Introduction

Based on the recent molecular phylogenetic analyses of small mammals, the last 10–20 million years were very important for establishing the current distributions of extant species (e.g., Michaux et al. 2002, 2003, 2004; Serizawa et al. 2000; Suzuki et al. 2003, 2004). Molecular studies of small mammals, which use specific ecological habitats, provide important clues for understanding how extant flora and fauna were established. In addition, they are also useful for understanding how environmental changes in the Late Tertiary and Quaternary affected the movements of lineages across areas and even across continents. Factors that affected speciation events must be considered from a biogeographic perspective. Among the various groups of small mammals, the wood mouse genus Apodemus is the most common rodent in temperate Eurasia and may be one of the best subjects for biogeographic study.

Species in the genus Apodemus are the most common rodents in the temperate zone of the Palaearctic region (Corbet 1978; Corbet and Hill 1992; Orlov et al. 1996), and more than 20 Apodemus species are now recognized, half from Asia and half from Europe (Musser and Carleton 2005). Apodemus species occur in a variety of habitats, including woodlands, forests, and grasslands, depending on resources such as acorns, insects, and other small invertebrates. Two or more species often coinhabit, showing resource partitioning among syntopic species, which is thought to be associated with the biogeographic history of this genus and its speciation processes. The framework for an Apodemus molecular phylogeny has been provided by analyses of the mitochondrial cytochrome b (cyt b) gene and the nuclear interphotoreceptor retinoid binding protein (IRBP) gene (Michaux et al. 2002; Serizawa et al. 2000; Suzuki et al. 2003), indicating four distinct lineages among the taxa examined to date: the Sylvaemus, the Apodemus, the A. argenteus, and the A. gurkha groups. These lineages are somewhat comparable to the subgeneric grouping system of Musser et al. (1996); however, Apodemus mystacinus, which is sometimes classified as the distinct subgenus Karstomys, was not included in the molecular analyses covering an entire set of the distinct lineages. Meanwhile, through ongoing morphological and molecular analyses, the taxonomic situation has improved somewhat, but there are still some difficult cases dealing with closely related species (see Musser and Carleton 2005).

In particular, the systematics of western Palaearctic species, included in the subgenus Sylvaemus, are complicated because of the description of several new species in southern Europe and the Middle East [A. alpicola, A. uralensis, A. fulvipectus, A. hermonensis (whiterbyi in Musser and Carleton 2005), A. arianus, A. hyrcanicus, A. ponticus, A. rusiges, and A. wardi (pallipes in Musser and Carleton 2005)]. The phylogenetic relationships among these taxa are still not solidly established, and the distribution of the two most widespread species, A. sylvaticus and A. flavicollis, in southeastern Europe and the Middle East is still unclear. Several subspecies of A. sylvaticus appear to be geographical forms of A. uralensis (Mezhzherin and Mikhailenko 1991). Vorontsov et al. (1992) proposed that four species inhabit the Caucasus and Transcaucasus based on morphological, chromosomal, and allozyme data: A. ponticus (previously considered a subspecies of A. flavicollis), A. hyrcanicus, A. uralensis, and A. fulvipectus. In western Anatolia, allozyme and morphological data indicate the presence of four Sylvaemus species (Filippucci et al. 1996): A. sylvaticus (rare and restricted to a small area near the coast of the Black Sea, previously considered a widely distributed species in Asia Minor), A. flavicollis (widely distributed, previously considered limited to eastern Anatolia and the Caucasus), A. hermonensis (widely distributed, previously known only from the type locality, Mount Hermon in Israel; Filippucci et al. 1989), A. uralensis (confined to the humid mountainous areas of northern Asia Minor). These results were confirmed by Macholán et al. (2001), extending the study to eastern Turkey, Armenia, and Iran, where populations of A. flavicollis, A. hermonensis, and A. cf hyrcanicus were found. The systematics of wood mice from Iran are still incomplete, and data from eastern Iran and central Asia are fragmentary. Musser and Carleton (2005) suggested that A. arianus inhabited Iran and A. wardi was distributed in Nepal, Kashmir, Pakistan, Afghanistan, and northwestern Iran. According to these authors, A. fulvipectus (in the north) and A. ponticus (in the northwest) probably also occur in Iran.

