Vicariant speciation resulting from biogeographic barriers in the Australian tropics: The case of the red‐cheeked dunnart (Sminthopsis virginiae)

Abstract Global biodiversity loss continues unabated, and in Australia, the rate of recent mammal extinctions is among the worst in the world. Meanwhile, the diversity among and within many endemic mammal species remains undescribed. This information is crucial to delineate species boundaries and thus inform decision‐making for conservation. Sminthopsis virginiae (the red‐cheeked dunnart) is a small, dasyurid marsupial found in four disjunct populations around the northern coast of Australia and New Guinea. There are three currently recognized subspecies, each occupying a distinct geographic location. Sminthopsis v. virginiae occurs in Queensland, S. v. rufigenis is distributed across New Guinea and the Aru Islands, and S. v. nitela has populations in the Top End of the Northern Territory and the Kimberley region of Western Australia. Previous molecular work has suggested the current subspecies definitions are not aligned with DNA sequence data, though the sampling was limited. We undertook a comprehensive genetic and morphological review of S. virginiae to clarify relationships within the species. This included mitochondrial (CR, 12S, and cytb) and nuclear (omega‐globin, IRBP, and bfib7) loci, and morphometric analysis of skulls and whole wet‐preserved specimens held in museums. Maximum Likelihood and Bayesian phylogenetic analyses resolved samples into two distinct clades, demarcated by the Gulf of Carpentaria in Australia's north. Sminthopsis. v. nitela was consistently separated from S. v. virginiae and S. v. rufigenis, based on the overall body and skull size and craniodental features, while S. v. virginiae and S. v. rufigenis were more difficult to distinguish from each other. Thus, we redescribed S. virginiae, recognizing two species, S. nitela (raised from subspecies) and S. virginiae (now comprising the subspecies S. v. virginiae and S. v. rufigenis). This study highlights the importance of recognizing cryptic mammal fauna to help address the gap in our knowledge about diagnosing diversity during a time of conservation crisis.


| INTRODUC TI ON
Rates of global biodiversity loss continue to accelerate in the 21st century as human impacts on natural ecosystems worsen.In Australia, the anthropogenic impacts on biodiversity have been profound.Changes brought about by European colonization have wrought devastation on various flora and fauna, in particular a suite of endemic, unique, and isolated mammals (Woinarski et al., 2015).
Currently, 149 mammal taxa and populations are listed as threatened under the Australian Environment Protection and Biodiversity Conservation Act 1999 (Department of Climate Change, Energy, Environment and Water, 2024), 40 of which are considered to be extinct (Burbidge, 2023).More recently, declining mammal populations have been documented in regions previously thought to be relatively undisturbed, such as northern Australia (Cremona et al., 2022;Woinarski et al., 2011), which is of particular concern as species in this region are typically understudied.Documenting the diversity of Australian mammals is now critical as the cumulative effects of multiple threats are realized, such as eroding natural habitats coupled with widespread bushfires and predation by introduced species such as feral domestic cats (Felis catus) and European red foxes (Vulpes vulpes).Among Australia's mammal fauna are the carnivorous dasyurid marsupials, comprising about 61 species (AMTC, 2023), many of which are restricted to peripheral or sensitive habitats and predicted to continue to decline during this century (Krajewski et al., 2024).The subfamily Sminthopsinae comprises almost half of the extant dasyurid species, and the majority of these belong to the genus Sminthopsis, one of only four currently recognized genera.Three of these: Antechinomys (two species), Ningaui (three species), and Sminthopsis (18 species) compose the tribe Sminthopsini; the fourth (Planigale) constitutes the tribe Planigalini (Krajewski et al., 2024).Although both morphological and molecular studies have consistently shown the monophyly of both tribes, several phylogenetic issues are in need of resolution, including the monophyly of Sminthopsis as currently recognized (Krajewski et al., 2012).Blacket et al. (2001) summarized the history of phylogenetic studies of Sminthopsins until that time and demonstrated that the precise interrelationships, and even the composition, of the three currently recognized genera were unclear since Antechinomys and Ningaui rendered Sminthopsis paraphyletic in all molecular studies.This, and all subsequent studies (Blacket et al., 2006;García-Navas & Westerman, 2018;Kealy & Beck, 2017;Krajewski et al., 2012;Westerman et al., 2016) have consistently demonstrated that the long-tailed dunnart, S. longicaudata, is sister to Antechinomys laniger, while the remaining species of Sminthopsis belong to one or other of two distinct species groups: "macroura" and "murina."Following analysis of DNA sequences from both mitochondrial and nuclear genes, Westerman et al. (2023) suggested that the long-tailed dunnart (previously S. longicaudata) be recognized as a second species of Antechinomys (now A. longicaudatus).Krajewski et al. (2012) extended the existing molecular dataset for Sminthopsins and incorporated some new characters based on penis morphology, demonstrating that the "macroura" species group was distinguishable not only by their DNA sequences but also by all members sharing a unique phallic morphology (Form 1, Woolley et al., 2007).However, status and interrelationships of some of the species currently recognized in this group (Sminthopsis crassicaudata, S. bindi, S. douglasi, S. macroura, and S. virginiae) remain unclear.
Although some species such as S. bindi and S. douglasi have relatively limited distributions, others are widespread across northern and central Australia and show variability not only in their external morphology but also in DNA sequence and allozyme (isozyme) profiles across the species' range (see Archer, 1981;Blacket et al., 2001;Umbrello et al., 2020).Some of these distinctive features have been used to designate currently recognized subspecies: two for S. crassicaudata (S. c. crassicaudata and S. c. centralis); at least three for S. macroura (S. m. macroura, S. m. froggatti, S. m. stalkeri) and three for S. virginiae (S. v. virginiae, S. v. nitela, and S. v. rufigenis).Some of the currently recognized subspecies were formerly recognized as distinct taxa before Archer (1981) synonymized them in his major review of the genus Sminthopsis.Thus, S. virginiae specimens collected from Cape York Peninsula, Queensland, were originally described as Phascogale virginiae de Tarragon, 1847; specimens from the Daly River region of the Northern Territory were named Sminthopsis nitela Collett, 1897; and specimens from the Aru Islands and the Trans Fly region of southern New Guinea were described as S. rufigenis Thomas, 1922.The genetic variation within Sminthopsis virginiae has been investigated in only one study (Blacket et al., 2001).Two genetic clades were resolved, Western Australia (WA) + Northern Territory (NT) and Queensland (Qld) + Papua New Guinea (PNG), using two mitochondrial loci (12S and Control Region [CR]) but only a small number of individuals.The authors concluded the divergence between the two clades was similar to that observed between other species of Sminthopsis, and as such, S. v. nitela was genetically distinct enough to warrant raising to species.In Archer's (1981) review of the genus Sminthopsis, he stated that "other than by its larger size" S. v. virginiae could not easily be distinguished from S. v. nitela.However, it differed from S. v. rufigenis in having females with eight nipples, instead of six, and for being generally larger and possessing a larger alisphenoid tympanic wing on the skull.Sminthopsis v. nitela differed from both other subspecies by smaller body size and the lack of an anterior cingulum on the upper molar teeth.However, Archer's (1981) investigations were muddied by the sample of "nitela" from NT including S. butleri (at the time, the latter was only known from WA), the inclusion of an undescribed S. bindi and museum specimen NTMU4340 (which was later In this study, we investigate the interrelationships between the three currently recognized subspecies of Sminthopsis virginiae, greatly expanding the DNA sequence dataset and conducting a detailed morphological analysis of samples across the known geographic range of the species.

