A Molecular Phylogeny of Stylodipus (Dipodidae, Mammalia): A Small Genus with a Complex History

Abstract

A range-wide phylogenetic/phylogeographic study of the three-toed jerboas of the genus Stylodipus is conducted using the mitochondrial cytb gene and fragments of several nuclear genes. The genus has been believed to include three species: S. telum (W Central Asia, SE Europe), S. andrewsi (E Central Asia), and S. sungorus (Dzungar basin). Our data support the dichotomy between S. andrewsi and the other taxa forming S. telum species group. Within the latter, both mtDNA and nuclear loci indicate a species-level divergence between S. telum and the S. t. birulae lineage (Zaisan depression, NE Kazakhstan), previously considered a subspecies of S. telum and here elevated to full species. S. sungorus is recovered as a close sister group to S. birulae on the basis of nuclear data but clustered with S. telum in the mitochondrial tree. The latter taxon is the most variable and includes two closely related eastern and western sublineages, separated by the Volga-Ural sands and joined by a more divergent S. t. karelini lineage (E Kazakhstan). The observed mitonuclear discordance is hypothesized to occur due to mtDNA introgression resulting from hybridization between S. sungorus and S. t. karelini, which highlights the important role of reticulations in the evolution of Dipodidae.

1. Introduction

The study of geographical structuring of genetic variation across the distribution range of a species is of inexhaustible interest because such data can be interpreted as an echo of changes in the historical range, described in terms of extinction across large areas, survival in small refugia, and subsequent re-colonization. Jerboas of Asian deserts and semi-deserts are an important model object for phylogeographic studies in the arid open landscapes due to their key role in this zone; however, their genetic diversity and demographic history are still known insufficiently. Such studies can serve as an important source of information on the dynamics of changes in the paleogeographic conditions of the Middle East and Central Asia, in particular, on the processes of aridization of Asia in the Pleistocene.

Among the three-toed jerboas (Dipodinae) and jerboas in general, genus Stylodipus is unique in its pattern of distribution, as it is endemic to a relatively narrow zone of Eurasian semi-deserts and northern deserts [1]. The range of the genus stretches across both eastern (Mongolian) and western (Kazakhstan) sectors of Central Asia, reaching the Black Sea region in the west. However, its range does not include most of the large sandy deserts located to the south, such as Karakum, Kyzylkum (except the northernmost part), Alashan (but the genus is present in Alashan mountains), and Ordos [1,2,3]. In contrast to most other dipodines (such as Dipus, Eremodipus, and Paradipus), Stylodipus species are not specialized psammophiles and instead have opportunistic soil affinities. They occur in stabilized sands near the western and northern range limits, but in the main part of its geographic range, these species avoid sands and hard clay soils and prefer sandy-gravel, sandy-loam, and loam soils [1]. Another feature unique among the three-toed jerboas (but shared with Eremodipus) is the ability of Stylodipus to store fat in the tail. In addition, representatives of the genus are characterized by relatively short hind limbs and, correspondingly, the least efficient ricochetal locomotion within its subfamily.

For a long time, it was believed that the western part of the range of genus Stylodipus was occupied by the fat-tailed three-toed jerboa S. telum Lichtenstein, 1823, while the eastern part was inhabited by the Mongolian three-toed jerboa S. andrewsi Allen, 1925. Later, the Dzungarian three-toed jerboa S. sungorus Sokolov and Shenbrot, 1987 was separated from S. telum (Lichtenstein, 1823) as a full species.

Thus, according to the current taxonomy, Stylodipus genus contains three species: first, S. andrewsi is found in semi-desert areas of the Mongolian Gobi and northern China (Inner Mongolia, northern Gansu, and Ningxia). Second, S. telum is a widely distributed species common in Kazakhstan from the Ural River to the Zaisan basin, also present in northern Kyzylkum, the lower Dnieper region, southeastern parts of European Russia, and possibly in northwestern China. Finally, S. sungorus inhabits the Dzungarian basin in southwestern Mongolia and northwestern China. According to the latest morphological revision [4], S. telum includes six subspecies: S. t. telum Lichtenstein, 1823; S. t. amankaragai Selevin, 1934; S. t. birulae Vinogradov, 1937; S. t. falzfeini Brauner, 1913; S. t. nastjukovi Shenbrot, 1991; and S. t. turovi Heptner, 1934. S. andrewsi and S. sungorus are considered monomorphic.

Previous molecular data have shown that the closest sister group to Stylodipus is the recently described Chimaerodipus Shenbrot, Bannikova, Giraudoux, Quéré, and Lebedev, 2017 from Ningxia (China) [5] and that S. andrewsi occupies a basal position within the genus [6]. However, there is still nearly no data on the genetic diversity and phylogeographic structure of Stylodipus species, with the exception of data by Rusin [7], limited to the western populations of S. telum.

