Genetic Structure and Hierarchical Population Divergence History of Acer mono var. mono in South and Northeast China

Knowledge of the genetic structure and evolutionary history of tree species across their ranges is essential for the development of effective conservation and forest management strategies. Acer mono var. mono, an economically and ecologically important maple species, is extensively distributed in Northeast China (NE), whereas it has a scattered and patchy distribution in South China (SC). In this study, the genetic structure and demographic history of 56 natural populations of A. mono var. mono were evaluated using seven nuclear microsatellite markers. Neighbor-joining tree and STRUCTURE analysis clearly separated populations into NE and SC groups with two admixed-like populations. Allelic richness significantly decreased with increasing latitude within the NE group while both allelic richness and expected heterozygosity showed significant positive correlation with latitude within the SC group. Especially in the NE region, previous studies in Quercus mongolica and Fraxinus mandshurica have also detected reductions in genetic diversity with increases in latitude, suggesting this pattern may be common for tree species in this region, probably due to expansion from single refugium following the last glacial maximum (LGM). Approximate Bayesian Computation-based analysis revealed two major features of hierarchical population divergence in the species’ evolutionary history. Recent divergence between the NE group and the admixed-like group corresponded to the LGM period and ancient divergence of SC groups took place during mid-late Pleistocene period. The level of genetic differentiation was moderate (FST = 0.073; G′ST = 0.278) among all populations, but significantly higher in the SC group than the NE group, mirroring the species’ more scattered distribution in SC. Conservation measures for this species are proposed, taking into account the genetic structure and past demographic history identified in this study.


Introduction
Northeast China contains the country's most important forest areas, but since the late 19th century forests in this region have been severely disturbed by wars, fires, deforestation, overexploitation and changes in land use to agriculture and grazing. Consequently, many tree species are now threatened [1]. Some conservation measures, such as designating areas as noncommercial forest zones and forbidding clear-cutting in degraded sites, have been applied in this region, but degradation of forests is still a serious threat [2][3]. The genetic diversity of historical lineages cannot be recovered if it is lost [4]. Thus, it is essential to understand the biological and genetic features of the tree species in a region's forests [5][6] in order to formulate effective conservation and restoration programs that will maintain the diversity required to adapt to environmental changes [4,[9][10]. This requires robust assessments of the genetic diversity, structure and demographic history of species [7][8]. Further, the focus must be on preserving the overall genetic diversity and natural evolutionary processes in metapopulations rather than the diversity within separate subpopulations [11]. Thus, large-scale genetic structure analysis covering species' ranges has been applied to numerous forest tree species, especially species that are economically, ecologically, and/or genetically important for a specific region [5][6]8,[11][12][13][14][15][16].
The genetic diversity and population structure of various tree species in Northeast China have been evaluated using several types of molecular markers, for example Phellodendron amurense (AFLP) [17], Quercus mongolica (RAPD, AFLP and ISSR) [18][19][20], Tilia amurensis (ISSR) [21], and Fraxinus mandshurica (SSR) [14]. The genetic diversity of Q. mongolica [19] and F. mandshurica [14] has been shown to decrease with increasing latitude. Hu et al. [14] postulated that the detected cline in F. mandshurica populations arose from northward expansion of the forest biomes in the mid-Holocene from a single refugium located in eastern China during the last glacial maximum (LGM) [22]. This putative post-LGM population expansion to the north from southern refugium is believed to have shaped the modern genetic structure of many species in temperate zones in the northern hemisphere [23][24]. However, it is still not clear whether the decrease in genetic diversity with increasing latitude in the northern populations is a common pattern for tree species in Northeast China. Moreover, no coalescence-based genetic analysis has been previously applied to plant species in Northeast China, thus it is uncertain whether the reductions in genetic diversity are related to changes in species distribution driven by climate changes during the LGM. Recent forest genetics studies have suggested that an ancient demographic event predating the LGM has continuing effects on modern genetic diversity and structure [25][26][27][28]. Therefore, it is essential to evaluate the past population demography over sufficient time to elucidate the evolutionary history of a tree species and apply the acquired knowledge when formulating forest management and conservation strategies. Recently, a very powerful and flexible approach, Approximate Bayesian Computation (ABC), has been developed to estimate demographic and historical parameters (e.g. effective population size, divergence time) and quantitatively compare alternative scenarios [29]. This approach can be used to address many important questions in ecology and evolutionary biology [29].
Acer mono (mono maple) is a common deciduous tree species that is widely distributed in east Asia, where there are many varieties [30][31][32], and an important companion tree species in the temperate mixed Korean pine and broadleaved forests of Northeast China [33]. It is an economically, ecologically, and culturally important species because of the unique properties of its wood, potential for production of sugar, oil, and medicinal materials, excellence for landscape planting and forestation of barren hills, and especially its attractiveness as an ornamental due to its vibrant red autumn foliage and fog-tolerance [34]. In China, A. mono is mainly distributed in Northeast China [32], but there are also scattered populations in forests of North China and the middle and lower reaches of the Yangtze River (Fig. 1). The natural resources of A. mono have decreased, and almost disappeared, at some sites due to: (1) harvesting of its wood for timber or fuel; (2) inappropriate forest management activities such as removal of all undergrowth species during thinning; and (3) of most importance recently, transplantation from natural forests to urban areas for use as ornamental, shade, or landscape trees. The conservation and sustainable management of natural populations of A. mono, as well as breeding and artificial planting of the species, are therefore now receiving more attention. However, few molecular ecology studies have addressed the genetic diversity and structure of natural populations of A. mono in China, although several have examined the reproduction and mating patterns of natural populations of the species in Japan [35][36][37].
We therefore conducted large-scale genetic analysis of A. mono using nuclear SSR markers to: 1) evaluate the wide-range genetic structure of 56 populations selected to cover its distribution in northeastern and southern China, 2) infer its population demographics using an ABC approach and 3) formulate conservation and management strategies for this natural genetic resource in Northeast China, taking into account its current genetic structure and inferred evolutionary history.