| Study species
Sminthopsis virginiae is a small (18-60 g), predominantly insectivorous marsupial that occurs in disjunct populations across northern Australia, southern New Guinea, and the Aru Islands (Figure 1).
They prefer tropical forest and savannah woodlands with heavy soils in swampy and riparian areas with dense vegetation (Bradley et al., 1987;Braithwaite & Lonsdale, 1987;Tate & Archbold, 1941).
The species' common name, the red-cheeked dunnart, alludes to the rufous fur patches on the sides of the head and neck, which distinguish it from all other Sminthopsis species, but the extent of this facial coloration and overall dorsal fur color is variable.Each of the three currently recognized subspecies occupies distinct, non-overlapping geographic regions (Krajewski et al., 2024;Woolley, 2023) (Figure 1).Differences in the nipple number of females, six in S. v. rufigenis and eight in S. v. virginiae and S. v. nitela, the presence or absence of an anterior cingulum on the upper molar teeth, and overall body size, are thought to be among the few diagnostic characters that can be used to distinguish between the three subspecies (Archer, 1981).

| Specimens sampled
In addition to two exemplars of each of the currently recognized species included in the "macroura" species group of Sminthopsis, we included 58 specimens of each recognized sub-species of S. virginiae to clarify genetic relationships within this species.
Specimens sampled are shown in Figure 1, and sample details and genes sequenced are listed in Appendix A; Table A1. Figure A1 F I G U R E 1 Map showing the distribution of Sminthopsis virginiae records (white circles) and samples used in this study (filled shapes) in northern Australia and southern Papua New Guinea (PNG), along with the localities from which type specimens were collected (yellow stars).The approximate species distribution is shown as shaded areas, colored by each subspecies.The approximate location of biogeographic barriers is shown with gray rectangles and labeled, as is the approximate location of the historical Lake Carpentaria (demarcated by dotted lines).Photos: S. v. nitela, Anders Zimny; S. v. virginiae, Eric Vanderduys; S. v. rufigenis, Pat Woolley.
shows on a map which samples were used for molecular and morphological studies.

| Molecular methods
Total genomic DNA was extracted from museum preserved liver, muscle, and ear tissue samples cryo-preserved at −80°C or stored in 70%-100% ethanol using Qiagen DNeasy® tissue and blood kits according to manufacturer's instructions, and genetic loci were amplified as described in Umbrello et al. (2017)  for IRBP (see Table A1)."Outgroups" for the molecular studies included the representatives of each species of the "macroura" group.
Data were separated into six gene partitions, three nuclear (IRBP, ω-globin, and bfib7) and three mitochondrial (cytb, 12SrRNA, CR) regions, each with its own model of sequence evolution as determined by jModeltest (Posada, 2008).In addition, we further partitioned protein-coding genes into their three codon positions (first, second, and third), and the two rRNA genes into stems and loops, to determine whether this affected resolution in the phylogenetic tree and if the topology of this tree differed from others.Branch lengths were not partitioned in our analyses.The general timereversible model with gamma-distributed rates and a fraction of invariable sites (GTR + G + I) was used for all partitions in RAxML.
For Bayesian analysis, best-fit models (see Table A2) were chosen using the Akaike Information Criterion as implemented in jModeltest 2.1.7(Posada, 2008).Node support was estimated by 1000 bootstrap pseudoreplications for RAxML.Bayesian analyses utilized random starting trees and two simultaneous runs of four Markov chains (one cold and three heated, using default heating values) applied for 5 × 10 6 generations, with sampling every 1000th generation.The first 1.25 × 10 6 generations were discarded from each run as burn-in.The remaining trees were used to construct a maximum clade credibility tree; posterior probabilities (PP) >0.95 were deemed as strong support and PP = 0.90-0.95as moderate support (see Kolaczkowski & Thornton, 2006).User-trees were tested using PAUP* (Swofford, 2003).
Average sequence divergence levels among samples of Sminthopsis virginiae and the outgroup taxa were determined using PAUP*.Kimura two-parameter (K2P) corrected distance values were calculated for all positions in cytb and transversions only for 12S (Kimura, 1980).Using these results, divergence estimates were calculated for cytb based on the estimated divergence from Krajewski et al. (2000) of 2.1% per million years for Sminthopsis which assumed a 25 million year date for the divergence of thylacinids and dasyurids, and for 12S following the method outlined in Springer et al. (1997) where divergence time (My) = (% divergence -0.0584)/0.0854where % divergence is the K2P value (for transversions only) in the 12S rRNA gene.
In this study, we have adopted the Unified Species Concept of De Queiroz (2007) which proposes reciprocal monophyly as providing evidence for species boundaries, and we tested for reciprocal monophyly using the Species Delimitation plugin in Geneious Prime (Masters et al., 2011).This tool assesses the strength of the phylogenetic evidence of species boundaries by applying Rosenberg's P AB statistic (Rosenberg, 2007), which was developed for calculating the probability of reciprocal monophyly under the null model of random coalescence.Rosenberg's P AB statistic is the probability that A putative species represented by a taxa will be monophyletic relative to a sister clade containing b taxa under the null model of random coalescence.
To investigate broad-scale phylogeographic patterns in the sequence data, statistical parsimony TCS haplotype networks (Clement et al., 2000) were constructed using the software program PopART (Population Analysis with Reticulate Trees; Leigh & Bryant, 2015) where each subspecies was well represented by the sequence data.
To obtain haplotypes for the nuclear ω-globin sequence data where individuals were heterozygotes we used the program PHASE v. 2.1.1 (Stephens et al., 2001) and created input files using the webtool SeqPHASE (Flot, 2010).We ran PHASE using default settings and only included phased haplotypes with confidence scores greater than 90%.PopART excludes gaps and unknown bases (N, ? and -) from the analysis, so these regions were removed from the alignments before building networks.Networks were built for each loci separately due to different individuals having coverage for different loci, with samples colored based on population location: WA, NT, Qld, or PNG.