The aim of the present study was to examine the phylogeographical structure of the species of the genus Stylodipus with a focus on the correlation between recognized morphological subspecies and genetic variability and to estimate the timing and geographic patterns of divergence among the major geographical lineages.

2. Materials and Methods

2.1. Sampling, DNA Extraction, Amplification, and Sequencing

Most samples used in the study are stored in the tissue collection of the Zoological Museum of Moscow State University (ZMMU). Samples from the Kherson region and Tsimla sands are kindly provided by M. Rusin. All vouchers (except those from the personal collection of M. Rusin) are stored in ZMMU.

The entire sample used in the main analyses (data set A) included 102 specimens of Stylodipus spp. from 35 localities (Figure 1, Supplementary Information S1, Table S1A). Among them, the original sample consisted of 39 specimens of S. telum, 34 specimens of S. andrewsi, and 22 specimens of S. sungorus from 31 localities across the distribution ranges of these species. For all specimens in this data set, we sequenced the complete cytochrome b gene (cytb); for the essential part of the sample, we obtained fragments of four nuclear loci: IRBP (interphotoreceptor retinoid-binding photoreceptor, exon), THY (thyrotropin gene, intron), CHY (substrate-regulated cyanide hydratase gene, intron), and DMP1 (dentin matrix acidic phosphoprotein 1, exon). Molecular dating and species tree reconstructions were performed using extended nuclear data (data set B) that included eight more loci (BRCA1, RAG1, GHR, SPTBN, OSTA, ROGDI, ABDH11, and AAT) sequenced in a few specimens of Stylodipus and the outgroups (Supplementary Information S1, Table S1B). Additional sequences of the cytb gene and nuclear loci were retrieved from GenBank (Supplementary Information S1, Table S2).

In most cases, DNA was extracted from muscle tissue preserved in ethanol or from small tissue biopsies of live-trapped animals. Additionally, 11 specimens of S. telum were represented by archive DNA extracted from museum skins or bones (the age of specimens ranging from 50 to 80 years).

Total DNA from the recent material was extracted from ethanol-preserved tissues of the kidney, muscles, and ear clippings using a standard protocol of proteinase K digestion, phenol-chloroform deproteinization, and isopropanol precipitation [8]. The cytb gene was amplified with primers by Montgelard et al. [9] and sequenced using primers specifically designed for this study. Primers for amplification and sequencing of the IRBP, BRCA1, RAG1, GHR, SPTBN, and THY loci and the corresponding PCR protocols were described in previous studies [10,11,12]. Primers and protocols designed for amplification and sequencing of the CHY, DMP1, OSTA, ROGDI, ABDH11, and AAT loci, as well as new internal primers for IRBP, THY, and cytb genes are given in Supplementary Information S1, Table S3.

DNA from dried skins and bones of museum specimens was purified directly using the DNeasy Blood and Tissue kit and QIAamp DNA Investigator kit (QIAGEN), following the manufacturer’s protocol and recommendations of Yang et al. [13] with the following modifications: lysis was performed overnight at 56 °C; elution buffer was heated to 56 °C before applying it to the column, after which the column was incubated at room temperature for 7–10 min before centrifugation at 14,000 rpm (Eppendorf Minispin benchtop centrifuge) for 1 min. Carrier RNA was used for extraction with the Investigator kit. DNA extracted from museum specimens was highly degraded; thus, for most historical specimens, we amplified short fragments using combinations of previously mentioned primers and additional primers designed specifically for this study (Tables S3 and S4). The PCR program for amplification of short fragments included the initial denaturation at 95 °C for 3 min, 45 cycles at 95 °C for 30 s, 52–62 °C (depending on primer used) for 30 s, 72 °C for 30 s, and a final extension at 72 °C for 6 min. All stages of the extraction process included a negative control run in parallel. To avoid contamination, the extraction and amplification of the DNA from the museum specimens were conducted in the ZMMU Laboratory of Historical DNA, exclusively equipped for work with the museum DNA specimens, where no previous work on fresh tissues had been performed. We ran aliquots (10 µL) of the extractions alongside a 100-bp ladder on a 1% agarose gel by electrophoresis.

The sequences obtained in this study can be accessed via GenBank (Accession numbers: OR650835-651236, Table S2).

