Molecules, morphology, and ecology indicate a recent, amphibious ancestry for echidnas

Abstract

The semiaquatic platypus and terrestrial echidnas (spiny anteaters) are the only living egg-laying mammals (monotremes). The fossil record has provided few clues as to their origins and the evolution of their ecological specializations; however, recent reassignment of the Early Cretaceous Teinolophos and Steropodon to the platypus lineage implies that platypuses and echidnas diverged >112.5 million years ago, reinforcing the notion of monotremes as living fossils. This placement is based primarily on characters related to a single feature, the enlarged mandibular canal, which supplies blood vessels and dense electrosensory receptors to the platypus bill. Our reevaluation of the morphological data instead groups platypus and echidnas to the exclusion of Teinolophos and Steropodon and suggests that an enlarged mandibular canal is ancestral for monotremes (partly reversed in echidnas, in association with general mandibular reduction). A multigene evaluation of the echidna–platypus divergence using both a relaxed molecular clock and direct fossil calibrations reveals a recent split of 19–48 million years ago. Platypus-like monotremes (Monotrematum) predate this divergence, indicating that echidnas had aquatically foraging ancestors that reinvaded terrestrial ecosystems. This ecological shift and the associated radiation of echidnas represent a recent expansion of niche space despite potential competition from marsupials. Monotremes might have survived the invasion of marsupials into Australasia by exploiting ecological niches in which marsupials are restricted by their reproductive mode. Morphology, ecology, and molecular biology together indicate that Teinolophos and Steropodon are basal monotremes rather than platypus relatives, and that living monotremes are a relatively recent radiation.

More than 99% of the ≈5,400 extant mammal species are therian (marsupials and placentals) (1). Monotremes, the only egg-laying mammals, are their living sister group and comprise just 5 extant species. One of these species is the semiaquatic, invertebrate feeding platypus (Ornithorhynchus anatinus) of eastern and southern Australia; the others are the terrestrial echidnas (Tachyglossidae), the short-beaked echidna or spiny anteater, Tachyglossus aculeatus of Australia and New Guinea, and three species of New Guinean long-beaked echidnas (Zaglossus bruijni, Z. attenboroughi, and Z. bartoni), which feed on worms and arthropod larvae. Fossil monotremes, such as Teinolophos trusleri and Steropodon galmani (2), along with their putative relatives, the insectivore-like ausktribosphenids (3–6), make up the bulk of the known Australian Cretaceous mammal fauna. Known monotreme diversity then contracts to only platypus-like taxa, subsequent to the arrival of marsupials from South America via Antarctica ≈71–54.6 million years ago (Ma) (7, 8). Monotrematum sudamericanum (9, 10) from the Palaeocene (≈61 Ma) of South America is known from two platypus-like distal femora and several molar teeth that closely match those of the extinct Australian platypus, Obdurodon (11, 12). The later appearance (≈25 Ma) of Obdurodon probably reflects the sparseness of earlier Tertiary mammal-bearing sites in Australia.

Fossil echidnas do not appear until the mid-Miocene (≈13 Ma) (13), despite excellent late Oligocene–Early Miocene mammal fossil records in both northern and southern Australia. This absence has tentatively been attributed in part to echidnas lacking teeth (14), which are the most common fossil remains from mammals. Alternatively, if the molecular dating studies that estimate the divergence of echidnas from platypuses at 17–35 Ma (15–22) are correct, then characters that clearly ally fossil taxa with echidnas would not be expected to have evolved until even more recently.

Molecular dating can play a pivotal role in inferring the evolutionary history of taxa with a sparse fossil record, such as monotremes (14, 23, 24). Recent reassignment of the 112.5–121 Ma Teinolophos and the ≈105 Ma Steropodon from outside of the monotreme crown group (platypuses and echidnas) specifically to the platypus lineage has profound implications (25). Teinolophos would be the oldest fossil unequivocally within any of the three mammalian crown groups (Monotremata, Marsupialia, and Placentalia). The fundamental morphological and ecological differences between platypuses and echidnas also would date back over 100 Ma, reinforcing the notion of monotremes being “living fossils,” a term that Darwin (26) first coined with reference to the platypus. Furthermore, it implies slow molecular evolution within monotremes that challenges current views of molecular evolutionary rates (25).