| Morphology
To assess the morphological variation among Sminthopsis virginiae subspecies, 93 vouchered specimens (which included 33 skulls, 16 skins, and 68 wet preserved specimens) were examined from the following institutions: Australian National Wildlife Collection, CSIRO (ANWC); Queensland Museum (QM); Western Australian Museum (WAM), the Museum and Art Gallery of the Northern Territory (MAGNT) and the Natural History Museum at the University of Oslo (NHMO) (see Figure A1 for the geographic distribution of material examined).Additionally, photographs of type specimens were examined from the Natural History Museum, UK and the Natural History Museum at the University of Oslo, Norway.Skulls and dentition were examined to identify and corroborate diagnostic features as described in Archer (1981).Similarly, pelt coloration was examined on dry skins and wet preserved specimens to identify any differences between the subspecies.Female spirit specimens were examined for pouch development, nipple number, and the presence of pouch young.
Tooth numbering followed Luckett (1993), and anatomical terminology of teeth and cranial features followed Beck et al. (2022).
Skull measurements are illustrated in Figure 2 and were taken from Archer (1981), Baker et al. (2015), and Umbrello (2018).The measurements used were as follows: ATL, alisphenoid tympanic process length; ATW, alisphenoid tympanic process width; BL, basicranial skull length, excluding incisors; CW, combined width of occipital condyles; DH, dentary height measured between M 2 and M 3 ; DL, dentary length, excluding incisors; FL, length of frontal midline suture; IBW, minimum width between auditory bullae; IFL, length of incisive foramen; IFMF, distance between the incisive foramen to the maxillopalatine fenestra; IOW, minimum width of interorbital constriction; LIML, maximum length of lower tooth row from I 1 to M 4 ; LML, maximum length of lower molar row, M 1-4 ; LPL, crown length of lower premolar row P 1-3 ; M 2 W, maximum width of upper M 2 , measured at crown; M 3 W, maximum width of upper M 3 , measured at crown; MFL, length of maxillopalatine fenestra; MW, maximum width across braincase; NL, maximum length of nasals; NWA, anterior width of nasals; NWP, width of nasals at the nasal/maxilla/ frontal junction; OBW, maximum outer width of auditory bullae; PML, length of premaxilla, measured as length of sutural contact between nasal bone and premaxilla; RWA, rostral width, from the outer edge of each upper canine; RWP, rostral width, across the medial surface of each infraorbital foramina; UCML, maximum length of upper tooth row from canine to M 4 ; UIL, maximum length of upper incisor series, I 1-4 ; UML, maximum length of upper molar row, M 1-4 ; UPL, crown length of upper premolar row P 1-3 ; ZW, maximum zygomatic width.The following external measurements were taken from preserved (wet) specimens: head-vent length, taken from the tip of the nose to the vent; tail-vent length, from the vent to the tip of the tail; ear length, measured across the longest part of the ear from the tip to the notch; pes length, from the heel to longest toe tip (minus claw); and head length, from the tip of the nose to the back of the cranium.All skull and external measurements are given in mm.
Skulls were categorized as adults if M4 was fully erupted and the deciduous premolar, dP3, had been lost and replaced by a fully emergent P3 (Archer, 1981).Individuals that did not fall into the were grouped for all analyses; while males are generally larger than females for many dasyurid species, the differences are slight for the smaller members of the genus (see body sizes reported in Baker & Gynther, 2023).
All measurements were recorded by hand with digital calipers to 0.01 mm (Table A3 and File S1), and statistical analyses were carried out in the computing program R (R Core Team, 2013) using a custom script that incorporated elements of the "MorphoTools2" package (Šlenker et al., 2022).Univariate statistics and analysis of variance (ANOVA) were conducted on the three subspecies for each measurement variable where data were normally distributed and homogeneity of variances tested with Bartlett's test (Bartlett, 1937).
One variable (ATL) did not have equal variances, so a nonparametric Kruskal-Wallis test (Kruskal & Wallis, 1952) was run instead.
Pairwise comparisons for significant ANOVA results were obtained with Tukey's post hoc test (Tukey, 1949) with correction for multiple comparisons.This allowed us to identify which measurements were most informative for identifying diagnostic features between the subspecies.
For multivariate analyses, variables with greater than 20% missing data were removed; this cut-off was chosen to reduce the amount of missing data while maximizing the number of measurement variables retained in the dataset.Any individuals with more than 15% missing data after this initial step were subsequently removed (these were ANWC M29861, 28% missing, MAGNT U3769, 20% missing, and WAM M4056, 16% missing).We accounted for the multicollinearity of variables via Pearson correlation coefficients, and JL and ZW were removed due to being highly correlated (>90%-95%) with BL.Hierarchical clustering using Euclidean distance and the UPGMA algorithm was conducted on the remaining dataset to see if the morphological distance was clustered according to subspecies.
The remaining missing data (only three variables M 2 W, M 3 W, and DH for one individual, QMJ15890) were imputed using predictive mean matching in the "MICE" package (Buuren and van Buuren & Groothuis-Oudshoorn, 2011) adopting the default settings, except that we increased the number of iterations from five to 50.To visualize the imputed craniodental measurement data, we conducted a Principal Component Analysis (PCA) using the pca.cor function in "MorphoTools2".In morphological datasets, a high proportion of the variance explained by PC1 is usually due to size (Berner, 2011;Jolicoeur, 1963), and accounting for the effect of size has been shown to help distinguish between morphologically cryptic dasyurid species as it allows for comparison on skull shape (Umbrello et al., 2023;Viacava et al., 2023).To account for size effects (specifically caused by isometric scaling) between species, we used log-shape ratios by calculating the geometric mean of all variables, per individual, as a proxy for skull size, dividing each variable by the geometric mean, then log-transforming this ratio (Mosimann, 1970;Onley et al., 2022;Viacava et al., 2023).The PCA was then re-run on the size-corrected data.
Differences in overall skull size between the subspecies were tested using ANOVA on the geometric means.This was followed by a Tukey's post hoc test to check the pairwise differences between groups with p-values adjusted for multiple group testing.To test how well individuals could be placed into the correct subspecies grouping, we used nonparametric k-nearest neighbor classificatory Discriminant Analysis due to the small sample sizes of Sminthopsis v. rufigenis and S. v. virginiae, nonequal covariances and deviation from multivariate normality.We ran this analysis on four different sets of data to assess the effect of size correction on correctly classifying specimens to subspecies and whether the genetic clades were better at diagnosing specimens than the current subspecies hypotheses.