2.2. Phylogenetic Analysis

All sequences were aligned by eye using Bioedit version 7.0.9.0 [14]. Heterozygous positions in nuclear gene sequences were coded using the IUB Ambiguity Codes. For the cytb dataset and the concatenated alignment of four nuclear genes, the phylogenetic trees were reconstructed under maximum likelihood (ML) and Bayesian inference (BI) criteria. Gene trees for individual nuclear loci were reconstructed using ML only. In these analyses, nuclear sequences were used as unphased genotypes.

In the combined analyses of the four nuclear genes, the final alignment consisted of 2739 bp, including 1011 bp of IRBP, 551 bp of CHY, 548 bp of THY, and 539 bp of DMPI. The data on two exonic markers (IRBP and DMPI) were partitioned into 1st + 2nd and 3rd codon positions while THY and CHY were left unpartitioned. In total, the concatenated dataset included 57 specimens of Stylodipus for which the sequences of all four genes were available. The composite outgroup consisted of Jaculus jaculus, Dipus sagitta, and Chimaerodipus auritus. Sequences from additional specimens were used in the reconstruction of individual gene trees only.

The final cytb alignment included 1140 bp for 93 specimens of Stylodipus. To root the trees, we used two sequences of Chimaerodipus auritus, which is the closest outgroup for Stylodipus. In all analyses, the cytb dataset was partitioned into three codon positions.

Maximum likelihood reconstructions were conducted in IQTree version 1.6 [15]. The ModelFinder routine [16] was used to determine the optimal partitioning scheme and best-fit substitution models for each subset under the BIC criterion. Clade stability was tested using Ultrafast Bootstrap [17] with 10,000 replicates.

Bayesian ultrametric trees were inferred in BEAST 1.10 [18,19], assuming gene-specific relaxed lognormal clocks and a birth–death model as the tree shape prior. Substitution models were set as in the ML analysis. The chain length was set to 50 million generations. Convergence was tested using Tracer v1.7 [20].

Estimation of the genetic p-distances was conducted in MEGA6 [21].

Sequences of the nuclear genes were also used as phased genotypes. For allelic phase reconstruction of nuclear genes, the PHASE module [22,23] implemented in the software DNAsp ver.5 [24] was used. Alleles with posterior probabilities below 0.8 were excluded from the analyses. Relationships among alleles were illustrated using median-joining networks constructed in POPART v.1.7 [25] under default options.

2.3. Molecular Dating and Species Tree Reconstruction

To reconstruct a calibrated species tree, we used the StarBEAST3 algorithm [26] implemented in BEAST2 [27]. This approach utilizes the species tree relaxed clock model [28]. Based on the results of the concatenated analysis, the following lineages were selected as taxonomic units for the species tree reconstruction: S. andrewsi, S. sungorus, S. t. birulae, S. telum-West, and S. telum-East. The latter two lineages consist of the specimens of S. telum occurring eastwards and westwards from the Volga River, respectively. Specimens of S. telum from East Kazakhstan (Chinghiztau and Balkhash region) were excluded since we suspected them of being under the influence of gene flow from neighboring lineages (see Results). Jaculus, Dipus, and Chimaerodipus were used as outgroups. The details of specimens’ assignment to taxonomic units are given in Supplementary Material Table S1B.

To reduce the time estimate errors, we extended the nuclear data set by including an additional eight loci (see above, data set B) that were only sequenced in a few specimens of Stylodipus and the outgroups (Supplementary Material Table S1B). The IRBP, THY, CHY, and DMPI alignments contained allelic data as produced by Phase; heterozygous sites in other genes were phased randomly. Each gene was allowed to have its own clock rate. Best-fit substitution models for each subset were determined in a supplementary analysis performed in ModelFinder (Table S5). Yule model was implemented as the tree-shape prior. The chain length was set to 200 million generations. Convergence was tested using Tracer v1.7.

To calibrate the tree, we used a set of secondary calibration points produced in an earlier phylogenetic study of Dipodinae [5]; the details on prior calibration densities are given in Table S6.

2.4. Biogeographic Paleo-Reconstructions

We based paleo-reconstructions of the distribution patterns on paleoenvironmental conditions. The main assumption of the employed method is that early stages of divergence between phylogenetic lineages occurred via geographic speciation. In other words, a unified geographic range of an ancestor was subdivided into isolated parts due to changes in climate, and divergence of phylogenetic lines started in these isolated parts as a result of local adaptations and the absence of gene flow between the isolates. According to the “niche conservatism hypothesis” [29], “ecological niches evolve little at or around the time of speciation events”. The consequence of this assumption is that species distribution modeling (SDM) of an ancestor and its close descendants can be based on environmental conditions in combined points of occurrences of these descendants. Previously, this approach was employed for paleo-reconstructions of distribution range history within Allactaginae [12].

