Introduction

Quercus is a widely distributed tree genus. Its species boundaries are as weak as those of Populus (Martinsen et al. 2001; Stettler et al. 1996), Betula (Johnsson 1945; Palme et al. 2004), and Eucalyptus (Field et al. 2011; Potts and Dungey 2004), and genetic admixtures, considered to be due to interspecific hybridization, have been reported frequently all over the world (Cavender-Bares and Pahlich 2009; Guichoux et al. 2013; Lepais et al. 2009; Matsumoto et al. 2009; Moran et al. 2012; Zeng et al. 2011). Thus, it can be used as a model genus for studying hybridization. However, some pairs of Quercus species, although they are morphologically and genetically differentiated from each other, share almost all their alleles (Moran et al. 2012; Muir and Schlotterer 2005; Valbuena-Carabana et al. 2005). In these cases, it is possible that a shared ancestral polymorphism rather than hybridization is responsible for the genetic admixture between the species. In the case of Quercus robur and Quercus petraea, this possibility has been discussed (Muir and Schlotterer 2005), and finally, it has been concluded that hybridizations have truly occurred (Abadie et al. 2012; Guichoux et al. 2013; Lexer et al. 2006; Scotti-Saintagne et al. 2004). As hybridization is common in the genus Quercus, there is no doubt that the evolutionary dynamics of the genus, including speciation and range expansion, have been affected by hybridizations (Petit et al. 2003; Zeng et al. 2011).

The genus Quercus can be separated into two subgenera—Quercus and Cyclobalanopsis—based on whether the cupule is imbricate-scaled or lamellate. Subgenus Quercus is found across the northern hemisphere and contains ca. 300 species, while Cyclobalanopsis is found from the Himalayas and Southeast Asia to China and Japan and contains ca. 150 species (Huang et al. 1999). The leaf morphologies of subgenus Cyclobalanopsis trees resemble each other more than those of subgenus Quercus (Yan and Zhe-Kun 2002). Unlike subgenus Quercus, there have, to our knowledge, been no studies of hybridization using genetic makers in subgenus Cyclobalanopsis.

Our study species, Quercus acuta Thunb. and Quercus sessilifolia Blume, both belong to subgenus Cyclobalanopsis and are late successional species with evergreen leaves; they are found in warm temperate forest. Q. acuta, with oblong or elliptic leaves, 2–4 cm long petiole, a long acuminate leaf tip, cuneate or rounded leaf base, and an entire leaf margin, occurs in Korea and Japan (Ohashi et al. 2006; Ohba 2006). Q. sessilifolia, with narrowly oblong or elliptic-oblanceolate leaves, 0.4–1.2 cm long petiole, acute leaf tip, cuneate or attenuate leaf base, and nearly entire leaf margin with sparse small teeth, is found in Taiwan, China, and Japan (Ohba 2006). These two species grow in a somewhat higher elevation zone than other Cyclobalanopsis species; Q. acuta prefers upper and Q. sessilifolia lower slopes (Ito et al. 2007). Previous studies that investigated the cpDNA haplotypes of species of the subgenus Cyclobaranopsis in Japan found that Q. acuta and Q. sessilifolia exhibited the same haplotype (Ohyama et al. 1999, 2001). Based on leaf morphology, Quercus x takaoyamensis Makino, which is the putative hybrid between Q. acuta and Q. sessilifolia, has been identified in a mixed stand of the two parental species (Kobayashi and Midorikawa 1959; Makino 1920; Yamashita et al. 1999). These previous studies reported that the leaf morphology of Q. x takaoyamensis was intermediate between Q. acuta and Q. sessilifolia and highly variable. However, no studies investigating the details of its morphological and genetic traits have been published, and it is still unclear whether there actually is hybridization between Q. acuta and Q. sessilifolia.

The main objective of this study was to clarify the existence and the degree of interspecific hybridization between Q. acuta and Q. sessilifolia in the wild. First, we investigated morphological and genetic differences between Q. acuta and Q. sessilifolia using five leaf morphological traits and eight microsatellite markers. Second, the degree of genetic admixing of the two species was examined by means of Bayesian admixture analysis. Third, we investigated whether the genetic admixture is only the result of shared ancestral polymorphism or a combination of shared ancestral polymorphism and interspecific hybridization. Finally, in order to quantify the degree of interspecific hybridization, the migration rate between the two species was estimated.