Filippucci et al. (2002) analyzed the genetic variation of 28–38 gene loci by protein electrophoresis in ten species of the genus Apodemus from Europe, the Middle East, and North Africa, and found high genetic distance in A. mystacinus between epimelas from Europe and mystacinus from Israel and Turkey, suggesting that the two taxa could represent distinct species and supporting Storch’s (1977) hypothesis of their ancient separation dating to the Pleistocene. The separation of the two taxa has been supported by molecular phylogenetic studies (Bellinvia 2004; Michaux et al. 2005). Apodemus mystacinus and A. epimelas appeared as sister taxa to other species of the Sylvaemus group (Bellinvia 2004; Michaux et al. 2002), in agreement with Musser et al. (1996), although the validity of the subgeneric name Karstomys, used by several authors (Rietschel and Storch 1974; Storch 1975) for the two former taxa, is not generally accepted.

Molecular analyses of the East Asian subgeneric group Apodemus (Serizawa et al. 2000; Suzuki et al. 2003) have indicated four distinct lineages represented by Apodemus agrarius, A. draco, A. peninsulae, and A. speciosus, in which the former two lineages include at least two (agrarius and chevrieri) and three species (draco, latronum, and semotus), respectively, forming species groups agrarius and draco, respectively. Liu et al. (2004) examined cyt b gene sequences, focusing on Chinese taxa, and showed that the Yunnan taxon A. ilex, which is often treated as a synonym of A. draco (Musser and Carleton 1993; Suzuki et al. 2003), possessed a taxon-specific mitochondrial lineage, as is seen in A. draco from other mainland areas and A. semotus from Taiwan.

Apodemus agrarius has a wide range across several geographic domains. A branch of A. agrarius is found in Europe, including Turkey. To determine the evolutionary episodes of the European population, a comparative study with Asian populations is necessary. Through biochemical analysis, Filippucci (1992) showed intraspecific variation in this species, as observed in other European Apodemus species. Although Nei’s genetic distance between two subspecies of A. agrarius (istrianus from northeastern Italy and agrarius from Yugoslavia, according to Krystufek 1985) is not very high (D = 0.027), populations from the two geographic areas displayed polymorphism at different loci (Italian population: Me-1, 6-Pgdh, Got-1, and Ada; Yugoslavian population: Idh-2, Got-1, Pgm-1, and Pgm-2), and 6-Pgdh was partially different between the two taxa. Currently, however, no data are available on the genetic relationships between Asian and European populations.

In this study, we conducted phylogenetic analyses with a larger set of both Asian and European species. We examined 17 species of Apodemus using five phylogenetic markers: the mitochondrial cyt b (1,140 bp) gene and the nuclear single-copy genes IRBP (1,152 bp), recombination activating gene 1 (RAG1, 1,002 bp), and von Willebrand factor (vWF, 1,192 bp). Exon regions of the nuclear genes IRBP, RAG1, and vWF are used to assess phylogenetic relationships across mammalian orders (Huchon et al. 1999, 2002; Springer et al. 1997; Stanhope et al. 1992, 1996). These nuclear genes are also useful for defining relationships among lower-level taxa (e.g., Chinen et al. 2005; Jansa and Voss 2000; Serizawa et al. 2000; Suzuki et al. 2000, 2003, 2004). We also examined the I7 gene (an olfactory receptor, 792 bp), a member of the olfactory receptor gene family, which is the largest gene family in the mammalian genome, with more than 1,000 members. This study is a good opportunity to assess the utility of olfactory receptor genes for phylogenetic inference. We also addressed the issue of genetic relatedness of Asian and European A. agrarius populations using both mitochondrial and nuclear gene variation.