Upon revising Luo and Wible's (27) data set, Rowe et al. (25) assigned Teinolophos to the platypus lineage, based primarily on characters related to the mandibular canal. In the platypus, Teinolophos, and certain other fossil monotremes, the canal is enlarged; in the platypus it contains the hypertrophied mandibular branch of the trigeminal nerve, which supports an extensive mechanosensory–electrosensory system (28). The mandibular canal is narrower in echidnas than in other monotremes, although still relatively larger than in most other mammals. But whether the condition in echidnas is primitive for monotremes or a partial reversal correlated with the reduction of the mandible to little more than elongate splints of bone remains unclear. All support for the platypus affinities of Teinolophos derives from mandibular characters. This is surprising given that the initial description of Teinolophos (29), based largely on a mandible, aligned it with stem therians rather than with monotremes, let alone platypuses. Full exposure of the molar revealed monotreme affinities (30) but suggested that Teinolophos was a stem monotreme, diverging before the common ancestor of platypus and echidnas underwent a size increase and a dietary shift that required relatively weaker bite forces.

One concern with the morphological data set of Rowe et al. (25) is that one character (enlarged mandibular canal) overlaps with others (hypertrophy of the mandibular canal; characters 425 and 440) and is also used to infer states for unfossilized characters (presence of a bill and electrosensory capability; characters 423 and 424). We address this nonindependence issue by removing redundant characters and indirect inferences [see supporting information (SI) Text]. The support for grouping Teinolophos and Steropodon with platypuses disappears when these changes are made. Furthermore, with the addition of 2 other characters discussed in the original description of Teinolophos as a monotreme but not included in the data set Rowe et al. (25), Teinolophos and Steropodon fall outside the monotreme crown clade.

Most previous molecular dating (15–22, 31–36) places the divergence between living Monotremata (echidna/platypus)—and thus the platypus lineage—within the Tertiary (<65.5 Ma), contradicting placement of Teinolophos (112.5–121 Ma) and Steropodon (≈105 Ma) along the platypus lineage. But all of the aforementioned studies either assumed a constant-rate molecular clock or were based on a single gene (34). Applying a relaxed-clock approach to a data set including 5 nuclear genes (2,793 nucleotides) (23), Rowe et al. (25) recovered broad credibility intervals of 51.6–130.8 Ma, with the uppermost extreme overlapping the age of Teinolophos; however, this consistency has its foundations in imprecision rather than in signal, and is equally consistent with a Tertiary divergence. The molecular dating estimates from the 5–nuclear gene data set were derived using calibrations taken from previous molecular dating studies, a practice that is problematic (37). An additional concern is that the upper estimate reported for Monotremata of 130.8 Ma (25) cannot be considered in isolation from the upper estimate of 322.8 Ma for the adjacent monotreme–therian divergence. There often is a strong correlation between credibility intervals for adjacent nodes on trees (38), and the 322.4 Ma upper estimate for crown mammals appears implausible, being twice as old as the oldest unequivocal fossils (39) and predating even the earliest fully terrestrial vertebrates (amniotes).

With a view to increasing the precision of molecular dating estimates for the echidna–platypus divergence, we add long sequences from 2 other nuclear genes (Rag1 and apob) to make a 7–nuclear gene data set (7,137 nucleotides) and analyze this alongside complete mitochondrial (mt) genome sequences. We use relaxed-clock dating methods, including up to 20 prior distributions for calibrations derived directly from the fossil record. We first estimate the age of the echidna–platypus divergence with no age constraints on this node, then examine how bounding the age of this node with Teinolophos (i.e., assuming that it is a stem platypus) effects estimates of molecular evolutionary rates.