Materials and methods

Study sites and sampling

We thoroughly searched habitats where Q. acuta and Q. sessilifolia occur in Gifu Prefecture, central Japan, and selected 15 sites where at least two individuals of one or both species were found (Fig. 1 and Table 1). Numbers of individuals were very low because the species are rare in this region. All trees taller than 2 m were selected for examination at each site, with the exception of Tsurusato (site number 14), where we examined only 30 individuals randomly selected from both species because they were so abundant. We sampled 3 to 4 shoots from 75 Q. acuta and 178 Q. sessilifolia individuals. Leaf specimens were collected for species identification and so that we could record the morphological traits. Three leaves picked from the sampled shoots were dried with silica gel and stored at room temperature while awaiting DNA extraction.

Fig. 1
figure 1

Locations of sampling sites

Full size image
Table 1 Sampling site information and the number of sampled individuals
Full size table

Measurement of morphological traits and data analysis

Initially, species statuses of individuals were identified subjectively based on the detailed observation of the specimens collected. Next, in order to confirm them objectively, leaf shapes, the length, and width of leaves (mm), petiole length (mm), shape of the leaf tip and base, and the absence or presence of leaf teeth were recorded for ten leaves per tree using the specimens collected. Leaf ratio was calculated by dividing length by width. The shape of the leaf tip was scored 0, 0.5, or 1, representing acute, intermediate, or acuminate, respectively. The shape of the leaf base was scored 0, 0.5, or 1, representing cuneate, intermediate, or obtuse, respectively. The status of leaf teeth was scored 0 or 1, representing without or with teeth, respectively. Five morphological traits, leaf ratio, petiole length, leaf tip, leaf base, and leaf teeth, were averaged for each individual tree. Principal component analysis (PCA) was conducted based on the five morphological traits, using the princomp function in R version 2.15.1 (R Development Core Team 2012). A correlation matrix was used for calculating eigenvalues to normalize the variables.

DNA extraction and microsatellite analysis

We selected seven genomic [QpZAG9, QpZAG110, and QpZAG119 (Steinkellner et al. 1997); QrZAG7, QrZAG20, and QrZAG101 (Kampfer et al. 1998); QM69-2M1 (Isagi and Suhandono 1997)] and seven EST microsatellite markers [QmC00141, QmC00898, QmC00963, QmC01133, QmC01368, and QmC02241 (Ueno et al. 2008); CR627959 (Ueno and Tsumura 2008)] developed for the subgenus Quercus. All were successfully amplified in our study species. Multiplex PCR was conducted using a Type-it Mnicrosatellite PCR Kit (QIAGEN, Hilden, Germany), following the manufacturers’ instructions, with a TaKaRa PCR Thermal Cycler (TaKaRa, Shiga, Japan). The length of the PCR products was scored with an ABI PRISMTM 310 Genetic Analyzer (Applied Biosystems, CA, USA), and allele size was determined using GENESCAN analysis software version 3.7 (Applied Biosystems, CA, USA). Quality of loci, deviation from Hardy-Weinberg equilibrium (HWE), genotyping error, and the fraction of null alleles for the 14 loci were checked using MicroChecker version 2.2.3 (Van Oosterhout et al. 2004). We also tested whether the loci were non-neutral, showing higher levels of interspecific genetic differentiation than expected if there was neutrality, using fdist2 (Beaumont and Nichols 1996). Finally, we used eight loci (QpZAG9, QrZAG7, QrZAG20, QrZAG101, QmC00141, QmC00963, QmC01368, and QmC02241; Table 2), which did not deviate from the assumption of HWE in both species, for which there was no evidence of genotype error or null alleles in both species, which satisfied the assumptions indicating selective neutrality and which contained enough genetic diversity to distinguish between the two species (gene diversity >0.1; see details in Supplementary material).

Genetic data analysis

Number of alleles and gene diversity of the eight loci were calculated using FSTAT version 2.9.3 (Goudet 2001). Genetic differentiations between the two species at the eight loci were tested by permutation tests not assuming HWE, implemented in FSTAT.