Materials and Methods

Samples and Sequence Data

In addition to the samples used in the previous studies (Chelomina and Suzuki 2006; Chelomina et al. 1998; Han et al. 1996; Sato and Suzuki 2004; Serizawa et al. 2000; Suzuki et al. 2003), the cyt b, IRBP, RAG1, I7, and vWF sequences of A. agrarius (Treviso, Venetia, Italy), A. hermonensis (Mount Hermon, Israel), and A. mystacinus (Masada Forests, Golan Heights, Israel) were determined (Table 1). We used the species name A. draco for individuals from Yunnan, China (sample codes HS 505, 1628; Suzuki et al. 2003) in this study, but it is notable that the taxon is sometimes classified as A. ilex and differs from A. draco from other parts of southern China (see Liu et al. 2004). RAG1 sequences were also determined for 13 ingroup taxa and an outgroup taxon, the mole rat Spalax ehrenbergi (Kerem Ben Zirma, Israel; sample code KBZ500). Sequences of other ingroup and outgroup taxa (Micromys minutus, Mus musculus, Rattus norvegicus, and Mesocricetus auratus) were obtained from the previous studies (Sato and Suzuki 2004; Serizawa et al. 2000; Suzuki et al. 2003, 2004).

Table 1 Apodemus species analyzed in the interspecies phylogenetic analysis
Full size table

Sequencing

Polymerase chain reactions (PCR) and direct sequencing of the cyt b (1,140 bp), IRBP (1,152 bp), and RAG1 (1,002 bp) genes were performed according to the previously described methods (Sato et al. 2004; Serizawa et al. 2000; Suzuki et al. 2000, 2003, 2004). A part of the olfactory receptor gene was amplified by PCR with the primer pair +Olf7-122 (5′-GGC CTA [C/T]GT GTT GGT G[T/C]T GAC-3′) and –OlfI7-916 (5′-CAG GTG CAG [C/A]GT GCG ACG TAG-3′), according to the mouse (Krautwurst et al. 1998) and rat (Buck and Axel 1991) I7 genes. Numbers in the primer names designate the position of the 3′ end of the primer in the published rat I7 sequence, and the prefixes + and – refer to the reading and complementary strands, respectively. A 0.8 kb portion of the fragment was then sequenced through the amplification of three segments with the following primer sets: R + SN-OlfI7-122 (5′-CAG GAA ACA GCT ATG ACC GGC CTA [C/T]GT GTT GGT G[T/C]T GAC-3′) and U-SN-OlfI7-533 (5′-TGT AAA ACG ACG GCC AGT CAC AGA AAA AGT GGT TGA TGG-3′); R + SN-OlfI7-524 (5′-CAG GAA ACA GCT ATG ACC TCT CGC CTG TCT TAC TGT GG-3′) and U-SN-OlfI7-916 (5′-TGT AAA ACG ACG GCC AGT CAG GTG CAG [C/A]GT GCG ACG TAG-3′). A 1.2 kb fragment of vWF was sequenced through the amplification of three segments with the following primer sets: vWF-F241-Mus (5′-TGT GAA CCT TAC GTG TGA AGC CTG-3′) and vWF-R843-rodent (5′-TCA TCC ACC CCA CTG AGC AGA AA-3′); vWF-F559-Mus (5′-CCC TCA GAG CTT CGG CGC AT-3′) and vWF-R1099-musculus (5′-ACG TCC ATG CGC TGG ATG ACC T-3′); vWF-F1030-musculus (5′-TCC ATG GTT CTG GAT GTG GT-3′) and vWF-R1507 (Huchon et al. 1999; in their paper, this primer was named W1). “F” and “R” in the primer names refer to forward and reverse, respectively. The numbers indicate the position of the 3′ ends in the published human sequence (Mancuso et al. 1989; accession no. M25851).

Sources for sequences obtained from databases are listed in Table 1. The new nucleotide sequences reported in this paper were deposited in the DDBJ, EMBL, and GenBank nucleotide sequence databases under accession nos. AB285442-AB285452 and AB303224-AB303285.