Results

Molecular divergence dates were derived using relaxed clocks and three alternative calibration schemes. The discussion that follows focuses on the scheme that incorporates the most calibrations (Table 1; see Materials and Methods), but dates derived using the other two schemes were very similar (Tables S3 and S4). All analyses indicate a mid-Tertiary origin for living Monotremata, the divergence between platypuses and echidnas. The 14-taxon combined nuclear (7,137 sites) plus mitochondrial (10,452 sites) data set, referred to herein as mtnuc₁₄, provides a median estimate of 32.1 Ma, with a 95% highest posterior distribution (HPD) of 18.5–47.8 Ma. As expected, the variances associated with the estimates for the individual nuclear and mitochondrial components are slightly larger, although the median estimates are similar. For nuc₁₄, this is 37.8 Ma. A median estimate of 27.7 Ma was provided by an expanded 88-taxon mitochondrial data set (mt₈₈), which also incorporates another 13 calibration prior distributions beyond the 7 used for the 14-taxon analyses. With both echidna genera included in mt₈₈, the median divergence estimate for Zaglossus–Tachyglossus is 5.5 Ma (95% HPD, 1.8–10.6 Ma).

Table 1.

Divergence age estimates from the BEAST analyses for Monotremata (platypus vs. echidnas), Tachyglossidae (Zaglossus vs. Tachyglossus), and Mammalia (Theria vs. Monotremata)

	Median	95% HPD
Monotremata
mtnuc14	32.1	18.5–47.8
nuc14	37.8	14.8–73.8
mt88	27.7	13.3–47.1
Tachyglossidae
mt88	5.5	1.8–10.6
Mammalia
mtnuc14	186.5	160.8–216.9
nuc14	203.4	163.5–252.2
mt88	168.8	145.8–196.2

The divergence times shown in Table 1 and Fig. 1 were inferred using BEAST (40). This Bayesian inference program has a number of desirable properties, including simultaneous phylogeny–branch-length–dating coestimation with incorporation of stochastic error, flexible models for calibration prior distributions, and relaxation of the assumption of rate correlation between adjacent branches. Nevertheless, BEAST assumes that rate variation follows a specific (appropriately plastic) lognormal distribution (41). Our results are not methodologically sensitive, however. The semiparametric, penalized likelihood method implemented by a different program, r8s (42), provided similar dates, with a best estimate of 31.3 Ma for the combined data set, mtnuc₁₄. Variance estimation in r8s is problematic (43), but even with a generous 4-lnL unit cutoff, the 95% confidence interval was 28.6–34.1 Ma.

Fig. 1.

Time scale of mammalian evolution (in Ma) as median (and 95% HPD) estimates from BEAST analyses for the combined nuclear and mitochondrial sequences (mtnuc₁₄). The (shaded) temporal window for the migration of marsupials into Australasia is 71–54.6 Ma. All nodes received Bayesian posterior probabilities of 1.00. Fossil calibrations were used at nodes indicated with an asterisk.

The analyses were repeated with the echidna–platypus divergence, enforced to be ≥ 112.5 Ma. This assessed the impact of assuming platypus affinities for Teinolophos (25) on rates of molecular evolution and echoes a recent study of substitution rates among reptiles (44). Fig. 2 shows substitution rates for platypus, echidna, and their (Monotremata) stem lineage relative to the ranges of rate estimates across sauropsids (reptiles and birds) and therian mammals, as well as the central 90% distribution across the tree (n = 26 for nuc₁₄, n = 174 for mt₈₈). The mitochondrial and nuclear patterns are similar. When the age of crown Monotremata is not constrained (Fig. 2 a and b), rates for all three monotreme branches fall well within both the therian range and the central 90% distribution, with the monotreme stem lineage rate close to the overall mean rate for both mt₈₈ and nuc₁₄.

Fig. 2.

Substitution rates in substitutions/1,000 sites/Ma for the platypus (black circles), echidna (dotted circles), and monotreme stem (white circles) lineages, estimated using BEAST, for mt₈₈ (a and c) and nuc₁₄ (b and d). The age of monotremes was either unconstrained (a and b) or bounded by the age of Teinolophos (c and d). The substitution rates for each monotreme branch are compared with ranges for therian mammals (right vertical bar), sauropsids (left vertical bar), the overall mean (horizontal bar), and central 90% distribution (shaded). The y-axis is log-scaled.

