Species Delimitation in the Genus Moschus (Ruminantia: Moschidae) and Its High-Plateau Origin

The authenticity of controversial species is a significant challenge for systematic biologists. Moschidae is a small family of musk deer in the Artiodactyla, composing only one genus, Moschus. Historically, the number of species in the Moschidae family has been debated. Presently, most musk deer species were restricted in the Tibetan Plateau and surrounding/adjacent areas, which implied that the evolution of Moschus might have been punctuated by the uplift of the Tibetan Plateau. In this study, we aimed to determine the evolutionary history and delimit the species in Moschus by exploring the complete mitochondrial genome (mtDNA) and other mitochondrial gene. Our study demonstrated that six species, M. leucogaster, M. fuscus, M. moschiferus, M. berezovskii, M. chrysogaster and M. anhuiensis, were authentic species in the genus Moschus. Phylogenetic analysis and molecular dating showed that the ancestor of the present Moschidae originates from Tibetan Plateau which suggested that the evolution of Moschus was prompted by the most intense orogenic movement of the Tibetan Plateau during the Pliocene age, and alternating glacial-interglacial geological eras.


Introduction
Speciation, extinction and migration are often driven by historical, ecological and biogeographic factors, which have played important roles in shaping global biodiversity by influencing regional differentiation [1,2]. In Asia, the uplift of the Tibetan Plateau was the most remarkable geological event because of the average altitude of the Tibetan Plateau, which was raised by approximately 3,000 m during the Quaternary period [3,4]. Associated with the uplift of the Tibetan Plateau, high mountains and deep valleys were generated, which profoundly accelerated ecological speciation events [5,6,7]. In the meantime, the uplift of the Tibetan Plateau led to the desertification of northern China because of the obstruction of the northward [22,23]. Up to now, there are seven species were that are described in Moschidae [24]. Among them, M. chrysogaster and M. leucogaster were found in 1839 [25]. Then, M. berezovskii [26], M. fuscus [27], M. anhuiensis [28,29] and M. cupreus [30] were described successively in more than one century. However, it remains disputed precisely how many species exist in Moschus. One of the most extreme opinions insist that just one species exist in Moschidae [31,32]. However, most scholars believe that there are multiple species in Moschus [33,34]. In addition to the above opinions, there remain some differing and controversial points. For example, Sheng et al. (2007) believed that there were only three species (M. moschiferus, M. berezovskii and M. chrysogaster) that existed in this genus, and that M. leucogaster and M. fuscus should be considered a subspecies of M. chrysogaster (M. c. leucogaster, M. c. fuscus), and that M. anhuiensis should be regarded as a subspecies of M. berezovskii (M. b. anhuiensis). Among those species, the debate with regard M. anhuiensis is the most complex. In 1982, three unknown specimens of musk deer were found in Huoshan, Anhui province, China, and it was recognized as the subspecies (M. m. anhuiensis) of M. moschiferus based on morphological data [29]. Later, some researchers thought that it should be considered a subspecies (M. b. anhuiensis) of M. berezovskii rather than M. moschiferus and as based on the characters of fur texture, stripes and the structure of the skull [33,35,36,37,38]. However, further studies based on molecular phylogeny suggested that it should be recognized as a separate species [28,34].
For a long time, it was believed that northern Asia was the origin center of Moschus [21], in part because the most ancient fossils of Moschus were found in Tunggur, Inner Mongolia [40,41]. Accordingly, M. moschiferus was regarded as the most primitive species in existence for musk deer. In addition, this contention was supported by several molecular phylogenetic studies [34,42,43]. Su et al. combined the data of current distributions, fossil records and molecular data, and then concluded that the historical dispersion of musk deer might be from north to south China [43]. However, others consider M. chrysogaster as the most primitive musk deer [44], which is also supported by molecular phylogenetic studies [45]. The existence of discrepancy makes the origin and evolutionary history of Moschus as an additional interesting question for further investigation.
Mitochondrial DNA (MtDNA) markers are chosen frequently to study evolutionary history, biological identification, taxonomy, biogeography and phylogeny [46]. The use of mtDNAs can improve the probability of congruence between the mitochondrial genetic tree and the species tree, which is useful in resolving relationships between recently divergent taxa [47]. Notably, single-gene phylogenies often differ dramatically from studies involving multiple datasets, suggesting that they are often unreliable [48]. Thus, the complete mtDNA genome were gradually used to construct reliable phylogeny for determining evolutionary relationships among species or higher taxa with accurate timescales [49,50,51].
In the present study, we set out to examine sequence variation of the complete mtDNA genome, as well as mitochondrial genomic structure and organization from inclusive species of Moschus to address the following emergent issues: (i) species delimitation for controversial species based on a reliable tree with time-scales, and (ii) the evolutionary history and speciation events of Moschus. Whether geological events and environmental changes including uplift of the Qinghai Tibetan Plateau, climatic oscillations in Pleistocene have affected its speciation evolution, is one of the key determining objectives of the current study.

