Introduction

Phylogeographic studies have provided significant insights into historical changes in the distribution and genetic structures of extant populations. In recent decades, there has also been increasing interest in inferring biogeographic histories from comparisons of phylogeographic patterns of genetic variations in related and co-distributed taxa (Avise 2000; Arbogast and Kenagy 2001). Such analyses have shown (inter alia) that the past and current patterns depend on the major factors involved. Some groups of related taxa display strong, common phylogeographic and population genetic structure because they have been largely affected by similar past geological events and climatic conditions (reviewed in Arbogast and Kenagy 2001). However, structures of other groups are highly complex because the taxa have been affected by biogeographic or ecological factors at least as strongly as by common historical events (e.g., Maliouchenko et al. 2007; Toyama and Yahara 2009; reviewed in Ohsawa and Ide 2011; Guicking et al. 2011; Iwasaki et al. 2012).

The genetic structure of forest trees, especially the geographic component, is particularly relevant for the management and conservation of genetic resources (Petit et al. 1993). In addition, molecular phylogenies have potential utility not merely for accurately defining target entities for conservation but also for predicting trends of long-term population processes, which may be related to translocation issues (Moritz 1995; reviewed in Hufford and Mazer 2003).

Previous studies suggest that introducing foreign genotypes into new environments could result in loss of local adaptation and lead to outbreeding depression, genetic swamping (Templeton 1986; reviewed in Hufford and Mazer 2003), mortality (Campbell 1979; Hamann et al. 2011; McKay et al. 2005), and/or threats to regional genetic and species diversity (Vander Mijnsbrugge et al. 2010). Thus, efforts to offset potential risks of translocation require careful consideration of seed sources and the establishment of “seed transfer zones” or regions, within which planting stocks (seeds and seedlings) can be moved with minimal detrimental effects on the average fitness of populations (Hufford and Mazer 2003). Most guidelines for seed zoning and transfer have been based on ecological classifications, climatic data, ecophysiological information, and/or results of provenance tests (Gomory et al. 1998). However, following advances in population genetic analyses, there has been growing awareness of the utility of geographic patterns of genetic variations in neutral molecular markers in the delineation of robust seed zones for genetic conservation and restoration activities (e.g., Gomory et al. 1998; Weels et al. 2003; Krauss and Koch 2004; Moncada et al. 2007; Keir et al. 2011; De Souza et al. 2012). Recently, genetic guidelines for transferring planting stocks have been proposed for a number of widely distributed broadleaf species in Japan based on results of several studies using molecular markers (Forestry and Forest Products Research Institute 2011, ISBN:978-4-902606-75-1; Tsumura and Suyama 2015). The guidelines include recommendations that transfer zones should be strongly delineated (i.e., transfers into or out of identified zones should be strongly avoided) when a species displays significant congruent geographic patterns of genetic differentiation in both nuclear and chloroplast DNA, while provisional boundaries between zones should be delineated when the geographic pattern is significant in only one genome. The latter does not necessarily entail strict transfer of planting stocks within distinct regions, but encourages consideration of the observed geographic structure. Broadleaf trees are mostly used for restoration activities, greening programs, and establishment of forest parks and green recreational areas in Japan, and such genetic guidelines would be particularly useful for planning and designing strategies for seed and seedling transfer.

There are larger volumes of oaks (Quercus spp.) in Japan than any other genera of broadleaf trees, and they are considered one of the most important tree groups in the country (Taoda 2005; Tani and Kawawata 2008). Their importance in Japan rests more on ecological, social, and esthetic values than on economic revenues, except in certain parts of the country where harvesting of some oaks in natural forests for commercial use is being promoted (Ohsawa et al. 2011). Some Quercus species comprise the main component of deciduous temperate forests of Japan (Kanno et al. 2004; Taoda 2005), which include four naturally occurring species of section Prinus; Quercus serrata, Q. crispula, Q. aliena, and Q. dentata (Kitamura and Horikawa 1951). These species all have differing distribution patterns and habitat preferences (Kanno et al. 2004; Okaura et al. 2007). Q. serrata and Q. crispula have wide distributions, while Q. aliena and Q. dentata have highly fragmented populations in Japan, and Q. serrata and Q. aliena are better adapted to warm regions, while Q. crispula and Q. dentata prefer cool climates.

According to phylogeographic structures of the four related species previously inferred from analyses of chloroplast DNA (cpDNA) variation, they all have two distinct lineages, separated in northeast and southwest regions of central Japan by major mountain chains (Kanno et al. 2004; Okaura et al. 2007). However, the studies were not based on sufficient genetic data for all congeneric species to support the generally inferred phylogeographic structure of these oaks, due to a lack of either sufficient polymorphism or sufficient numbers of surveyed populations. Moreover, genetic markers with multiple modes of inheritance should ideally be examined to elucidate the phylogeographic structure of a species fully, due to associated variations in levels of gene flow. Many previous studies have already shown that a species may have varying geographic structure in its genomes. Therefore, it is essential to consider the genetic structures of different genomes of focal species, especially when developing genetic guidelines for designating seed and seedling transfer zones, such as those of the Forestry and Forest Products Research Institute (FFPRI) mentioned above. Thus, in this study we assessed the geographical patterns of genetic variations in Q. aliena and Q. serrata, by analyzing both cpDNA sequences and nuclear microsatellite markers from expressed sequence tag sites (EST-SSRs). CpDNA has been used in many phylogeographic studies because it retains strong signatures of population history, due to its slow evolution and lack of recombination (Hillis and Moritz 1990). A major strength (and weakness) is that it is maternally inherited, thus it reflects the genetic structure and history of single lineages. In contrast, the analysis of nuclear microsatellite markers can reveal the genetic structure and population history at bi-parentally inherited loci. The high variability of these markers enables higher resolution analysis of genetic structure, with the power to detect loss of genetic (allelic) diversity, which is useful for predicting potential long-term population genetic risks associated with translocation. The main purpose of this study was to obtain genetic information to facilitate the development of genetic guidelines for transfer of planting stocks of the two species.