Applying a hard bound of ≥ 112.5 Ma to the age of crown Monotremata forces the substitution rates on the monotreme branches to opposite extremes of the distributions (Fig. 2c and d). For both mt₈₈ and nuc₁₄, the rate estimates (substitutions/1,000 sites/Ma) for the echidna (0.44 and 0.41) and platypus (0.33 and 0.26) branches are the lowest in the trees, well below the lowest among all other mammals (0.78 and 0.50), including those with similar putative life history correlates of evolutionary rates, such as body size, generation time, and mass-specific metabolism (e.g., xenarthrans, strepsirrhine primates, and many marsupials) (45, 46). They are even below those in the reptiles examined here, and even further below the mean rates for the tree (2.00 and 0.99). Conversely, the mt₈₈ and nuc₁₄ substitution rate estimates for the monotreme (echidna + platypus) stem lineage are very high (4.11 and 1.81), similar to those for murid rodents and at approximately the upper 5th percentile for the distribution of rates among all of the branches (Fig. 2c and d). The extraordinary (4.5- to 12.5-fold) rate decelerations from the monotreme stem lineage to the platypus and echidna lineages in the bound analyses are approximately double to triple the magnitude of the next-steepest deceleration across the entire tree, for both nuc₁₄ and mt₈₈.

The 63-taxon morphological data sets include modern (Ornithorhynchus, Tachyglossus) and fossil monotreme taxa from the Tertiary (Obdurodon) and the Cretaceous (Teinolophos, Steropodon). Modifications to the data set used by Rowe et al. (25), and the analyses that follow herein, are fully described in SI Text. These modifications involve the elimination of redundant characters and speculative (unfossilized) character-state codings (related to the width of the mandibular foramen/canal) from among characters 423, 424, 425, and 440. In doing so, the data set is reduced to 439 characters, and is referred to as morphol₄₃₉. The addition of 2 characters derived from the original description of Teinolophos as a monotreme (30), body size and mandibular aspect ratio, provides an alternative data set referred to as morphol₄₄₁. Alternative resolutions for Teinolophos and Steropodon relative to the modern platypus and echidna were tested using backbone constraints, as well as bootstrapping in PAUP*4.0b10 (47).

Morphol₄₃₉ produced 9 shortest trees (L = 1,790), all with both Teinolophos and Steropodon along the platypus lineage (Table 2); however, trees only a single step longer placed both Teinolophos and Steropodon outside an echidna–platypus clade (i.e., outside living monotremes). With the more character-inclusive morphol₄₄₁, support for the latter phylogeny increased; 20 shortest trees (L = 1,806) all placed both Teinolophos and Steropodon outside an echidna–platypus clade (Table 2). Two extra steps were required to place Teinolophos along the platypus lineage, and trees one step longer placed Steropodon along this lineage.

Table 2.

Maximum parsimony (shortest) trees that group the living platypus with echidna, Teinolophos (with or without Steropodon), and Steropodon (with or without Teinolophos)

	Platypus with echidna	Platypus with Teinolophos	Platypus with Steropodon
Morphol439
Unconstrained	1,7911	1,7902	1,7903
Australosphenida-constrained	1,7944	1,7985	1,7946
Morphol441
Unconstrained	1,8061	1,8082	1,8073
Australosphenida-constrained	1,8094	1,8155	1,8106

Backbone constraints were used to define these alternatives for morphol₄₃₉ and morphol₄₄₁, both with the remainder of the tree topology unconstrained or with monotremes constrained to group within Australosphenida (″Australosphenida-constrained″). Shortest trees for each of these 4 analyses are in bold; note that the first analysis placed both Teinolophos and Steropodon along the platypus lineage, to the exclusion of the echidna. Superscript numbers indicate the relevant constraint on relationships within monotremes satisfied by each tree (see SI Text).