| Molecular analyses
Our Maximum Likelihood and Bayesian analyses of the partitioned concatenated datasets provided a well-resolved, monophyletic group comprising all specimens of Sminthopsis virginiae (Figure 3 The topology of this tree did not differ from the single gene trees (Figure A2), as all showed a separation of Sminthopsis virginiae samples into two distinct clades: one comprising specimens from the Northern Territory and Western Australia ("nitela") and one comprising those from Queensland and Papua New Guinea ("virginiae" + "rufigenis").Within this latter clade, there was no obvious distinction between exemplars of the two currently recognized subspecies, and a user-tree specifying such a separation was a significantly worse fit to the data (Table 1).
We evaluated the fit of our data to alternative phylogenetic relationships using the SH (Shimodaira & Hasegawa, 1999) and KH (Kishino & Hasegawa, 1989) tests implemented in PAUP*4.0b10(Swofford, 2003).These included: (1) that the currently recognized subspecies "virginiae" (Qld) and "rufigenis" (PNG) were distinct entities (i.e., reciprocally monophyletic), and (2) that within "nitela," Northern Territory individuals were genetically distinct from Kimberley (WA) ones, as might be expected if the Ord Basin was effective in separating the two lineages over time.The results (Table 1) clearly show that the first of these maximum likelihood trees is a significantly worse fit to the data (p < .001***for KH, wtd KH, SH, wtd SH, and AU tests) than the tree shown in Figure 3, suggesting that

| Mitochondrial divergence
To further test the differences between currently recognized species and subspecies of Sminthopsis virginiae, we examined their differentiation using Kimura 2 Parameter (K2P) corrected distances derived from cytb gene sequences (Table 2).Corrected pairwise K2P differences between S. virginiae and the outgroup Sminthopsis species ranged from ~14%-20%.Within S. virginiae, K2P values for cytb were lower between S. v. rufigenis and S. v. virginiae (1.2%) compared to S.
v. nitela (8%-8.2%),and values between the WA and NT populations of S. v. nitela (1.4%) were comparable to those between the eastern subspecies (1.2%).These correspond to average divergence times of ~3.9 million years (range = 2.88-5.11)for the Eastern and Western virginiae clades, and ~ 0.6 million years within each population, based on a 2.1% divergence per million years for dasyurids at cytb (Krajewski et al., 2000).For 12S, the K2P values were much smaller, being calculated based on transversions only, following the method described in Springer et al. (1997).The 12S K2P values reveal similar estimates, with divergences between the NT and WA, and Qld and PNG groups approximately 0.6 million years, and divergence between the Eastern and Western clades occurring 1.8-3.9 million years ago (where divergence time (My) = (% divergence -0.0584)/0.0854).

| Species delimitation
The species delimitation analysis showed that nitela consistently formed a monophyletic clade, as did the combined rufigenis + virginiae samples, but not rufigenis nor virginiae on their own (Table 3).The ratio of intra-distance versus inter-distance showed low levels of genetic differences within the nitela sample relative to rufigenis + virginiae, but higher levels of within sample differences between rufigenis and virginiae, and for 12S these were equal to the differences between the two samples.Similarly, the probability of sorting samples of nitela into the correct clade was much higher (81%-95%) than sorting samples of rufigenis and virginiae (43%-79%).The p-values for Rosenberg's P AB were all significant for the nitela versus rufigenis + virginiae clades.

| Haplotype network analyses
The length of the fragments used in the 12S network analysis was 567 bp, for CR it was 386 bp, for cytb, it was 389 bp, and for ωglobin, it was 757 bp.All four networks resolved the samples into two groups consistent with the phylogenetic analyses (Figure 4; Figure A3).Some haplotypes were shared between NT and WA populations of Sminthopsis v. nitela in the 12S, CR, and ω-globin data, and shared haplotypes were present between S. v. rufigenis and S.