Materials and methods

Characteristics of the study species

Quercus aliena and Q. serrata co-occur in secondary forests in temperate zones of Japan (Kanno et al. 2004; Iwabuchi et al. 2006; Okaura et al. 2007). Q. serrata is generally more abundant, but they have different patterns of geographic distribution across Japan. Q. aliena is more common in southwestern Japan, but mostly occurs in the Chugoku region (Fig. 5c), and is sporadically distributed (mostly in forests on alluvial plains around lakes and rivers) in northeastern Japan (Nozaki et al. 2001; Kanno et al. 2004). Q. serrata is widely distributed from the Ishikari lowlands (43° 32′ N, 141° 55′ E) in the southern tip of the northern island Hokkaido to Kyushu (31° 34′ N, 130° 48′ E) in southern Japan, except for populations at the northern limit, which are relatively small and fragmented (Kitamura and Horikawa 1951; Kanno et al. 2004). Outside of Japan, the two species are naturally distributed and co-occur in secondary forests in Korea (You et al. 1995; Lee and You 2012) and China (Kitamura and Horikawa 1951; Tani and Kawawata 2008).

We analyzed a total of 12 populations of Q. aliena and 44 populations of Q. serrata distributed across almost the entire ranges of the species in Japan (see Table 1, Fig. 5a–c) and for detailed information about the populations, Online Resource 1). Due to limited numbers of individuals in some populations of Q. aliena, only nine populations (n = 212) were used for genotyping nuclear microsatellite markers, while all 12 populations (n = 89) were used for sequencing cpDNA. For analysis of Q. serrata, 43 populations (n = 1032) were genotyped by nuclear microsatellite markers, while cpDNA of 44 populations (n = 350) was sequenced.

Table 1 Estimates of genetic diversity, inbreeding coefficients, and results of tests for bottlenecks in populations of Q. aliena and Q. serrata based on 30 EST-SSR markers
Full size table

DNA extraction, microsatellite genotyping, and sequencing of chloroplast DNA regions

DNA was extracted using a DNeasy Plant Mini Kit (Qiagen, Hilden, Germany). Each individual was genotyped at 30 simple sequence repeats of expressed sequence-tagged sites (EST-SSR) loci. All the loci except three (see Online Resource 1) were amplified using primers developed from Quercus mongolica var. crispula (Ueno et al. 2008; Ueno and Tsumura 2008; and this study); Castanopsis sieboldii var. sieboldii (Ueno et al. 2009a); and Fagus crenata (Ueno et al. 2009b) and a multiplex PCR Kit (Qiagen, Hilden, Germany). The (7-μL) reaction mixtures contained 3.0 μL Qiagen Master Mix, 0.2 μM of each forward and reverse primer, and 10 ng/μL of genomic DNA. The temperature program, provided by a GeneAmp 9700 cycler (Applied Biosystems), consisted of 15 min initial denaturation at 95 °C, followed by 35 cycles of denaturation at 94 °C for 30 s, annealing at 58 °C for 1 min and 30 s, extension at 72 °C for 1 min, and a final extension at 60 °C for 30 min. The products from each locus were separated using a 3100-Avant Genetic Analyzer (Applied Biosystems), and their sizes were scored using Genotyper 3.2 (Applied Biosystems).

Four polymorphic and informative non-coding regions (Online Resource 3a) screened from a total of 43 cpDNA regions (Online Resource 3b) in the preliminary analysis were sequenced and analyzed in samples of both species. Previously reported primers were used to amplify three of the analyzed regions: the trnT (UGU) –trnL (UAA) 5′exon spacer (Taberlet et al. 1991), rps16 intron (Shaw et al. 2005), and rpL32-trnL (Shaw et al. 2007). The sequences of forward and reverse primers used to amplify the 3′ to_rps2 fragment (5′-GTCATATATTTGATCCCGCC-3′ and 5′-AACCGGAACTAGTCGGATG-3′, respectively) have not been previously published. The PCR conditions were the same as those used for the EST-SSR amplification, except that each region was amplified in 10-μL reaction mixtures, containing 5 μL Qiagen Master Mix, and the annealing temperature was reduced to 57 °C. The products were purified using ExoSAP-ITTM (GE Healthcare Limited) and sequenced in both directions using a BigDye Terminator Sequencing Kit (PE Biosystems). The sequencing products were then purified by ethanol precipitation and analyzed in a 3100 Genetic Analyzer (Applied Biosystems). Sequence data were assembled and manually edited using Sequencher 10.4.1 (Gene Codes Corporation).