Furthermore, all of the aforementioned trees united the Chinese “primitive triconodont” Hadrocodium with monotremes, contrary to the majority opinion (5, 6, 24, 27, 48) that monotreme affinities lie with the Gondwanan clade, Australosphenida. The original incarnation of the present morphological data sets (27) favored uniting monotremes with Australosphenida, specifically with ausktribosphenids, an Australian group contemporaneous with Teinolophos. It is notable, then, that even without any modification, reanalysis of Rowe et al. 's original matrix (25), constraining Australosphenida, causes Teinolophos to fall outside the echidna–platypus clade. Reanalysis of morphol₄₃₉ with monotremes constrained to group with australosphenidans revealed 24 shortest trees (L = 1,793), 12 of which excluded both Teinolophos and Steropodon from the echidna–platypus clade and 12 of which placed Steropodon on the platypus lineage (Table 2). Teinolophos was outside the echidna–platypus clade in all of these trees; the shortest trees that placed it on the platypus lineage were 4 steps longer. Similar analysis of morphol₄₄₁ also revealed 24 shortest trees (L = 1,808), all of which placed both Teinolophos and Steropodon outside an echidna–platypus clade. Seven extra steps were required to place Teinolophos along the platypus lineage, while trees a single step longer placed Steropodon on this lineage (Table 2).

Bootstrapping also reveals a consistent signal that Teinolophos is basal rather than a crown monotreme. Analyses of morphol₄₃₉ and morphol₄₄₁, with and without constraining Australosphenida all produce bootstrap consensus trees placing Teinolophos outside an echidna–platypus clade (Fig. 3A). Support for this arrangement ranges from 61% to 99%, being greatest with the additional characters and the Australosphenida constraint. Teinolophos falls outside the echidna–platypus clade in all 4 analyses; Steropodon is similarly positioned in 3 of the 4 analyses (Fig. 3B). Bootstrap support for all nodes increases when the phylogenetically unstable Steropodon is pruned from the primary bootstrap trees. Thus, there is clear evidence that Teinolophos is a basal monotreme, and weaker evidence for a similar position for Steropodon. It is notable that Steropodon is the least complete taxon and shares <3% of the characters in the matrices with both the echidna and either Ornithorhynchus or Obdurodon. As such, more complete material from Steropodon is needed to confidently resolve its position among other monotremes.

Fig. 3.

Maximum parsimony bootstrap trees for extant (echidna and platypus) and extinct [Teinolophos (>112.5 Ma), Steropodon (≈105 Ma), and Obdurodon (from ≈25 Ma)] monotremes, based on analyses in which monotremes were either unconstrained phylogenetically or constrained to group within Australosphenida. Bootstrap percentages are shown for 4 analyses: morphol₄₃₉ unconstrained/morphol₄₄₁ unconstrained/morphol₄₃₉ constrained/morphol₄₄₁ constrained. (A) The reduced bootstrap consensus after Steropodon is pruned. (B) The bootstrap consensus for all taxa. An asterisk indicates that the clade was enforced. A total of 58 nonmonotreme taxa are not shown.

Overall, the molecular divergence dating strongly indicates that the Cretaceous Teinolophos and Steropodon lie outside crown monotremes. The morphological data also favor this view, but less decisively.

Discussion

Teinolophos trusleri is the oldest known monotreme (121–112.5 Ma), as indicated by its distinctive molar structure (30). Although recently aligned with platypuses (25), earlier studies (6, 24, 27, 30) considered Teinolophos to be basal to crown (living) monotremes, consistent with its age and general morphology. Teinolophos resembles typical small insectivorous basal mammals (from which monotremes presumably evolved) in its small adult body size (≈40–180 g), deep twin-rooted molars, and deep mandibles with substantial coronoid and angular processes for strong muscle attachments. In contrast, platypuses and echidnas are derived with respect to this morphotype, being relatively large (≈700 g–18 kg) specialists on soft-bodied invertebrates, with shallow multiple-rooted molars (absent in adults of extant species) and shallow mandibles with small/vestigial coronoid and angular processes.

