Mitogenome of the extinct Desert ‘rat-kangaroo’ times the adaptation to aridity in macropodoids

The evolution of Australia’s distinctive marsupial fauna has long been linked to the onset of continent-wide aridity. However, how this profound climate change event affected the diversification of extant lineages is still hotly debated. Here, we assemble a DNA sequence dataset of Macropodoidea—the clade comprising kangaroos and their relatives—that incorporates a complete mitogenome for the Desert ‘rat-kangaroo’, Caloprymnus campestris. This enigmatic species went extinct nearly 90 years ago and is known from a handful of museum specimens. Caloprymnus is significant because it was the only macropodoid restricted to extreme desert environments, and therefore calibrates the group’s specialisation for increasingly arid conditions. Our robustly supported phylogenies nest Caloprymnus amongst the bettongs Aepyprymnus and Bettongia. Dated ancestral range estimations further reveal that the Caloprymnus-Bettongia lineage originated in nascent xeric settings during the middle to late Miocene, ~ 12 million years ago (Ma), but subsequently radiated into fragmenting mesic habitats after the Pliocene to mid-Pleistocene. This timeframe parallels the ancestral divergences of kangaroos in woodlands and forests, but predates their adaptive dispersal into proliferating dry shrublands and grasslands from the late Miocene to mid-Pleistocene, after ~ 7 Ma. We thus demonstrate that protracted changes in both climate and vegetation likely staged the emergence of modern arid zone macropodoids.

sometime between 1902 and 1905 ( Fig. 1D) has been reidentified 22 , and various unsubstantiated live sightings made 17,23,24 , with the most recent in 2011 24 and 2013 17 prompting unsuccessful surveys for the species in 2018 and 2019 17 . Caloprymnus campestris has otherwise been classified as Extinct by the IUCN (https:// www. iucnr edlist. org/) since 1994, with the probable cause being over-predation by feral dogs, cats and foxes 25 .
At latest count, only 25 specimens of C. campestris are catalogued in museums worldwide 22 . This dearth of research material has led to uncertainty about potoroid interrelationships 26 , as well as the concomitant chronicle of their arid zone evolution. Here, we therefore analyse the first complete mitochondrial (mt) genome of C. campestris, which augments the 12S rRNA (AY245615) and partial cytochrome b (AY237246) gene sequences 27 already available from GenBank (https:// www. ncbi. nlm. nih. gov/ genba nk/). Our novel dataset is used to construct a comprehensive phylogeny of crown potoroid species and subspecies within Macropodoidea. We also apply molecular clock calibrated ancestral range estimations to infer both the timing and context of macropodoid habitat change over the last ~ 25 Ma.

