Received: 21 March 2018 | Revised: 25 April 2018 | Accepted: 10 June 2018 DOI: 10.1111/zsc.12303

ORIGINAL ARTICLE

Phylogeny and taxonomic reassessment of , Dipus (Rodentia, Dipodinae), in inland Asia

Jilong Cheng1,2 | Deyan Ge1 | Lin Xia1 | Zhixin Wen1 | Qian Zhang1 | Liang Lu3 | Qisen Yang1

1Key Laboratory of Zoological Systematics and Evolution, Institute of Zoology, Chinese Abstract Academy of Sciences, Beijing, China Deserts and arid regions are perceived to have low biological and genetic diversity, 2University of the Chinese Academy of which has partially influenced the identification of psammophilic species, especially Sciences, Beijing, China taxonomic inferences based on morphology alone. Recent studies of Dipus sagitta 3 State Key Laboratory for have revealed clear, deep genetic divergence beyond the species level within this Infectious Disease Prevention and Control, Collaborative Innovation monotypic genus. To clarify the taxonomy of Dipus, we examined a morphometric Centre for Diagnosis and Treatment of dataset consisting of 191 voucher specimens covering nearly the entire distribution Infectious Diseases, National Institute of the genus to explore skull variation using traditional morphological measurements for Communicable Disease Control and Prevention, Chinese Centre for Disease and geometric morphometric analysis. Phylogenetic relationships within Dipus using Control and Prevention, Beijing, China two mitochondrial genes (n = 383) and six nuclear genes (n = 106) were assessed by

Correspondence Bayesian inference and maximum‐likelihood procedures. We used a “candidate spe- Qisen Yang, Key Laboratory of Zoological cies approach” with the divergent mtDNA phylogenetic groups and subspecies iden- Systematics and Evolution, Institute of tified in previous studies as the starting point and analysed the candidates using five Zoology, Chinese Academy of Sciences, Beijing 100101, China. species delimitation methods and two validation methods. Our findings indicate that Email: [email protected] Dipus can be divided into four phylogenetic groups that include two species: the Funding information Deasyi group (D. deasyi), Sagitta group, Sowerbyi group and Turanicus group Grants of Key Laboratory of Zoological (D. sagitta). According to the morphological analyses and the examined specimens, Systematics and Evolution of the Chinese Academy of Sciences, Grant/Award pelage colour varies significantly with season and age, making it unsuitable as a di- Number: Y229YX5105 agnostic characteristic of the subspecies. Furthermore, measurements of body size and skull size require a large number of specimens to reach statistical significance and obtain reliable results. Geographical distributions should be considered first when identifying species or subspecies due to the disjunct habitats of Dipus. Our molecular analyses revealed the long‐neglected potential diversity in arid regions and improved the efficiency of species/subspecies identification. We also found that individuals from more humid areas or higher altitudes were larger, whereas individu- als from drier areas possessed longer appendages and larger tympanic bulla.

1 | INTRODUCTION Isolation of population fragments accelerates genetic and morphologic divergence, which ultimately leads to the emer- Geographic isolation is commonly considered an important gence of new species (Wang, Wan, Lim, & Yue, 2016). The step in allopatric speciation. Discontinuous landscapes have mid‐latitude arid zone of Eurasia provides a good natural ex- important consequences for the evolution of the species within ample of this phenomenon (the topography of this area is de- them (Krystufek, Klenovsek, Amori, & Janzekovic, 2015). tailed in Supporting Information Figure S1). It encompasses

630 | © 2018 Royal Swedish Academy of Sciences wileyonlinelibrary.com/journal/zsc Zoologica Scripta. 2018;47:630–644. CHENG et al. | 631 unique, complex landscapes with huge mountains and riv- et al., 2010) and greater than the average genetic distance ers that partition the overall arid zone into several large between sister species (9.55% on average; Baker & deserts (Cheng et al., 2017; Zhang, 2012). However, com- Bradley, 2006). Therefore, cryptic species diversity might pared to forest ecosystems, such as the Himalaya–Hengduan exist within Dipus. An integrative approach combining mor- Mountains in South‐west China (Wen et al., 2016), deserts phology and molecular data is essential for identifying and and arid regions are perceived as having low biological and describing species‐level diversity because it avoids both the genetic diversity, which partially influences the identification underestimation of species richness and the camouflaging of of psammophilic species (Brito et al., 2014; Durant et al., true phylogenetic relationships due to parallelisms and con- 2012), especially taxonomic inferences based on morphology vergence based on morphology alone as well as the increase alone (Petrova, Tesakov, Kowalskaya, & Abramson, 2016). in species numbers due to over‐splitting based on molecular Molecular species delimitation, whose use has expanded in data alone (Petrova et al., 2016). recent years, offers an integrative taxonomic approach that To revise the taxonomy within Dipus in the present study, relies on both molecular data and morphology to identify and we collected pictures of the skulls of specimens covering as describe species‐level diversity. much of the entire distribution of the genus as possible to The genus Dipus is distributed in the desert, semi‐des- explore intragenus skull variation by geometric morphomet- ert and steppe of the Don River from south‐eastern Europe rics (GM; Figure 1). We also collected morphological data of through Central Asia to north‐eastern China, and it has the body size and measured skull indices to understand morpho- largest geographical range among the Dipodinae, even among logical diversity. In addition to using molecular data from pre- Palaearctic desert (Lebedev et al., 2018; Shenbrot, vious studies, we added three new nuclear genes and applied Sokolov, Heptner, & Koval’skaya, 2008; Wilson & Reeder, different species delimitation methods to assess the stability 2005). Dipus has a specialized extended hind limb and sal- and congruence of the detected lineages. We attempted to (a) tatorial locomotion, making them well adapted for the desert reassess the taxonomy within the genus Dipus, (b) reassess the environment. In addition, Dipus was found to have a male‐bi- taxonomic status of D. deasyi Barrett‐Hamilton, 1900 and (c) ased dispersal characteristic (Shenbrot et al., 2008) . The first reconstruct the phylogenetic relationship within Dipus. Dipus fossils were found from the middle of the Late Miocene (7.50–9.10 Ma) in Central Inner Mongolia (Qiu & Li, 2005; Qiu, Wang, & Li, 2006); similarly, Pisano et al. (2015) es- 2 | MATERIALS AND METHODS timated the divergence time of Dipus at 8.66 Ma (7.67– 10.29 Ma) using molecular data. Dipus tended to spread from eastern parts in China and Mongolia to the western parts in 2.1 | Molecular analyses Kazakhstan during its evolutionary history (Lebedev et al., 2018). Dipus was long considered the least divergent in the 2.1.1 | DNA isolation, Dipodinae phylogenetic line and monotypic, including only amplification and sequencing the northern three‐toed jerboa, Dipus sagitta Pallas, 1773 We collected 383 sequences of cytochrome b (Cytb), 148 (Shenbrot et al., 2008; Wilson & Reeder, 2005). However, sequences of cytochrome oxidase subunit I (COI), 53 se- the taxonomy within Dipus is still somewhat ambiguous. quences of breast cancer protein 1 (BRCA1), 57 sequences Shenbrot et al. (2008) suggested that geographic variation in of exon 10 of the growth hormone receptor (GHR), 106 se- the northern three‐toed jerboa was largely manifested in the quences of exon 1 of the interphotoreceptor retinoid‐bind- dimensions and proportions of the skull and the fur colour. ing protein (IRBP) and 54 sequences of a portion of the However, after examining specimens from the eastern half recombination‐activating protein 1 (RAG1) from GenBank of the range of the species, Ma, Wang, Jin, and Li (1987) (accession numbers KX399483–KX399764, JX891491– reported that morphological measurements and pelage colour JX891500, JX891502–JX891511, JX891517–JX891524 and varied significantly with season and age due to wide distribu- MF535659–MF535907). All sequences came from 92 locali- tion and unstable characteristics of the subspecies. Moreover, ties covering almost the entire distribution of Dipus (Cheng the range of the variation within intra‐subspecies was even et al., 2017; Lebedev et al., 2018; Liao et al., 2015). In addi- greater than the average variation within inter‐subspecies. tion, we added two new nuclear genes [cannabinoid receptor Recent phylogeographic studies by Cheng et al. (2017) 1 (CNR1) and recombination‐activating protein 2 (RAG2)] and Lebedev et al. (2018) both revealed strong divergence and more sequences of the BRCA1 gene from 35 localities. within the monotypic genus Dipus in both mitochondrial and The new sequence dataset generated here is available in the nuclear genes, as supported by well‐resolved Bayesian trees GenBank repository with Accession numbers MG557704– and deep genetic distances in some clades (>10%). This dif- MG557728. Because of the shallow genetic divergence of ferentiation appears notably higher than that detected from nuclear genes, we included only one or two specimens per intraspecific studies of rodents (2.09% on average; Ben Faleh locality in the nuclear gene analyses. Detailed information on 632 | CHENG et al.