| Cranial morphology
All variables were normally distributed for each subspecies, with the largest measurements recorded for the Sminthopsis v. rufigenis specimens (Table 4).One-way ANOVAs showed significant differences between subspecies, except for the length of the nasals (NL and PML) and width of the auditory bullae (ATW and OBW) (see Table A4).Of these, 22 of the 27 measurements tested showed significant differences between S. v. rufigenis and (see Figure 5a).Principal Components Analysis (PCA) showed over half of the variation in the data (55%) was explained by PC1 with PC2 explaining 9% (Figure 5b); 70% of the variation in the data was explained by only the first three PCs.Almost all measurement variables overwhelmingly contributed to PC1 and indicated size, with greater PC1 scores = larger skull size.PC2 indicated the length and width of the auditory bullae measurements (Figure A4).Analysis on skull size (as the geometric mean) also showed that S. v. nitela is significantly smaller than the other two subspecies (ANOVA p-value = .00002),TA B L E 3 Results from the species delimitation analysis run in Geneious on the concatenated tree and single gene trees with sufficient sample sizes: Cytb, 12S, and omega globin.and S. v. rufigenis and S. v. virginiae could not be separated by size (pvalue = .84)(Figure 5c).
Removing size as a variable from the remaining data and rerunning the PCA did not improve the separation of the three subspecies in ordinal space, but 70% of the variation in the data was explained by the first 6 PCs with variables contributing more evenly to the ordinal space (Figure A4).K-nearest neighbor discriminant analysis was slightly more successful at correctly classifying individuals to subspecies in the raw data (80% correct) than the dataset with size removed (76.7% correct), but it performed better at classifying the Sminthopsis v. virginiae samples when size was removed (Table 5).Overall, the analysis performed better when individuals were grouped according to the genetic clades, "nitela" versus "rufigenis" + "virginiae" with the raw data   Archer (1981) proposed that Sminthopsis v. nitela could be distinguished from S. v. rufigenis and S. v. virginiae on size and the absence of an anterior cingulum on the upper molars.We examined specimens for the presence of an anterior cingulum as well as the presence, shape, and degree of development of other molar tooth cusps.We found an anterior cingulum to be present on all upper molars (M 1-4 ) in all individuals from the three subspecies (except WAMM4056), but the size differed.In S. v. nitela, the anterior cingulum on M 1 tended to be reduced compared to S. v. rufigenis and S.

| Qualitative cranial examinations
v. virginiae.An anterior cingulum was present on M 2-4 for all specimens examined.However, we observed several other characters that more consistently differed between the subspecies.The protoconule on M 1 tended to be well-formed in S. v. rufigenis and S.
v. virginiae, whereas it was reduced in S. v. nitela.Where visible in the absence of dental wear, the stylar cusp B on M 1 varied in size and connected to the paracrista in S. v. nitela, but it was always large and did not connect to the paracrista in S. v. rufigenis and S. v. virginiae.
The entoconid on M 3 tended to be larger in S. v. rufigenis and S. v. virginiae when compared to S. v. nitela.The posthypocristid on M 4 was absent in all S. v. nitela specimens examined, whereas it was present and well-defined in S. v. rufigenis and S. v. virginiae, where visible in the absence of tooth wear.A postorbital process was absent from all S. v. nitela specimens except ANWCM29861 and ANWCM8861, whereas it was present in almost all S. v. rufigenis and S. v. virginiae examined.

| External morphology
As with the cranial dataset, Sminthopsis v. nitela was found to be significantly smaller than both S. v. rufigenis and S. v. virginiae for all measurements (Figure 6, Table 6), whereas S. v. rufigenis and S.
v. virginiae could not be distinguished based on size.Individuals from New Guinea (S. v. rufigenis) had the largest pes (hindfoot) and head length, two variables less prone to distortion on wet preserved specimens, but this was not significantly different to individuals from Queensland.Examination of dry study skins from each subspecies (S. v. nitela = 9, S. v. rufigenis = 3, and S. v. virginiae = 1) showed no obvious or unique differences in pelage color, tail thickness, or footpad morphology (Figure 7).Slight variations in the depth and spread of the facial russet coloration and the width and continuity of the central facial stripe were observed across specimens both within and between subspecies.Variation in the robustness of individuals was observed across the species distribution, with the stockiest individuals being from PNG and Qld, with the NT also having some large, robust males.In contrast, most individuals from WA were smaller, with more gracile snouts.

| DISCUSS ION
In this study, we have undertaken the first combined morphological and genetic review of the variation within the tropical dasyurid marsupial species, Sminthopsis virginiae.Building on the work of Blacket et al. (2001) and Archer (1981) we have examined and sequenced multiple exemplars from across the species' distribution to comprehensively assess the diversity within the taxon.Our results consistently resolved the species into two morphologically and genetically distinct clades: one comprising S. v. nitela individuals from Western Australia and the Northern Territory, the other comprising exemplars of both S. v. virginiae (Queensland) and S. v. rufigenis (Papua New Guinea).

| Molecular divergence in Sminthopsis virginiae
Using allozyme loci, Blacket et al. (2001) found fixed differences between subspecies "nitela" (n = 10) and "rufigenis" (n = 6), but these were less clear-cut and using a much smaller sample than that sequenced here.Using additional mitochondrial and nuclear gene loci, we have corroborated the genetic distinctness of S. v. nitela from the other two currently recognized subspecies of S. virginiae.In all of our phylogenetic analyses in which individual genetic loci were weighted appropriately, or in which 3rd codon positions of proteincoding genes were weighted and the mitochondrial 12S rRNA locus was partitioned as Stems and Loops, the resulting tree always resolved two distinct clades: one comprising all Northern Territory and Western Australian specimens ("nitela") and one comprising the Queensland ("virginiae") + Papua New Guinean ("rufigenis") specimens, of which the latter two subspecies were intermingled rather than forming two distinct lineages.Our results suggest that although Queensland and Trans-Fly Plains (PNG) populations of S. virginiae have been separated from one another by the formation of Torres Strait ~9000 years ago, genetically they are indistinguishable and are not monophyletic entities.In contrast, the user-tree positing genetic differences between NT and WA samples of "nitela" was not significantly different from the best tree (Table 1).
The multiple exemplars of the three currently recognized subspecies in our data thus suggest a marked genetic divergence between Sminthopsis v. nitela (Kimberley and Northern Territory) and S. v. virginiae (Queensland)/S.v. rufigenis (Papua New Guinea).These "western" and "eastern" geographic groupings are separated from one another by the Carpentaria Barrier or Basin (Edwards et al., 2017), an area of unsuitable habitat for multiple species in northern Australia and a well-known biogeographic barrier (Catullo et al., 2014;Cremona et al., 2021;Eldridge et al., 2014;von Takach et al., 2022).Due to the mitochondrial divergence (~8.2% K2P differences) reported here, we hypothesize that S. v. nitela diverged from their Cape York (Qld) and PNG conspecifics ~3.9 million years ago (ma) in the early-mid Pliocene, assuming a rate of 2.1% divergence per million years for dasyurids at cytb (Krajewski et al., 2000).We acknowledge that divergence estimates calculated on single locus data may not be reflective of the true population divergence time and that it is difficult to obtain robust divergence times for Sminthopsins due to a lack of suitable fossils available in this group (Krajewski et al., 2024).The rate we have used for cytb assumes a divergence between thylacines and dasyurids of 25 mya; however, more recent studies have estimated that divergence occurred ~40 mya (Mitchell et al., 2014;Westerman et al., 2016), which suggests our divergence values are conservative.
We noted much smaller genetic differences between Sminthopsis v. nitela populations from the Kimberley and those from the Northern Territory, suggesting a much more recent divergence (0.67 ma;