These inferred ecological shifts correspond well with the stratigraphic record and the phylogeny provided by our expanded morphological data set, which places Teinolophos basal to all other monotremes (Fig. 3). A >5-fold size (mass) increase occurs along the branch leading to Steropodon plus crown monotremes, and a further shift to relatively weak bite forces can be inferred along the branch leading to crown monotremes (platypuses and echidnas).

Our molecular clock analyses (Fig. 1) combine relaxed-clock methods, direct fossil calibration priors, and multiple genes for monotremes and are in full agreement with the morphological phylogeny and the foregoing ecological inferences. When the age of the monotreme crown (echidna–platypus divergence) is unconstrained, each of the highly parametric (BEAST) and semiparametric (r8s) molecular divergence best estimates place this divergence between 27 and 38 Ma, with a 95% HPD for the combined data of 19–48 Ma (Table 1). These dates strongly exclude the possibility that the ≥ 112.5 Ma Teinolophos and the ≥ 105 Ma Steropodon belong within crown monotremes. The estimate for the monotreme–therian divergence from the combined data (186 Ma; 95% HPD, 161–217 Ma) falls squarely within the Early Jurassic and is highly consistent with the stratigraphic record. Candidate stem monotremes and stem therians are both known from ≈166 Ma (39), whereas unquestioned outgroups to crown mammals, such as the morganucodontans, had diverged before the end of the Triassic (200 Ma). Furthermore, the rates of both mt and nuclear evolution in the monotreme stem and crown lineages fall comfortably within the variability across other land vertebrates, including therian mammals (Fig. 2 a and b).

When the analyses are repeated with the proposed 112.5 Ma echidna–platypus divergence imposed, the extensive amount of molecular change on the monotreme stem needs to be compressed into a short time, whereas the small molecular divergence between the platypus and echidna needs to be reconciled with a very lengthy interval. This results in inferred mitochondrial and nuclear evolutionary rates in stem monotremes that are among the highest in land vertebrates (Fig. 2c and d), comparable to the extreme rates in murid rodents, which have compromised mutational repair activity (49). Conversely, the inferred rates in branches leading to the platypus and to the echidna are extraordinarily slow, below those of all other amniotes considered (including reptiles). Slower rates have been associated with large body mass, long generation times, and low metabolism, although their effects typically explain only a small proportion of the variability (50–52). Living monotremes are not large and do not have long life cycles for mammals, and their mass-specific metabolism is similar to many of the other included mammals (45, 46) and indeed is higher than that of ectothermic reptiles.

It was previously recognized that a crown position for Early Cretaceous monotremes implies extraordinarily slow molecular evolution within living monotremes (25). But that study did not discuss how that phylogenetic hypothesis simultaneously (and paradoxically) also implies extremely rapid rates along the monotreme stem lineage. The implied rate decrease (fast in stem monotremes, slow in crown monotremes) for both the nuclear and mitochondrial sequences is 2- to 3-fold greater than decreases across any other node in the tree. A recent study of substitution rates among placental mammals reveals that dramatic rate decreases are far less likely than dramatic rate increases (53), perhaps because the mechanisms involved in DNA copying and repair are far easier to break than to fix.

The consilience of evidence from molecular dating, morphology, ecology, and the temporal order of monotreme fossils indicates that hypertrophy of the mandibular canal in Teinolophos (25) does not reflect close affinities with platypuses, but rather is a primitive monotreme feature, secondarily reduced in echidnas. This reduction could be correlated with functional constraints (e.g., stress tolerance) associated with the general reduction in echidnas of the mandibles to little more than elongate splints. In echidnas, the beak-like rostrum is used to lever soil (54), and the number of specialized electroreceptors is at least 10-fold fewer compared with the platypus (55). The relatively wider mandibular canal of Teinolophos is consistent with extensive mechanoreception–electroreception capabilities, but further evidence is required to confirm this and also to infer whether it is associated with platypus-like, mole-like, or other behavior.