FIGURE 1 Sampling sites and extant distribution of Dipus, with major geographical features within the distribution area labelled. Locality codes and coordinates are presented in supplementary Supporting Information Table S1. The shape of the symbol marking the sampling sites represents the analyses used for each site. The colour of each sampling site indicates individuals belonging to corresponding subclades identified by molecular analyses directly or deductively. The potential correspondences between subspecies and subclades are also labelled with different colourful boxes. The red hollow hexagons indicate the typical localities of subspecies in a previous study: a, D. deasyi; b, D. s. aksuensis; c, D. s. nogai; d, D. s. innae; e, D. s. austrouralensis; f, D. s. turanicus; g, D. s. ubsanensis; h, D. s. lagopus; I, D. s. megacranius; j, D. s. usuni; k, D. s. sagitta; l, D. s. zaissanensis; m, D. s. bulganensis; n, D. s. halli; o, D. s. fuscocanus; p, D. s. sowerbyi the sampling localities is provided in Supporting Information was employed independently to select the best fit model of Table S1. base‐pair substitutions for each gene partition based on the Total genomic DNA was extracted using a TIANamp Bayesian information criterion (BIC), and we used MrBayes Genomic DNA Kit (DP304; Tiangen Biotech, Beijing, China) v.3.2.5 (Ronquist et al., 2012) to estimate the topologies. according to the manufacturer’s instructions. All amplifica- Two parallel runs, one cold and three heated, of Markov tions were conducted in 25‐μl reactions containing approxi- chain Monte Carlo (MCMC) analyses were performed for mately 25 ng of extracted DNA, 200 μm each dNTP, 0.2 μm 20 million generations or more, with trees sampled every each primer, 0.75 unit of LA Taq polymerase and 2.5 μl of 1,000 generations to produce convergence (SD < 0.01). The 10× buffer. The amplifications were performed using the fol- first 25% of the Markov chain samples (N = 20,000) were lowing PCR profiles: initial denaturation at 94°C for 3 min discarded as burn‐in, and the remaining samples were used followed by 35 cycles with denaturation at 94°C for 30 s, to generate majority rule consensus trees. Maximum like- annealing at 55–57°C (depending on the primers) for 30 s lihood phylogenies were inferred with RAxML v.8.2.10 and polymerization at 72°C for 1 min or 2 min (depending (Stamatakis, 2014) using a GTRGAMMAI evolution model on the length of the target), and a final extension at 72°C for and 1,000 bootstrapping replicates. Final trees were arranged 10 min. The primers and annealing temperatures along with and formatted with FigTree v1.4.2 (available at http://tree. the corresponding primary references are listed in Supporting bio.ed.ac.uk/software/fig-tree/). Information Table S2. 2.2 | Species delimitation 2.1.2 | Phylogenetic analysis To test species hypotheses for the clades of the northern three‐ Phylogenies were reconstructed using Bayesian inference toed jerboa, two different analyses were conducted using the analysis (BI) and maximum likelihood (ML) approaches different datasets (Carstens, Pelletier, Reid, & Satler, 2013; with combined mitochondrial genes and separate and com- Razkin, Gomez‐Moliner, Vardinoyannis, Martinez‐Orti, & bined nuclear genes. Each dataset was partitioned by gene, Madeira, 2017). First, five different species discovery meth- allowing each one to evolve at different rates. The program ods were used on the mitochondrial data to discover uncon- jmodeltest 2.1.7 (Darriba, Taboada, Doallo, & Posada, 2012) firmed candidate species; species validation methods were CHENG et al. | 633 then used on all molecular datasets to validate the candidate of substitutions. We implemented the PTP method in the ma- species. jority consensus tree from the above‐concatenated Bayesian analysis of the final dataset in the bPTP web server (http:// species.h-its.org/ptp/) for 500,000 generations, with thin- 2.2.1 | Molecular species discovery ning = 100 and burn‐in = 10%. The bPTP web server ran the As both genetic distances and P ID(Liberal) require a priori original ML version of the PTP as well as an updated version, species designation, we set 4–14 “candidate species” (i.e. which adds Bayesian support to delimited species in the input clades) based on a combination of phylogenetic topologies tree (Bayesian implementation of the PTP model or bPTP). from the gene trees (BI and ML trees; see the Results). To analyse genetic distance, we examined the mean intraspe- cific and interspecific Kimura two‐parameter (K2P) distance 2.2.2 | Molecular species validation and the p‐distance calculated for each “candidate species” in We used the program bp&p version 3.2 (Yang & Rannala, MEGA 6.04 (Tamura, Stecher, Peterson, Filipski, & Kumar, 2010) and *BEAST (Heled & Drummond, 2010) to validate 2013) and added telum sequences as a reference. the candidate species discovered with the mitochondrial The species delimitation plugin (Masters, Fan, & Ross, 2011) and nuclear sequence data. In bp&p, we used the A11 and in GENEIOUS 9.1.4 (Kearse et al., 2012) was used to ob- A01 algorithms to explore different species delimitation tain P ID(Liberal) statistical values and to calculate the mean models and species phylogenies. We ran reversible‐jump probability of the intra/inter genetic distance ratios for these MCMC (rjMCMC) analyses for 500,000 generations (with candidate species (Xu, Liu, Chen, Li, & Kuntner, 2015). a five generation sampling interval) with a burn‐in phase of The following three methods do not require assigning 100,000. The priors were set to θ ~ G (20, 2, 000) and τ ~ G a priori information of candidate “species”. We first used (20, 1, 000), and the other divergence time parameters were the Automatic Barcode Gap Discovery (ABGD) method assigned to the Dirichlet prior (Yang & Rannala, 2014). (Puillandre, Lambert, Brouillet, & Achaz, 2012) to calculate Each analysis was run twice with different starting seeds all pairwise distances in the dataset, evaluate intraspecific di- to ensure consistency. Concatenated mtDNA and nDNA vergences and sort the terminals into candidate species based sequences were analysed using *BEAST in BEAST 1.8.2 on the calculated p values. We performed ABGD analyses (Drummond, Suchard, Xie, & Rambaut, 2012), and the to- online (http://wwwabi.snv.jussieu.fr/public/abgd/) using three pology was obtained from BI and ML trees. The substitu- different distance metrics: Jukes–Cantor (JC69; Jukes, Cantor, tion models were unlinked, and the substitution parameters & Munro, 1969), K2P (Kimura, 1980) and simple distance were set according to the jmodeltest results. We chose the (p‐distance; Nei & Kumar, 2000). We analysed the data using Yule process tree priors model; the uncorrelated relaxed Pmin: 0.001, Pmax: 0.1, Steps: 10, X (relative gap width): 1 log‐normal clock was set for all loci. The mtDNA muta- or 1.5, with the other parameters set to default values. The tion rates were set to 2%, and nuclear genes were estimated Generalized Mixed Yule Coalescent method (GMYC) uses a according to the mtDNA. The MCMC chains were run for ML framework to delimit species based on ultrametric tree, 200 million generations with sampling every 5,000 genera- and it estimates the transition point on a tree, before which tions. The