Phylogenetic Analysis

Trees were constructed using the neighbor-joining (NJ; Saitou and Nei 1987), maximum parsimony (MP), and maximum likelihood (ML) methods using PAUP 4.0b10 (Swofford 2003). For the NJ analysis, we used Kimura’s two-parameter distances (d; Kimura 1980) in single-gene datasets. For MP and ML analyses, ten heuristic searches were conducted with the tree-bisection reconnection option, in which the input order of taxa is randomized. The best-fit model of DNA evolution for ML analysis was chosen on the basis of Akaike’s information criterion (AIC), according to the suggestion of Posada and Buckley (2004) that the AIC-based model selection is superior to the hierarchical likelihood ratio tests, using PAUP 4.0b10 (Swofford 2003) and Modeltest version 3.7 (Posada and Crandall 1998). Robustness of clades of phylogenetic trees was assessed using bootstrap analyses with 1,000, 1,000, and 100 resamplings for the NJ, MP, and ML analyses, respectively. The data were also analyzed by Bayesian inference (BI) using MrBayes 3.1.2 (Ronquist and Huelsenbeck 2003), in which the appropriate model was selected with AIC in Modeltest, according to the suggestion of Posada and Buckley (2004).

Divergence times were estimated by the method first described by Thorne et al. (1998), in which the node dates were estimated in a Bayesian framework allowing changes in evolutionary rate over time, using PAML version 3.14 (Yang 1997), paml2modelinf, estbranches, and multidivtime according to the description by Rutschmann (2005). The superiority of using the combined dataset to estimate divergence time, as suggested by Yang and Yoder (2003), enabled us to date divergence times with the combined four sequences of rodents. First, we obtained single-gene trees by ML analysis with models and parameters determined by AIC (with Modeltest and PAUP). Model parameters [unequal nucleotide frequencies, transition/transversion rate, and rate heterogeneity among sites for the F84 + G model; F84, Felsenstein 84 implemented in Phylip (Felsenstein 1993)] were then estimated for each single-gene tree with PAML. The obtained model parameters and a single Bayesian tree from the combined dataset, as described above, were subsequently used for estimating branch lengths in this Bayesian tree and the variance–covariance matrix of the parameters (with estbranches). From the information above, Markov Chain Monte Carlo analyses were performed to approximate the posterior distributions of substitution rates and divergence times (with multidivtime).

Results

Nuclear Gene Sequences

We constructed single-gene NJ and MP trees for 13 species of Apodemus with data from IRBP, RAG1, I7, and vWF (Fig. 1). The number of variable sites for ingroup comparisons was 174, 125, 72, and 177, respectively. We did not detect any signals of gene duplication in our data, including those from I7, where gene duplication is known to occur. We also conducted phylogenetic inference with the combined IRBP + RAG1 + I7 + vWF data using the NJ, MP, ML, and BI methods (Fig. 2). In the ML and NJ analyses, we used a TrN + I + G model (-ln L = 12457.3728; A = 0.2253, C = 0.2818, G = 0.2696, T = 0.2254; gamma shape parameter = 0.8033; assumed proportion of invariable sites = 0.4695). There were 529 variable sites, and the resultant trees were likely to be more reliable than the single-gene trees.

Fig. 1
figure 1

Phylogenetic relationships among Apodemus species, based on the nuclear genes of IRBP (1,152 bp; a), RAG1 (1,002 bp; b), I7 (792 bp; c), and vWF (1,192 bp; d). The phylogenetic trees were constructed using the NJ method, with various combinations of outgroup taxa, including Micromys minutus and Rattus norvegicus. Bootstrap analyses were performed with NJ and MP methods (1,000 replicates for both methods). The nodes supported by high-bootstrap values (>90% by both methods) are marked with closed triangles

Full size image
Fig. 2
figure 2

Phylogenetic relationships among Apodemus species, based on the concatenate sequences of IRBP (1,152 bp), RAG1 (1,002 bp), I7 (792 bp), and vWF (1,192 bp). The phylogenetic trees were constructed using the NJ method. Bootstrap analyses were performed with NJ (1,000 replicates), MP (1,000 replicates), and ML (100 replicates) methods. The resultant values are shown at each node (NJ/MP/ML; as percentages, only if >50%). Bayesian inference posterior probability values were shown for the well-supported nodes. Divergence times estimated using the software Multidivtime (Thorne et al. 1998) are shown at each of the key nodes with 95% confidence intervals. Notably, the present subgeneric grouping system (I–IV) is in good agreement with the current taxonomic view of Musser and Carleton (2005), which is the modified version of Musser et al. (1996)