The 61 Ma Monotrematum is the oldest known Tertiary monotreme, with teeth and femora very similar to those of undoubted fossil and living platypuses. The mid-Tertiary origin for crown monotremes inferred in all of our molecular analyses indicates that Monotrematum is a late stem monotreme, and thus that the immediate ancestors of living monotremes already exhibited a platypus-like morphology (9–11, 56). Therefore, echidnas essentially would be derived, terrestrial platypuses. Compelling evidence for secondary derivation of terrestrial habits from semiaquatic ancestors has been offered previously only for the evolution of elephants (57).

A number of aspects of echidna biology are consistent with an origin from a platypus-like ancestor with such traits as aquadynamic streamlining (58), dorsally projecting hind limbs acting as rudders (59), and locomotion founded on hypertrophied humeral long-axis rotation, which provides a very efficient swimming stroke (60). In echidnas, traits that are potentially homologous with these are dorso-ventral compression, reversed hind-foot posture, and “front-wheel drive” locomotion based on humeral long-axis rotation. Each of these traits would be highly anomalous if derived directly from a more generalized terrestrial insectivore morphotype (typical of basal mammals). The embryologic presence in echidnas of the marginal cartilage that contours the bill of platypuses (58) similarly suggests that a bill (rather than a beak or snout) is ancestral for crown monotremes. Other features of echidnas also suggest a substantial, relatively recent ecological shift. Despite now lacking teeth, relaxation of selection has yet to result in degraded sequences for the tooth matrix protein amelogenin (61). Similarly, the ankle spurs, which are venomous only in platypuses, are retained in many echidnas, despite their derived hind-limb morphology that ensures that they are nonfunctional (54). Finally, the absence of echidna-like fossils before 13 Ma is consistent with mid-Tertiary origins.

From the diversity of monotremes and other mammals (e.g., ausktribosphenids) that composed the Australian Early Cretaceous faunas, only the lineage leading to platypuses and echidnas demonstrably survived the marsupial invasion (Fig. 1). Competitive displacement by marsupials is consistent with surviving monotremes occupying ecological niches that may have been subject to the least competition from marsupials. Ausktribosphenids, which did not survive, appear to have been generalized insectivores (4, 29) ecologically similar to many Early Tertiary marsupials. Conversely, platypuses and echidnas exploit ecological niches in which marsupials are restricted by their reproductive mode. After birth, marsupial young begin an extended period of fixation onto a nipple. This both compromises the ability of the mother to forage in aquatic environments and provides an early developmental constraint on the evolution of an anteater-style “beak” (62). The only (remotely) ecologically analogous marsupials to the platypus and echidnas, the semiaquatic yapok and termite-eating numbat, are far more restricted in their aquatic and fossorial activities compared with their monotreme counterparts.

Monotremes and therian mammals diverged by the Early Jurassic; however, the popular notion of monotremes as “living fossils” is belied by such morphologically and ecologically distinct forms as platypuses and echidnas diverging only in the mid-Tertiary, more recently than the earliest divergences within most placental and marsupial orders (35). The semiaquatic-to-terrestrial niche shift and associated adaptive radiation of echidnas represents a recent expansion of niche space despite potential competition from marsupials. This contradicts the assumption of arrested morphological and molecular evolution that continues to be associated with monotremes (25). It further suggests that in certain niches, oviparous reproduction in monotremes confers advantages over marsupials, a view consistent with present ecological partitioning between monotremes and marsupials.

Materials and Methods

The primary molecular data set consists of 7 nuclear genes and complete mitochondrial (mt) genome protein-coding and RNA-coding DNA sequences for 7 placental mammals, 3 marsupials, both the platypus and short-beaked echidna, and 2 outgroup representatives (lizard and chicken). The nuclear genes are acetylcholinergic receptor M4 (chrm4), dopamine receptor type 1A (drd1a), α-2B adrenergic receptor (adra2b), proto-oncogene C-MOS (c-mos), sex-determining transcription factor SOX-9 (sox9), recombination-activating gene RAG1 (rag1), and apolipoprotein B (apob). The mitochondrial data set was expanded to include 88 taxa for analyses in which sampling was not limited to taxa available for the nuclear data set. All sequences were aligned in Se-Al 2.0a9 (63), and sites of ambiguous homology were excluded. GenBank accession numbers are provided in Table S1 and Fig. S1, and alignments are available online at TreeBASE.