Ethical Statement
In the present study, collection of samples was performed within a long-term investigation project on Moschus and all samples were from individuals that died naturally and were found Species Delimitation in Moschus and Its High-Plateau Origin during field work. This investigation project and the sample collection were approved by the Forestry Administration. Our experimental procedures complied with the current laws on animal welfare and research in China, and were specifically approved by the Animal Research Ethics Committee of Anhui University.
We extracted total DNA using a standard proteinase K/phenol-chloroform protocol [59]. An EasyPure PCR Purification Kit (TransGene) was used to purify each DNA extraction. Twenty-one pairs of primers(S2 Table) were designed using Primer Premier 5.0 [60] based on M. moschiferus (NC013753) and Rangifer tarandus (AB245426). The product size of above primer pairs ranged from 635 bp to 1400 bp. Polymerase chain reaction (PCR) was performed using a reaction mixture (25 μL) containing 1 μL genomic DNA (concentration 10-50 ng/μL), 2.5 μL 10×buffer, 1 μL of 2.5 mM MgSO 4 , 2 μL of 2 mM dNTPs, 2.5 U Taq polymerase (Meridian Bioscience, Singapore) and 0.3 mM of each of the primers. Pure molecular biology grade water was added to reach the appropriate volume. The amplification protocol included an initial denaturation step of 95°C for 5 min; this was followed by 32 cycles of denaturation at 95°C for 30 s, primer annealing at 55°C for 30 s, and an extension at 72°C for 80 s, with a final extension at 72°C for 10 min. PCR products were purified using an EasyPure PCR Purification Kit (TransGene), and sequenced using previous primers and the BigDye Terminator v3.0 Ready Reaction Cycle Sequencing Kit (Applied Biosystems) following the manufacturer's instructions on an ABI Prism 3730 automated sequencer. In addition, several different methods (e.g. BLAST search and translation test method) had been adopted to exclude potential nuclear mitochondrial pseudogenes [61].

Sequence Analysis
Sequences were assembled by Seqman II (DNAStar, Madison, WI, USA) and checked by visual inspection to ensure the accuracy of variable sites identified by the program [62]. Protein-coding genes were identified by comparison with known complete mtDNA sequences of Ruminantia using Sequin 11.0. The 22 tRNA genes were identified using the software package tRNA Scan-SE 1.21 (http://lowelab.ucsc.edu/tRNAscan-SE 1.2.1). In addition, the DOGMA annotation software was used to check annotated genes [63]. All assembled and annotated Moschus mitochondrial genomes are available at GenBank, accession numbers are given in S1 Table. Moreover, A+T content was calculated using MEGA 5.05 [64]. Strand asymmetry was calculated using the formulae AT skew = [A−T]/[A+T] and GC skew = [G−C]/[G+C] [65], for the strand encoding the majority of the protein-coding genes. The complete alignment of nucleotides of the four Moschus mtDNAs was used to effect sliding window analyses using DnaSP v. 5 [66]. A sliding window of 300 bp and steps of 10 bp were used to estimate nucleotide diversity (π) for the entire alignment. Nucleotide diversity for the entire alignment was plotted against midpoint positions of each window, and gene boundaries as indicated. We calculated uncorrected genetic distances corrected by GTR + I + G for each mitochondrial gene separately, and did so using MEGA 5.05 [64].
Five different datasets were generated for different analyses. Dataset 1 composed of complete mtDNA, except the control region, from 14 species in Ruminantia, including four Moschus species, M. chrysogaster, M. moschiferus, M. berezovskii and M. anhuiensis (S1 Table). In Dataset 2, complete mtDNA, with the exception of the control region of the above defined four species in Moschus were included. In Dataset Table). In Dataset 5, all mitochondrial protein gene were included, except ND6 gene (S1 Table). We selected the best fit model of evolution for these datasets using MrModeltest 1.0 b [67] based on the AIC criterion.