Bayesian admixture analysis was performed using STRUCTURE version 2.3.4 for estimating the genetic admixture rates of individuals (Falush et al. 2003; Pritchard et al. 2000). To determine the optimal number of clusters (K), ten independent runs were replicated for each value of K from 1 to 12, and the ΔK statistic was calculated using the CorrSieve package in R (Campana et al. 2011; Evanno et al. 2005). Independent runs were performed with a burn-in period of 500,000 and 500,000 subsequent Markov chain Monte Carlo (MCMC) steps. The ten independent results for each K value were merged into one using CLUMPP version 1.1.2 (Jakobusson and Rosenberg 2007).

In order to determine whether the observed genetic admixture is just the result of genetic drift or a combination of genetic drift and gene flow between the two species, relative roles of drift and gene flow were assessed using the 2MOD program that was based on coalescent theory (Ciofi et al. 1999). The QrZAG101 locus was excluded because it consisted of compound microsatellite repeats, the evolutionary processes of which are complex (Bull et al. 1999), and so, it was not ideal for the coalescent analysis. Seven loci were thus used in this analysis. 2MOD evaluates (1) the drift model, which assumes that drift is the sole reason for population divergence since a certain time, i.e., the observed genetic admixture was created by only shared ancestral polymorphisms, and (2) the gene flow model, which assumes immigration-drift equilibrium of gene frequencies within populations, i.e., the observed genetic admixture was created not only by shared ancestral polymorphisms but also by interspecific hybridization. Three independent MCMC simulations were conducted with 100,000 steps; 10,000 steps were discarded as the burn-in period and every fifth step was sampled. Ultimately, we used 54,000 samples, and the probabilities of each model [P (drift) and P (gene flow)] were calculated.

In order to quantify the interspecific gene flow, BayesAss version 3.0 (Wilson and Rannala 2003) and Migrate-n version 3.3.2 (Beerli 2006; Beerli and Felsenstein 2001) were used. Both programs are able to detect asymmetric migration events. The former can estimate the recent migration rate (over the last several generations), and the latter can estimate the historical migration rate based on coalescent theory. In the BayesAss analysis, asymmetric migration rates were estimated by Bayesian methods. Three replicate runs consisting of 1 million burn-in steps and a further 2 million MCMC steps were performed with different random number seeds, and posteriors were sampled every 2,000 steps. We confirmed that approximately the same posterior distributions were obtained from each of the three replicate runs. Trace plots for all parameters were produced, and their convergences were checked. The 95 % higher posterior density (95 % HPD) was calculated using the coda package in R (Plummer et al. 2006). In the Migrate-n analysis, because this is based on coalescent theory, we used seven loci, excluding the QrZAG101 locus, as was the case with the 2MOD analysis. The mutation-scaled effective population size (θ = 4N eμ) and migration rate (M = m/μ) were estimated by Bayesian methods. N e, m, and μ represent the effective population size, migration rate and mutation rate, respectively. Prior uniform distributions ranging from 0 to 100 and from 0 to 120 were used for θ and M, respectively, with starting values estimated from F ST. Relative mutation rates were estimated from the data, and the Brownian motion model was employed because the exact stepwise mutation model was extremely time-consuming and did not converge. The Metropolis-Hasting algorithm was used to generate posterior distributions. Three replicate runs consisting of 1 million burn-in steps and a further 200,000 MCMC steps were performed with different random number seeds, and posteriors were sampled every 100 steps. We confirmed that approximately the same posterior distributions were obtained from each of the three replicate runs. Trace plots for all parameters were produced, and their convergences were checked. A static heating scheme using 32 chains with the exponential increase heating term option was employed. Posterior distributions of parameters over all loci were calculated by averaging the randomly drawn values from the posterior distributions of each of the seven loci as empirical distributions, adjusting mutation rates among the loci using estimated relative mutation rates. The 95 % HPD was calculated, as was the case with the BayesAss analysis. Symmetric migration (migration rates were symmetrical), one-way migration, and no migration models were compared to the full model (asymmetric migration) to determine the direction of interspecific gene flow. The Bezier approximation and harmonic mean method were used to marginalize the likelihood, and we calculated log Bayes factors against the most parsimonious model [log (marginal likelihood of the most parsimonious model)–log (marginal likelihood of the model)] and determined which model was the best fit for the data (Beerli and Palczewski 2010).

