses using mitochondrial genome sequences do not strongly support this assertion (Springer et al., 1994; Osborne et al., 2002; Nilsson et al., 2004). In our study, the QP tree (Fig. 2) indicated that Phalangeridae (Trichosurus + Phalanger), Macropodidae (Macropus + Lagorchestes) and Vombatiformes (Vombatus + Phascolarcotos) form a monophyletic group. But the more sophisticated ML analysis supports the monophyly of Phalangerida with 83% BP by the aa analysis, and 84% by the nt analysis of first and second positions in codons (Fig. 3). Thus, our analysis of the complete mitochondrial genome supports the monophyly of Phalangerida.
In addition, our ML analysis with mitochondrial aa sequence supports a close relationship between Phalangeridae and Macropodoidea. Osborne’s mitochondrial ND2 analysis indicated that Phalangeridae is a sister group of Macropodoidea (Osborne et al., 2002). Our analysis of mitochondrial genome aa sequence partially supports a close relationship between Phalangeridae and Macropodidae, (and, at the same time, that Phalangerida is monophyletic). Therefore our analysis is consistent with Flannery’s hypothesis 1.
Familial relationships in Phalangerida remain ambiguous, although some molecular analyses have attempted to clarify relationships among kangaroos (Macropodidae) and possums (Phalangeridae, Petauridae, Pseudocheiridae, Acrobatidae, Tarsipedidae and Burramyidae) (Springer et al., 1994; Osborne et al., 2002). Our analysis of aa sequences supports the monophyly of Acrobatidae (Distoechurus) and Petauridae (Petaurus + Dactylopsila), but the codon sequence analysis does not. The elucidation of relationships among possum families is difficult because a large amount of molecular data are not yet available. More possum DNA sequences, in addition to other diagnostic information (for example, SINE (Short Interspersed Repetitive Element) insertions), are still needed to resolve phylogenetic relationships among the Diprotodontian families.
|
Index