Phylogenetic Analyses
Eighteen species in the Ruminant (Dataset 1) group were used to reconstruct a phylogenetic study using Maximum Likelihood (ML) and Bayesian methods, with Tragulus kanchil used as an outgroup (Fig 2) [50]. Before reconstructing the phylogenetic trees, sequence alignment was carried out using Clustal X 1.8 software [68], followed by manual adjustment. All ambiguous regions, i.e., those that involved ambiguity for the gap positions, were excluded from the analyses to avoid erroneous hypotheses of primary homology. ML analyses were performed in RAxML version 8 [69] and a general time reversible model of nucleotide substitution under the Gamma model of rate heterogeneity (i.e., GTRCAT). Support for internal branches for the best-scoring tree was evaluated via the bootstrap test with 1000 iterations. Bayesian inference of phylogeny was performed using the MrBayes 3.1.2 software program (http://mrbayes.csit.fsu.edu/index. php) [70], with the same best fit substitution model used as that selected for the ML analysis. MrBayes analyses simultaneously initiate two Markov Chain Monte Carlo (MCMC) model runs to provide additional confirmation of convergence of posterior probability distributions. Analyses were run for 10,000,000 generations. Chains were sampled every 1000 generations. When the average standard deviation of split frequencies reached a value less than 0.01. The first 10% of the total trees were discarded as ''burn-in" and the remaining trees were used to calculate Bayesian posterior probabilities. MrBayes analyses were also implemented for Dataset 3, Dataset 4 and Dataset 5 with the GTR + I model (Fig 3 and S1 Fig). We assumed that the tree topology that was derived from analyses of the entire mitochondrial sequence was the true mitochondrial gene tree.
To determine the support provided by each gene or region to this topology, we ran ML, which was based on the bootstrap test with 1000 iterations on each gene or region separately in MEGA 5.05(S2 Fig) [64]. In addition, to obtain the estimated best fit evolution model for each mitochondrial region, we performed analyses separately as described above using the MrModeltest 1.0 b software program [67] in Paup Ã 4.0b10 [71], which was based on the AIC criterion.

Species Delimitation
The species delimitation in Moschus was implemented in BPP v. 3.0 [72,73] based on reversible-jump Markov Chain Monte Carlo (rjMCMC) model sampling in Dataset 2. The guide tree was collected from the phylogenetic analyses. We assessed the impact of ancestral effective population size and time of divergence on species delimitation by testing a range of different prior distributions for θ and τ 0 . First, we fixed the τ 0 (1: 100), and θ acted as the variable. Low values for priors (i.e., at a frequency of 1:10) generally infer a large population sizes and deep divergence, whereas higher values infer small population sizes and shallow divergence for θ and τ 0 , respectively. Secondly, we fixed θ (i.e., at a ratio of 1:2000), and τ acted as the variable. Based on the results (Fig 4), we fixed θ (1:2000) and τ (1:10) for the prior distributions (Table 1). Each analysis ran three independent chains of 500 000 steps, sampling every fifth step, with 100 000 burn-in steps. In addition, for Dataset 3, we included two other species, M. leucogaster and M. fuscus to probe whether both of these species are independent species. The prior distributions for θ and τ in the context of this analyses were identical to that described for the first part of the analysis (Table 1).