convergence of the MCMC chains was examined all nodes reflect species diversification events and after which in Tracer 1.6, and the first 25% were discarded as burn‐in. all nodes represent a population‐coalescent process (Pons et al., 2006; Razkin et al., 2017). We recovered a new mito- chondrial tree by BEAST 1.8.2 after removing identical se- 2.3 | Morphological analyses quences from the combined mitochondrial dataset because zero‐length terminal branches hinder likelihood estimates 2.3.1 | Taxon sampling (Fujisawa & Barraclough, 2013; Razkin et al., 2017). All the All the specimens for morphological analyses used in the present parameter settings were from a previous study (Cheng et al., study are preserved at the Institute of Zoology of the Chinese 2017). The GMYC tests were run using the web server (http:// Academy of Sciences (IOZCAS), the Zoological Museum of species.h-its.org/gmyc/) under the single‐threshold model be- Moscow State University (ZMMSU), the Zoological Institute cause prior studies have shown that the output of the multiple‐ of the Russian Academy of Sciences in Saint Petersburg (ZISP) threshold model is no better than that of the single‐threshold and the Natural History Museum (London). model (Fujisawa & Barraclough, 2013). The Poisson tree process (PTP) method (Zhang, Kapli, Pavlidis, & Stamatakis, 2013) is similar to the GMYC method in that it seeks to iden- 2.3.2 | Traditional morphometric methods tify significant changes in the rate of branching in a phylo- We explored differences in body size among four phyloge- genetic tree, but rather than using time to estimate branching netic groups (see the Results). The morphological data in- rates as in the GMYC model, the PTP directly uses the number cluded the body mass (BM), body length (BL), tail length 634 | CHENG et al. (TL), hind foot length (FL) and ear length (EL) of 174 These landmarks are showed in Supporting Information suitable voucher specimens from 46 localities. Juveniles Figure S2 and defined in Supporting Information were excluded from the analysis; these were identified Table S4. based on the sutures on the cranium and the occlusal sur- The raw datasets for each of the above three views were face of the tooth row (Lu et al., 2015). Eighteen measure- examined to test whether specimens greatly deviated from ments (Supporting Information Table S3) were taken on the average values. Morphological information (including each skull from 152 suitable voucher specimens using cal- size and shape) was extracted from these three datasets lipers with an accuracy of 0.01 mm according to the guide using a generalized full Procrustes fit (Goodall, 1991; of measurement in Xia, Yang, Ma, Feng, and Zhou (2006) Rohlf & Slice, 1990), and all configurations were scaled and Yang, Xia, Ma, Feng, and Quan (2005). to the same size after variations in size and orientation To characterize the differences among the four phylo- were removed. We later analysed the divergence in cra- genetic groups, standard statistics including the mean and nium size and shape by defining phylogenetic groups ob- standard error were applied. The pairwise differences in tained from molecular phylogenetic analysis (see Results). body size and skull measurements between phylogenetic Principal component analysis (PCA) was conducted to groups were tested by single‐factor analysis of variance visualize shape differences between individuals, and we (ANOVA) using both the least significant difference and used canonical variate analysis (CVA) to test the differ- Tukey’s test. If the result of the homogeneity of variance ences between different phylogenetic groups. All shape test for the measurement was not significant, Dunnett’s analyses were performed in MorphoJ 1.06d (Klingenberg, test was used. All calculations were performed using log 2011). 10‐transformed measurements to linearize age variation (Lissovsky, Yatsentyuk, & Ge, 2016). These analyses were performed using SPSS Statistics version 21.0 (SPSS, 3 | RESULTS Chicago, IL, USA). 3.1 | Molecular analyses For the new nuclear DNA sequences, BRCA1 (808 bp), CNR1 2.3.3 | Photography and digitization (1,078 bp) and RAG2 (1,032 bp) were obtained from 41, 34 A total of 191 skulls of the voucher specimens from 59 and 42 individuals, respectively. The best model accord- localities that covered most of the species distribution ing to the BIC criterion was determined to be GTR+I+G, (Figure 1) were included in the GM analyses (Adams, GTR+G, HKY+I, HKY+G, GTR+G, HKY+G, and HKY+I Rohlf, & Slice, 2004). For each skull, 3 standardized 2D for mtDNA, BRCA1, CNR1, GHR, IRBP, RAG1 and RAG2, images (Supporting Information Figure S2) were collected respectively. using a Canon PowerShot S5IS (Japan) with a macro‐focus- All pairs of trees resulting from Bayesian and ML ing lens: the cranial dorsal view (including the nasal, fron- analyses of the mtDNA genes showed very similar to- tal and parietal bones), the cranial ventral view (including pological structures (Figure 2), but ML bootstrap values the jugal, upper tooth row, auditory bulla and the palatine, (BS) were always lower than the corresponding Bayesian basisphenoid and basioccipital bones) and the mandible posterior probability values (PP*100). The phylogenetic lateral view (including the condylar, coronoid, and angular topologies agreed on four highly supported (PP > 0.90, processes and the lower tooth row). A scale was set in some BS > 60) phylogenetic groups, namely, the Deasyi images using a ruler with cm as the unit of measure. Images group, Turanicus group, Sagitta group and Sowerbyi were standardized for specimen position, camera lens plane group (Figure 2). The Deasyi group was monophyletic and the distance between the lens of the camera and the and placed as a sister to all other groups (PP = 1.00, specimen. BS = 100). The Turanicus group was subdivided into two subclades supported by PP = 0.89, BS = 73. The phylogenetic relationships in the Sagitta group contained 2.3.4 | Geometric morphometrics subclades IIIb, IIIc, IIId and IIIe, which were solidly To digitize evenly distributed points along the contours supported (PP > 0.95, BS > 77). Subclade IIIa1+2 had of structures, MakeFan8 (Sheets, Zelditch, & Swiderski, an extremely low support (PP = 0.59, BS = 36). The 2004) was used to establish guidelines before digitiza- Sowerbyi group was the most diversified group, with six tion. Landmark digitization was conducted with tpsDig2 subclades, IVa–f, with intermediate to high support. 2.30 software (Rohlf, 2017). We selected and digitized 9 The tree for the combined six nuclear genes was not landmarks in the dorsal view of the cranium (DVC), 25 completely in agreement with the mtDNA tree (Supporting landmarks in the ventral view of the cranium (VVC) and Information Figure S3). The Deasyi group had deep diver- 21 landmarks in the lateral view of the mandible (LVM). gence as a monophyly with PP = 1.00, BS = 100. The other CHENG et al. | 635 three groups showed significant paraphyly and clustered to- came from different groups. Similar clustering was ob- gether with low support. However, it is notable that IIb and served for subclades IIIa and IVc–f (Supporting Information IIIb clustered together, even though these two subclades Figure S3). 636 | CHENG et al.