Analyses of nuclear microsatellite data

The genetic diversity of microsatellite loci at each locus and over all loci, across populations in each species, was estimated in terms of the total number of alleles (A) and Nei’s unbiased estimate of gene diversity within populations statistic (H S; Nei 1987). Across loci, the genetic diversity within populations in each species was assessed in terms of the unbiased expected heterozygosity (H E; Nei 1987) and allelic richness (AR; El Moussadik and Petit 1996; Petit et al. 1998). The F IS value of each population was calculated to measure the degree of inbreeding, and significant deviations of F IS values from 0 were determined by permutation tests, with Bonferroni correction (Rice 1989). FSTAT 2.9.3.2 (Goudet 2002) was used for calculating all these statistics and significance tests.

The genetic differentiation at each locus and over all loci, across populations in each species, was estimated by the fixation index, F ST (Wright 1951; Weir and Cockerham 1984). Population differentiation was tested, without assuming random mating, by calculating log-likelihood G statistics through randomizations of genotypes among samples (Goudet et al. 1996). The degree of genetic differentiation was also evaluated by the standardized G ST, G′ST (Hedrick 2005).

To examine geographical trends in genetic diversity, the relationships between genetic diversity (AR and H E) and latitude of populations were assessed by correlation analyses and one-tailed Student’s t tests implemented in Microsoft Excel (2010). Theoretically, genetic bottlenecks cause losses of both alleles and heterozygosity (Nei et al. 1975; Maruyama and Fuerst 1985; Allendorf and Leary 1986). However, if a population has been subjected to a recent bottleneck, its allelic diversity will theoretically have declined more than its heterozygosity, which will be larger than values expected at mutation-drift equilibrium (H EQ) since H EQ is calculated from the observed number of alleles at focal loci and population size (Nei et al. 1975; Maruyama and Fuerst 1985; Cornuet and Luikart 1996). To detect signatures of bottlenecks in studied populations, we used the program Bottleneck 1.2 (Piry et al. 1999) and applied the one-tailed Wilcoxon’s sign-rank test (Luikart et al. 1998) to assess excess heterozygosity in each population. P values associated with bottleneck tests were calculated by performing 10,000 permutations under both the infinite allele model (IAM) and stepwise mutation model (SMM).

According to the key assumption underlying the isolation by distance model, genetic differentiation should be stronger between geographically distant populations than between proximal populations (Rousset 1997). To test this assumption, we assessed the relationship between populations’ genetic similarity, expressed as F ST/(1 − F ST) (Rousset 1997), and the natural logarithm of linear distances between them (converted from pairwise geographic coordinates by the Geographic Distance Matrix Generator 1.2.3; Ersts 2006), using the GENEPOP 4.0.10 web platform (Raymond and Rousset 1995; Rousset 2008). The degree and significance of this relationship, for both species, were tested using Mantel tests (Mantel 1967) with 10,000 permutations implemented in GENEPOP.

Genetic structure at the individual level was inferred by model-based clustering analysis implemented in Structure 2.3.3 (Pritchard et al. 2000), first using data pertaining to both species, and then separately for each species to detect genetic structure at species and population levels. The model assumes the existence of K ancestral genetic clusters, to which individual multilocus genotypes are probabilistically assigned through Markov chain Monte Carlo simulations and the K is sought that optimizes the log probabilities of the data (LnPD), and thus provides the best model of observed genetic structure. We analyzed our data following the admixture model of ancestry, with prior information on the location of populations (LOCPRIOR; Hubisz et al. 2009), and correlated allele frequencies to improve the clustering for closely related populations (Falush et al. 2003). Twenty replications were performed for each value of K between 1 and 9, with 100,000 iterations after 50,000 burn-in iterations. However, no clear indication of the optimal K for each species was obtained as LnPD progressively increased as K increased (data not shown). Thus, we calculated ΔK values (Evanno et al. 2005), using Structure Harvester (Earl and vonHoldt 2011), and assumed that the K with the highest ΔK value provided the best indication of clustering for each species. The genetic relationship between populations in each species was explored by generating neighbor-joining (NJ) trees (Saitou and Nei 1987) based on D A distances (Nei et al. 1983) between populations, using Populations 1.2.30 (Langella 1999). The robustness of the relationships was assessed by bootstrap percentages from 1000 replicates.

Analyses of cpDNA data

Haplotype classification was solely based on nucleotide substitutions because there were numerous gaps due to large indels in one of the cpDNA regions (rpL32-trnL).The haplotype frequencies and diversities within populations (Nei 1987) were computed using DnaSP 5.10 (Librado and Rozas 2009). The total and mean diversities (H T and H S) and relative differentiation (G ST = D ST/H T, where D ST is the absolute differentiation computed as H T − H S; Nei 1973) in each species were estimated, following assumptions of Pons and Petit (1995), by CONTRIB 1.02 (Petit et al. 1998).

To examine the phylogenetic relationships of haplotypes, minimum spanning networks were calculated, using Prim’s (1957) algorithm as implemented in Arlequin 3.5 (Excoffier and Lischer 2010) and illustrated using HapStar 0.5 (Excoffier et al. 2005; Prim 1957).