Divergence Time Analyses
To estimate divergence times in Moschus, we applied a Bayesian MCMC method (Dataset 1) based on mitochondrial genomes, which employs a relaxed molecular clock approach, as   implemented in BEAST 1.7.4 [74]. We assumed a relaxed uncorrelated log normal model of lineage variation and a Yule Process prior to the branching rates based on the HKY+I +G model, and as recommended by MrModel test 1.0 b [67]. Four replicates were run for 10,000,000 generations with tree and parameter sampling that occurred every 1,000 generations for the first 10% of samples that were discarded as burn-in. All parameters were assessed by visual inspection using Tracer v. 1.5 [75]. The tree was generated and visualized with TreeAnnotator v. 1.6.1 [76] and FigTree v. 1.3.1 [77], respectively. All calibration points were derived from Hassanin et al. (S3 Table) [50]. In addition, more species (for example adding M. leucogaster and M. fuscus) with relatively shorter sequences (i.e., containing 12S and Cyt b, and contained in Dataset 3) were also used to estimate divergence times in Moschus, and were based on the calibration points of the analysis described above. Relaxed uncorrelated log normal models of lineage variation and the Yule Process were set by basing them on the GTR + I model as recommended by MrModeltest 1.0 b [67].

Genomic Organization and Gene Arrangement
The complete mtDNA sequence of Moschus ranged from 16351 to 16354, and contained 13 protein-coding genes (i.e., ATP6, ATP8, COI, COII, COIII, ND1, ND2, ND3, ND4, ND5, ND6, ND4L, and Cyt b), two rRNAs (i.e., 12S rRNA and 16S rRNA), 22 tRNAs and a control region. The base composition of mtDNAs was A (29.6%), G (14.8%), C (32.8%), and T (22.9%) (S3 Fig, S4 Table), and the percentage of A and T (52.5%) was slightly higher than that of G and C. The heavy DNA strand (H-strand) carried most of the genes; i.e., 12 protein-coding genes, two rRNAs, and 14 tRNAs. While ND6 and eight tRNAs were located on the L-strand. The arrangement of the whole mitochondrial genome of Moschus matched known typical mammalian patterns [78]. The total length of the 13 protein-coding genes was 11,315 bp, which represented about 69.2% of the entire mitochondrial genome in Moschus.
The longest gene was ND5 (1,821 bp), which was located between tRNA Leu (CUN) and ND6, and the shortest was ATP8 (201 bp), which was located between tRNA Lys and ATP6. Eleven of the 13 protein-encoding genes ended with a complete (TAA) or an incomplete (T or TA) stop codon; the latter was presumably completed as TAA by post-transcriptional polyadenylation [79,80]. However, Cyt b ends with AGA, and ND2 ends with AGA. Ten of the protein-coding genes started with an ATG codon and ND2 and ND3 started with an ATA codon. Notably, ND5 in Moschus began with ATT while all other species in our study displayed ATA as the start codon.
Sliding window analysis of the nucleotide alignment of four species in Moschus provided an indication of nucleotide diversity (π) within and between mitochondrial genes, which revealed significant regional variation across the alignment (Fig 5). The plot readily showed a high  Overall mean distances of each mitochondrial gene based on the bootstrap method in the Kimura 2-parameter model showed a lack of uniformity. For example, for the Control region, Cyt b and ND3 nearly obtained a doubled value as compared with that of the 12S rRNA and 16S rRNA. This phenomenon is also supported by the relative evolution rate that was calculated by Beast (S4 Fig) and the estimated model parameters (S4 Table). Uncorrected genetic distances among four species ranged from 1.4% (M. berezovskii-M. anhuiensis) to 4.8% (M. chrysogaster-M. berezovskii) (Fig 6).

Phylogenetic Reconstructions
Topologies recovered from the maximum likelihood (ML) and Bayesian inference (BI) analyses of complete mtDNA for 18 Ruminant species (Dataset 1) were highly congruent, with only slight differences found in the bootstrap support or posterior probability values for most nodes (Fig 2). The 17 species were divided into three major lineages in which each represented Cervidae, Bovidae, and Moschidae. The lineage of Bovidae is a sister group to Moschidae. In Moschidae, there are four clades for which there is a high level of Bayesian posterior probabilities and ML support. Each clade represents one species. M. chrysogaster is at the base of the tree and M. berezovskii and M. anhuiensis are sister groups of the inner clade (Fig 2). In addition, the phylogenetic analyses based on mitochondrial gene have similar topologies (S1 and S2 Figs).
Topologies recovered from Bayesian inference (BI) of 12S and Cyt b for six Moschus species (Dataset 3) obtained highly posterior probability values for most nodes (Fig 3). Those six species were divided into two major lineages: M. leucogaster, M. fuscus and M. chrysogaster that were clustered into one lineage (A) and the other three species that were clustered into the other lineage (B) (Fig 3).  (Fig 3).