| Connection across the Torres Strait
In contrast to the marked genetic differences observed between western "nitela" and eastern "virginiae + rufigenis" samples, only minimal differences (~1.2%, see Table 2) were noted between specimens of the currently recognized subspecies Sminthopsis v. virginiae from the Cape York Peninsula (CY) and S. v. rufigenis from the Trans-Fly region of Papua New Guinea.This suggests that the genetic differences between these two genetic lineages are not much greater than the differences observed between individuals within each of the two subspecies.Most likely, these two taxa, currently separated by the Torres Strait, have not long diverged from one another, or there has been gene flow between the two populations during periods when the two regions were contiguous at times of lower sea levels (Voris, 2000).
Glacial cycles during the Pleistocene caused periodic exposure of land surfaces on the Sahul Shelf between Australia and New Guinea that were covered by savanna woodland including Lake Carpentaria (Figure 1) (Reeves et al., 2008;Torgersen et al., 1983).Such habitats would have potentially allowed periodic exchange (and isolation of populations) of terrestrial fauna favoring savannah and woodland habitats, including Sminthopsis virginiae, between the two land masses (Lavery & Leung, 2023;Wüster et al., 2005).The Sahul Shelf was progressively inundated due to sea-level rise after the last glacial maximum and a final land bridge between Australia and New Guinea was last cut about 9000 years ago with the formation of the Torres Strait, thus preventing further genetic exchange between the populations isolated on one or other of the land masses (Voris, 2000).

| Morphological divergence in Sminthopsis virginiae
Our comprehensive morphological analyses strongly accord with the genetic data in grouping western "nitela" as divergent from a combined grouping of eastern "virginiae" + northern "rufigenis".
In his major morphological revision of Sminthopsis, Archer (1981) recognized three subspecies of S. virginiae corresponding to isolated populations across northern Australia and southern Papua New Guinea (Figure 1).Archer (1981)  and development of an anterior cingulum on the upper molar teeth.
This feature is in fact present in all three subspecies on all upper molars (M 1-4 ), but varies between them, being reduced on M 1 in S.
v. nitela.However, we found that S. v. nitela could be clearly distinguished from S. v. rufigenis and S. v. virginiae based on various cranial and external measurements, but that S. v. rufigenis and S. v. virginiae could not be distinguished from each other based on the cranial features we investigated.
Archer also noted that specimens of nitela were "somewhat diverse and may represent more than one form" (Archer, 1981: 141).
Some of these individuals have subsequently been recognized as separate species-S.bindi and S. butleri-and other specimens examined by Archer (1981) are exemplars of S. macroura and another dasyurid genus, Pseudantechinus.In the present study, we found no indication of multiple forms of S. v. nitela.

| Taxonomic implications
Little is known about the capacity of "nitela," "virginiae," and "rufigenis" individuals to interbreed and produce viable offspring.The disjunct distributions of the three subspecies (see Figure 1) suggest that hybridization would not occur naturally.Wild populations of "nitela" also appear to have a different reproductive strategy to that seen in "virginiae" and "rufigenis" in being seasonal breeders with reproduction occurring from July to February (Morton  et al., 1987).In contrast, the Qld and PNG forms of S. virginiae are thought breed year round (Taplin, 1980;Woolley, 1994).We note that although Woolley (2023) reported that a male "nitela" was able to sire offspring when crossed in the laboratory with a female from Queensland ("virginiae"), there is no evidence that the progeny was fertile.In contrast, NT ("nitela") males never produced offspring when mated with PNG ("rufigenis") females (P. A. Woolley, pers.

comm.).
We have demonstrated reciprocal monophyly and morphological differentiation of Western Australian and Northern Territory specimens of Sminthopsis virginiae from their Queensland and Papuan New Guinean conspecifics.Our genetic and morphological data strongly suggest that "nitela" individuals are sufficiently distinct from specimens of the currently recognized subspecies "virginiae" and "rufigenis" to warrant reinstatement as a full species under the Unified Species Concept (De Queiroz, 2007)-Sminthopsis nitela-as originally proposed by Collett (1897) and suggested by Blacket et al. (2001).But what of continued recognition of the currently recognized subspecies S. virginiae virginiae and S. v. rufigenis?These "lineages" are not reciprocally monophyletic in our phylogenetic analyses and differ in only a few morphological characters.However, the two subspecies are separated geographically by the Torres Strait and are unlikely to naturally come into contact.They also differ in nipple number between females (6 in S. v. rufigenis and 8 in S. v. virginiae), which is a character that often discriminates taxa within dasyurids (Baker & Dickman, 2018).Taken together, we propose the subspecific status of Queensland and Papua New Guinean S. virginiae specimens be retained, in the interest of recognizing the two geographically distinct populations.The taxonomic revision of S. virginiae is formalized in the section below.

Type details
Neotype, QMJ15890 adult male skull, and wet specimen, Herbert Vale, Queensland, designated by Archer (1981).The original type specimen described by de Tarragon is considered lost (Collett, 1887a), and no type locality was listed, but a subsequent description of the species based on a specimen collected from Herbert Vale, Queensland was published by Collett (1887aCollett ( , 1887b)).We have examined the neotype (see Figure 8b).

Remarks
Here, we include two forms under this species, Sminthopsis v. virginiae and S. v. rufigenis based on our combined molecular and morphological analysis.It differs from all other species of Sminthopsis by the combination of the following characters: rufous cheeks and dark facial stripe, conspicuously enlarged oval apical granules on the interdigital footpads, thin tail, large body size (although smaller than S. douglasi and S. psammophila) and the presence of continuous anterior cingula on the upper molars (see Archer, 1981).Sminthopsis virginiae can be distinguished from S. nitela by genetic divergence, larger size, and other craniodental characteristics detailed below.
Both subspecies were accepted as valid taxa by Archer (1981) and subsequent authors (e.g., Jackson et al., 2015) and are considered to be aseasonal breeders (Taplin, 1980;Woolley, 1994).Females of the two subspecies can be distinguished by nipple number, six in S.
v. rufigenis and eight in S. v. virginiae.A full list of material examined can be found in Table A1.Note, while we did not include genetic material from the Aru Islands in this study, this form was considered to be the same as those from Papua New Guinea by Archer (1981).