Inference of historical and demographic history by coalescence simulation using microsatellite genotype data

To infer the evolutionary history underlying the observed significant genetic differentiation at nuclear microsatellites between northeast (NE) and southwest (SW) populations of Q. aliena (see “Results”), we used the coalescence simulation implemented in DIYABC 1.0 (Cornuet et al. 2010). DIYABC can analyze more complex evolutionary scenarios, involving both recent and ancient historical events (Cornuet et al. 2008, 2010), than STRUCTURE, which is more reflective of recent admixture (Cornille et al. 2012). The data subset for this analysis was drawn from eight individuals of each population in each cluster, except the SHG population due to moderate admixture and population substructure observed in the STRUCTURE analysis. We evaluated four scenarios (Fig. 6) that could explain the most probable divergence history of the SW and NE populations of Q. aliena, including time of divergence and associated changes in demography brought by bottleneck events. Assumptions on time and associated bottlenecks were inferred from published records and/or reviews about pollen fossil records of oaks, as well as glacial and post-glacial re-colonization routes, with the assumed northward expansion of broadleaved trees in Japan, including oaks (reviewed in Okaura et al. 2007), and the associated reduction in population sizes during periods of retreat and expansion of populations in response to changes in climate.

The four scenarios differ in the presumed time of divergence of the SW and NE populations, that is, the presumed divergence time (td) in scenarios 1 and 2 was older than that in scenarios 3 and 4. In terms of associated bottlenecks, all scenarios had considered the occurrence of bottleneck (whether before or after split) in the history of the present SW and NE populations, with presumed bigger population sizes. Scenarios 1 and 3 had one bottleneck event, but the bottleneck in scenario 1 is assumed to be longer than in scenario 3. On one hand, scenarios 2 and 4 had two bottleneck events, with the more recent one (t1 to tbk and td1 to tbk, respectively) being more severe than the earlier, leading to further decrease in effective population sizes of SW and NE (i.e., N1bk and N2bk, respectively). Between scenarios 2 and 4, scenario 4 assumes that the second bottleneck from time t to tbk, which is assumed to be more severe than the earlier one, happened after the population split into SW and NE populations, resulting to population sizes N1bk and N2bk, respectively.

Specifically, the following assumptions for each scenario were inferred. In scenario 1, the SW and NE populations experienced a period of one historically long bottleneck (from time td2 to tbk) after the split of ancestral population, NA. Each population then recovered into its present population sizes, N1 and N2, respectively. In scenario 2, the SW and NE populations experienced two major bottleneck events since split of ancestral population (NA) in which the most recent (occurred from t1 to tbk) was more extreme than the initial bottleneck (occurred from td2 to t1), causing further decrease of population size after the split into SW and NE populations (population sizes are N1a and N2a, respectively), and each continue to experience further bottleneck (population sizes are N1bk and N2bk, respectively). In scenario 3, the SW and NE populations first experienced bottleneck (from time t to td1) before their split. At time td1, each population recovered into its present population sizes (N1 and N2, respectively). In scenario 4, however, the more recent bottleneck causing further decrease in population sizes is due to the split of the ancestral population that underwent bottleneck (Na).

To determine which mutation model and prior distribution of parameters (i.e., effective population size and times of split and bottleneck) provide the best fits to the observed data, we ran simulations under both generalized stepwise mutation (GSM) and stepwise mutation model (SMM) assumptions, with uniform and log-uniform priors. The GSM model considers two mutation parameters: the mutation rate (μ) and geometric distribution (P) of the number of repeated motifs involved in mutation. Uniform prior distribution was assigned for the mean μ (\( \overline{\mu} \)) and mean P, and values for each locus were drawn from a gamma distribution. Single-nucleotide insertion-deletion mutation (SNI), with a mean mutation rate of μSNI, was included to take into account alleles that have not followed strict stepwise mutations in motif size, again using values drawn from a gamma distribution (Online Resource 3). For SMM-based simulations, the mean P and locus-specific values were set to 0 since the repeating unit involved in mutation under SMM is always 1. We generated 106 simulated data for each scenario by calculating the following summary statistics: mean number of alleles (A) across loci, mean genetic diversity (H E) across loci, mean size variance (V) across loci, mean Garza-Williamson M index, classification index (LIK), and F ST between samples. Following the results of pre-evaluation of scenario-prior combinations under the two mutation models and subsequent model-checking (see “Results” and Online Resources 5, 6, 7, and 8), we computed the historical demographic parameters based on scenario 4 under the GSM model, with uniform priors, using 103 simulated data closest to the observed dataset, after logit transformation of parameter values. Measures to assess bias and precision of the analysis and obtained parameters were also calculated.

Results

Genetic diversity and differentiation at nuclear microsatellite loci

Estimates of microsatellite diversity within populations of each species in terms of average AR and H E were 6.1 and 0.571, respectively, in Q. aliena and 5.6 and 0.590, respectively, in Q. serrata (Table 1). There were significant deviations of F IS from 0 in one population of Q. aliena and five populations of Q. serrata (Table 1), but they were not significant (P > 0.05) after Bonferroni correction. The 30 nuclear microsatellite loci were polymorphic across populations in both studied species (Online Resource 2). The average number of alleles (A) and gene diversity (H S) in Q. aliena were 9.2 and 0.567, respectively, and 12.0 and 0.591, respectively, in Q. serrata. Over all loci, estimates of genetic differentiation among populations (F ST and G’ST) varied among loci in each species (Online Resource 2). F ST values obtained for Q. aliena and Q. serrata were 0.113 and 0.014, respectively, and G′ST values were 0.369 and 0.055, respectively.