Species Delimitation
The species delimitation based on Dataset 2 (M. chrysogaster, M. moschiferus, M. berezovskii and M. anhuiensis) showed that the probability values of four separate species exceeded 0.95, which demonstrated that the four species could be regarded as valid species. The topology [i.e., (Mc, (Mm, (Mb, Ma)))] attained an approval probability greater than 0.69, while the other three kinds of topologies were less than 0.25. Further, the six species can be regarded as valid species in the species delimitation of Dataset 3 because of the probability values of six separate species higher than 0.95. The topology, ((Ml, (Mf, Mc)), (Mm, (Mb, Ma))), attained an approval probability of 0.54. Moreover, the other three kinds of topology had a probability of less than 0.25. These two highly approved topologies were similar to the above phylogenetic study (Figs 2 and 3).

Mitochondrial Genome Annotation and Features
The mtDNA genome of four species in Moschus is nearly identical to those of other ruminants in many respects, in which there are no introns, no long intergenic spacers, and only a few overlapping sequences [80,81,82]. The overall mean base composition was: A, 34.0%; C, 25.0%; G, 12.9%; and T, 28.1%. The A + T content (62.1%) was higher than the C + G content (37.9%), which indicated a strong AT bias, which was similar to other ruminants in our studies (58.8-63.7%). Guanine (G) is the rarest nucleotide; the percentage of the other three bases were roughly equal to each other (S4 Table), similar to other vertebrate animals [78,80,83].
GC and AT skews are a measure of compositional asymmetry. In amniote mtDNA, GCskew values are all negative (G<C), while AT-skews are positive (A>T) [65]. In mtDNA of Moschus, the GC-skew (−0.3175 to −0.3245) and the AT-skew (0.0915-0.0968) values were in accord with this principle (S4 Table). In sliding window analysis, the largest value in the curve was 0.129 (Pi) and the largest value of the number of variable positions per unit length of the gene was CR (0.062).
When we compared the 11 ruminant species from four families with respect to the predicted initiation and termination codons of 13 mitochondrial protein-coding genes ( Table 2), we Table 2. Predicted initiation and termination codons for 13 mitochondrial protein-coding genes in 11 species in Artiodactyla.

Gene
Predicted initiation and termination /TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAG  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA   ATP6  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA   ATP8  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/T  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA   CO2  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAG  ATG/TAA  ATG/TAA  ATG/TAA  ATG/TAA  ATG/ Species Delimitation in Moschus and Its High-Plateau Origin found that most protein-coding genes used ATG as the start codon, and only a few species started with the GTG, ATT or ATA sequences. Stop codons were also similar across different species with TAA, TA-and T-occurring most frequently. Notably, ND5 in Moschus showed a differential pattern, which started with ATT and ended with TAA. Others in the Artiodactyla, started with ATA and ended with TAA. Based on our phylogenetic study, we inferred that the mutation (synonymous transition) had occurred in the Moschus ancestor.

Species Delimitation
Although seven potential Moschus species were not all included in this study, our results still obtained reliable information of species delimitation in Moschus. The six species (i.e., M. leucogaster, M. fuscus, M. chrysogaster, M. moschiferus, M. berezovskii and M. anhuiensis) in the present study were all recognized as a separate species (Figs 2 and 3., Table 1). Therefore, multiple species exist in Moschus [33] rather than only one species [31,32]. Among the six species analyzed in our study, the taxonomic status of M. anhuiensis was the most controversial. It was considered as a sub-species of M. moschiferus [29] or M. berezovskii [33,35,38]. In 1999, Li et al. set it as a valid species based on the Cyt b sequence (367 bp) and morphological variation [28]. Further, the later phylogenetic study also supported it as a separate species, and was considered the sister group of M. chrysogaster and M. berezovskii [34]. Our study supported the notion that M. anhuiensis was a valid species (Table 1), and it was the sister group of M. berezovskii (Figs 2 and 3). This result could be supported by the discontinuity of the distribution area between M. berezovskii and M. anhuiensis (Fig 1).