Materials and methods
Samples and sequencing. We obtained non-formalin-fixed liver samples from a male Caloprymnus campestris (Museums Victoria, Melbourne, Australia [NMV] C8981) that was collected in 1834 from Mulka cattle station in northeastern South Australia (Fig. 1B). Our DNA extraction, PCR amplifications, sequencing and alignment procedures followed Westerman et al. 27,28 . Whole genome libraries were prepared with the Nextera DNA flex library kit (Illumina, CA), incorporating 50 ng of input DNA per sample. Sequencing was performed on the Illumina MiSeq platform using 2 × 300 bp V3 chemistry to generate 4,445,476 read pairs and  Phylogenetic and molecular clock analyses. Phylogenetic relationships within Macropodoidea were examined using a mitogenome dataset including representatives of all potoroid species, together with Hypsiprymnodon moschatus and multiple species-level exemplars for selected macropodid genera (see Supplementary Table S1). The Northern common cuscus, Phalanger orientalis (Phalangeridae), and Western pygmy possum, Cercartetus concinnus (Burramyidae), were added as non-macropodoid outgroups. To accommodate for recognised gene incongruence 32 , we then compared these results with analyses of nuclear (n), and combined mitogenome/mtDNA/nDNA sequence datasets derived from GenBank, which integrated an expanded taxon sample of all potoroid species and subspecies (see Supplementary Tables S1 and S2). The mitogenomes were treated as a single partition, or alternatively sub-partitioned into 12S/16S rRNA stems and loops, pooled 1 st , 2 nd and 3 rd protein codon positions, and 3 rd codon positions with RY coding to allow for heterogeneity and saturation. A General Time Reversible gene partition model, gamma distribution and variable site proportions were determined using jModelTest 33 36 with node support calculations based on 1000 bootstrap pseudoreplicates (%) and Bayesian Posterior Probabilities (BPP), respectively. Maximum likelihood used a GTR + I + Γ partition model, while non-dated Bayesian MCMC analyses were run for 6 × 10 6 generations with a sample frequency of 1000, eight chains, default temperature of 0.2, and burn-in fixed at 6 × 10 4 . Time-trees were constructed in BEAST 2.2.1 36 with relaxed clocks and the minimum-maximum node age constraints listed in the Supplementary Information. Up to 95% of the normal prior distributions were assigned to the interval between minimum and maximum, with 2.5% to each tail. Gamma priors (shape = 1; scale = 1) were assigned to the "ucld.mean" parameter for each partition. MCMC analyses were run for 65 × 10 6 generations with a burn-in of 10 × 10 6 generations and sampling every 10 × 10 3 generations. ESS values were > 200 for all estimated parameters. TreeAnnotator 2.2.1 (https:// www. beast2. org/ treea nnota tor/) was used to summarise the tree sample with mean node heights.
Ancestral area analyses. Distributional areas were optimised onto the time-calibrated BEAST consensus tree and analysed using the R package BioGeoBEARS 37 to compare alternative biogeographical range models, and a Bayesian Binary MCMC (BBM) approach 38,39 to reconstruct ancestral ranges in RASP 4 40 . Area codes (Supplementary Table S4) followed standard units 6 but were refined to represent a generalised vegetation map 41 : A = humid forest (rainforest and/or 'wet' sclerophyll dominant) prevalent throughout eastern coastal Australia, western Tasmania and New Guinea; B = woodland ('dry' sclerophyll dominant) prevalent throughout northern, eastern and southwestern inland Australia and northeastern Tasmania; C = shrubland (Acacia and chenopodiaceous shrubland dominant) prevalent throughout central and central-western Australia; and D = grasslanddesert (arid grasslands and/or desert dominant) prevalent in central and central-northwestern Australia. The maximum number of ancestral areas was restricted to three because this equalled the maximum number of areas occupied by our terminal taxa at any given node.
BioGeoBEARS comparisons proceeded with likelihood ratio testing of 'Jumping dispersal events (+ J)' , which have been considered inappropriate for dispersal-extinction-cladogenesis (DEC) models 42 . However, the three parameter Bayesian inference of historical biogeography for discrete areas (BAYAREALIKE) + J model (P = 0.0006) received overwhelmingly highest support (AICc = 199.6; AICc_wt = 0.98) for conferring best statistical likelihood on our data (Supplementary Table S5). Finally, we accommodated for connectivity by designating a dispersal multiplier of '1' for adjacent areas (A-B-C) 41 versus non-adjacent areas (A-D) 41 , which were assigned a value of '0.5' .
Our BBM analyses utilised 10 MCMC chains with default temperature 0.1, and run over 5 × 10 6 generations with sampling frequency and burn-in fixed at 1000. Model settings included 'Gamma(+ G)' for among-site rate variation, and 'Fixed (JC)' for state frequencies.