There were small but significant negative correlations between genetic diversity (AR and H E) in Q. aliena and Q. serrata populations and their latitude (Fig. 1), and in both species H E declined slightly less with increases in latitude than AR, as expected due to the weaker sensitivity of H E to changes in population size in a short time (Maruyama and Fuerst 1985; Cornuet and Luikart 1996). Evidence of significant recent population bottlenecks was detected in seven populations of Q. aliena and 15 populations of Q. serrata (Table 1), but associated probabilities were mostly weak and not significant after Bonferroni correction (P > 0.05), except for three populations of Q. aliena (MYG, SHG, and FKK).

Fig. 1
figure 1

Relationship between genetic diversity, in terms of AR and H E, and latitude of populations of a Quercus aliena and b Q. serrata based on 30 EST-SST markers

Full size image

Genetic structure revealed by nuclear microsatellite analysis

Genetic differentiation in Q. aliena and Q. serrata displayed significant IBD patterns (Fig. 2), but the degree of IBD was stronger in Q. aliena (Fig. 2a) than in Q. serrata (Fig. 2b). F ST values between the isolated HKD population of Q. serrata and the other populations were substantially higher than the values for all other pairs of populations, but excluding HKD from the analysis did not lessen the IBD (R 2 = 0.1756, P < 0.001). The STRUCTURE analyses of the two species (separately or together) indicated that 2 was the optimal number of ancestral genetic clusters, since ΔK was highest at this value (data not shown). This seems biologically reasonable given the reported differentiation between northern and southern lineages of oaks (Kanno et al. 2004; Okaura et al. 2007; FFPRI 2011). At K = 2, there were strong signals of the geographic genetic structure in Q. aliena, clearly separating NE and SW populations (Fig. 3B (a)). Average F ST values of the genetic clusters representing the NE and SW populations were 0.092 and 0.110, respectively. Further subdivision was observed at K = 3, with the SHG population mainly forming a different cluster (Fig. 3B (b)), but the NE-SW trend remained strong. In contrast, the genetic structure of Q. serrata at K = 2 did not reveal distinct clustering of populations in accordance with the predicted differentiation of NE and SW groups. Instead, there was a rather smooth cline in fractions of membership of individuals to each of the two ancestral genetic clusters across the species’ distribution (Fig. 3B (c)). At K = 3, a moderate genetic substructure was only observed in populations in Kyushu (Fig. 3B (d)). In addition, the HKD population tended to occupy a separate position in both inferred clusters (at K = 2). In accordance with the STRUCTURE analysis, the NJ trees indicated strong genetic relationships among the NW and SE clusters of Q. aliena populations (Fig. 4a), and consistently weak relationships within these groups of Q. serrata populations (Fig. 4b).

Fig. 2
figure 2

Relationship of genetic similarity expressed as F ST/(1 − F ST) and the natural logarithm of geographic distances between populations of a Q. aliena and b Q. serrata based on 30 EST-SSR markers

Full size image
Fig. 3
figure 3

Results of STRUCTURE analysis based on 30 EST-SSR markers using data for both Q. aliena and Q. serrata (A) and each species separately (B). Proportion of membership coefficients of Q. aliena (a, b) and Q. serrata (c, d) individuals for the inferred clusters with K = 2, the optimal number of clusters according to △K values (c, d), and K = 3 (b, d). Each vertical bar shows the average obtained from 20 replicates. Population names are indicated in Table 1

Full size image
Fig. 4
figure 4

Neighbor-joining tree based on D A distances among populations of a Q. aliena and b Q. serrata using 30 EST-SSR markers. Nodes supported by greater than 50% bootstrap values from 1000 replicates are shown

Full size image

Estimation of historical and demographic parameters

Pre-evaluation of scenario-prior combinations showed that microsatellite datasets of Q. aliena similar to observed datasets could be generated with uniform priors, under both GSM and SMM models, for scenarios 3 and 4, but not for scenarios 1 and 2 (Online Resources 6a and 6b). However, scenario 4 was consistently supported with considerably higher posterior probability than scenario 3 under both models (Online Resources 7a and 7b), although confidence in scenario 4 was blurred due to higher frequencies of type I and II errors than under scenario 3 (Online Resource 7b). However, in the model-checking (Online Resource 8a and 8b) only scenario 4 under the GSM model did not have mis-specified summary statistics (Online Resource 8b), indicating a better fit of this scenario. This suggests that the SW and NE populations separated after the predicted bottleneck. Under this scenario, the estimated time of divergence (td) of the nuclear genome of NE and SW populations of the species was 785 (median) generations ago and their divergence is associated with severe bottleneck (Table 2 and Fig. 5). A very similar estimate (median td = 781) was obtained after removing the six loci with imperfect repeat motifs, but allele distributions are consistent with a stepwise mutation pattern (data not shown).