The occurrence records of Stylodipus spp. and Chimaerodipus auritus used in the species distribution modeling (SDM) were obtained from our field observations, ZMMU collections, and other collections accessed via the Global Biodiversity Information Facility (GBIF, https://www.gbif.org (accessed on 2nd January 2022)), and literature sources (Supplementary Information S1 Figure S1).

The environmental data used for SDM were 30 arc-second grids (approximately 1 km resolution) of climate. Current climate data (BioClim 1–19) were obtained from WORLDCLIM Version 2.0 [30]; paleo-climate data for intervals 20,000–5400 kya were obtained from the Oscillayers dataset [31]. Paleo-climate data originally available at 2.5 arc-minutes were upscaled to 30 arc seconds.

The study area or the “landscape of interest” significantly affects SDM results [32,33]. To define the study area of a species, we clipped environmental variables using a rectangle mask covering the area of the occurrence points plus 4 degrees in each direction. The data for SDM were prepared using ArcGIS Desktop: Copyright © 1995–2020 Esri. The SDMs were built using MAXENT 3.4.0 [34], available at https://biodiversityinformatics.amnh.org/open_source/maxent/ (accessed on 24 November 2020). The model for current environments was constructed with default MAXENT settings as these settings were demonstrated to be the most appropriate for wide-ranging data [35]. We used the MAXENT logistic output, which provided estimates of relative habitat suitability [33]. Most bioclimatic variables are intercorrelated, so using all variables in an analysis can lead to an overpredicted model. To avoid overprediction, we calculated the matrix of pair-wise correlations of 19 bioclimatic variables for the study area using the “Band Collection Statistics” tool from the standard Arc Toolbox, Spatial Analyst Tools (ArcGIS Desktop 10.8.1). From the obtained correlation matrix, we chose the highest value among correlation coefficients, built two models with either of the two highest correlated variables alternatively removed in each, and estimated the model’s performance using the area under the receiver-operating characteristic curve (AUC) test. We then removed the variable whose presence in the model provided a lower AUC value from the data set. This operation was repeated with the next pairs of highly correlated variables until the removal of a variable from the model resulted in the abrupt decline of the AUC value. The resulting model with the final reduced set of environmental variables for the current time was projected onto paleoenvironmental data sets. For each node, we used divergence time intervals estimated from molecular dating. As a result, we achieved the set of models of relative habitat suitability (RHS) for each 10 kya within the analyzed time periods. To define the areas of real species occurrence, the original model values, ranging continuously from 0 to 1, were transformed to a binary 0 or 1 using a threshold value. The threshold value was chosen to be equal to the “maximum training sensitivity plus specificity”, as this threshold was experimentally demonstrated to provide optimal results [36].

In addition to the SDM-based reconstructions, we performed ancestral area estimation using the DIVA algorithm as implemented in RASP ver. 4.0 [37]. The analysis was conducted based on the species tree topology with the addition of S. t. karelini. Seven biogeographic areas were defined based on the distribution of Stylodipus species and lineages: Mongolian Gobi, northern China deserts, Dzungaria, Zaisan basin, East Kazakhstan, Central and West Kazakhstan, and Southeastern Europe. Maximum range size was fixed to two areas. Constraints on dispersal were specified by enumeration of allowed ancestral ranges, which included all pairs of adjacent areas.

3. Results

3.1. Cytb Tree and Mitochondrial Variability

In the cytb tree (Figure 2), S. andrewsi is placed as a divergent sister group to other species with a p-distance between the two clades of ≈12.6%. Sequences of S. telum sensu lato are paraphyletic relative to S. sungorus because of the position of two haplogroups distributed in East Kazakhstan, namely in the Zaisan basin (subspecies S. t. birulae) and in the northern foothills of Chingiztau (subspecies S. t. karelini). The birulae lineage is the earliest offshoot from the sungorus-telum clade (cytb p-distance = 7.2%), thus being more distant from S. telum (p-distance = 7.2%) than S. sungorus (p-distance = 4.6%). The mitogroup of S. t. karelini is a very close sister branch to S. sungorus (p-distance = 1.6%). The S. sungorus + S. t. karelini clade is placed as the sister group to the clade comprising all other haplotypes of S. telum s. l. The latter clade is subdivided into eastern and western lineages, which differ by 1.93% and consist of the specimens collected eastwards and westwards from the Volga River, respectively. This partition separates a group of western subspecies (S. t. falzfeini from the lower Dnieper region, S. t. turovi from the lower Don) and eastern subspecies: S. t. telum (=proximus) from West and Central Kazakhstan, S. t. amankaragai from Naurzum isolate in North Kazakhstan, and S. t. nastjukovi from North Kyzylkum, Mangyshlak, and South Balkhash region. The above results putatively indicate that the S. t. birulae and S. t. karelini lineages could be distinct entities separate from S. telum proper.