Ethics Statement
No specific permission was required for sampling activities in most populations because we sampled in the wild and only a few leaves were collected from each tree, except for some populations located in National Nature Reserves (listed after the corresponding populations in the ''Sampling and study sites'' section), in which the sampling activities were permitted by the relevant authorities. The field studies did not involve disturbance of any endangered or protected species.

Sampling and Study Sites
We collected leaves of Acer mono var. mono from 48 natural populations, covering almost its entire distributions in Northeast China (hereafter NE populations), and eight in the rest of China (hereafter southern China, SC), where there are scattered, lowdensity populations (Fig. 1).

DNA Extraction and Genotyping
DNA was extracted using a modified CTAB method [38]. Nuclear SSR markers were selected from primers previously developed for Japanese A. mono [35] and other Acer species [39][40][41]. Seven of these markers (Am116, Am118, Am258, Am340, Am607 and Am742 [35], and Aca24 [39]) were found to be sufficiently stable and polymorphic and therefore used in this study.
Primers were mixed to a final concentration of 2 mM, except for primer Am258 (4 mM). The target sequences were then amplified in 5 mL reaction mixtures containing 2.5 mL of 26 Qiagen Multiplex PCR Master Mix, 0.5 mL of the primer mixture, 1.5 mL H 2 O, and 0.5 mL template DNA by PCR (95uC for 15 min, followed by 30 cycles of 94uC for 30 s, 58uC for 90 s, and 72uC for 60 s, with a final extension at 60uC for 30 min) using a TaKaRa TP600 Thermal Cycler. All PCR products were genotyped using an ABI 3130 Genetic Analyzer and the results were analyzed using GeneScan analysis software and Genotyper software version 3.5 (Applied Biosystems).