FIGURE 2 Bayesian mtDNA gene tree for 217 terminals of Dipus, with the results of four different species delimitation approaches, including morphology. Numbers below branches show posterior probabilities and bootstrap support. Values beside clades names show mean interspecific genetic distances, calculated as Kimura two‐parameter/p‐distance. The correspondence between each subclade and geographical distribution is as follows: I, Tarim Basin and south‐west of Qaidam; IIa, North Caucasus, IIb, Turan Plain and west Kazakhstan; IIIa, North‐west Great Lake depression, IIIb, Aral Karakums and south Lake Balkhash, IIIc, Irtysh valley, IIId, West Zaisan depression, IIIe, Junggar Basin and Mongolian Dzungaria; IVa, East Mongolian Plateau and north‐east China, IVb, Central Mongolia and East Gobi, IVc, Ordos Plateau, IVd, Turpan Basin, IVe, Great Lake Basin, IVf, North‐east Qaidam, Hexi Corridor, Alashan, Transaltai Gobi

in 11 clusters and 22 entities with different confidence inter- 3.2 Species delimitation | vals of 7–17 and 16–30, respectively (Supporting Information Table S8), 2 of which included only one or two specimens 3.2.1 Molecular species discovery | (Figure 3). The ML obtained with the GMYC model was sig- Based on our priori hypotheses, the four groups showed ap- nificantly higher than the likelihood observed in the null model: proximately 10% interspecific distance among each other GYMC model = 884.1897, null model = 857.8582, likelihood (Supporting Information Table S5). The lowest interspecific ratio = 52.6631 and LR test <0.0001. The ML and Bayesian distances were 9.06% (K2P) between the Sowerbyi group solution PTPs identified 29 and 30 “species”, respectively, 20 and Turanicus group and 8.31% (p‐distance) between the of which included only one or two specimens (Figure 3). The Sagitta+IIIa group and Turanicus group, whereas the highest PhyloMap visualization of the PTP species delimitation result interspecific distance was 13.45/11.91% (K2P value on the showed six distinct taxa (Figure 2), consistent with previous left of divide, p‐distance value on the right of divide, same analyses. Consensus delimitation for the discovery of 11 can- means of expression in following cases) between the Deasyi didate species was obtained after grouping or removing candi- group and Sagitta+IIIa group. In total, the Deasyi group date entities containing less than five specimens (Razkin et al., showed a higher mean interspecific distance (12.91/11.54%; 2017), except IIa and IIIc (Figure 3). Figure 2) than the other phylogenetic groups; the Sowerbyi group had shorter mean genetic distances (8.57–9.59/7.83– 8.69%). The results based on the BI tree indicated high P 3.2.2 | Molecular species validation ID(Liberal) values of ≥0.95 (0.85–1.00) with four phyloge- The BP&P analysis identified 14 candidate “species” as netic groups or six subdivisions, which separated the Sagitta distinct clades (A11: PP > 0.85). The species tree of the group into two subclades (Supporting Information Table S6; combined mtDNA and nuclear genes showed an inconsist- Figure 2). The intra/inter value for the Deasyi group was <0.1, ent structure and low support compared to the mtDNA tree indicating shallow divergence within this group. However, (Figure 3). The Deasyi group was delimited with a posterior the Sagitta+IIIa group showed the highest intra/inter value probability of 1.00, whereas the other groups yielded different (0.49), clearly indicating deep genetic distance within this results. However, the *BEAST tree showed a phylogenetic group. The Turanicus and Sowerbyi groups showed interme- tree similar to that of mtDNA tree, except for the position of diate intra/inter values of 0.23 and 0.22, respectively. the Turanicus group (Figure 3). The Deasyi group was also The ABGD results agreed on 5–7 candidate “species” delimited with PP = 1.00. This relatively low support value (Supporting Information Table S7) using different parame- and variable topological structure in the two analyses is due ter combinations and the initial 12–14 partition. The Deasyi, to the weak divergence and slow evolutionary rate of the nu- Turanicus and Sowerbyi groups were considered independent clear genes. Only the Deasyi group could be differentiated candidate “species”, whereas the Sagitta group was separated from the other taxa. into IIIb/IIIc+d+e or IIIb+c+d+e; IIIa was separated into IIIa1/ IIIa2 or not (Figure 2). Different numbers of groups were ob- tained for recursive partition with differing prior maximal dis- 3.3 | Morphological analyses tances (P): 13–25 (P = 0.0010), 13–22 (P = 0.0017), 13–16 The general variation in the body size of the four groups (P = 0.0028), 13–14 (P = 0.0046), 9–10 (P = 0.0077), 5–7 is given in Supporting Information Tables S9 and S10. (P = 0.0129) and 5–7 (P = 0.0215). According to the sugges- The Sagitta group had the largest body size, with an aver- tions of Puillandre et al. (2012), initial and recursive partitions age BW of 86.64 ± 18.37 g (53–123 g, n = 48) and BL of were stable for P = 0.0129 (Razkin et al., 2017). We considered 125.69±10.15 mm (100–145 mm, n = 49), whereas the the accurate result is 5–7 candidate “species” (P = 0.0129). The Deasyi group possessed the largest appendages, with an following two analyses suggested more candidate “species” be- average TL of 173.42±10.96 mm (144–201 mm, n = 31), cause they identified additional subclades within the four phy- HL of 62.90±2.07 mm (58–67 mm, n = 31) and EL of logenetic groups. The single‐threshold GMYC model resulted 20.77±2.08 mm (17–25 mm, n = 31). Based on the ANOVA CHENG et al. | 637

FIGURE 3 Summary of the results of the species delimitation analyses (including both discovery and validation approaches) represented on a phylogenetic tree of the concatenated dataset. Grey boxes represent resultant candidate species (identified in discovery analyses); consensus discovery is in colourful boxes. Black lines along the GMYC and PTP indicate that less than four individuals were included in these candidate groups. The ultrametric tree was based on mtDNA data. The coalescent species tree was estimated by BPP and *BEAST based on all sequences. Posterior probabilities are shown at the nodes

results (Supporting Information Table S10), TL and EL of (Figure 4a). The VVC view showed the most clear variation the Deasyi group and HL of the Sowerbyi group were signifi- among the phylogenetic groups (p < 0.0001 in MD and PD; cantly different (p < 0.05). Figure 4b), except PD (p = 0.1670) between the Sagitta and For the skulls, only specimens of the Deasyi, Sagitta Turanicus groups. In the LVM view, the Turanicus group and Sowerbyi groups were measured. Six significantly vari- showed a largely overlapping area with the Sowerbyi group able indices (LBBO, GMB, LPF, WPF, LTB and ALCT) (Figure 4c), which was confirmed by PD (p = 0.1230). were screened out among the three groups (Supporting Information Tables S11 and S12). The Sagitta group had the largest LBBO of 10.81 ± 0.77 mm (9.48–12.23 mm, 4 | DISCUSSION n = 29) and GMB of 20.92 ± 0.66 mm (19.32–22.71 mm, n 4.1 | Phylogenetic relationships and = 29), but the smallest LPF of 5.16 ± 0.51 mm (4.18– Dipus 6.15 mm, n = 29). The Deasyi group had the smallest LBBO taxonomic reassessment in of 10.30 ± 0.42 mm (9.19–11.24 mm, n = 36) and GMB of The northern three‐toed jerboa, Dipus, has long been con- 19.06 ± 1.42 mm (16.85–21.52 mm, n = 36), but the largest sidered the least diversified in the Dipodinae phylogenetic LTB 9.82 ± 0.39 mm (9.02–10.88 mm, n = 36) and ALCT line, which includes only one species, D. sagitta (Pallas, 6.01 ± 0.30 mm (5.47–6.66 mm, n = 36). The ANOVA also 1773). This species is highly specialized to desert habitat resulted in the most significant variation between Deasyi and and is highly distinct (both genetically and morphologically) the other two groups (Supporting Information Table S12). from its closest living relatives (Stylodipus). In part due to A total of 191 intact skull voucher specimens were its fragmented distribution and unstable characteristics for available for the four phylogenetic groups (Deasyi group, partitioning subspecies, the traditional taxonomy of Dipus n = 35; Sagitta group, n = 55; Sowerbyi group, n = 94; has been problematic. In the current study, we used a “can- and Turanicus group, n = 7). We analysed skull morpho- didate species approach” with the divergent mtDNA clades logical divergence among the groups. No outliers were and subspecies identified in previous studies as the starting identified in DVC, VVC and LVM. Mahalanobis distances point (Cheng et al., 2017; Lebedev et al., 2018; Shenbrot (MD) and Procrustes distances (PD) were detected in CVA. et al., 2008). Because the nuclear and mitochondrial markers Significant shape variation between the Deasyi group and should be in concordance to document the presence or ab- the other groups was observed in all views (p < 0.0001 in sence of species and to avoid overestimation of species num- MD and p < 0.05 in PD; Figure 4a–c). In the DVC view, ber caused using only molecular data (Bezerra et al., 2016), all shape variations were significant (p < 0.01 in MD and morphometric analyses and molecular phylogenetic analyses p < 0.05 in PD), despite the partially overlapping area were combined. 638 | CHENG et al.

FIGURE 4 Variation in the shape and principal component analysis of the dorsal and ventral views of the cranium and lateral views of the mandible in voucher specimens. Canonical variant plots of the dorsal, ventral and mandible cranium are given in a, b and c, respectively. Principal component plots of the dorsal, ventral and mandible cranium are given in d, e and f, respectively. The data points on these plots are coloured according to taxonomic status based on the phylogenetic tree as well as traditional subspecies delimitation. The ellipses give 90% confidence equal frequencies for the four phylogenetic groups