The intraspecific variability of S. andrewsi (mean intra-group p-distance π = 0.45%) and S. sungorus (π = 0.07%) is significantly lower than that of each of the two sublineages of S. telum s. l. (western, 0.82%, and eastern, 0.7%). In contrast to S. telum, these species lack a pronounced intraspecific structure. However, within S. andrewsi, we can identify two poorly differentiated mitochondrial sublineages (≈0.8% cytb p-distance), which occur in the western (the Valley of the Gobi lakes, Great Lakes Basin) and eastern (East and South Gobi) populations, respectively (Supplementary Information S2, Figure S2).

3.2. Nuclear Trees and Networks

Results of the allelic phase reconstruction are presented in Supplementary Information S2, Table S7. The relationships among alleles and individual gene trees reconstructed from unphased genotypes are shown in Figure 3 and Supplementary Information S2 and Figure S3, respectively. All nuclear genes support the separate position of S. andrewsi relative to other species, which is consistent with the mitochondrial data. In most cases, S. sungorus forms a compact lineage, which, however, tends to cluster with S. t. birulae. S. telum is polymorphic, and in some gene trees, it is paraphyletic relative to S. sungorus + S. t. birulae. Western and eastern lineages of S. telum have both unique and shared alleles. S. t. karelini is highly polymorphic, some alleles are as in S. telum (THY), some are shared with S. sungorus (DMPI), and some are specific and divergent (IRBP). This pattern may be a product of the past or recent hybridization.

3.3. Analysis of Nuclear Concatenation

The ML analysis of the dataset, including all specimens of S. t. karelini (n = 3, locality 3 in Figure 1, Supplementary Information S1, Table S1A), which is suspected to be of hybrid origin, produced a tree with non-monophyletic S. telum sensu stricto (Supplementary Figure S4), which could be an artifact due to the presence of hybrid specimens. When only one specimen of S. karelini was retained (the one with the fewest heterozygous sites and the longest sequences), the monophyly of the S. telum clade was restored (Figure 4). In the latter tree, S. andrewsi is again placed as the sister lineage to the other species (p-distance = 1.53%); however, S. sungorus was recovered here as the sister group to S. t. birulae and not to S. telum. Reciprocally monophyletic S. sungorus and S. t. birulae are well divergent from S. telum (p-distance = 0.66%) and less so between themselves (0.26%). The specimen of S. t. karelini is placed as the sister group to S. telum, albeit with poor support. Within S. telum, the monophyly of the western subclade is well supported; however, this is not the case for the eastern clade. The latter appears as paraphyletic in the ML tree, while in the Bayesian tree, its monophyly is violated by the position of a single specimen from the Balkhash region.

3.4. Species Tree and Divergence Times

The topology of the species tree (Supplementary Figure S5) is concordant with the phylogeny obtained from nuclear concatenation. Posterior probabilities for all clades were equal to 1.0. The splits among Stylodipus lineages were dated to the Early and Middle Pleistocene (

Table 1. Ages of splits among Stylodipus lineages (mean and 95% HPD) as estimated by STARBEAST3 based on sequences of 12 nuclear loci.

Split	Age, MYA (Posterior Mean)	95% HPD, MYA (Highest Posterior Distribution)
S. andrewsi/S. telum species complex	2.23	1.64–2.8
S. telum/S. sungorus + S. birulae	1.04	0.66–1.43
S. sungorus/S. birulae	0.42	0.20–0.66
S. telum East/S. telum West	0.20	0.09–0.32

). To estimate the times of the splits that are present in the cytb tree only, we obtained a crude estimate of the cytb substitution rate, which was approximated by dividing the cytb divergence (node height in the BEAST tree) by the time estimate from the nuclear species tree. In particular, we used the ages of splits between S. t. birulae and S. telum (~1 Mya) and between western and eastern lineages of S. telum (~200 kya), which produced the cytb substitution rate estimates of 4.6% and 5.5% per My, respectively. Based on these values, the divergence time between mitotypes of S. sungorus of S. t. karelini was estimated at 140–170 kya, while the split between S. telum and the karelini/sungorus lineage was calculated to be 500–600 kya old. However, these estimates should be treated with caution, taking into account potential rate decay, which is typical for mtDNA (e.g., [38]).

3.5. Biogeographic Paleo-Reconstructions

The main parameters of the obtained models for each node are provided in Supplementary Information S2, Table S8.