Genetic Diversity within Populations
The FSTAT 2.9.3.2 software package [42] was used to evaluate the genetic diversity of each population by calculating the total number of alleles detected (N A ), range of allele size (R AS ), allelic richness (A R ), expected heterozygosity (H E ), and fixation index (F IS ). In addition, significant deviations from Hardy-Weinberg equilibrium, as indicated by deviations of F IS from zero, and genotypic disequilibrium for all locus pairs in each population, were tested by randomization using FSTAT. Rare allelic richness (RA R ; [8]) and private allelic richness (PA R ; [8]) were calculated. Pearson Correlation Analysis, implemented in SPSS 13.0 for Windows [43], was applied to determine possible correlations between genetic diversity parameters (A R , RA R , PA R and H E ) and geographical gradients (of altitude, latitude and longitude). Since clear genetic differentiation was detected between the NE group (Populations 1-48) and the SC group (Populations 51-56) in the NJ tree and by STRUCTURE analysis (see Results), average values of A R , H E , F IS and F ST for each group and their differences between the two groups were tested using permutation tests in FSTAT. RA R and PA R of each group and the differences between the two groups were tested by One-way ANOVA using SPSS.

Genetic Differentiation and Structure
The genetic differentiation among populations was evaluated by calculating F ST values [44], and their significance was tested by comparison to 95% and 99% confidence intervals acquired from 1000 bootstraps. Pairwise F ST values were also calculated, and the significance of pairwise population differentiation was tested by randomizing multilocus genotypes between the pairs using FSTAT. The source of genetic variation was determined by analysis of molecular variance analysis (AMOVA) [45], using GENALEx6.1 [46]. Standardized genetic differentiation (G9 ST ) was then calculated following Hedrick [47].
Isolation by distance (IBD) [48] was evaluated with GenAlEx 6.4 [46] using the method of Rousset (1997) [49], which tests for significant association between pairwise population differentiation (F ST /(12F ST )) and the natural logarithms of direct minimum geographic distance between populations. The genetic relationships among populations were evaluated by generating a Neighbor-Joining (NJ) tree based on the D A genetic distances [50], using Populations 1.2.30 beta software [51]. Statistical confidence in the topology of the tree was evaluated using 1000 bootstraps derived using the same software. The NJ tree was reconstructed on a topographic map using Mapmaker and GenGIS2 software packages [52]. Individual-based genetic structure was evaluated by Bayesian clustering as described by Pritchardet al. [53] using STRUCTURE 2.3.3 (hereafter, STRUCTURE analysis) [54]. We assumed a pre-assigned number of genetic clusters (K) ranging from 1 to 20. All runs involved 100,000 Markov Chain Monte Carlo (MCMC) generations, after a burn-in period of 100,000 iterations according to Hubisz et al. [55], based on the LOCPRIOR model described by Hubisz et al. [55], an admixture model and the correlated allele frequencies model (hereafter, the F-model) described by Falush et al. [56]. Ten runs were performed for each value of K. The optimum number of K was evaluated by calculating DK values [57] and applying the methods suggested in the original paper describing STRUC-TURE analysis [53], which involves evaluating genetic structure by comparing mean log likelihoods of postulated models. Bar charts for the proportions of the membership coefficient of each individual in STRUCTURE analysis over 10 runs for each K were summarized using CLUMPP software [58] and visualized using DISTRUCT software [59].  Fig. 2C) and use of all individuals (1483 in total) would have been computationally expensive without substantially increasing the amount of information acquired. We examined four simple population demography scenarios (Fig. 3A), taking into account the results of the NJ tree and the STRUCTURE analysis. In these scenarios, t# and N# refer to timescale (expressed as generation time) and effective size of the corresponding populations (Pops1, 2, 3 or the ancestral population) during each time period (e.g. 0-t1, t1-t2). The scenarios are as follows:

Population Demographic Analysis
Scenario 1 (isolation with admixture model): Pop2 was generated by admixture of Pops 1 and 3 at t1, then Pop1 merged with Pop3 at t2. Based on the results of STRUCTURE analysis, this scenario was expected to be the most likely one. In all scenarios, Pop2 was specified as the population to be traced back to an ancestral population (as required for DIYABC analysis). In addition, we assumed a bottleneck in the past at t3, which we positioned before the divergence point in all scenarios. We assumed past population growth and set the ancestral population size (Na in scenario 1 and Nb in scenarios 2-4) to be smaller than that of other populations. In DIYABC, we employed the higher mutation rate in the generalized stepwise mutation model (GSM) [62] with the lower mutation rate of single nucleotide indels (SNI) for modeling the mutation of SSRs. Default prior values were used for all these parameters (Table S1).
The mean values of expected heterozygosity (H E ) and number of alleles (N A ) were used as the summary statistics for each of the three populations, while for summary statistics for each pair of populations from the three groups we used H E , N A , genotype likelihood and F ST . A million simulations were performed for each scenario. After the total of three million simulations, the mostlikely scenario was evaluated by comparing posterior probabilities using logistic regression. The goodness of fit of the scenario was assessed by the option ''model checking'' with principal component analysis (PCA) in DIYABC, which measures the discrepancy between model and real data.
B) Demographic inferences in South China. Because the STRUCTURE analysis indicated that DK was highest at K = 4 and detected clear genetic structure (see Results, Fig. 2D), we defined four populations: SC1 (Pops 51 and 52), SC2 (Pops53 and 54), SC3 (Pop 55) and SC4 (Pop 56) to simplify demographic analysis of the SC populations. We examined the hierarchical split models (scenarios 1-3) and simple split model (scenario 4) using DIYABC v2.0 as shown in Fig. 3B. In the hierarchical split model, the most ancient split was set between SC3 and the others because SC3 was the most highly genetically differentiated of the SC populations according to the NJ tree (Fig. 2B). After the split from SC3, all permutations of pairs of SC1, 2 and 4 populations were set in the hierarchical split models in scenarios 1-3. The effective population sizes of SC1, 2, 3 and 4 were designated N1, N2, N3 and N4, respectively (setting N1. N2. N3. N4, according to their genetic diversity). Default prior values were used for all these parameters (Table S2), and the same approach was applied as for the whole-range analysis described above.

Genetic Diversity of Populations
In total, 223 alleles were scored over the seven loci among the 1702 sampled A. mono individuals. The allelic diversity was highly variable across loci, with N A ranging from 14 to 45 (32612 on average), A R from 4.824 to 16.778 (10.86763.981 on average), and H E from 0.587 to 0.861 (0.73760.110 on average) ( Table 1). The fixation index (F IS ) for individual loci ranged from -0.116 to 0.223 and the average over all loci was 0.053. No significant genotypic linkage disequilibrium for pairwise-loci was detected within any population (P.0.05).
The genetic diversity of each population is presented in Table 2. Values of N A , R A , P A , A R , RA R , PA R and H E ranged from 39 to 84, 8 to 55, 0 to 7, 5.181 to 11.210, 1.002 to 6.031, 0 to 1.293 and 0.535 to 0.840, respectively (Table 2). Although significant deviation of the F IS value from zero was detected in 11 of 56 populations, there was no geographic pattern in the distributions of F IS values ( Table 2). When the correlations between genetic variation parameters (A R and H E ) and geographical gradients (latitude, longitude and altitude) were evaluated, significant positive correlation was detected only between H E and latitude (P,0.05, Table 3). However, when geographical pattern of genetic diversity was analyzed within each of the NE and SC regions, different trends were detected: A R significantly decreased with increasing latitude within NE populations (P,0.01), while both A R and H E increased together with latitude within SC populations (P,0.05, Table 3).