Shenbrot et al. (2008) divided D. sagitta (i.e. Dipus) whole Dipus genus was clearly divided into four significant into the “Sagitta” and “Lagopus” groups based on the phylogenetic groups (i.e. the Deasyi group, Sagitta group, structure of the hind limbs (subdigital skin pad size and Sowerbyi group and Turanicus group; the origins of these lobe number), the colour of the inner hairs of the tuft on the names are discussed in the next section; Figures 1, 2 and 4). foot and the shape of the tail flag as well as the genitalia There is a clear segmentation of the geographic boundaries of and karyotype along the Irtysh River and boundary moun- the four groups in the broad mid‐latitude arid zone of Eurasia tains between China and Kazakhstan. Although karyosys- (Figure 1). The basic reason for the genetic barriers is that the tematic methods are particularly useful in the case of taxa suitable habitats of Dipus are interrupted by the formation that lie between the species level and the subfamily‐family of geographical boundaries (Cheng et al., 2017). Among the level, karyosystematics is of little help in differentiating four groups, the Deasyi group occupies the Tarim Basin and intraspecific and higher taxa and karyotype variability is the southern and western Qaidam Basin in China; the Sagitta a common phenomenon within rodent populations/species group is distributed from the Altay and Tianshan Mountains (McKenna, 1986; Vorontsov, 1979). Furthermore, phylo- westward to the Aral Sea; the Sowerbyi group occupies the geographic studies of D. sagitta (i.e. Dipus) have shown Mongolian Plateau and adjacent areas; and the Turanicus that only continuously large mountains can act as genetic group is distributed on most of the Turan Plain and lowland barriers, but the average elevation of the boundary moun- around the Caspian Sea. tains is approximately 1,000–2,000 m, which is not high The BI and ML trees highly support the four groups enough to obstruct genetic communication between the (PP > 0.93, BS > 61), and the K2P and p‐distance for each two side areas (Cheng et al., 2017). The individuals dis- group ranged from 9.06/8.31% to 13.06/11.91%. Notably, tributed in the Junggar Basin also presented longer genetic this differentiation level has been observed in other inter- distances from the individuals from the eastern distribution specific studies of rodents (Ben Faleh et al., 2012; Petrova, area (Figures 1 and 2). Therefore, this two‐group division Zakharov, Samiya, & Abramson, 2015). Because of the large cannot reflect the true differentiation. differentiation within the Sagitta group, P ID(Liberal) and Based on the results of molecular and morphological ABGD yielded a more detailed partition method, with 6 (P analyses, we proposed a new division method in which the ID ≥ 0.95; 0.84–1.00) and 5–7 subdivisions, respectively. It CHENG et al. | 639 is not inconsistent with the four groups; on the contrary, these sequences covering nearly the whole distribution of Dipus, two analyses supplemented additional information about the we reviewed the taxonomy within Dipus, including species divergence level. As it is suggested to choose delimitations and subspecies, particularly the relationship between genetic that are concordant among different species‐delimitation clades and traditional subspecies. analyses (Carstens et al., 2013), consensus delimitation of GMYC and PTP eliminated candidate groups/entities with 4.2.1 | Taxonomic status of Dipus deasyi, few specimens and selected 11 candidate species (Figure 3). D. s. aksuensis However, the structure of the nuclear genes revealed some Barrett‐Hamilton, 1900 and in paraphyletic subclades; only the Deasyi group was supported Deasyi group as a monophyly with deep divergence. This might be caused The Deasyi group showed robust differentiation from the by the male‐biased dispersal characteristic. The nuclear gene other groups in all analyses. Genetic variation of the Deasyi clades from neighbouring areas tended to cluster together, group was highly supported in the BI and ML trees and the rather than clades from the same phylogenetic group. For ex- genetic distance (11.54/12.91%) in mtDNA, consistent with ample, IIb and IIIb, IIIa and IVe in bp&p and IIa, IIb and IIIb the acknowledged range of interspecific genetic variation in the *BEAST species tree clustered together (Figure 3). (Baker & Bradley, 2006). The nuclear gene tree and all spe- Although Dipus appeared in the middle of the Late cies delimitation and validation analyses suggested that the Miocene (7.50–9.10 Ma) according to fossil records (Qiu & Deasyi group is a monophyly with high support. Furthermore, Li, 2005; Qiu et al., 2006), compared to other rodent taxa that morphological measurements agreed that the appendages experienced explosive radiation (Lv, Xia, Ge, Wu, & Yang, were significantly longer in the Deasyi group than the other 2016; Parada, Pardinas, Salazar‐Bravo, D’Elia, & Palma, groups; significant shape variation between the Deasyi group 2013), the extant species have a relatively low evo- and other groups was also observed in all views. It is reason- lutionary rate and a low level of diversification (Pisano et al., able to regard the Deasyi group as a separate species from the 2015). In the fossil Dipus species, changes in tooth size and other groups in Dipus. tooth crown height have been found, whereas morphological The Deasyi group mainly occupies the Tarim Basin and characteristics have remained similar since the late Miocene the southern and western Qaidam Basin in China (Figure 1). (Liu, Zhang, Cui, & Fortelius, 2008). In the GM analysis, The type localities of Dipus deasyi, Barrett‐Hamilton, 1900; the Turanicus group showed largely visualized overlap with D. s. aksuensis Wang, 1964 and D. s. fuscocanus Wang, the others in all views in CVA. In addition, the terrain of this 1964, are located in this area. We compared specimens from arid zone is relatively simple compared to the mountains in these localities, including (a) the only holotype specimen of mid‐ or low‐latitude regions, which favours the male‐biased D. s. fuscocanus without its skull; (b) six specimens, includ- dispersal and genetic communication that results in morpho- ing the holotype of D. s. aksuensis; and (c) type specimens logical conservatism and low divergence of nuclear genes of D. s. deasyi. We found that the pelage colour of D. s. fus- (Cheng et al., 2017; Shenbrot et al., 2008). cocanus was more similar to that of D. s. sowerbyi, and the In summary, we suggest that the whole genus Dipus should body size of D. s. aksuensis was truly smaller than that of be divided into four phylogenetic groups including two spe- D. s. deasyi. However, considering that the significant vari- cies: the Deasyi group (i.e. D. deasyi), with strong evidence ation in pelage colour and skull size is influenced by several from mtDNA, nuclear DNA and morphological data (dis- factors, it would be hasty to describe a new subspecies based cussed in the next section) and the Sagitta group, Sowerbyi on so few specimens. Furthermore, molecular data showed group and Turanicus group (i.e. D. sagitta). The deep diver- that samples from Aksu (locality 51; type locality of D. s. ak- gence in mtDNA among groups in D. sagitta already reaches suensis) do not cluster together as a monophyly in the Deasyi the interspecies level in rodents. However, due to the limits group. Samples from Heshuo (locality 54), which is near Korla of morphological data (especially for the Turanicus group) (type locality of D. s. fuscocanus), clustered as a monophyly and the paraphyletic structure from the nuclear gene tree, we in the Sowerbyi group in all molecular analyses. Therefore, conservatively regard the three groups as D. sagitta. we considered that (a) D. deasyi Barrett‐Hamilton, 1900 is the valid name for the new separate species and (b) D. s. ak- suensis is conspecific and a junior synonym of D. deasyi. 4.2 | Genetic divergence and subspecies Dipus deasyi Dipus was long considered a remarkably similar taxonomy of external form with inconstant identification characteristics Previous taxonomic studies of Dipus mainly used traditional from other taxa in Dipus. However, in the original descrip- morphological measurements and body shapes; nearly 17 tion, Barrett‐Hamilton (1900) noted that “…the external names of subspecies, including synonyms, were established. appearance resembles D. loftusi Blanford, but the colour of Detailed information on the subspecies with some modifica- the upper surface is richer and not so brown; the exact tint tions is provided in Table 1. Combined with the molecular being somewhere between ‘Ecru drab’ and ‘Fawn colour’. 640 | CHENG et al. TABLE 1 List of the distribution information and correspondence with subclades from the phylogenetic tree of taxa in Dipus, including species and subspecies levels

Subclade in Potential matching Phylogenetic group phylogenetic tree species/subspecies Distribution Revision & comments Deasyi group I D. deasyi Tarim Basin and southern and western Valid; syn. D. s. aksuen- Qaidam Basin sis Wang, 1964 Turanicus group IIa D. s. nogai Dagestan, Kalmykia, Astrakhan region Valid; need more (right bank Volga), left bank of North sample Don IIb D. s. austrouralensis Sandy deserts and semi‐deserts of Need more sample Ural‐E'mba interfluve D. s. innae Sandy deserts and semi‐deserts of Need more sample; syn. Volga‐Ural interfluve D. s. kalmikensis Kazantseva, 1940 D. s. turanicus Turan Plain and west Kazakhstan Valid Sagitta group IIIa1 D. s. ubsanensis Tuva Valid IIIa2 D. s. ssp.1 North‐west Great Lake depression Need more samples IIIb D. s. lagopus Sands of Greater and Lesser Barsuki, Valid Aral Karakums, sands of lower stream of Turgai(Tosynkum) and Sarysu (Sarysu Muyunkums, Zhetykonur) D. s. megacranius Chui Muyunkums Need more samples D. s. usuni Sandy massifs of southern Lake Need more samples Balkhash area and Ili basin IIIc D. s. sagitta Winding pine forests of right bank of Valid Irtysh in semi‐palatinsk and Pavlodar regions of Kazakhstan and south of Altai territory IIId D. s. zaissanensis North‐western part of Zaisan Hollow Valid IIIe D. s. bulganensis Junggar Basin (North‐western China, Valid Dzhungaria) and Mongolian Dzungaria Sowerbyi group IVa D. s. halli Eastern Mongolia, North‐eastern China Valid (Sandy land of Horqin, Hulun Buir and Hunshandake) IVb D. s. ssp.2 China (north of the Yinshan Mountains Need more morphologi- and Daqingshan Mnountains); cal sample Mongolia (Central and Eastern Gobi) IVc D. s. ssp.3 Erdos Plateau (Mu Us Sandy Land, Need more morphologi- Kubuqi Desert) cal sample and compare with D. s. sowerbyi IVd D. s. fuscocanus North‐eastern Kashgaria Need more sample IVe D. s. ssp.4 Valley of great lakes, basin of Gobi Need more morphologi- lakes, Gobi Altai cal sample IVf D. s. sowerbyi North‐east Qaidam, Hexi Corridor, Need to compare with Alashan, Transaltai Gobi D. s. ssp.3