Table 2 Estimates of demographic parameters based on 30 EST-SSR markers obtained using coalescence simulation based on the most probable evolutionary scenario (scenario 4) underlying the divergence of the nuclear genome of NE and SW populations of Q. aliena under the GSM model, with uniform prior distribution of parameters
Full size table
Fig. 5
figure 5

Map of Japan a showing the four main islands (Hokkaido, Honshu, Shikoku, and Kyushu) and five regions (Tohoku, Kanto, Chubu, Kinki, and Chugoku). The ranges of natural distribution of Quercus aliena and Q. serrata are shown as gray areas on b, c, respectively. The geographic distributions of chloroplast haplotypes detected in d Q. aliena and e Q. serrata. The sizes of pie graphs are proportional to the number of analyzed individuals in each population as indicated in Table 1. Minimum spanning networks for chloroplast haplotypes in Q. aliena and Q. serrata are shown in f, g, respectively. The sizes of circles are proportional to the frequency of observed haplotypes in each species, except for gray-shaded circles (less than 5%). Small black circles indicate vectors for missing haplotypes

Full size image

Genetic variation and differentiation of cpDNA

The combined length of the four examined cpDNA regions in the two species was 3340 bp, with a total of 15 nucleotide substitution sites (Online Resource 3). The rpL32-trnL region was the most polymorphic, with nine substitution sites. Nucleotide diversity (π) of the four regions was 0.00068 in Q. aliena and 0.00075 in Q. serrata. The sequences have been deposited in the DNA Data Bank of Japan (Online Resource 4).

In total, 20 haplotypes were detected in the two species (Fig. 5d, e): eight in Q. aliena, 19 in Q. serrata, and 7 common to both species. In Q. aliena, haplotype 2 was the most common (42%) and the dominant haplotype among northeastern populations, while the least common (1%) was haplotype 20, a private haplotype of the MIE population and the only Q. aliena haplotype not found in Q. serrata (Fig. 5d and Online Resource 3). Similarly, haplotype 2 was the most common (29.1%) haplotype of Q. serrata and the dominant haplotype among northeastern populations of the species, while haplotypes 10 and 19 were the least common (0.3%) and only detected in southwest Japan. Eleven haplotypes were private to specific Q. serrata populations, but one of them (haplotype 15) was also present (and more frequent) in Q. aliena. Within-population haplotype diversities of Q. aliena and Q. serrata ranged from 0.000 to 0.833 and 0.000 to 0.821, respectively (Table 1). However, geographic locations of haplotypes in both species were not clearly reflected in the haplotype networks (Fig. 5f, g). Average values for diversity within populations (H S) and in the total population (H T) of Q. aliena were 0.171 and 0.816, respectively, and the relative genetic differentiation (G ST) was 0.791. Corresponding estimates for Q. serrata were 0.213, 0.849, and 0.750, respectively.

Fig. 6
figure 6

Tested evolutionary models underlying the divergence of NE and SW populations of Q. aliena based on 30 EST-SSR markers. Scenarios 1 and 2 assume that populations split before the predicted bottleneck, while scenarios 3 and 4 assume the split occurred after the predicted bottleneck. Scenarios 2 and 4 differ from 1 and 3, respectively, by further assuming that each of the populations experienced a bottleneck following the LGM (as explained in the “Materials and methods” section). Parameters are defined in Table 2 and their relationships, including prior distributions, are indicated in Online Resource 7a and 7b. NA is bounded by broken lines to represent an unknown relationship with the present populations

Full size image

Discussion

Genetic diversity and structure of Q. aliena and Q. serrata

Previous observations on Q. crispula in Japan have detected significant, but moderate, latitudinal reductions in allelic diversity of populations from south to north (Ohsawa et al. 2011; A. Matsumoto et al. in preparation). We observed the same geographical cline in Q. aliena and Q. serrata in this study (Fig. 2), but it appears to be weaker than in Q. crispula. However, the clines are considerably weaker in all three of these oaks than in several other deciduous broadleaf species examined, such as F. crenata (Hiraoka and Tomaru 2009a), which co-occurs with oaks in cool temperate forests. This clinal reduction of within-population genetic diversity in species has been associated with the northward expansion of deciduous forests in Japan following the LGM. However, not much substantial evidence of recent bottlenecks was detected either in this study or the cited study on Q. crispula. The weak clinal reduction in AR towards the north may be related to the smaller spatio-temporal distributions of refugial populations of oaks during the LGM than those of other taxa, including F. crenata, which reportedly had multiple refugia along both the Pacific Ocean and Japan Sea Side of Japan during the LGM (Tsukada 1982a, b; Hiraoka and Tomaru 2009a). However, it is surprising that the loss of allelic diversity in Q. aliena was not more pronounced, given the strong fragmentation of populations, especially in northeastern Japan. This may reflect a small difference in effective population sizes of “refugial” and “founding” populations of Q. aliena during the LGM. In accordance with this hypothesis, our coalescence simulations indicate that the effective sizes of the assumed “founder populations” of NE and SW populations, N1bk and N2bk, respectively, were of similar magnitude and both probably recovered at considerable rates after their split. This is evident from the large changes from N1bk (founder of Q. serrata) to N1 (present population), and from N2bk (founder of Q. aliena) to N2 (present population) (Table 2). However, the prior-posterior distribution for N1 generally overlapped and skewed towards the upper boundary (data not shown), suggesting that our observed data lacked sufficient information to estimate this parameter as precisely as other parameters.