Full size image

Based on this dataset (Figs. 1, 2), we noted three groupings (designated here as species groups) of species with close phylogenetic relationships and low genetic distances: (1) A. agrarius and A. chevrieri (agrarius species group); (2) A. draco, A. latronum, and A. semotus (draco species group); and (3) A. alpicola, A. sylvaticus, A. flavicollis, A. hermonensis, A. uralensis, and A. wardi (sylvaticus species group, which corresponds to the subgenus Sylvaemus) (Musser et al. 1996). Together with four monospecific groups (A. argenteus, A. gurkha, A. peninsulae, and A. speciosus) and the A. mystacinus species group with two species (see Introduction), eight species groups were thus recognized in the Apodemus species examined.

Among the eight species groups, a monophyletic relationship with high support was recovered for four ancient lineages of the A. agrarius group (agrarius, semotus, peninsulae, and speciosus), which was suggested by Serizawa et al. (2000). The clustering of A. mystacinus with the sylvaticus species group had high bootstrap support in the RAG1 (96–97%), vWF (72–93%), and IRBP + RAG1 + I7 + vWF (97–100%) analyses. Overall, as illustrated in Fig. 2, the Apodemus species examined can be grouped into four distinct lineages (Clades I–IV), those represented by A. sylvaticus (I) and A. agrarius (II), together with the two distinct lineages of A. gurkha (III) and A. argenteus (IV).

In the single-gene analyses, no substantial clustering patterns were observed for the members of Clade I (sylvaticus species group). In the concatenated trees (Fig. 2), however, five prominent lineages appeared: (1) A. alpicola, (2) A. flavicollis/A. ponticus, (3) A. hermonensis, (4) A. sylvaticus, and (5) A. uralensis/A. wardi. The relationships among the five lineages were entirely obscured with our current data on nuclear genes.

In Clade II, which roughly corresponds to the subgenus Apodemus (Musser et al. 1996), A. speciosus consistently had a basal position in the separate datasets, as well as in the concatenated dataset (85–100%). Relationships among the remaining three species groups are rather ambiguous. A close affinity was seen between the draco species group and A. peninsulae in the IRBP (Fig. 1) and concatenated (Fig. 2) datasets, although the supporting values were extremely low.

Cyt b Gene Sequences

The cyt b sequences were used to assess the phylogenetic relationships within the three species groups represented by A. sylvaticus (sylvaticus species group), A. draco (draco species group), and A. agrarius (agrarius species group), where the taxonomic situation is rather ambiguous. We abandoned the effort to solve the species group relationships with the cyt b sequences because of the heavy saturation in such a deep lineage comparison (e.g., d > 0.12; Suzuki et al. 2003; but see May-Collado and Agnarsson 2006). We constructed an NJ tree with our sequences and those from the literature and conducted bootstrap analyses with the NJ and MP methods (Fig. 3).

Fig. 3
figure 3

Phylogenetic relationships focusing on three species groups of agrarius, draco, and sylvaticus, based on the mitochondrial cyt b gene sequences. The phylogenetic trees were constructed using the NJ method with distances inferred from Kimura’s two-parameter distances (Kimura 1980). Bootstrap analyses were performed with NJ and MP methods (1,000 replicates for both methods). The resultant values are shown at each node (NJ/MP; as percentages, only if >50%). Taxonomic names of some species are likely to be unstable, and species names are temporally presented on the species clades. Apodemus hermonensis, for example, should be A. whiterbyi according to Musser and Carleton (2005)

Full size image

We added the A. fulvipectus sequence (AY179491; Reutter et al. 2003) to our analysis for the sylvaticus species group. Six distinct groups, consisting of (1) A. alpicola, (2) A. flavicollis/A. ponticus, (3) A. hermonensis, (4) A. sylvaticus, (5) A. uralensis/A. wardi, and (6) A. fulvipectus, were noted, with genetic distances between the groups ranging from 0.09 to 0.13 (Fig. 3). The sister species relationship between A. flavicollis and A. ponticus from Georgia was detected with weak support. Apodemus wardi from Nepal clustered with the A. uralensis clade, as in the nuclear gene analysis (see also Suzuki et al. 2003). The uralensis clade consisted of three lineages, (1) from eastern territories [Pakistan, Xinjiang (China), and Altai (Russia)], (2) A. wardi from Nepal, and (3) from western territories [Turkey, Kazakhstan, Moldova, Chelyabinsk (Russia)], with genetic distances ranging from 0.07 to 0.08.