Results

Morphological data

Although the distributions of each of the five leaf traits differed between the two species, there was some overlap for all of them (Fig. 2). However, using a combination of the first two principal components (PC) derived on the basis of the five leaf traits, the two species could be distinguished (Fig. 3). Although there was little overlap for PC 1, the distributions for the two species were very close. Distributions for PC 2 overlapped completely.

Fig. 2
figure 2

Distributions of five leaf morphological traits of Quercus acuta (black) and Q. sessilifolia (white)

Full size image
Fig. 3
figure 3

Distribution of principal components (PC) based on the five leaf morphological traits. Circles and squares represent individuals of Quercus acuta and Q. sessilifolia, respectively. Genetic statuses were estimated by the program STRUCTURE, based on the genotypes of eight microsatellite markers. P Qac and P Qse are the probabilities of belonging to Q. acuta and Q. sessilifolia, respectively

Full size image

Genetic differentiation and genetic admixing

For the eight loci, although the ranges of the alleles overlapped and the two species shared most of the alleles (Fig. 4), genetic differentiations were highly significant for all eight loci (Table 2). Numbers of alleles of Q. acuta were lower than those of Q. sessilifolia for all eight loci. Gene diversities of Q. acuta were also lower than those of Q. sessilifolia for six of the eight loci.

Fig. 4
figure 4

Distributions of alleles in Quercus acuta (black) and Q. sessilifolia (white) at the eight loci

Full size image
Table 2 Estimates of genetic diversities and differentiation between Quercus acuta (N = 75) and Q. sessilifolia (N = 178) at eight microsatellite loci
Full size table

In the Bayesian admixture analysis, the average value of the log probability of data plateaued at K = 4 to 8, then decreased with increasing values of K (Fig. 5). ΔK peaked at K = 2 and K = 4, indicating that these K values were appropriate to consider (Fig. 6), as secondary peak of the ΔK statistic may indicate the existence of within-species genetic structure. If such genetic structure is ignored, it can reduce the potential for species assignment. Thus, we used the results of K = 4 in the following analyses. Because the admixture coefficients q 41 and q 42, and q 43 and q 44 roughly correspond to Q. acuta and Q. sessilifolia, the sum of q 41 and q 42, and the sum of q 43 and q 44 could be considered to represent the probability of an individual belonging to Q. acuta (P Qac = q 41 + q 42) or Q. sessilifolia (P Qse = q 43 + q 44), respectively (Fig. 6). The values of P Qac and P Qse varied greatly within each species (Fig. 7). The average values of P Qac and P Qse for Q. acuta were 0.938 and 0.062, while those for Q. sessilifolia were 0.131 and 0.869, respectively. Of the individuals Q. acuta and Q. sessilifolia, 11 and 24 %, respectively, had a probability of less than 0.9 of being correctly assigned to their species. These two species thus showed similar levels of genetic admixing. P Qac and P Qse values for individual trees of the two species were compared with the results of the PCA (Fig. 3). However, individuals that showed genetic admixing were uniformly distributed across PC 1 and PC 2, and no clear relationships could be detected.

Fig. 5
figure 5

Changes in the log probability of data and ΔK statistic from the Bayesian clustering analysis using the program STRUCTURE

Full size image
Fig. 6
figure 6

Posterior distributions of membership of each genetic cluster (q; K = 2 and 4) estimated by the program STRUCTURE. Lower and upper bars indicate individuals of Quercus acuta and Q. sessilifolia, respectively. Numbers under the bar plot indicate sampled sites shown in Fig. 1 and Table 1

Full size image
Fig. 7
figure 7

Posterior distributions of the probability of belonging to Quescus acuta (1-probability of belonging to Q. sessilifolia)

Full size image

Population history and migration

The 2MOD population history test strongly supported a gene flow model rather than a drift model [P (gene flow) = 1.000, P (drift) = 0.000], indicating that interspecific hybridization significantly contributed to the observed genetic admixture between the two species.