Skull resembles that of D. lagopus Licht., but the teeth side of the anteriority concaves inward lightly, which re- are more massive and their pattern less complicated….” sults in three obtuse convex angles on the lateral side and Wang and Yang (1983) also noted that “… in specimens a total of five dental lamina…”. Ma et al. (1987) compared from South Xinjiang, the anterior fold of the first upper D. deasyi to D. s. sowerbyi and D. s. bulganensis (regarded molar is moved more to the lateral side, and the lingual as D. s. zaissanensis by the authors); according to their CHENG et al. | 641 description, although the pelage colour showed variation, support values of PP and BS in the phylogenetic tree. In ad- all specimens of D. deasyi showed lighter pelage (sandy dition, the phylogenetic position of IIIa was different from colour) and longer tails. that in a recent study by Lebedev et al. (2018). It is difficult After we reviewed a series of specimens, including to attribute IIIa to Sagitta group or Sowerbyi group or in- type specimens of D. deasyi and D. s. sowerbyi, we also dependent phylogenetic group. To avoid the Sagitta group observed differences in the number of dental lamina in being paraphyly, we temporarily do not process the position the first upper molar, although this unique characteristic of subclade IIIa, even if IIIa and Sagitta group presented can be influenced by tooth abrasion with age. In the skull, a monophyletic structure on the phylogenetic tree. Further D. deasyi has the smallest LBBO and GMB but the largest investigation and analysis need to be conducted to confirm LTB and ALCT, resulting in a sharper shape at the palate the position of IIIa and its relationship with Sagitta group (Supporting Information Table S11). A recent similar study and Sowerbyi group. by Lebedev et al. (2018) also suggested the taxonomic sta- tus of D. deasyi. 4.2.3 | Diversification and taxonomy within the Sowerbyi group 4.2.2 | Taxonomic discussion in the The Sowerbyi group was the taxon with the highest ge- Turanicus group and Sagitta group netic divergence; only three subspecies were identified in Two clades are separated within the Turanicus group and past research (Wilson & Reeder, 2005). The attributable correspond to the following subspecies: IIa – D. s. nogai, relationships between subclades and subspecies are IVa – western coast of the Caspian Sea; IIb – D. s. turanicus, D. s. halli, East Mongolian Plateau and north‐east China; D. s. innae and D. s. austrouralensis, the Turan Plain and IVb – D. s. ssp.2, Central Mongolia and East Gobi from west Kazakhstan. The Turanicus group was named after the north slope of Yinshan Mountains; IVc – D. s. ssp.3, the dominant taxon, D. s. turanicus. The Syr Darya ap- Ordos Plateau; IVd – D. s. fuscocanus, Turpan Basin; IVe – pears to have acted as a potential genetic barrier between D. s. ssp.4, Great Lake Basin; IVf – D. s. sowerbyi, North‐ the Turanicus and Sagitta groups (Figure 1); however, the east Qaidam, Hexi Corridor, Alashan, Transaltai Gobi. The clustered nuclear genes of IIb and IIIb reject this hypoth- Sowerbyi group takes its name from D. s. sowerbyi, which esis (Figure 3 and Supporting Information Figure S3). was named earlier than the other two subspecies. Morphological analyses also indicated no significant dif- In the present study, IVa showed robust position and high ferentiation between the Turanicus group and the other support in most analyses; the mtDNA genetic distance be- groups. Therefore, we still regarded the group as D. sag- tween IVa and the other clades in the Sowerbyi group varied itta, and more molecular samples and morphological speci- from 3.08% to 3.46% (Cheng et al., 2017), consistent with the mens are needed for analysis. acknowledged range of intraspecific genetic variation (Baker The Sagitta group is subdivided into four subclades, & Bradley, 2006). Dipus sagitta halli is widely acknowledged including IIIb – D. s. lagopus, D. s. megacranius and as a subspecies and is distributed in the easternmost areas of D. s. usuni, Aral Karakum and south coast of Lake Balkhash; the mid‐latitude arid zone of Eurasia (Shenbrot, 1991), con- IIIc – D. s. sagitta, Irtysh valley; IIId – D. s. zaissanensis, sistent with the distribution range of IVa. Furthermore, speci- west Zaisan depression; IIIe – D. s. bulganensis, Junggar mens sampled from localities belonging to IVa had the largest Basin and Mongolian Dzungaria. The Sagitta group was body size in our morphological data (not show), which is named after the nominate subspecies D. s. sagitta. IIIb shows also an important morphological diagnosis of D. s. halli. a large distribution area that includes the type localities of The living environment of D. s. halli provides more food three subspecies. The low divergence within IIIb is possible and a moderate environment compared with areas further in- due to the lack of obvious geographic barriers; the large mor- land, which could support more species. Increased interspe- phological variation among the three subspecies might be cific competition could promote larger body size (Du et al., caused by different climate factors (Supporting Information 2017; Ochocinska & Taylor, 2003). Therefore, we consider Figure S1). IIIc and IIId are distributed in very narrow areas, D. s. halli a valid subspecies. which may have acted as refugia during the Quaternary cli- Specimens from the areas where subclades IVb, IVc, mate change. IIIe is widespread with high haplotype diver- IVe and IVf are distributed have long been recognized as sity and low nucleotide diversity (Cheng et al., 2017), and the subspecies D. s. sowerbyi. Subclade IVd corresponds D. s. bulganensis matched this subclade. to subspecies D. s. fuscocanus. However, the phylogenetic Subclade IIIa includes IIIa1 – D. s. ubsanensis and analysis revealed surprising intraspecific diversification as IIIa2 – D. s. ssp.1 distributed in the North‐west Great Lake well as increased taxonomic doubt, especially the Central depression. The status of IIIa2 requires further investiga- Inner Mongolia region where three subclades coexisted tion. It needs to be explained separately because of its low here (Figure 1). The primary problem is that there are two 642 | CHENG et al. subclades, IVc and IVf, coexisting in the type locality of Du and Yongbin Chang for their help with both the fieldwork D. s. sowerbyi. This subspecies was first described based and software analyses. We thank Gregory I. Shenbrot for his on specimens from Yulinfu, Shensi, China (now north of suggestions and prior work on . We also thank the re- Yulin City, Shaanxi, China; near locality 46). Second, sub- searchers who submitted sequences from previous studies to clade IVf covers the widest ecological range with regard GenBank. This research was supported by grants from the to humidity and elevation, and the variation in elevation is Key Laboratory of Zoological Systematics and Evolution of from below sea level to the Qinghai–Tibetan Plateau (over the Chinese Academy of Sciences (No. Y229YX5105). 