Phylogenetic Relationship and Origin of Present Musk Deer
Historically, M. moschiferus was regarded as the primitive species in Moschus. Vislobokova et al. proposed that Moschidae originated from the north of Asia, since M. grandaevus and M. primaevus were collected from Oligocene formations in Mongolia, and were regarded as the most primitive fossil in Moschus [21]. Presently, only M. moschiferus is distributed through the north of Asia, thus it was regarded as the most primitive musk deer [21]. This viewpoint was strengthened by some additional molecular phylogenetic studies [34,42,43]. However, other investigators in the field contended that M. chrysogaster was more primitive than M. moschiferus based on morphological or molecular data [44,45]. In this study, two main evolutionary lineages (A, B) were disclosed in Moschus (Figs 2 and 3). One of them, was referred to as lineage A, which contained M. leucogaster, M. fuscus and M. chrysogaster, and were distributed in the Tibetan Plateau margin (Fig 1). Moreover, lineage B was composed of M. moschiferus, M. berezovskii and M. anhuiensis, which were distributed in the area that was located around the Sichuan basin, Qinling and the adjacent areas, as well as the Dabie mountain, east Asia and the far east area (Fig 1). However, in our study, M. moschiferus failed to occupy the most basal branches of the phylogenetic tree (Figs 2 and 3). Therefore, M. moschiferus was not the most primitive musk deer. Molecular dating very clearly showed that speciation events of Moschus occurred during 4.42 Ma to 0.84 Ma (Figs 2 and 3). According to the topology, lineage A was distributed in the Tibetan Plateau margin, branched off from the most common ancestor of musk deer (about 4.42 Ma ago) and then followed the bifurcation forming the M. moschiferus lineage and the lineage that clustered M. berezovskii and M. anhuiensis (i.e., around 3.50 Ma before present). The remaining speciation events happened rather recently (i.e., less than 2.0 Ma). During 8 Mya-2.6 Mya, the average altitude of the Tibetan Plateau was raised by approximately 3,000 m, and was associated with high mountains and deep valleys that were generated in a way that separated several major rivers that ran in parallel [3,4]. The uplift of the Tibetan Plateau obstructed the northward flow of warm, wet-air from India across the mainland [8,9,10]. In addition, the Ice Age began about 2.8 million years ago [11]. Some studies have shown that ecological speciation events were profoundly accelerated by those geographic barriers, climatic changes and alternating glacial-interglacial periods [5,6,7,12,13]. In brief, our results suggested that the variance patterns of genetic structures of Moschidae may have resulted from: (1) the uplift of the Tibetan Plateau followed by increased aridification, and desertification in northern China in the Middle Pleistocene age, (2) the monsoon and the existence of Qinling Mountains and Liupan Mountains, and (3) the glacial cycles from the late Middle Pleistocene to the early late Pleistocene age.
Previously, Deng et al. proposed that the Tibet Plateau may represent a primitive form area for some Megaherbivores and large predators like Pseudois nayaur (Bovidae, Artiodactyla), Coeloonta thibetana (Rhinocerotidae, Perissodactyla), Equus kiang (Equidae, Perissodactyla), Panthera uncia (Felidae, Carnivora) [11]. Therefore, on the basis of the above discussion, we contended that the ancestor of the present musk deer might have originated from the Tibetan Plateau. Combining the data of current distributions and phylogenetic results of this study (Figs 1, 2 and 3), we suggested that the most direct ancestor of present musk deers were distributed in the Tibet Plateau margin or adjacent mountains that are located around the Sichuan basin.
During the past 5 million years, the orogenic movement and climate change prompted its divergence and speciation in Moschus. Lineage A, which is composed of M. leucogaster, M. fuscus and M. chrysogaster, represented the clade that inhabited the forested marginal area of the Tibet Plateau, and occupying land from Kashmir to Qinghai, China. Lineage B represented the clade that expanded to the far east (M. moschiferus), as well as the mountains that are located around the Sichuan Basin (M. berezovskii) and the Dabie mountain (M. anhuiensis). Although earliest musk deer fossils of the Genus Moschus (i.e., M. grandaevus, M. primaevus) were found in the north of Asia [21], we believe that they may not be the most direct ancestor of the present musk deer.