Results and discussion
The Caloprymnus campestris mitogenome (16,866 bp) is ordered with 13 protein-coding genes, two ribosomal (r)RNA genes, 21 transfer (t)RNAs, and a non-coding AT-rich control region, which follows the typical configuration for marsupials 43,44 . The tRNAs are arranged around the origin of the L strand (A-C-W-O L -N-Y) and intersected between the NADH2 and COX1 genes. Substitution of the anticodon GCC for tRNA ASP (trnD) is also consistent with RNA-editing 45 .
Maximum likelihood and Bayesian analyses of our mitogenome dataset produce unanimous resolution of Macropodoidea with Potoroidae as the sister to Macropodidae (Supplementary Figures S1-S6). This pivotal higher-level grouping accords with other crown macropodoid phylogenies 12,46-49 , and warrants a new taxonomic definition 50 Table S6), but decisively rejected (P < 0.0001***). Taxonomically, therefore, we conclude that the original classification of Gould's Desert 'bettong' 19 as generically consistent with Bettongia is feasible, but defer any formal nomenclatural amendment pending a detailed morphological re-evaluation. Our maximum likelihood, Bayesian and time-tree analyses of the nDNA (Supplementary Figures S7-S9) and combined mitogenome/mtDNA/nDNA datasets ( Fig. 2; Supplementary Figures S10-S12) yield broadly compatible topologies, with the basal divergence of potoroids and macropodids, and subsequent split between potoroines and bettongines both occurring from the latest Oligocene to earliest-middle Miocene ( Table 2; Supplementary Table S7). Notably, this concurs with divergence times derived using different dating methods and constraints 12,[46][47][48][49][50]56 . Furthermore, while our BioGeoBEARS and BBM ancestral range estimations correlate the latest Eocene (or mid-Eocene using nDNA: Supplementary  Supplementary Figures S13 and S14). These potentially included 'mallee-like' 57 sclerophyll communities, which propagated throughout central Australia from the early to middle Miocene 41 .
Bettongia is karyotypically conservative, retaining the 2n = 22 chromosomal number of most macropodoids 69,70 . Conversely, chromosomal fission in P. longipes has produced 2n = 24, while fusions (and inversions) in P. tridactylus and P. gilbertii manifest unusual reductions to 2n = 12♀, 13♂ 71 . Aepyprymnus rufescens, on the other hand, exhibits a unique karyotypic increase to 2n = 32, which is the highest for any marsupial 71 , and presumably reflects its independent evolution since the later-early to early-late Miocene (nDNA favouring a younger later-middle to early-late Miocene range: Table 2; Supplementary Table S7). Although the chromosomal arrangement of C. campestris is unknown, our robustly supported (bootstrap = > 90%; BPP = 1) earliest-middle to early-late Miocene split from Bettongia (

Conclusions
Our characterisation of the complete mitogenome for Caloprymnus campestris provides an ecological diversification timescale for bettongs and potoroos within the context of crown macropodoid evolution. Most importantly, we show that the unambiguously monophyletic C. campestris-Bettongia lineage probably originated with the onset of increasingly arid intracontinental climates during the middle to late Miocene 41,57-60,74 , corresponding with the deepest divergences of Australia's arid zone biota around ~ 15 Ma 1 . This contrasts with the largely late Miocene to Pleistocene radiation of kangaroos, whose abundance in modern arid zone habitats has been attributed to grazing adaptations and the spread of grasslands during the Pliocene and Pleistocene 3,9,12 . Clearly, therefore, the appearances of Australia's distinctive arid zone macropodoids were staged over some ~ 3-6 Ma (based on minimum-maximum confidence interval differences for C. campestris versus Osphranter rufus: Table 2), and likely occurred in response to a complex interplay of abiotic and biotic drivers involving both climate and vegetation change.
Unfortunately, little is known about the biology of C. campestris or other extinct 'Desert bettongs' , such as Bettongia anhydra 63 , and the Nullarbor dwarf bettong 75 , Bettongia pusilla 76 . Nonetheless, early eye-witness reports state that C. campestris inhabited sparsely vegetated gibber plains 16 . The diet of C. campestris is also uncertain 23 , but might have been varied 16,23 similar to the extant arid zone Bettongia lesueur 77 and Bettongia penicillata 64 , which consume a range of plant matter, fungi and insects 78,79 . Caloprymnus campestris was thus probably an important 'ecosystem engineer' 63 whose tragic loss is compounded by dramatic range reductions and the Near Threatened (Bettongia gaimardi, B. lesueur, Potorous tridactylus), Vulnerable (Potorous longipes), Endangered (Bettongia tropica), Critically Endangered (B. penicillata, Potorous gilbertii), or Extinct (B. anhydra, C. campestris, Potorous platyops) IUCN Red listings (https:// www. iucnr edlist. org/) for 10 out of the 11 named non-fossil crown potoroids. The extinction susceptibility of C. campestris was presumably exacerbated by its limited distribution (only four recognised collection 22 , and 13 potential sighting localities 17 within a ~ 350 km radius) and desert specialisation, which when coupled with habitat modification and the introduction of exotic species via European pastoralism 80 , underscores the extreme conservation sensitivity of Australia's unique arid zone marsupials and the urgent need to document their now dwindling multi-million-year evolutionary histories.
Ethical approval and informed consent. No live animal subjects were used for experiments in this study. All extinct animal tissues were obtained and their use approved by the La Trobe University Animal Ethics Committee (AEC). All experiments were performed in accordance with institutional guidelines and regulations.