3,000 m). Specimens coming from north and east Qaidam Basin have larger bodies than those from Hexi Corridor. ORCID This partly verifies the viewpoint that the measurements and pelage colour show significant seasonal variation, and the Jilong Cheng https://orcid.org/0000-0003-0959-0521 intra‐subspecies variations were even larger than the average Qisen Yang https://orcid.org/0000-0001-9843-2378 inter‐subspecies variations (Ma et al., 1987). Third, after we checked the voucher specimens from the localities of IVb, IVc, IVd and IVf, our preliminary morphological data indi- REFERENCES cated unstable pelage colour and no significant divergence in GM analysis. Therefore, we temporarily attributed IVb‐f Adams, D. C., Rohlf, F. J., & Slice, D. E. (2004). Geometric mor- phometrics: ten years of progress following the ‘revolu- to D. s. ssp.2, D. s. ssp.3, D. s. fuscocanus, D. s. ssp.4, and tion’. Italian Journal of Zoology, 71, 5–16. https://doi. D. s. sowerbyi , respectively. The status of these subclades re- org/10.1080/11250000409356545 quires additional sampling efforts in the future. Baker, R. J., & Bradley, R. D. (2006). Speciation in and the genetic species concept. Journal of Mammalogy, 87, 643–662. https://doi.org/10.1644/06-MAMM-F-038R2.1 5 CONCLUSION Barrett‐Hamilton, G. (1900). On a small collection of mammals obtained | by Captain Deasy in South Chinese Turkestan and Western Tibet. Proceedings of the Zoological Society of London 68 Through an objective species delimitation approach includ- , , 196–197. Ben Faleh, A., Cosson, J. F., Tatard, C., Ben Othmen, A., Said, K., ing multilocus gene and morphological data, we propose & Granjon, L. (2010). Are there two cryptic species of the Dipus that can be divided into four phylogenetic groups in- lesser jerboa jaculus (Rodentia: Dipodidae) in Tunisia? cluding two species: the Deasyi group (D. deasyi); Sagitta Evidence from molecular, morphometric, and cytogenetic data. group, Sowerbyi group and Turanicus group (D. sagitta). Biological Journal of the Linnean Society, 99, 673–686. https:// According to the morphological analyses of the specimens doi.org/10.1111/j.1095-8312.2010.01374.x we examined, we suggested that pelage colour shows sig- Ben Faleh, A., Granjon, L., Tatard, C., Boratyński, Z., Cosson, J. nificant variation with season and age and is therefore un- F., & Said, K. (2012). Phylogeography of two cryptic spe- Jaculus Biological suitable as a diagnostic characteristic of the subspecies. cies of African desert jerboas (Dipodidae: ). Journal of the Linnean Society, 107, 27–38. https://doi. Measurements of body and skull size require a large number org/10.1111/j.1095-8312.2012.01920.x of specimens to achieve statistical significance and reliable Bezerra, A. M. R., Annesi, F., Aloise, G., Amori, G., Giustini, L., & results. Geographical distributions should be considered first Castiglia, R. (2016). Integrative taxonomy of the Italian pine when identifying Dipus species or subspecies due to the dis- voles, Microtus savii group (Cricetidae, Arvicolinae). Zoologica junct habitat of the genus. Our molecular analyses revealed Scripta, 45, 225–236. https://doi.org/10.1111/zsc.12155 the long‐neglected potential diversity in arid regions and Brito, J. C., Godinho, R., Martinez‐Freiria, F., Pleguezuelos, J. M., facilitated more efficient species/subspecies identification, Rebelo, H., Santos, X., … Carranza, S. (2014). Unravelling bio- diversity, evolution and threats to conservation in the Sahara‐ but additional morphological data, particularly from Central Sahel. Biological Reviews, 89, 215–231. https://doi.org/10.1111/ Asia, are needed to fully resolve the taxonomic status of the brv.12049 Turanicus group, Sagitta group and Sowerbyi group within Carstens, B. C., Pelletier, T. A., Reid, N. M., & Satler, J. D. (2013). this genus. We also found a tendency of individuals from How to fail at species delimitation. Molecular Ecology, 22, 4369– more humid areas or higher altitudes to be larger, whereas 4383. https://doi.org/10.1111/mec.12413 individuals from drier areas possessed longer appendages Cheng, J., Lv, X., Xia, L., Ge, D., Zhang, Q., Lu, L., & Yang, Q. (2017). and larger tympanic bulla. Experimental studies are needed Impact of orogeny and environmental change on genetic diver- Dipus sagitta to properly address issues related to adaptive evolution. gence and demographic history of (Dipodoidea, Dipodinae) since the Pliocene in inland east Asia. Journal of Mammalian Evolution, 000, 1–14. https://doi.org/10.1007/ ACKNOWLEDGEMENTS s10914-017-9397-6 Darriba, D., Taboada, G. L., Doallo, R., & Posada, D. (2012). jmodeltest We thank the editors and anonymous reviewers for providing 2: more models, new heuristics and parallel computing. Nature valuable comments. We sincerely thank Xue Lv, Yuanbao Methods, 9, 772. https://doi.org/10.1038/nmeth.2109 CHENG et al. | 643 Drummond, A. J., Suchard, M. A., Xie, D., & Rambaut, A. (2012). Liu, L., Zhang, Z., Cui, N., & Fortelius, M. (2008). The Dipodidae (jer- Bayesian phylogenetics with BEAUti and the BEAST 1.7. boas) from loc. 30 of baode and their environmental significance. Molecular Biology and Evolution, 29, 1969–1973. https://doi. Vertebrata PalAsiatica, 46, 124–132. org/10.1093/molbev/mss075 Lu, L., Ge, D., Chesters, D., Ho, S. Y. W., Ma, Y., Li, G., … Liu, Q. Du, Y., Wen, Z., Zhang, J., Lv, X., Cheng, J., Ge, D., … Yang, Q. (2017). (2015). Molecular phylogeny and the underestimated species The roles of environment, space, and phylogeny in determining diversity of the endemic white‐bellied rat (Rodentia: Muridae: functional dispersion of rodents (Rodentia) in the Hengduan Niviventer) in Southeast Asia and China. Zoologica Scripta, 44, Mountains, China. Ecology and Evolution, 7, 10941–10951. 475–494. https://doi.org/10.1111/zsc.12117 https://doi.org/10.1002/ece3.3613 Lv, X., Xia, L., Ge, D., Wu, Y., & Yang, Q. (2016). Climatic niche con- Durant, S. M., Pettorelli, N., Bashir, S., Woodroffe, R., Wacher, T., servatism and ecological opportunity in the explosive radiation De Ornellas, P., … Baillie, J.E.M. (2012). Forgotten biodiver- of arvicoline rodents (Arvicolinae, Cricetidae). Evolution, 70, sity in desert ecosystems. Science, 336, 1379–1380. https://doi. 1094–1104. https://doi.org/10.1111/evo.12919 org/10.1126/science.336.6087.1379 Ma, Y., Wang, F., Jin, S., & Li, S. (1987). Gliers (rodents and lago- Fujisawa, T., & Barraclough, T. G. (2013). Delimiting species using morphs) of northern Xinjiang and their zoogeographical distri- single‐locus data and the generalized mixed yule coalescent ap- bution. Beijing, China: Science Press. proach: a revised method and evaluation on simulated data sets. Masters, B. C., Fan, V., & Ross, H. A. (2011). Species delimitation Systematic Biology, 62, 707–724. https://doi.org/10.1093/sysbio/ ‐ a geneious plugin for the exploration of species boundar- syt033 ies. Molecular Ecology Resources, 11, 154–157. https://doi. Goodall, C. (1991). Procrustes methods in the statistical analy- org/10.1111/j.1755-0998.2010.02896.x sis of shape. Journal of the Royal Statistical Society Series B McKenna, M. C. (1986). Evolutionary relationships among rodents: a (Methodological), 53, 285–339. multidisciplinary analysis. Science, 231, 166–168. https://doi. Heled, J., & Drummond, A. J. (2010). Bayesian inference of species org/10.1126/science.231.4734.166 trees from multilocus data. Molecular Biology and Evolution, 27, Nei, M., & Kumar, S. (2000). Molecular evolution and phylogenetics. 570–580. https://doi.org/10.1093/molbev/msp274 Oxford, UK: Oxford University Press. Jukes, T. H., Cantor, C. R., & Munro, H. (1969). Evolution of protein Ochocinska, D., & Taylor, J. R. E. (2003). Bergmann's rule in shrews: molecules. Mammalian Protein Metabolism, 3, 132. geographical variation of body size in Palearctic Sorex species. Kearse, M., Moir, R., Wilson, A., Stones‐Havas, S., Cheung, M., Biological Journal of the Linnean Society, 78, 365–381. https:// Sturrock, S., … Drummond, A. (2012). Geneious basic: an in- doi.org/10.1046/j.1095-8312.2003.00150.x tegrated and extendable desktop software platform for the or- Parada, A., Pardinas, U. F. J., Salazar‐Bravo, J., D'Elia, G., & Palma, ganization and analysis of sequence data. Bioinformatics, 28, R. E. (2013). Dating an impressive Neotropical radiation: mo- 1647–1649. https://doi.org/10.1093/bioinformatics/bts199 lecular time estimates for the Sigmodontinae (Rodentia) provide Kimura, M. (1980). A simple method for estimating evolutionary rates insights into its historical biogeography. Molecular Phylogenetics of base substitutions through comparative studies of nucleotide and Evolution, 66, 960–968. https://doi.org/10.1016/j. sequences. Journal of Molecular Evolution, 16, 111–120. https:// ympev.2012.12.001 doi.org/10.1007/BF01731581 Petrova, T. V., Tesakov, A. S., Kowalskaya, Y. M., & Abramson, Klingenberg, C. P. (2011). MorphoJ: an integrated software package N. I. (2016). Cryptic speciation in the narrow‐headed vole for geometric morphometrics. Molecular Ecology Resources, 11, Lasiopodomys (Stenocranius) gregalis (Rodentia: Cricetidae). 