Population Genetic Structure
The F ST among all populations was 0.073, with 95% and 99% confidence intervals (CI) ranging from 0.056 to 0.093 and 0.051 to 0.1, respectively, and 66.35% of pairwise-F ST values were significant (P,0.05). The overall value of F ST corresponded to a standardized genetic differentiation (G9 ST ) of 0.278, suggesting moderate population structure. Separate analysis of the genetic variation in NE and SC populations, based on results of the STRUCTURE analysis and the NJ tree (see below), showed that H E and RA R values were significantly higher for NE than SC population (H E = 0.743 and 0.676, RA R = 3.825 and 2.799, respectively; Table 4). In contrast, PA R and F ST values were significantly higher for SC than NE populations (PA R = 0.309 and 0.006, F ST = 0.207 and 0.029, respectively), but no significant differences in A R was detected ( Table 4). The F IS values of both groups were almost zero (0.052 in NE and 0.042 in SC) and there was no significant differentiation between them ( Table 4). The G9 ST values for the NE and SC groups were 0.114 and 0.722, respectively, suggesting strong genetic structure in the SC group. Thus, 96.4% of the pairwise-F ST values were significant (P,0.05) in the SC group while the proportion was much lower (56.2%) in the NE group. Moreover, separate analysis of the pattern of genetic diversity within the groups showed that genetic variation decreased with increasing latitude in NE populations, but increased in SC populations ( Table 3). Analysis of molecular variance based on these two groups showed that within-population variation, variation among populations in each region and between-region variation accounted for 76%, 16% and 8% of the total detected genetic variation, respectively (P = 0.01). Mantel tests indicated that there was significant isolation by distance in all populations (R 2 = 0.4241,P = 0.000), but not within either the NE or SC populations (R 2 = 0.009,P = 0.090 for NE populations; R 2 = 0.081,P = 0.060 for SC populations).
According to the whole-range STRUCTURE analysis, the probability of the LnP(D) data increased progressively with increases in K up to K = 8, where it started to plateau (Fig. S1). However, the results for K .4 showed complex admixture patterns and none of the clusters were specific to any populations or regions. Since DK indicated that the optimal value for K was 2 (Fig.  S1), in the following text discussion of population genetic structure is mainly based on results acquired with K = 2. In this case, all NE populations clustered into one group, and populations 51-56 clustered into the other group. Populations 49 and 50, located in Northern China, adjacent to the regions occupied by both groups, appeared to possess admixed-like features of the two groups ( Fig. 2A, K = 2). However, with values of K = 3 and 4, although genetic structure was not clear over all NE populations, these two populations were assigned to a specific cluster. The NJ tree showed similar genetic structure to that detected by the STRUCTURE analysis, with two main groups corresponding to the NE and SC populations (Fig. 2B, C). Populations 49 and 50, for which STRUCTURE analysis suggested admixture when K = 2, fell between the two groups ( Fig. 2B, C). In addition, most of the nodes in the NE populations were poorly supported by the bootstraps due to weak population structure, but most of the nodes for Populations 49 and 50, and SC populations, were supported by higher bootstrap values (Fig. 2B), indicating strong genetic structure. Because of higher differentiation, STRUCTURE analysis on the 6 populations in SC was conducted. These six populations divided into four clusters, according to the highest DK (Fig. 2D) [57], and when K = 6 they were clearly divided into six population-specific clusters (Fig. S2A, K = 6). Moreover, separate STRUCTURE analysis of NE populations detected weak genetic structure among populations associated with the mountain ranges, notably the 12 populations located in the Xiaoxing'anling Mountains (1-12) were almost entirely dominated by a single cluster (Fig. S2B, K = 2).

ABC-based Inferences of Population Demography
In DIYABC analysis, based on the whole range STRUCTURE analysis, scenario 1, (the isolation with admixture model) was expected to be the most likely. However, the highest posterior probability was found for scenario 2 (the hierarchical split model 1), and the value (0.7341, 95% CI: 0.7079-0.7602) was much higher than those for scenario 1 (0.2557), 3 (0.0032) and 4 (0.0071) (Table 5A). In scenario 2, the median values of effective population size of N1 (Pop1), N2 (Pop2), N3 (Pop 3), and Nb (ancestral population) were 6570, 3700, 8700 and 1470, respectively (Table S3). Estimated median times of recent divergence (t1 for Pop 1 and Pop2) and ancient divergence (t2 for southern Pop3 and the others) were 681 and 2960 generations ago, respectively. Assuming the generation time of A. mono to be 30 years, the divergence times of t1 and t2 correspond to 20,430 and 88,800 years ago, respectively. The median time for a pronounced increase in population size (t3) was estimated at 6480 generations ago (194,400 years with a generation time of 30 years). However, the posterior distribution pattern suggested that this estimate was    (Table S4). This was corroborated by the PCA, which yielded a large cloud of data from the prior and observed datasets, centered around a small cluster from the posterior predictive distribution (Fig. S4).