The A. draco lineage from the Yunnan Plateau was distinct both from A. semotus from Taiwan and A. draco from other southern Chinese areas, with average genetic distances of 0.08–0.10 (Fig. 3), as noted by Liu et al. (2004).

We compared cyt b sequences to assess the phylogeographic episodes of A. agrarius from Asia and Europe and found that sequence variability was relatively low (0.01 between Italy and Germany) and high (up to 0.03 among China, Korea, and Primorye). The European sequences showed close relationships with a portion of the Asian sequences, namely those from Primorye, Far East Russia, with genetic distances of 0.01. In addition, individuals from Germany and Italy possessed similar cyt b sequences, which is somewhat concordant with a previous allozyme analysis (Filippucci 1992).

Discussion

Taxonomic Status of Some Species

Various taxonomic views for Apodemus have been challenged by morphological and molecular phylogenetic approaches (e.g., Filippucci et al. 2002; Michaux et al. 2002; Musser et al. 1996; Serizawa et al. 2000; Suzuki et al. 1994, 2003) and have greatly improved to a new standard (Musser and Carleton 2005). The taxonomy may need further revision, however, in some species such as the southern Chinese “A. ilex” and the European “A. fulvipectus.”

Yunnan’s draco-like taxon, A. ilex, possesses a distinct cyt b gene sequence (e.g., ∼8% genetic distance) from A. draco sequences from the other locations, making A. ilex likely to be a valid species (Liu et al. 2004). However, mitochondrial DNA sometimes shows high levels of genetic divergence among local populations, regardless of similarity in nuclear gene markers, partly due to female philopatry and partly due to random lineage sorting. Since each of the nuclear gene sequences examined here differed among draco, ilex, and semotus, these gene sequences and other single-copy genes should be useful markers for future phylogeographic studies to better understand the evolutionary history of species group draco.

The taxonomic situation in species group sylvaticus seems rather confusing. On the basis of the data presented here (Figs. 1, 2), there seem to be six distinct lineages, including one that represents A. fulvipectus. If we consider only the molecular data, A. wardi from Nepal is likely to be conspecific with A. uralensis. Thus, the species harbors three geographic populations, from the west (Turkey, Kazakhstan, Moldova), east (Pakistan, western China, Altai), and Nepal, on the basis of the mitochondrial dataset. A close relationship between Kazakhstan and Moldova individuals is seen in the nuclear gene sequences (NJ analysis in Fig. 1). The mitochondrial sequence of A. ponticus is distinct from but closely related to that of A. flavicollis (Fig. 3). This trend is also evident in the analysis of the combined nuclear gene sequences (Fig. 2). Overall, our current nuclear and mitochondrial gene data show differences in the major species lineages of the sylvaticus group, even though differentiation is low, providing reproducible patterns in both datasets. However, our datasets, especially for the nuclear genes, are likely insufficient to settle lower taxonomic problems (such as the status of A. ilex, A. ponticus, and A. wardi); further analyses with more nuclear gene markers are needed.

Phylogenetic Framework of the Apodemus Species

The distinctness of these eight species groups is prominent in the sequence analyses of all nuclear genes and their concatenated sequences (Fig. 2). Given previous phylogenetic studies and the present one, using various markers including morphological and molecular characters (e.g., Filippucci et al. 2002; Liu et al. 2004; Michaux et al. 2002; Musser et al. 1996; Serizawa et al. 2000; Suzuki et al. 2003), the genus Apodemus consists of eight distinct lineages: (1) the agrarius species group, (2) A. argenteus, (3) the draco species group, (4) A. gurkha, (5) the mystacinus species group, (6) A. peninsulae, (7) A. speciosus, and (8) the sylvaticus species group.