Geographic distribution
This species occurs in two disjunct populations separated by the Torres Strait, S. v. rufigenis in the Aru Islands and Trans-fly region of New Guinea, and S. v. virginiae in Queensland on the Cape York Peninsula and along the north-eastern coast of Australia (Figure 1).

Diagnosis
Morphological description of Sminthopsis virginiae is included in Collett (1887aCollett ( , 1887b)), and an extensive description is provided by Archer (1981).We provide a brief diagnosis.Sminthopsis virginiae is a large species of Sminthopsis with striking russet-orange coloration to the sides of the face and a dark central stripe on the head.The tail is always longer than the head-body length and does not become incrassated (i.e., fattened at the base) like other members of the "macroura" species Apical granules on the hindfoot pads (underside of the foot) are large and elongate, sometimes appearing striated (see Archer, 1981).
Sminthopsis virginiae differs from S. nitela by larger size; a well-developed anterior cingulum on M 1 ; protoconule and a large stylar cusp B on M 1 that does not connect to the paracrista; a larger entoconid on M 3 ; and a present and well-defined posthypocristid on M 4 .A postorbital process is often present and well developed (which is rare in S. nitela).

Recommended common names
We propose the common names as follows: eastern red-cheeked dunnart for both Sminthopsis virginiae and S. v. virginiae, and northern red-cheeked dunnart for S. v. rufigenis.These names delineate the various regional distributions of each form relative to S. nitela in the west, see below.

| Sminthopsis nitela Collett, 1897
Etymology Collett (1897) gave no indication on the meaning behind the species name, "nitela."In Latin, it means "brightness" or "splendor," and this may allude to the bright orange fur patches on the sides of the face.

Type details
Collett (1897) described the species from four specimens all collected from the Daly River area, NT in July and October 1894 by Knut Dahl (Figure 1).Three of these reside in the Natural History Museum, University of Oslo, Norway (Wiig & Bachmann, 2013).

Remarks
Differs from all other species of Sminthopsis in the same ways that S. virginiae differs.We recognize Sminthopsis nitela as a separately evolving lineage from S. virginiae (see the unified species concept of De Queiroz, 2007), from our comprehensive evidence of molecular and morphological distinctiveness, geographic isolation, and differences in breeding biology (Morton et al., 1987).Sminthopsis nitela differs by 8.09%-8.23%divergence at the mitochondrial cytb and 0.9%-1.6%at 12S loci from S. virginiae.

Geographic distribution
Confined to several localities in the Kimberley, WA, including Mitchell Plateau, Kalumburu, and Sir Graham Moore Island, and widespread in the Top End of the NT, including the Tiwi islands (see Figure 1).

Diagnosis
See description by Collett (1897) and further details in Archer (1981).
Like Sminthopsis virginiae, but generally smaller in all aspects of ex- , Phylogenetics, Taxonomy, Zoology identified as Pseudantechinus bilarni, by P. G. Horner in 1996), as well as specimen AM M4403, a S. macroura.
at the Western Australian Museum.Primer details and PCR cycling are available as supplementary methods (see File S2).Bi-directional Sanger DNA sequencing was carried out at the Australian Genome Research Facility (Perth, Western Australia).Alignment of sequence reads was performed in Geneious Prime (Kearse et al., 2012) as described in Umbrello et al. (2017) and Westerman et al. (2023).We amplified 181 new DNA sequences from three mitochondrial (Control Region [CR] n = 29, l = 558 bp, 12S rRNA [12S] n = 50, l = 972 bp, Cytochrome b [cytb] n = 43, l = 1146 bp) and three nuclear (omega globin [ω-globin] n = 41, l = 822 bp, β fibrinogen intron 7 [bfib7] n = 15, l = 1454, interphotoreceptor binding protein [IRBP] n = 4, l = 1068 bp) loci.The new cytb, ω-globin, and IRBP gene sequences were checked for authenticity by looking for premature stop codons, frameshift mutations, indels, etc., as well as for indications of pseudogene amplification.GenBank accession numbers for the novel sequences generated in this study include PP851402-PP851404, PP853397-PP853441, PP854425, and PP900178 for 12S, PP858677-PP858705 for CR, PP887888-PP887930 for cytb, PP858706-PP858743 and PP898284-PP898286 for ωglobin, PP887931-PP887945 for bfib7 and PP898280-PP898283 Illustration of the Sminthopsis virginiae, based on ANWC M29776, cranium (a,b) and dentary (c) with measurements and teeth labeled.adult class were excluded from the morphometric analyses.Where wet-preserved bodies were present, sexual maturity was confirmed by the development of the external reproductive structures (i.e., presence and size of testes [males] or pouch and nipple development [females]), and juvenile or subadult individuals were not included in analyses.In total, we measured 19 S. v. nitela, seven S. v. rufigenis, and seven S. v. virginiae extracted skulls, which represented all the available material in Australian institutions.Due to the small number of individuals available of S. v. rufigenis and S. v. virginiae, sexes ), which consisted of two well-resolved subgroups: one comprising all individuals from the Northern Territory (NT) and the Kimberley region of Western Australia (WA); the other comprising all individuals from Cape York Peninsula, Queensland (Qld) together with those from the Trans-Fly region of southern Papua New Guinea (PNG).Our tree thus resolved the currently recognized subspecies S. v. nitela as sister to a lineage comprising specimens of S. v. virginiae and S. v. rufigenis, though neither of the latter was themselves monophyletic.Both clades were well supported with Bootstrap support >94% for RAxML or Bayesian Posterior Probability of 1.00.
Queensland and Trans-Fly Plains (PNG) populations of S. virginiae are genetically indistinguishable and are not monophyletic entities.The user-tree positing genetic differences between NT and WA samples of "nitela" was not significantly different from the best tree (p > .05for KH, wtd-KH, SH, and wtd-SH tests), suggesting that NT and WA "nitela" specimens are not clearly genetically divergent.Thus, our mitochondrial and nuclear DNA sequence data suggest that Sminthopsis virginiae currently comprises two genetically F I G U R E 3 Bayesian tree of concatenated nuclear and mitochondrial gene loci partitioned by gene locus and codon position (RY).Sample subspecies identification: Northern Territory (NT) Sminthopsis virginiae nitela populations (pink); Western Australia (WA) S. v. nitela populations (yellow); Queensland (Qld) populations of S. v. virginiae (green); Papua New Guinea (PNG) populations of S. v. rufigenis (blue).Numbers at nodes are Bayesian Posterior Probabilities (BPP) >0.81.distinct groups: the first of these comprises individuals currently recognized as S. virginiae nitela; the second comprises individuals currently recognized as belonging to S. v. virginiae and S. v. rufigenis, but which are genetically indistinguishable from one another based on the genetic markers examined.
S. v. nitela, 17 were significantly different between S. v. virginiae and S. v. nitela and only one measurement was significantly different between S. v. virginiae and S. v. rufigenis (FL, length of the frontal suture).Hierarchical clustering analysis sorted the samples into two clusters, one being most Sminthopsis v. nitela together with one S. v. virginiae (QM JM15890), and the other cluster comprising all S. v. rufigenis, most S. v. virginiae and a single S. v. nitela individual (ANWC M8861) 64) NANote: For each pair of "species" or clades, whether they are monophyletic (Mono) is stated, the ratio of intra to inter genetic distance (Intra/Inter) between the species and the mean probability (Prob correct) with confidence intervals of placing an individual in the correct clade, and Rosenberg's P AB value.Not calculated for groups that are not monophyletic.F I G U R E 4 Parsimony (TCS) haplotype network of Sminthopsis virginiae (a) 12S sequences (n = 50, l = 567 bp) and (b) omega globin haplotypes (n = 50, l = 737 bp).Refer to the key within the figure for subspecies designation of the samples, tick marks represent nucleotide differences between unique haplotypes.