Genetic Diversity of A. mono Populations
In Northeast China, although the forest zone is threatened, nSSR studies have shown that several tree species maintain moderate to strong genetic diversity, similar to that of A. mono (H E = 0.73860.058), e.g. Fraxinus mandshurica (H E = 0.74660.051) [14] and Juglans mandshurica (H E = 0.756) [63]. Significant positive correlations between genetic diversity (H E ) and latitude was detected over all the 56 populations (Table 3). However, as both A R and H E showed significant positive correlation with latitude in SC group (Table 3), the smaller population size and the greater geographical distances between populations in SC than in NE may have strongly contributed to this overall pattern. Thus, different trends were detected within NE populations and A R values significantly decreased with increasing latitude. Although it was not significant, H E also showed the similar negative correlation with latitude. As mentioned in the Introduction, the reductions in genetic diversity with increasing latitude observed in NE populations of Q. mongolica [19] and F. mandshurica [14] were also found in A. mono, indicating that it may be a common trend for tree species in this region. According to previous studies [64], the genetic diversity of most species declines from central to marginal populations. We therefore inferred a probable route for the expansion of A. mono, from the central section of China toward the north and south. However, the lower diversity, especially the low rare allelic richness in populations 1 and 53-56, may not be mainly due to margin effects. For population 1, extremely low winter temperature may contribute to low genetic diversity by hindering the growth of seedlings and young saplings [14], or even killing them, thereby reducing the population size and rare allelic richness. For populations 53-56, besides the patchy and scattered distribution of the species in their region, another factor contributing to lower rare allelic richness may be landscape fragmentation (one of the major factors causes of dramatic loss of biodiversity globally) by blocking gene flow among populations, altering the biogeographical environments and diminishing available habitats [65].
Although inbreeding will increase homozygosity in the short term and affect diversity in the longer term [66], neither H E nor A R in the 11 populations for which F IS values significantly deviated from 0 ( Table 2) were significantly lower than others. Thus, reduction of genetic diversity induced by inbreeding was not suggested in this study. Instead of inbreeding, demographic factor such as Wahlund effect would be better explanation to higher values of F IS detected in this study.