Among the eight major lineages, the two lineages confined to Europe and western Asia, A. mystacinus and A. sylvaticus, were sister groups with high support, as previously denoted by Michaux et al. (2002; cyt b + 12S rRNA) and Filippucci et al. (2002; 34 allozyme loci), but not in accord with Liu et al. (2004; cyt b), who suggested that A. mystacinus is closely related to East Asian lineages such as A. agrarius. Contrary to the ambiguous phylogenetic inference from the mitochondrial gene sequences, our combined sequence dataset firmly supports (100–97%) the monophyly of A. mystacinus and the sylvaticus species group.

Considering the Europe/Central Asia and agrarius/draco/peninsulae/speciosus groups together with the two remaining Asian lineages, A. gurkha and A. argenteus, we recognize four anciently diverged lineages. The concatenated dataset suggests monophyly of the former three lineages and the basal position of A. argenteus with low support, which may imply a geographic affinity in the diversification of the four ancient lineages. Notably, this view supports the subgeneric grouping of Musser et al. (1996), with a minor revision excluding A. gurkha from the Apodemus group and making it monospecific (Fig. 2; Musser and Carleton 2005). Given that a weak trend was seen for monophyly of A. argenteus and A. gurkha in our previous study (Suzuki et al. 2003), this issue needs to be addressed further in future studies.

Evolutionary Views of Apodemus

The phylogenetic data obtained here (Figs. 1–3) provide insight into several evolutionary topics related to Apodemus, including multiple radiation events, factors determining body size, and the recent westward expansion of A. agrarius.

From our current phylogenetic study (Fig. 2), the first radiation of the Apodemus species is itself separated in three steps: the first one, appearing around 6 million years ago, and two others, around 5 million years ago, which gave rise to agrarius, draco, peninsulae, and speciosus and to mystacinus and sylvaticus. Around 2 million years ago, a second evolutionary event gave rise to more recent species, and most recently, agrarius, which colonized the whole Palearctic region. These remarkable radiation events were probably primed by global changes that occurred 6–7 million and 2–3 million years ago (e.g., Cerling et al. 1997). The existence of old lineages in Asia implies that the forest ecosystem has been maintained for a long evolutionary period and that wood mice accommodated the ecosystem at the initial stage of radiation.

Excluding the broad range of mystacinus/sylvaticus, three geographic regions can be considered for Apodemus evolution, i.e., China, Nepal, and the Japanese archipelago, which hosted three ancient lineages (II, III, and IV, respectively; Fig. 2), as noted previously (Serizawa et al. 2000; Suzuki et al. 2003). It is currently unknown why the second step of the radiation occurred in East Asia around 1 million years after the first step. In the second radiation, the four old lineages of A. agrarius, A. draco, A. peninsulae, and A. speciosus emerged (Fig. 3), probably in China and Japan. From an ecological perspective, this would be correlated with habitat capacity, which is associated with species richness in small mammals (McShea et al. 2003). In the case of Apodemus, two species could co-occur in the same general habitat. A possible scenario is encompassed by two processes: the first one is allopatric speciation, and the second is range overlapping. From the genetic perspective, it is possible to presume that it takes 1–2 million years for speciation to occur in the first process, which allowed range overlapping without erosion by genetic hybridization in the second process. One or two million years after the first colonization event of A. argenteus, the second colonization event of A. speciosus could have occurred in Japan. In the case of the three ancient lineages of A. agrarius, A. draco, and A. peninsulae, which are considered to have diverged simultaneously (Fig. 2), they would have been in three different geographic areas somewhere in the continental part of East Asia, although we cannot specify the exact locations, followed by range overlapping with one another. In Asia, thus, we currently see the following consequences: A. agrarius and A. peninsulae in North China, Korea, and Far East Russia; A. agrarius and A. draco in southern China; A. draco (or A. ilex) and A. latronum in West Yunnan; A. chervieri and A. draco (or A. ilex) in East Yunnan; and A. argenteus and A. speciosus in Japan. The idea that the subsequent events can be explained by a range overlap (or coexistence of two species) due to habitat capacity applies to the Yunnan wood mice A. latronum and A. draco (or A. ilex), where two closely related lineages overlapped; A. latronum and A. draco were allopatrically or parapatrically speciated, and the latter expanded into the range of the former. This trend can also be seen in the evolutionary patterns of subtropical Eurasian Mus (Suzuki et al. 2004), in which lineage differentiation occurred in the subgenus Mus during a short period of evolutionary time at the initial stage of the radiation.