353–357. https://doi.org/10.1111/j.1755-0998.2010.02924.x Zoologica Scripta, 45, 618–629. https://doi.org/10.1111/ Krystufek, B., Klenovsek, T., Amori, G., & Janzekovic, F. (2015). zsc.12176 Captured in ‘continental archipelago’: phylogenetic and envi- Petrova, T. V., Zakharov, E. S., Samiya, R., & Abramson, N. I. (2015). ronmental framework of cranial variation in the European snow Phylogeography of the narrow‐headed vole Lasiopodomys vole. Journal of Zoology, 297, 270–277. https://doi.org/10.1111/ (Stenocranius) gregalis (Cricetidae, Rodentia) inferred from jzo.12274 mitochondrial cytochrome b sequences: an echo of pleistocene Lebedev, V. S., Bannikova, A. A., Lu, L., Snytnikov, E. A., Adiya, Y., prosperity. Journal of Zoological Systematics and Evolutionary Solovyeva, E. N., … Shenbrot, G. I. (2018). Phylogeographical Research, 53, 97–108. https://doi.org/10.1111/jzs.12082 study reveals high genetic diversity in a widespread desert ro- Pisano, J., Condamine, F. L., Lebedev, V., Bannikova, A., Quere, J. P., dent, Dipus sagitta (Dipodidae: Rodentia). Biological Journal Shenbrot, G. I., … Michaux, J. R. (2015). Out of Himalaya: the of the Linnean Society, 123, 445–462. https://doi.org/10.1093/ impact of past Asian environmental changes on the evolutionary biolinnean/blx090 and biogeographical history of Dipodoidea (Rodentia). Journal Liao, J., Jing, D., Luo, G., Wang, Y., Zhao, L., & Liu, N. (2015). of Biogeography, 42, 856–870. https://doi.org/10.1111/jbi.12476 Comparative phylogeography of Meriones meridianus, Dipus Pons, J., Barraclough, T. G., Gomez‐Zurita, J., Cardoso, A., sagitta, and sibirica: potential indicators of the Duran, D. P., Hazell, S., … Vogler, A. P. (2006). Sequence‐ impact of the Qinghai‐Tibetan Plateau uplift. Mammalian based species delimitation for the DNA taxonomy of unde- Biology‐Zeitschrift für Säugetierkunde, 81, 31–39. https://doi. scribed insects. Systematic Biology, 55, 595–609. https://doi. org/10.1016/j.mambio.2015.05.002 org/10.1080/10635150600852011 Lissovsky, A. A., Yatsentyuk, S. P., & Ge, D. (2016). Phylogeny and Puillandre, N., Lambert, A., Brouillet, S., & Achaz, G. (2012). taxonomic reassessment of pikas Ochotona pallasii and O. argen- ABGD, automatic barcode gap discovery for primary species tata (Mammalia, Lagomorpha). Zoologica Scripta, 45, 583–594. delimitation. Molecular Ecology, 21, 1864–1877. https://doi. https://doi.org/10.1111/zsc.12180 org/10.1111/j.1365-294X.2011.05239.x 644 | CHENG et al. Qiu, Z., & Li, Q. (2005). Restudies in Sminthoides schlosser, a fossil Wen, Z., Yang, Q., Quan, Q., Xia, L., Ge, D., & Lv, X. (2016). Multiscale genus of three‐toed jerboa from China. Vertebrata PalAsiatica, partitioning of small β‐diversity provides novel insights 43, 24–35. into the Quaternary faunal history of Qinghai‐Tibetan Plateau and Qiu, Z., Wang, X., & Li, Q. (2006). Faunal succession and biochro- Hengduan Mountains. Journal of Biogeography, 43, 1412–1424. nology of the Miocene through Pliocene in Nei Mongol (Inner https://doi.org/10.1111/jbi.12706 Mongolia). Vertebrata PalAsiatica, 44, 164–181. Wilson, D. E., & Reeder, D. M. (2005). Mammal species of the world: Razkin, O., Gomez‐Moliner, B. J., Vardinoyannis, K., Martinez‐Orti, A., a taxonomic and geographic reference. Baltimore, MD: Johns & Madeira, M. J. (2017). Species delimitation for cryptic species Hopkins University Press. complexes: case study of Pyramidula (Gastropoda, Pulmonata). Xia, L., Yang, Q., Ma, Y., Feng, Z., & Zhou, L. (2006). A guide to the Zoologica Scripta, 46, 55–72. https://doi.org/10.1111/zsc.12192 measurement of mammal skull III: Rodentia and Lagomorpha. Rohlf, F. J. (2017). Tps series. Department of Ecology and Evolution, Chinese Journal of Zoology, 41, 68–71. State University of New York, Stony Brook, NY. Retrieved from Xu, X., Liu, F. X., Chen, J., Li, D. Q., & Kuntner, M. (2015). Integrative http://life.bio.sunysb.edu/morph/. taxonomy of the primitively segmented spider genus Ganthela Rohlf, F. J., & Slice, D. (1990). Extensions of the procrustes method for (Araneae: Mesothelae: Liphistiidae): DNA barcoding gap agrees the optimal superimposition of landmarks. Systematic Biology, with morphology. Zoological Journal of the Linnean Society, 39, 40–59. https://doi.org/10.2307/2992207 175, 288–306. https://doi.org/10.1111/zoj.12280 Ronquist, F., Teslenko, M., van der Mark, P., Ayres, D. L., Darling, Yang, Z., & Rannala, B. (2010). Bayesian species delimitation using A., Hohna, S., … Huelsenbeck, J. P. (2012). MrBayes 3.2: effi- multilocus sequence data. Proceedings of the National Academy cient bayesian phylogenetic inference and model choice across a of Sciences of the United States of America, 107, 9264–9269. large model space. Systematic Biology, 61, 539–542. https://doi. https://doi.org/10.1073/pnas.0913022107 org/10.1093/sysbio/sys029 Yang, Z., & Rannala, B. (2014). Unguided species delimitation using Sheets, H. D., Zelditch, M., & Swiderski, D. (2004). Morphometrics DNA sequence data from multiple loci. Molecular Biology and software: IMP‐integrated morphometrics package. Department Evolution, 31, 3125–3135. https://doi.org/10.1093/molbev/ of physics, Canisius College, Buffalo, NY. Retrieved from http:// msu279 www3.canisius.edu/~sheets/morphsoft.html Yang, Q., Xia, L., Ma, Y., Feng, Z., & Quan, G. (2005). A guide to Shenbrot, G. I. (1991). Geographical variation of the three‐toed the measurement of mammal skull I: basic measurement. Chinese brush‐footed jerboa, Dipus sagitta (Rodentia, Dipodidae). II, Journal of Zoology, 40, 50–56. Subspecific differentiation in the eastern Kazakhstan, Touva and Zhang, L. (2012). Palaeogeography of China‐the formation of Chinese Mongolia. Zoologichesky Zhurnal, 70, 91–97. natural evironment. Beijing, China: Science Press. Shenbrot, G. I., Sokolov, V. E., Heptner, V. G., & Koval'skaya, Y. M. Zhang, J., Kapli, P., Pavlidis, P., & Stamatakis, A. (2013). A general spe- (2008). Mammals of Russia and adjacent regions: jerboas. New cies delimitation method with applications to phylogenetic place- Delhi, India: Amerind Publishing Co., Pvt. Ltd. ments. Bioinformatics, 29, 2869–2876. https://doi.org/10.1093/ Stamatakis, A. (2014). RAxML version 8: a tool for phylogenetic anal- bioinformatics/btt499 ysis and post‐analysis of large phylogenies. Bioinformatics, 30, 1312–1313. https://doi.org/10.1093/bioinformatics/btu033 Tamura, K., Stecher, G., Peterson, D., Filipski, A., & Kumar, S. (2013). SUPPORTING INFORMATION MEGA6: molecular evolutionary genetics analysis version 6.0. Additional supporting information may be found online in Molecular Biology and Evolution, 30, 2725–2729. https://doi. org/10.1093/molbev/mst197 the Supporting Information section at the end of the article. Vorontsov, N. N. (1979). Karyosystematics. The Great Soviet encyclo- pedia. Farmington Hills, MI: The Gale Group. Retrieved from https://encyclopedia2.thefreedictionary.com/karyosystemati-cs. How to cite this article: Cheng J, Ge D, Xia L, et al. Wang, L., Wan, Z. Y., Lim, H. S., & Yue, G. H. (2016). Genetic vari- Phylogeny and taxonomic reassessment of jerboa, ability, local selection and demographic history: genomic evi- Dipus (Rodentia, Dipodinae), in inland Asia. Zool Scr. dence of evolving towards allopatric speciation in Asian seabass. 2018;47:630–644. https://doi.org/10.1111/zsc.12303 Molecular Ecology, 25, 3605–3621. https://doi.org/10.1111/ mec.13714 Wang, S., & Yang, G. (1983). Fauna of rodent in Xinjiang. Urumqi, China: Xinjiang People's Publishing House.