Chimaerodipus–Stylodipus split (Supplementary Figure S6(A1,A2)). Among the set of models for the time interval 5.4–1.64 Mya, the single non-subdivided area of suitable habitats that covered northwestern, central, and southeastern Mongolia and adjacent areas of China was found only once, 4.97 Mya. At all other times, suitable habitats were presented by two isolated areas, in northwestern Mongolia and Chinese Dzungaria on one hand and southeastern Mongolia and adjacent Inner Mongolia on the other. The gap between these two areas in the Gobi Lakes Valley varied from 300 to 350 km. During the aforementioned time interval of 5.4–1.64 Mya, models predicted suitable habitats in Kazakhstan, but these areas had no connections with Dzungaria.

Stylodipus andrewsi–S. sungorus + telum split (Supplementary Figures S6(B1) and S5(B2)). Models demonstrated that during the time interval of 2.8–2.35 Mya, connections between Dzungaria and Mongolian Great Lakes’ Hollow appeared and disappeared repeatedly. The connection existed 2.79, 2.77–2.75, 2.68-–2.67, 2.64–2.63, 2.47, 2.39, and for the last time, 2.35 Mya.

S. sungorus–S. telum split (Supplementary Figure S6(C1–C3)). Models demonstrated that during the time interval of 5.4–0.66 Mya, connections between Dzungaria and Kazakhstan appeared only once, at 1.07 Mya, through the Dzungarian Gate.

S. sungorus–S. t. birulae split. Models demonstrated that the widest distribution over an uninterrupted patch of suitable habitats was possible at 0.49 Mya. During the time interval from 0.47 to 0.43 Mya, this previously unified area of suitable habitats had contracted and divided into two isolated patches, one in the Zaisan Depression and the second in the western part of Chinese Dzungaria. This isolation was disrupted when the gap between two isolated parts closed, and the distribution expanded eastward into Mongolian Dzungaria during the time periods of 0.42–0.36, 0.38–0.28, and 0.2–0.2 Mya. However, it had constricted again with isolation recovered at 0.35 and 0.27–0.25 Mya. (Supplementary Figure S6(D1,D2)).

S. t. telum–S. t. falzfeini split (Supplementary Figure S6(E1–E4)). Models demonstrated that the widest unfragmented range existed during interglacials, whereas the narrowest distribution with fragmentation of suitable habitats into several isolated patches was associated with glacial maxima. A wide continuous range was possible at 0.24–0.2 and 0.12–0.08 Mya. During the former time interval, there was a patch of suitable habitats in the Lower Dnieper River area, but this patch was not connected with the main distribution range; connection with the Lower Dnieper River area was established only at 0.12 Mya, when the species could disperse there and establish current population. At the same time (0.12 Mya), connections between Kazakhstan and Dzungaria emerged. Fragmentation of the range started at 0.19 Mya when the range was divided into two parts: western (central Kazakhstan and area between the Volga and Don rivers) and eastern (eastern Kazakhstan). During other time intervals (0.18–0.13 and 0.07–0.02 Mya), the range was split into three parts: northern Cis-Caucasus, central Kazakhstan, and eastern Kazakhstan. The results of the ancestral area estimation in RASP [38] (Supplementary Figure S7) are generally consistent with the SDM-based paleo-reconstructions.

4. Discussion

4.1. Genetic Diversity in Genus Stylodipus and Implications for Its Taxonomy

Our results suggest that the taxonomic structure of Stylodipus is more complex than previously believed. Consistenct with previous studies [5,6], S. andrewsi is recovered as a relatively distant sister group to the S. telum/S. sungorus clade, which can be regarded as the S. telum species complex since it includes several divergent allopatric lineages. An unexpected result within this species complex is the isolated position of the Zaisan lineage, recognized previously as a subspecies S. t. birulae. This lineage appears to be highly divergent from S. telum proper based on both nuclear and mitochondrial data. The age of the split between the two lineages is estimated to have occurred as early as 1 Mya; such a level of divergence potentially corresponds to a rank of full species, being comparable to those between Dipus deasyi/D. sagitta s.l. [10,39] or between Scarturus elater/Sc. heptneri and between Scarturus euphratica/S. williamsi [12,40].

According to the nuclear data, S. t. birulae is a relatively close sister group to S. sungorus, having separated from the latter approximately 400 kya. This is in line with the pattern of distribution of the two forms: S. sungorus is documented to occur in the eastern and central parts of Dzungaria, while S. t. birulae is restricted to the Zaisan depression. These two areas were continuously connected by a corridor suitable for desert species such as Meriones meridianus [41]. However, the Zaisan and Dzungarian three-toed jerboas are clearly distinct, being well differentiated not only genetically but also morphologically. S. sungorus has a significantly larger body and skull size and wider molars than S. t. birulae and S. telum s.str. (Supplementary Table S9). Moreover, the morphological difference between S. sungorus and other Stylodipus taxa pronouncedly exceeds intraspecific differences observed between subspecies of S. telum, warranting recognition of S. sungorus as a species [42]. At the same time, according to the range-wide morphometric study [4], S. t. birulae also differs from all subspecies of S. telum by the larger skull size and relatively larger bullae (Supplementary Table S9).