Body size change within an evolutionary lineage is a major topic in mammalogy (e.g., Adler 1995; Millien 2004; Musser et al. 1996). A finely resolved phylogenetic tree for the Asian species (Fig. 2) shows the evolutionary changes of body size in Apodemus. In Asia, where Apodemus species have long evolutionary histories, thereby showing evolutionary consequences for stationary states, there appears to be a difference in body size in the two sympatric species in southern mainland China (i.e., A. agrarius and A. draco) and the Japanese archipelago (A. speciosus and A. argenteus; Musser et al. 1996). Since the lineages with large (agrarius, speciosus) and small (draco, argenteus) body sizes are not monophyletic (Fig. 2), body size divergence may have occurred independently in southern China and Japan. This parallel trend can be explained by diversifying selection through niche segregation between the sympatric taxa, in which ecotype segregation of ground and arboreal dwellers (Sekijima and Sone 1994) and/or preferences for different acorn sizes (Shimada 2001) may be involved in selection. Additionally, in Yunnan, A. latronum and A. draco co-occur, and these taxa appear to differ in body size (Musser et al. 1996) despite their recent divergence from a common ancestor (Fig. 2). Interestingly, the Taiwanese endemic A. semotus is larger than its closely related taxon A. draco, which occurs solely at higher elevations and thus departs from the range of A. agrarius residents at lower elevations (Adler 1996; Musser et al. 1996). It is possible that A. semotus evolved a larger body size without niche segregation, although we must consider other factors such as insular syndrome. In contrast, the situation in North China, Korea, and Far East Russia is likely to be different; the coexisting species A. agrarius and A. peninsulae do not show a substantial difference in body size. This may be due to their habitat preferences; A. agrarius is often regarded as a generalist meadow species (e.g., Bryja and Rehák 1998; Kostenko 2000), and A. peninsulae is a typical forest species in Korea and Far East Russia. The European Apodemus (sylvaticus species group) may provide information to better understand the evolutionary factors determining body size when there is a clearer phylogeny.

We analyzed the phylogenetic relationships among local populations of A. agrarius, hoping to learn about the factors shaping dispersal patterns in wood mice. We compared individuals from Europe (Italy) to those from China, Russia, and Korea and found a close relationship among those from Far Eastern Russia and Europe. Since the A. agrarius lineage appears to have originated in Asia, the European lineage is likely the descendant of recent colonizers from Asia. Our data suggest that some genetic elements localized in northern Asia contributed to the mitochondrial DNA in Europe. The question remains as to why A. agrarius was selected for further radiation events in the last part of the biogeographic development of Apodemus, namely the transcontinental dispersal of A. agrarius from Asia to Europe. It is possible that this occurred because (1) an omnivorous species has more chances for dispersal, (2) an ancestral A. agrarius population was situated in a locality accessible to Europe at the time of dispersal, (3) A. agrarius adapted to a special habitat that was newly established, or (4) A. agrarius is always dispersing into new areas. In our opinion, a plausible explanation is that A. agrarius inhabits shrub habitats rather than forests, at least in Far East Russia, which may be a source area of European A. agrarius (Fig. 3). Considering the low genetic distance of 0.01 between Europe and Far East Russia, the time of the migration from the east to the west may have been around 200,000 years ago, assuming the evolutionary rate of the cyt b in Apodemus to be 0.024 per site per million years (Suzuki et al. 2003). The Quaternary ice ages, when shrub fields developed across the continent, would have been a good time for A. agrarius to migrate to Europe.