In the detection of recent migration events by BayesAss, the medians (95 % HPD) of the posterior distributions of the migration rates from Q. sessilifolia to Q. acuta and from Q. acuta to Q. sessilifolia were 0.140 (0.099–0.187) and 0.086 (0.060–0.116), and their 95 % HPDs overlapped.

In the detection of historical migration events by Migrate-n, the most parsimonious model was that based on symmetric migration; models with different migration parameters were compared against the most parsimonious model using the log Bayes factor (LBF; Table 3). For all the models, LBFs were larger than 21.3 and 128.5 for the Bezier approximation and harmonic mean methods, respectively. When the Bayes factor for each model compared to the most parsimonious model is >150 (LBF against the most parsimonious model was greater than log (150) ≈ 5.0), support for the most parsimonious model is overwhelming (Link and Barker 2010). In our analysis, since LBF > > 5 in both methods, the symmetric migration model was strongly supported.

Table 3 The most parsimonious model (in bold) and other candidate models and their log marginal likelihoods (LML) and log Bayes factors (LBF) when compared to the most parsimonious model
Full size table

Medians (95 % HPD) of the posterior distribution of mutation-scaled population size (θ = 4N eμ) in Q. acuta and Q. sessilifolia were 2.433 (2.070–2.958) and 18.52 (15.30–21.49), respectively (Fig. 8a). The median (95 % HPD) of the posterior distribution of mutation-scaled migration rate (M = m/μ) was 14.19 (12.86–15.84; Fig. 8b). Posterior distributions of the effective number of migrants per generation (N e m) were calculated by θ × M/4, and medians (95 % HPD) of the values from Q. sessilifolia to Q. acuta and from Q. acuta to Q. sessilifolia were 8.843 (7.362–11.15) and 71.98 (56.64–84.72), respectively (Fig. 8c).

Fig. 8
figure 8

Posterior distributions of mutation-scaled effective population size (θ = N eμ; a), mutation-scaled migration rate (M = m/μ; b), and the effective number of migrants per generation (N e m; c)

Full size image

Discussion

What caused the genetic admixture?

Although Q. acuta and Q. sessilifolia were genetically and morphologically differentiated, they were genetically admixed in the selectively neutral genomic regions studied. Interspecific hybridization is one of the possible causes of this. In the genus Quercus, genetic admixing has often been reported over its distribution range and is considered to be the result of hybridization (Cavender-Bares and Pahlich 2009; Guichoux et al. 2013; Lepais et al. 2009; Matsumoto et al. 2009; Moran et al. 2012; Zeng et al. 2011). In the studied species, most of the alleles at each locus were shared between the two species, and the ranges of the alleles were almost entirely overlapping. These distributions of alleles suggest that this admixture may be the result of a shared ancestral polymorphism (Muir and Schlotterer 2005). If this admixture was created by only shared ancestral polymorphism, the underlying mechanism could be considered to be pure genetic drift after speciation. We compared pure drift model with a gene flow model (drift + gene flow), and the gene flow model was strongly supported. Despite the fact that these two species are phylogenetically very close to each other (Ohyama et al. 1999, 2001) and, therefore, the possibility that the shared ancestral polymorphism is the main cause of this genetic admixture that cannot be completely rejected, the result of the model comparison indicated that the interspecific gene flow significantly affected the genetic admixing of the two species. As with the species in our study, the European oaks Q. robur and Q. petraea exhibit genetic admixing, with most of their alleles shared, and this has been attributed to shared ancestral polymorphism at one time (Muir and Schlotterer 2005). However, increasing evidence based on genome scans and detailed artificial pollination experiments (Abadie et al. 2012; Guichoux et al. 2013; Lepais et al. 2013; Lexer et al. 2006; Scotti-Saintagne et al. 2004) supports the possibility that hybridization accounts for shared polymorphisms between these oak species. The results of previous studies and our study suggest that genetic admixing with the sharing of many alleles as a result of hybridization can be considered a common occurrence in the genus Quercus (Moran et al. 2012; Muir and Schlotterer 2005; Valbuena-Carabana et al. 2005).