Taking into account the degree of relatedness between S. t. birulae and S. sungorus, one may treat them either as close sister species or as subspecies within a single polymorphic species, which, following the principle of priority, would be named S. birulae. The relatively recent divergence time between the two taxa is nevertheless compatible with the species rank, as illustrated by cases in other groups of rodents [43,44]). Furthermore, the genetic data show no evidence of gene flow between S. t. birulae and S. sungorus: the shared alleles are found in CHY only (which can be explained by retention of an ancestral variant). Based on the sum of these points, we believe that S. t. birulae and S. sungorus can hardly be considered conspecific and, therefore, the Zaisan three-toed jerboa as significantly divergent from all recognized species should be elevated to the full species status, Stylodipus birulae (this name is used hereafter).

The mtDNA reconstructions are not fully congruent with the nuclear concatenate or species tree since, in the cytb tree, S. sungorus is closer to S. telum than to S. birulae. Moreover, S. sungorus haplotypes render S. telum paraphyletic, as they are very close (cytb p-distance = 1.6%) to those of S. t. karelini from the northern slopes of Chingiztau (East Kazakhstan), while all other S. telum haplotypes form a separate diverse clade. A likely explanation for this case of mitonuclear discordance is the introgression of mtDNA from S. t. karelini into S. sungorus.

The position of S. t. karelini is the most controversial, as alleles of the four studied nuclear genes show various affinities: some are closer to S. telum sensu stricto, some are shared with S. sungorus (but not S. birulae), some are well divergent from all others. To clarify this situation, one would need an additional sampling of populations combined with a significantly increased sampling of loci, which can be achieved using population genomic approaches. Meanwhile, we can hypothesize that S. t. karelini was once a separate branch that originated as an early offshoot from the stem of S. telum (as illustrated by the position of its mitotypes) but subsequently experienced hybridization with S. telum s.str. and S. sungorus. Taking into account the location of the distribution range of S. t. karelini (i.e., between S. telum s.str. and S. sungorus), this explanation appears more plausible than the alternatives implying incomplete lineage sorting.

The high rank of the S. birulae lineage endemic to the Zaisan depression highlights the important role played by this refugial area in the conservation of biodiversity throughout the Pleistocene. Noteworthy, the Zaisan region harbors specific divergent lineages of the lesser jerboa (Scarturus elater zaisanicus) [12,40], the northern three-toed jerboa (Dipus sagitta zaissanens [10], and the Eversmann’s hamster (Allocricetulus eversmanni pseudocurtatus [45]; it is also a refuge for the relict westernmost populations of Eolagurus luteus, which is otherwise extinct across West Central Asia [40,46].

The results once again support the importance of reticulate events in the evolution of species complexes of jerboas, as was previously demonstrated for Dipus [39]. This point is illustrated by the proposed hybridization episode between S. sungorus and S. t. karelini—the two taxa that had separated more than 800 ky before their secondary contact, if our time estimates are correct.

4.2. An Evolutionary Biogeographic Scenario for Stylodipus

The hypothetical scenario of the phylogenetic history of the genus that we suggest here implies the following sequence of events (Figure 5). The early history of the genus was likely associated with East Central Asia as follows from the fact that the earliest fossil Stylodipus was found in the Lower Pliocene of Mongolia [47]. In the Early Pleistocene (Gelasian), the common stem of Stylodipus split into two branches (eastern and western) ancestral to S. andrewsi and the S. telum species group, respectively. This event was probably associated with the subdivision of the ancestral range into Mongolian and Dzungarian segments. In the late Calabrian (ca. 1 Mya), the ancestor of S. telum branched off from the S. birulae/S. sungorus clade. Supposedly, at that moment, the former lineage dispersed from the Dzungar basin into adjacent areas in East Kazakhstan and thus began colonization of West Central Asia. In contrast, the S. birulae/S. sungorus lineage persisted in Dzungaria throughout the subsequent history. The results of paleo-range reconstructions allowed us to conclude that Stylodipus was absent from West Central Asia before 1.0 Mya. At the same time, a connection between Dzungaria and East Kazakhstan must have existed prior to that time, based on the earlier dispersal to Kazakhstan of the ancestor of Scirtodipus and its subsequent isolation therein. Subsequently (Middle Pleistocene, ca. 470–430 kya), the S. birulae/S. sungorus clade diverged into two lineages, with S. birulae likely being restricted to the Zaisan depression thereafter. Plausibly, the S. t. karelini lineage separated from the rest of S. telum by approximately 500–600 kya (as follows from the mtDNA data if we accept the Chingiztau haplotypes as authentic). At the Late/Middle Pleistocene boundary (≈150 kya), S. karelini came into contact with S. sungorus, which resulted in a gene exchange, including introgression of mtDNA into S. sungorus (with subsequent fixation) and, possibly, introgression of a fraction of nuclear alleles into S. karelini. At some moment(s) (unclear when), hybridization events between S. t. karelini and S. telum could have occurred, which resulted in the introgression of nuclear but not mitochondrial haplotypes. In the late Middle Pleistocene (ca. 200 kya), the western clade of S. telum emerged, possibly via colonization of areas westward from the Volga River; mitochondrial variation among fat-tailed jerboas of this region was previously discussed by Rusin [7]. Generally, the obtained models demonstrated that expansions of geographic ranges were associated with hot and wet periods, whereas range contractions took place during cold and dry periods.

