The two recently sequenced marsupial genomes of the South American opossum (Monodelphis domestica) [5] and a kangaroo, the Australian tammar wallaby (Macropus eugenii), provide a unique opportunity to apply a completely new approach to resolve marsupial relationships. The insertion patterns of retroposed elements, pieces of DNA that are copied via RNA intermediates and pasted randomly elsewhere in the genome, have successfully resolved the more than 130 million-year-old branch of therian mammals [6] and early placental mammalian divergences [7] as well as relationships within other mammalian orders [8]. Because the insertion sites are effectively random and parallel insertions or exact excisions are very rare [9], the shared presence of retroposed elements at identical orthologous genomic locations of different species, families, or orders is a virtually homoplasy-free indication of their relatedness. Thus, the interpretation of retroposon markers is simple and straightforward: the presence of one of these elements in the orthologous genomic loci of two species signals a common ancestry, while its absence in another species signals a prior divergence [10]. No other sequenced mammalian genome has shown as high a percentage of discernible retroposed elements as marsupials (52%) [5], an extremely large number of possible informative markers.

Three different search strategies (see Materials and Methods) revealed ~217,000 retroposon-containing genomic loci. Highly conserved exonic primers were generated for 228 loci and experimentally tested on a small set of species. After carefully screening the sequences, we selected 32 loci based on criteria outlined in the Materials and Methods section for amplification in 20 marsupial species (Table S2). We carefully aligned and analyzed approximately 440 marsupial sequences to reveal 53 informative markers (Figure 2, Table 1).

Ten of the phylogenetically informative markers accumulated in the metatherian genome since their split from placental mammals, approximately 130 million years ago (MYA) [12],[13], and before the earliest divergence of the modern marsupial mammals, 70–80 MYA [3],[14]. All ten are absent in other mammals, significantly confirming the monophyly of marsupials (p=2.0×10⁻⁵; [10 0 0] [15]).

The other 43 phylogenetically informative retroposon markers provide significant support for most of the basal splits within marsupials. The earliest marsupial divergence was previously impossible to resolve based on sequence data, which could not distinguish between Paucituberculata and Didelphimorphia as the sister group to Australidelphia [14],[16]–[19]. We identified two markers (MIR3_MarsA) in the South American shrew opossums (Paucituberculata) that were also present in all Australidelphia but absent in Didelphimorphia (Figure 2). Albeit not significant (p=0.1111; [2 0 0]), this is the first molecular support for the earliest branching of Didelphimorphia, establishing it as the sister group to the remaining six marsupial orders. However, as significant support for this important marsupial branch requires three or more conflict-free markers [15], we attempted to find additional retroposons for the marsupial root. To find the third marker for the supported topology (Figure 2), a MIR3_Mars element present in kangaroo plus Paucituberculata but absent in opossum, we recovered ten additional loci from in silico screening; two contained the previously detected markers and eight contained new retroposons. Unfortunately, experimental verification showed that the absences of MIR3_Mars in opossum were due to non-specific deletions. On the other hand, we also did not find any loci with MIR3_Mars elements present in opossum plus Paucituberculata but absent in kangaroo, which would have supported the alternative of a close relationship between Didelphimorphia and Paucituberculata. We then screened for markers that would support the alternative hypothesis of Paucituberculata being the sister to all marsupials by performing an exhaustive in silico pre-screening for orthologous MIR3_Mars elements present in short introns of opossum and kangaroo. Starting from ~6,000 potentially informative loci, we selected 39 highly conserved MIR3_Mars-containing introns. However, experimental verification showed that all of the elements were also present in the order Paucituberculata (Rhyncholestes), thus supporting the monophyly of marsupials (data not shown), but not the basal divergence.

Assuming that, in the entire genomes, there are more than just the two detected diagnostic insertions for the root, an expanded search including larger introns and conserved intergenic regions is required to find significant support for this branch. Such relaxed search conditions are expected to provide a huge number of additional markers spread over the entire marsupial tree, but will require extensive additional computational and experimental work.

Molecular estimates have placed the earliest divergences of Marsupialia in the Late Cretaceous, 65–85 MYA [3],[4],[14]. To resolve placental mammalian Cretaceous divergences [20], large amounts of sequence data were crucial to gain sufficient phylogenetic signal, which is a plausible explanation for the difficulties encountered in trying to resolve this branch in previous marsupial investigations [3],[4],[14]. However, morphological data have revealed several characters from the skull and postcranium, supporting Didelphimorphia as the sister to all marsupials [21], consistent with our two molecular markers.

Leaving the base of the tree for the time being, 13 of the original 53 markers were present in the South American Microbiotheria and the four Australasian orders but not in either Didelphimorphia or Paucituberculata, significantly supporting the monophyly of Australidelphia [1] (p=6.3×10⁻⁷; [13 0 0]; Figure 2). The large number of phylogenetically informative markers indicates a long phylogenetic branch and/or a high degree of retroposon activity and fixation in the ancestral Australidelphia lineage. The branch separating Australidelphia from Didelphimorphia and Paucituberculata is one of the strongest supported and evolutionarily longest inter-ordinal branches in the marsupial tree [3],[4]. The fossil Australian marsupial Djarthia murgonensis is the oldest, well-accepted member of Australidelphia. Thus, combined with the lack of old Australidelphian fossils from South America, the most parsimonious explanation of the biogeography of Australidelphia is of an Australian origin [22]. However, the poor fossil record from South America, Antarctica, and Australia does not exclude that Djarthia, like Dromiciops, could be of South American origin and had a pan-Gondwanan distribution. Additional fossils from Australia or South America will shed more light on the early Australidelphian relationships and their biogeography.