Evolutionary models for the analyses of the combined data set (mtnuc₁₄) are partitioned to reflect the primary sources of genomic and functional variation among the sequences: mt first and second codon positions, mt third codon positions, mt RNA_stems, mt RNA_loops, nuclear first and second codon positions, and nuclear third codon positions. Within the nuclear data, 3 gene-wise partitions were used to ensure that genes with differing taxon sets were modeled separately. These were rag1 (complete taxon sampling), apob (lizard missing), and chrm4/drd1a/adra2b/c-mos/sox9 (elephant missing), within each of which codon position partitioning is maintained.

In accord with concerns for signal loss in mt sequences (64), we tested the relative ability of likelihood models to appropriately correct for multiple substitutions at sites (saturation). RY coding (A,G → R; C,T → Y) was required to appropriately correct for saturation among the mt protein-coding first and second positions and RNA_loops (see SI Text and Table S2). After RY coding, saturation remained problematic for the mt protein-coding third positions, and so these were excluded from our analyses. Standard nucleotide coding was retained for the nuclear sequences and the mt RNA_stems.

Model selection for the molecular data sets used ModelTest 3.7 (65), with the most general available model suggested by either the Akaike information criterion or likelihood ratio test used. The strict molecular clock hypothesis was rejected at P < .01 for each data set, under likelihood ratio clock tests (performed in PAUP*4.0b10) (47). Relaxed-clock divergence times and rates of evolution were estimated under Bayesian inference in BEAST 1.4.8 (40) and under penalized likelihood in r8s 1.70 (44); see SI Text for full details.

To provide temporal calibration, we used prior age distributions (in Ma) for 7 nodes on the 14-taxon nuclear and combined-data trees: Amniota (305–330.4), Sauropsida (255.9–299.8), Theria (124.6–167), Marsupialia (58.5–84), Australasian marsupials (25.5–71.2), Tethytheria (52–71.2), and Cow-pig (52.4–65.8). Prior age distributions were used for a further 13 nodes on the 88-taxon mitochondrial trees: Xenarthra (58.5–71.2), Primates (55–71.2), Rodentia (55.6–71.2), Shrew-hedgehog (61.5–101.5), Perissodactyla (54–58.7), Carnivora (39.7–65.8), Whippomorpha (49–61), Dasyuromorphia (23–45.6), Galloanserae (66–86.5), Penguin-albatross (61–74), Petauroidea (25.5–54.6), Alligatoridae (64–80), and Archosauria (239–250.4). Three alternative calibration schemes were used. The above scheme, which uses the most calibrations, was devised by M.J.P. To test the robustness of the dating estimates to calibration variability, two additional calibration schemes were used, one devised by M.S.Y.L. and the other taken directly from a recent review of calibration points by Benton et al. (66). Fossil record justifications and references for the primary and alternative calibration schemes are provided in SI Text. Notably, median estimates among the 3 calibration schemes for the echidna/platypus divergence differ by only 0.1–3.5 Ma across the mtnuc₁₄, mt₈₈, and nuc₁₄ data sets (see SI Text and Table S4).

Analyses in which Teinolophos was used for calibration adopted the results of Rowe et al. (25) for timing. The minimum and maximum hard bounds were the minimum stratigraphic age of Teinolophos (112.5 Ma) and the upper confidence interval of their molecular estimate (130.8 Ma).

Maximum parsimony trees from the morphological data sets (morphol₄₃₉ and morphol₄₄₁) were inferred under 100 random addition heuristic searches in PAUP 4.0b10. Backbone constraint trees used for evaluating alternative hypotheses of relationships are provided in SI Text. Bootstrap support for clades was based on 1,000 replicates.