Characteristics of hybridization between the two oaks

In both species, admixed individuals exhibited a very variable admixture rate. Moreover, genetic admixtures were observed even at the site where only Q. sessilifolia grew. As indicated by the results of the BayesAss and Migrate-n analyses, the genetic admixture observed in this study could have been the result not only of recent hybridization but also of several occurrences of historical hybridization and, thus, introgression. Admixed individuals exhibited a range of morphological traits. This could be because no loci used in this study violate the assumption of selective neutrality since, although introgression can occur between the selectively neutral genomic regions as a result of hybridization, this cannot occur in the genomic regions affected by natural selection that produced functional differences, even when interspecific gene flow occurs frequently (Butlin 2010; Via 2009). Thus, the two species could remain morphologically distinct. Most of the populations in the study region consisted of small numbers of individuals. This may have limited intraspecific pollination, and interspecific hybridization can be an effective way to ensure the reproduction. If this is so, the estimated admixture rates in this study may differ from rates in other regions where there are enough individuals of each species to minimize levels of hybridization.

A putative hybrid of the two species, Q. x takaoyamensis, is known, identified by its intermediate leaf morphology (Makino 1920). It has been reported that the leaf morphology of Q. x takaoyamensis is not stable and is highly variable among individuals (Kobayashi and Midorikawa 1959; Yamashita et al. 1999). Not all individuals with the intermediate genetic admixture rate exhibited intermediate morphological traits (PC 1), and thus, there was no clear congruence between the leaf morphology and genetic status. Although hybrid individuals between Q. acuta and Q. sessilifolia, i.e., Q. x takaoyamensis, are sure to exist in the wild as shown by the result of our genetic analyses, it is difficult to distinguish them from pure Q. acuta and Q. sessilifolia individuals on the basis of leaf morphology.

No significant asymmetric migration between the two species was detected in association with either recent or historical migration events. Generally, there are some differences in the strength of barriers to interspecific gene flow between species, and therefore, asymmetric hybridization occurs (Arnold 1997; Lewis and Crowe 1958; Tiffin et al. 2001). In the genus Quercus, some cases of asymmetric hybridization have been reported: For example, in Q. robur and Q. petraea, there were differences in fertilization success (Steinhoff 1993), and in Q. robur, Q. petraea, Quercus pubescens, and Quercus pyrenaica, post-zygotic barriers were the key determinant of the direction of hybridization (Lepais et al. 2013). In our study, although migration rates were symmetric across the whole region, at some sites (1, 6, and 14), the two species exhibited different levels of admixing. A more detailed study is required to reveal the direction of hybridization between Q. acuta and Q. sessilifolia.

In the coalescent-based historical migration analysis, although the mutation-scaled migration rate (M = m/μ) was fixed between the two species, the mutation-scaled effective population sizes (θ = 4N eμ) were different, and the value of θ for Q. sessilifolia was ca. 8 times larger than that for Q. acuta. As a result, effective numbers of migrants per generation (N e m) were different, and the values from Q. sessilifolia to Q. acuta and from Q. acuta to Q. sessilifolia were 8.843 and 71.98, respectively. Sewall Wright’s one migrant per generation rule states that, in ideal populations, one migrant per generation is enough to prevent complete population differentiation (Wright 1931). However, in real populations with factors in play other than those accounted by genetic theory, one to ten migrants per generation are considered necessary (Mills and Allendorf 1996; Wang 2004). The values of N e m observed in this study are equal to (N e m into Q. acuta) or larger than (N e m into Q. sessilifolia) 10, suggesting that there is sufficient hybridization to prevent complete genetic differentiation between the species and that the genetic condition of most alleles being shared between the species has been maintained. In order to confirm hybridization between the two species, artificial pollination experiments or paternity and parentage analyses to directly detect recent hybridizations are required.

Conclusions

Q. acuta and Q. sessilifolia exhibited genetic admixing, and we consider that the cause is not simply shared ancestral polymorphism but also interspecific hybridization. Hybridizations have occurred several times, and as a result, there has been introgression. Although these two species are genetically and morphologically differentiated, genetic admixture rates do not correspond to the observed morphology. This suggests that genomic regions related to morphology have been strongly affected by natural selection. N e m has been sufficient to prevent complete genetic differentiation, and the sharing of most of the alleles has been maintained by introgressive interspecific gene flow.