The sharp genetic differentiation between NE and SW populations of Q. aliena at nuclear microsatellites and the cpDNA distribution observed in the present study is generally consistent with findings of previous phylogeographic studies on Japanese oaks, namely, Quercus crispula and Q. serrata (Kanno et al. 2004; Okaura et al. 2007). This differentiation is also congruent with the reported geographic structure of variation in the length and epidermis of the NE and SW populations’ acorns, which are characterized as “large and glabrous type” and “small and hairy type,” respectively (Kanno and Suzuki 2005). In addition, acorn characteristics are reportedly intermediate in populations of Shiga Prefecture, along the boundary of the two groups, which also have moderate admixture of the two ancestral genetic clusters of nuclear DNA. Similarly, geographic patterns of variations at neutral markers and both morphological and chemical properties are congruent in populations of several other Japanese tree species, e.g., Cryptomeria japonica and F. crenata (Yasue et al. 1987; Tsumura et al. 2014; Hagiwara 1977; Tomaru et al. 1997, 1998). However, an intra-specific sub-division of Q. aliena into two varieties (vars. aliena and pellucida) by Ohba (2006), based on density of stellate hairs on abaxial leaf surfaces does not follow the geographic pattern (Kanno and Suzuki 2005). Therefore, the discrepancy between genetic backgrounds of populations revealed by this study and this morphological variation challenges the suitability of the current intra-specific classification of Q. aliena.

In contrast to the consistent patterns in Q. aliena, geographic patterns of variations at nuclear and chloroplast markers were strikingly inconsistent in Q. serrata. However, the geographic distribution of its cpDNA haplotypes displays a strong north-south trend and is very similar to the pattern reported by Okaura et al. (2007), including the suggested boundary between NE and SW clusters along the Itoigawa-Shizuoka Tectonic Line (ISTL). According to a previous study, a barrier created by volcanic activity of a mountain chain in the middle of the western side of the Fossa Magna (a back-arc rift basin of Tertiary sediments, which mirrors the ISTL) caused the initial separation of populations currently located at each side of the ISTL (reviewed in Ohsawa and Ide 2011). Effects of geographic barriers on patterns and discontinuities of genetic structure or haplotypes have also been documented in other tree species (e.g., Ferris et al. 1993; Rossetto et al. 2007). Recent evidence indicates that genomes of very few Japanese plant species, except F. japonica (Hiraoka and Tomaru 2009a) and Quercus spp., are differentiated across the ISTL. However, Ohsawa and Ide (2011) suggest that differentiation or phylogeographic patterns created by physical barriers formed by factors such as the ISTL tend to disappear rapidly in evolutionary terms, especially at nuclear loci.

Inconsistencies in geographic patterns of genetic structure of nuclear genomes have also been reported in several related and co-distributed species (e.g., Elaeocarpus largiflorens and E. angustifolia, Rossetto et al. 2007; F. crenata and F. japonica, Hiraoka and Tomaru 2009a, b). The stark contrast in genetic structures of nuclear genomes between Q. aliena and Q. serrata likely stemmed from varying distributions of their populations. Strong fragmentation of Q. aliena, especially in northern Japan, may have been the main cause of the stronger differentiation of its nuclear genome. Theoretical expectations regarding effects of fragmentation include reductions in genetic diversity and increases in inter-population divergence (Ohsawa et al. 2006), both of which were observed in Q. aliena relative to Q. serrata. Genetic differentiation (G′ ST) of the nuclear genome is considerably higher in Q. aliena (0.369) than in Q. serrata (0.055) and other widely distributed deciduous broadleaf species in Japan, such as F. crenata (based on F ST = 0.027, an analogous of G ST; Hiraoka and Tomaru 2009a) and Q. crispula (G′ST = 0.090; Ohsawa et al. 2011). IBD patterns are also more pronounced in Q. aliena than in Q. serrata, indicating that genetic drift could be strongly relative to gene flow in Q. aliena. Lower levels of IBD in Q. serrata are most likely due to the greater density of mother trees associated with its broader distribution and dominance in secondary forests, especially over Q. aliena, which can be attributed to its wider ranges of light and moisture tolerance (Lee and You 2012).

Inferences on phylogeographic structure and historical demography of Q. aliena and Q. serrata

Phylogeographic studies on Quercus species based on cpDNA as earlier mentioned (Kanno et al. 2004; Okaura et al. 2007) revealed two distinct lineages (NE and SW) for these oaks and that they have contrasting evolutionary histories, which is suggested to have been separated into northeast and southwest regions by major mountain chains of central Japan. According to Okaura et al. (2007), local populations of the four related oak species (section Prinus) mentioned in this study often share cpDNA haplotypes by chloroplast capture or introgression through natural hybridization among the related oak species. This observation is very clear in our two studied oaks, with almost completely shared cpDNA haplotypes. Even among the genus Quercus, with more than 300 species, natural hybridization and introgression between closely related species have been described but morphological and ecological boundaries between species are clear (e.g., Whittemore and Schaal 1991; Dumolin-Lapegue et al. 1997; Ferris et al. 1998; Belahbib et al. 2001 reviewed in Okaura et al. 2007). Thus, oaks are suggested to be “large sets of broadly sympatric species exchanging genes” (Van Valen 1976 as reviewed in Kremer and Petit 1993) but manages to remain morphologically distinct. These are likely the same mechanisms shaping the structure of cpDNA of the four mentioned native oaks. Molecular markers such as AFLP (Matsumoto et al. 2009) had identified the distinct genetic differences of these species, although hybrid individuals had been detected. In our separate study using a fairly similar set of nuclear microsatellites as used in this study, we observed a very clear genetic differentiation among these species and found that Q. aliena and Q. serrata are more closely related than any of the two other native oaks (unpublished data but preparation for publication is underway).