It should be emphasized that the results of paleo-reconstructions should be treated with caution, considering the uncertainty in the divergence time estimates used. Range modeling was based only on paleo-climate data due to the absence of other types of paleo data. There are still some potentially important factors (substrate type and biotic factors) that are not incorporated in SDM, and hence, the predicted distributions may be just a crude approximation of the past reality. However, the species of the genus Stylodipus are not strong substrate specialists [46]; thus, the exclusion of substrate type from our models should result in only a minor overestimation of the distribution range size (due to the inclusion of some areas with unsuitable habitats). We also did not incorporate biotic interactions such as interspecific competition in our models directly, but competition among closely related species was accounted for in the interpretation of the modeling results, whereas competition among non-related species of jerboas was found to be mostly insignificant [48].

4.3. Intraspecific Variation and Phylogeographic Patterns

Our genetic data are generally consistent with the morphological view that S. andrewsi and S. sungorus are monomorphic. This is hardly surprising in the case of the latter species, given the small size of its range. However, the reason for the lack of variation within a relatively wide-range S. andrewsi is less clear. Its shallow phylogeographic structure stands in sharp contrast to the pattern observed within S. telum, thus suggesting different Pleistocene histories for these two species. At the same time, we can hypothesize that the presence of two weakly divergent mitochondrial sublineages in S. andrewsi may be a sign of range fragmentation in the past and the group’s survival in two Late Pleistocene refugia located in the northwestern and southeastern parts of the semi-desert belt of East Central Asia. Earlier, a similar, albeit not identical, pattern was discovered in another species endemic to Mongolian semi-desert, Spermophilus pallidicauda [49].

Considering morphologically polymorphic S telum, it is evident that current molecular data corroborate fewer lineages than is recognized in the morphological revision by Shenbrot [4]. In particular, mitochondrial data support two sublineages, western and eastern, which are likely separated by the Volga−Ural sands. The nuclear results are more ambiguous; nevertheless, they demonstrate the same tendency. Among the subspecies recognized by Shenbrot [4], S. t. falzfeini and S. t. turovi belong to the western lineage, while S. telum s.str., S. t. amankaragai, and S. nastjukovi belong to the eastern lineage. At the same time, the molecular data suggest a separate position of S. t. karelini, thus contradicting the cited revision, according to which it should be synonymized with the nominal form. The nature of the eastern/western subdivision as well as the absence of S. telum throughout the larger part of the Volga–Ural interfluvial region, can be explained by the impact of the Middle–Late Pleistocene (i.e., Khazarian and Khvalynian) Marine transgressions. However, it remains unclear as to why, in contrast to other arid-dwelling species such as Scarturus elater, Allactaga major, Meriones spp., etc., the fat-tailed jerboa failed to recolonize this area in the latest Pleistocene/Holocene. Morphological differentiation between S. telum s.str and northern S.t. amankaragai or southern S. t. nastjukovi may, in fact, be associated with adaptation to contrasting environments rather than with significant genetic divergence.

5. Conclusions

Our study of the genus Stylodypus provides further examples of deep genetic subdivisions within widely distributed Palearctic desert rodents. Based on its results, we propose a full species rank for the narrow-range Zaisan three-toed jerboa S. birulae. To establish the taxonomic status of another cryptic lineage, S. t. karelini, which might be a product of reticulate evolution, an additional study is warranted. The ancestral range reconstructions identify East Central Asia (Mongolia, Northwestern China) as the center of origin of the genus and suggest that its range evolution progressed via consecutive westward expansions into West Central Asia and, finally, Southeastern Europe.