F
I G U R E 5 (a) Hierarchical clustering dendrogram of Euclidean distance in cranial measurements between individual Sminthopsis virginiae specimens measured in this study.(b) Principal Components Analysis of cranial measurements for the three subspecies.(c) Comparison of skull size (as the geometric mean) for each of the three subspecies, p values <.001 = ***; ns >0.05.TA B L E 5 Percent of specimens correctly classified into subspecies or genetic clade using k-nearest neighbor discriminant analysis on the raw measurement data and the size-corrected data.
These estimates are similar to those reported for Melithreptus honeyeaters(Toon et al., 2010) and the common wallaroo(Eldridge et al., 2014), but are much older than divergences between species of Petaurus gliders (87-405 thousand years ago,Cremona et al., 2021) and the antillopine wallaroo, which exhibited some gene flow across the barrier(Eldridge et al., 2014).
<1.3%), which is separated by the Ord Basin, an area long recognized as being a biogeographic boundary for vertebrate species occurring across the Kimberley and Top End of the NT(Catullo et al., 2014;Edwards et al., 2017;Potter et al., 2012).

F I G U R E 7
Photographs of dry preserved specimens of Sminthopsis virginiae representing the three subspecies, (a) ANWCM8862 S. virginiae nitela, (b) ANWCM739 S. v. virginiae, and (c) ANWCM29790 S. v. rufigenis.Specimens all photographed on the same scale (not shown).

F
I G U R E 8 (a) Sminthopsis virginiae in life from Queensland, Australia (Photo: Eric Vanderduys), (b) cranium and dentary of the neotype QMJ15890.
ternal and craniodental proportions with occasional large males from the NT population reaching sizes similar to S. v. virginiae individuals.Differs from S. virginiae in possessing a reduced anterior cingulum and F I G U R E 9 Photograph of Sminthopsis nitela lectotype NHMO-DMA-31007, (a) mounted specimen, (b) cranium and dentary.Note that fur coloration has faded in the mounted specimen.Photos: Lars Erik Johannessen, NHMO.A PPEN D I X A TA B L E A 1 Details of specimens and tissue samples of Sminthopsis examined in this study including outgroup samples and GenBank accession numbers for each gene sequenced.

F
Map of sample localities for Sminthopsis virginiae with type of material examined coded as genetic data only (triangle), morphological data only (circle), or as genetic and morphological data available from the same specimen (diamond).F I G U R E A 2 Single gene trees, (a) β fibrinogen intron 7, (b) omega globin, (c) Control Region, (d) Cytochrome (b,e) 12 s rRNA.Samples from the Kimberley, WA yellow; NT, red; Qld, green and PNG, blue.F I G U R E A 3 Sminthopsis virginiae mitochondrial haplotype networks for (a) control region (n = 35, l = 386 bp) and (b) cytochrome b (n = 41, l = 389 bp).Tick marks indicate nucleotide changes between samples haplotypes.
Probabilities of specified user-tree tests in PAUP*.Average Kimura 2 Parameter corrected genetic distances (as percentages) between outgroup taxa and subspecies of Sminthopsis virginiae.
Summary of cranial measurements (in mm) of Sminthopsis virginiae measured in this study.
TA B L E 4Note: Number of individuals (n), mean and S.D. (x¯ ± σ), and range.
Summary of external measurements of Sminthopsis virginiae specimens measured in this study (except for weights which were taken from museum databases).Measurements are given as mean ± standard deviation (x ± ) and in mm and weights in grams.
Archer (1981).v.nitela was morphologically distinguishable from the other two subspecies in being smaller and in lacking a continuous anterior cingulum on the upper molar teeth.Our results concord withArcher (1981)in that S. v. nitela is indeed smaller than S. v. rufigenis and S. v. virginiae; however, with our larger sample, we observed substantial variation in the presenceTA B L E 6Note: No/very little data was available for weights from S. v. rufigenis and S. v. virginiae.
Models of sequence evolution obtained for the Bayesian analyses of the Sminthopsis sequence data.External measurement data (in mm) for Sminthopsis virginiae specimens in this study.ANOVA post hoc test scores for each of the cranial measurements, except for ATL for which a Kruskal-Wallis test was run instead.Results of k-nearest neighbor discriminant analysis, showing how many samples were classified into the correct and incorrect groups.