However, Kanno et al. (2004) postulated that the NE populations were established by northward expansion of SW populations through a refugium in Kanto region following the LGM. In contrast, Okaura et al. (2007) concluded that the haplotypes in northeast Japan probably originated from populations in the Eurasian continent that migrated through a northern land bridge connecting northern Hokkaido. The haplotype network in the present study is consistent with establishment of the northern populations from a refugium around central Honshu. More specifically, the YNS population of Q. serrata, containing haplotype 8, from which most haplotypes of NE and SW populations likely originated, could be in part of that refugium. However, the presence of haplotype 2 in the MIE population of Q. aliena, which originated from the inferred ancestral haplotype 11 (Fig. 5a), suggests that this haplotype did not evolve independently from southern populations. This haplotype could be rare in southern parts of Japan and have established in northern Japan through a small number of founding individuals.

In forest tree species, the generation time is always a big question in natural forest (Petit and Hampe 2006). Some consider generation as the time of the start of first flowering (or first reproduction), while others use a range of values based on observed differences in maturity and reproduction of individuals of a species. For example, Okumura and Oki (1992) reported a range of minimum age of reproductive maturity of several oaks, varying from 20 to 40 years. Most of the studies used their mutation rate to estimate the generation time (Cavender-Bares et al. 2011) and sometimes used a specific formula (Yang et al. 2016) or ecological observation data (Tsuda et al. 2015). There is still no concrete method to estimate the generation time in natural forest trees. However, even in the same genus, several generation times were reported for oaks, which varied from 30 to 220 years. Bognoli et al. (2016) used 30 and 50 years for the generation time in Mediterranean oak; Ortego et al. (2015) used 50 years for scrub oak; Yang et al. (2016) hired 80 years for Chinese oaks; and Cavender-Bares et al. (2011) estimated the generation time for oaks (Quercus series Virentes) between 120 and 220 years. However, less than 40 years based on Okumura and Oki’s reported minimum age of reproductive maturity for oaks for generation time is too short to use in our case. Thus, adapting the range of 50–220 years, for example, suggests that the divergence time (or td based on the median value of 785) of Q. aliena would be a number of times earlier than the presumed period of northward expansion of broadleaved trees in Japan following the LGM. This is if the equivalent of one generation is 50 to 220 calendar years from the present (or around 39,250 to 173,000 calendar years). Since generation time of studied oaks and related species has not been clearly estimated yet, there is one likely scenario supported by this study that Q. aliena, which might be true for the related oaks, established into SW and NE populations following sequential historical population bottlenecks.

Implications of geographic genetic structure of Q. aliena and Q. serrata for seed and seedling transfer

Although molecular markers generally reflect neutral variation, molecular marker-based assessment of genetic variation is reasonable for developing practical genetic guidelines for designating seed transfer zones for conservation and restoration activities, since methods for detecting locally adaptive traits, such as reciprocal transplant experiments, generally require considerable time and labor (Hufford and Mazer 2003). As early as the 1990s when isozymes and fragment length polymorphisms such as amplified fragment length polymorphism and restriction fragment length polymorphism (AFLP and RFLP) were the main genetic markers in analysis of population genetic structure, polymorphisms of these markers have already been suggested to delineate seed transfer zones for a given species (e.g., Fagus sylvatica in Czech Republic (Gomory et al. 1998); Acacia rostellifera and A. cochlearis in southwest Australia (Krauss and He 2006) and several other plant species in Australia mentioned in the paper; Avenella flexuosa and Agrostis mertensii in Norway (Jørgensen et al. 2016)). To date, with the advancement of genetic tools to elucidate even the most cryptic genetic structure of populations, a number of studies in many countries have strongly supported the power and practicality of molecular markers for seed zoning or delineating local provenances, with some of them already adapted in the current strategies for seed collection for restoration or conservation activities for the species.

In accordance with the FFPRI criterion for transferring species with strong and consistent genetic structure at two examined genomes, as described in the “Introduction” section, we show that the NE and SW populations of Q. aliena represent well-differentiated “gene pools” in both nuclear and cpDNA genomes; thus, the areas they cover should represent valid transfer zones for the species. In 2015, Tsumura and Suyama (2015) has published a book (in Japanese) that collates geographic genetic structures of various forest tree species in Japan for establishing guidelines for seed and seedling transfer, including broadleaf leaf, pines, and larch species in Japan. Our proposal on NE-SW seed transfer zones for the studied oaks as earlier reported in brief in Tsumura and Suyama (2015) is similar to the proposed genetic guidelines for transferring planting stocks of the closely related oak Q. crispula and several other deciduous broadleaf species as published in the said book. Although the geographic genetic structure of cpDNA and nuclear microsatellites of Q. serrata are inconsistent, we suggest that the strong NE-SW geographic structure of cpDNA should be considered, in line with the FFPRI criteria for species with such inconsistency. A previous study reported that variation in acorn traits of Q. serrata tends to correlate with changes in geographic variables and environmental conditions (e.g., Iwabuchi et al. 2006). This may help (but not necessarily) to restrict the transfer of planting stocks in congruence with the cpDNA structure. Our proposal, however, does not in anyway discredit the use of climate, vegetation, topography, edaphic factors, and many other environmental variables as criteria for delineating seed zones in a conventional strategy. In the absence of current guidelines for native oaks in Japan, the proposed guidelines will be useful as logical framework for restoration of these ecologically important species.