Four markers significantly support the monophyletic grouping of the four Australasian orders to the exclusion of Microbiotheria (p=0.0123; [4 0 0]; Figure 2). Several studies have presented evidence for the monophyly of the Australasian orders; these have typically been based solely on nuclear protein-coding genes such as ApoB, BRCA1, IRBP, RAG1, and vWF [4],[17],[19], albeit with relatively low support values. By contrast, other sequence-based studies, relying completely or partially on mitochondrial data, find the South American order Microbiotheria nested within the Australasian orders [3],[16],[23]. Thus, the two competing hypotheses, Microbiotheria nested within or outside Australasian orders, have confounded the search for a reliable marsupial phylogeny.

Two studies tried to combine the nuclear and mitochondrial data using different approaches to achieve a larger dataset with higher probability of resolving the marsupial phylogeny [14],[18]. Only R/Y-coding, removing of sites [18], or partitioning [14] reduced possible artefacts from the mitochondrial data enough to reach a topology consistent with the retroposon markers. However, both studies gave low support for the position of Microbiotheria, illustrating the difficulties in resolving a short branch using sequence data under difficult conditions, such as possible nucleotide composition bias problems and randomization of fast evolving sites. The support from two independent sources of phylogenetic information, our retroposon markers and nuclear genes [4],[17],[19], invalidates the mitochondrial results [3],[16],[23]. Complete mitochondrial genomes can give misleading signals, as was demonstrated for the incorrect position of Monotremata among mammals [24], and can even mislead phylogenetic reconstruction when mixed with nuclear data.

The position of Microbiotheria has been intensely debated since the cohort Australidelphia was first suggested based on tarsal evidence [1]. After decades of uncertainty derived from molecular and morphological data, we have uncovered four independent diagnostic retroposon insertions that finally place the South American order Microbiotheria at its correct place in the marsupial tree (Figure 2). Therefore, we propose the new name Euaustralidelphia (“true Australidelphia”) for the monophyletic grouping of the four Australasian orders Notoryctemorphia, Dasyuromorphia, Peramelemorphia, and Diprotodontia.

The relationship among the four Australasian orders is not resolved, and of special interest is the phylogenetic position of the marsupial mole, Notoryctes typhlops, which has been debated for a long time [3],[4],[14],[17]–[19],[21],[23]. The marsupial mole is the only burrowing marsupial and is found in the deserts of Australia. The eyes of the marsupial mole are vestigial and the fore- and hind limbs are morphologically derived due to the burrowing lifestyle. The derived morphology and the fact that the marsupial mole is the single species in the order Notoryctemorphia have complicated attempts to resolve its phylogenetic position relative to the other three Australian orders. Most analyses of molecular sequence data find the marsupial mole closely related to the orders Dasyuromorphia and Peramelemorphia, but the support values are generally weak [3],[4],[14],[17]–[19],[23], and the exact phylogenetic position relative to the other two orders is yet to be determined. During the retroposon screening one marker was found supporting a grouping of Notoryctes, Dasyuromorphia, and Peramelemorphia (p=0.3333 [1 0 0]). The single retroposon marker is in agreement with the results from the sequence data. Extended screening of retroposons can provide additional evidence for the position of the marsupial mole among marsupials and which of the orders, Dasyuromorphia or Peramelemorphia, is the sister group.

Of the original 53 markers, 18 of them provide significant support for the monophyly of each of the five multi-species marsupial orders: five for Didelphimorphia (p=0.0041; [5 0 0]), three each for Paucituberculata, Dasyuromorphia, and Diprotodontia (p=0.037; [3 0 0]), and four for Peramelemorphia (p=0.0123; [4 0 0]). Four of the remaining markers provide non-significant support for various intra-ordinal relationships of Diprotodontia (Figure 2). Two of them support the division between Vombatiformes (wombats and koala) and Phalangerida (kangaroos, possums) (p=0.1111; [2 0 0]), challenging the results from mitochondrial sequence-based studies ([3], but see [25]), and one marker each supports the grouping of the possums Tarsipes and Pseudocheirus and that of the kangaroos Macropus and Potorous (p=0.3333 [1 0 0]). One final marker supports the grouping of the Didelphis and Metachirus.

The outstanding advantage of using retroposon presence/absence data for phylogenetic reconstructions is the low probability of insertion homoplasy. Independent parallel insertions of identical elements or exact deletions are extremely rare [9], but nevertheless not completely negligible, especially after genome-wide in silico screening of rare informative loci. LINE1-mobilized elements, in particular, show a slight preference for a TTAAAA consensus insertion motif [26], but on the other hand, such elements are rare in the deep phylogenetic branches of marsupials (Figure 1; Figure S1). Excluding the more frequent near identical insertions or unspecific deletions requires careful aligning and interpretation of orthologous informative markers (see Materials and Methods and Dataset S1).

Another possible source of errors is incomplete lineage sorting (polymorphism during speciation) or ancestral hybridization that can affect any marker system. Particularly short internal branches of a tree (rapid speciation) and biased in silico pre-screening for potential phylogenetically informative loci are exposed to such effects [27].

The available genomes of the opossum and the kangaroo placed us in the advantageous situation of independently pre-screening two distant branches of the marsupial tree. All 53 experimentally verified markers confine a phylogenetic tree free of any marker conflicts. Fourteen of them were randomly inserted as a second marker in specific loci. For most internal branches we found significant support for the underlying prior hypothesis by three or more markers with a clear rejection of alternative hypotheses.

Given the limitations just mentioned, the retroposon marker system identified a clear separation between the South American and Australasian marsupials. Thus, the current findings support a simple paleobiogeographic hypothesis, indicating only a single effective migration from South America to Australia, which is remarkable given that South America, Antarctica, and Australia were connected in the South Gondwanan continent for a considerable time.

Dataset S1

Figure S1

Figure S2

Figure S3

Table S1

Table S2

Table S3