Genetic Structure and Demographic History of A. mono Populations
The F ST and G9 ST values indicated that differentiation was much lower among NE populations (F ST = 0.029; G9 ST = 0.114) than among SC populations (F ST = 0.201; G9 ST = 0.722). For the NE populations, the relatively short geographic distances between populations and the continuous distribution of mountains provide a more uniform landscape that probably facilitated gene exchange among populations [69], explaining the higher diversity and lower differentiation observed. The weak genetic structure and reductions in genetic diversity with latitude among the NE populations suggest that they expanded from a single lineage. Interestingly, similar observations (no significant substructure and latitudinal reduction of genetic diversity among mountain ranges in Northeast China) have also been reported for an ash species, F. mandshurica [14]. The patterns of both species could be explained by reduction of genetic diversity to the north due to expansion from a single refugium in the last ice age. STRUCTURE analysis focused solely on NE populations detected weak genetic structure among mountain ranges (Fig. S2B, K = 2). More specifically, the clustering pattern shown by the barplots indicated that the 12 populations located in the Xiaoxing'anling Mountains (1-12) were almost entirely dominated by a single cluster. Thus, the weak genetic structure in NE populations was probably due to the mountain ranges limiting gene flow, as they often do for tree species [70]. Since A. mono is an insect-pollinated species [71], and its large winged seeds are dispersed by wind, the potential for gene flow may be more sensitive to geographic barriers than that of wind-pollinated and wind-dispersed species such as birch (Betula maximowicziana), which was examined by Tsuda et al. [70]. In addition, the STRUCTURE analysis and NJ tree separated all populations into three groups (NE and SC populations, and those from adjacent regions). The analysis indicated that the two populations (49 and 50) located in northern China were admixed by STRUCTURE analysis when K = 2, but they appeared as an additional, independent genetic group in the results with K = 3 and 4. Since the isolation by admixture scenario was not supported by the ABC analysis, we propose that these two populations constitute an independent group. The demographic history of plant species associated with the LGM and species refugia in Northeastern China has been discussed [14,63], but no coalescent-based genetic analysis has been previously applied. Thus, the results of ABC analysis obtained in this study provide valuable foundations for addressing the demographic history of A. mono over a long time scale. The estimated timing of t3 (a median value of 6480 generations, corresponding to 194,400 years) was not robust (Fig. S3), but since the effective population size increased to 3700 (N2) from 1470 (Nb) at t3, the results clearly indicate that the population expanded rapidly at or around this time.
The ancient divergence of Pop3 at t2, may have been caused by the Taihang Mountains acting as a barrier. These Mountains have a north-south orientation (36-40uN, 112-115uE) and form a major natural boundary in eastern China. The southern Taihang Mountain marginal fault zone has been a major boundary of neotectonic deformation since the mid-late Pleistocene, dividing northern China latitudinally, and the south region containing the deformation system that includes the south Qinling Mountains [72]. This division might explain not only the distribution of cpDNA haplotypes of Juglans mandshurica in the south and north of the Taihang and Qinling Mountains (Fig. 1) [63], but also the divergence of Pop3. In further support of this hypothesis, maternally-inherited mtDNA patterns of Pinus tabulaeformis are reportedly divided into three geographical groups, separated latitudinally by the Yellow River and longitudinally by the Taihang Mountains [12]. Pop1 and Pop2 were separated by the Northeast Plain, the largest Plain in China with a total area of 350,000 km 2 (40u 259 to 48u 409N, 118u409 to 128uE), and their estimated divergence time was estimated at 20,430 a BP. The Northeast Plain was covered by coniferous forest/steppe during the early late Pleistocene and broad-leaved forest during the middle late Pleistocene, but at the end of the Pleistocene (42,000 , 20,300 a BP), because of the colder and drier climate, most of the Northeast Plain was covered by sand, and the vegetation changed to open forest/steppe [73]. Thus, Pop1 and 2 appear to have diverged during this period, when the species occupied refugia before the LGM. STRUCTURE analysis divided SC populations (Pop3) into four subgroups. The strong genetic structure in these populations was probably caused by a combination of the greater geographic distances separating them [16,67] and the presence of geographic barriers such as discontinuous mountain ridges [68][69] and large plains and valleys [25] (Fig. 1). These factors presumably isolated the populations and thus promoted population differentiation by limiting the potential for gene exchange. DIYABC analysis of the SC populations provided estimated divergence time (t1, t2 and t3)   of each subgroup (SC1, SC2, SC3 and SC4) but the estimate for t3 (189,900 years) was not robust (Fig. S5). SC2 and SC4 apparently diverged from SC1 at 88,800 a BP (t2), possibly due to a major uplift of the Qinling Mountains. These mountains form a major boundary between north and south China, ranging west-east (32-35uN, 103-113uE) and rose strongly during the Quaternary, especially the Late Quaternary (100,000 a BP), when they reportedly rose up to 5.4 mm per year [74][75]. Thus, if we accept t3 as accurate, we can assume that the Qinling Mountains rose several tens or hundreds meters during the t3 to t2 period, which could have been a major contributor to the divergence of SC1 and SC2.
Implications for Genetic Breeding, Conservation, and Silviculture of A. mono A shelter wood system of forest management has been found to have little or no impact on genetic diversity and mating in Fagus sylvatica [76]. However, extensive logging can significantly alter the level and distribution of genetic variation in Acer saccharum [77]. The cited studies indicate that silvicultural activities may affect genetic diversity, but in ways that vary according to species and management type.
Natural resources of A. mono in China are under increasing pressure due to its use as a source of wood and various other materials. Hence, specific natural varieties are being selected for artificial planting and silviculture in Northeast China and elsewhere. We only used neutral genetic markers in the present study, but the information obtained is potentially highly informative for the breeding, conservation and silviculture of A. mono, especially for the designation of conservation units. Since clear genetic differentiation was found between NE and SC populations, these two groups should be treated as independent conservation units. Although admixture was found for Populations 49 and 50 in the STRUCTURE analysis when K = 2, the ABC-based analysis suggested that these two populations were not generated by the admixture of NE and SC populations. Thus, populations in this region of northern China located between NE and SC populations should also be treated as an independent conservation unit. Seeds and nursery stocks of local provenance should be used in reforestation efforts in each area to avoid nonlocal introductions and genetic homogenization [11,78]. Moreover, since strong genetic structure was detected in SC populations, further local-scale studies in this region are needed to develop appropriate strategies for conservation and the designation of natural reserves. In addition, ex situ conservation efforts, which could involve collecting seeds from populations in this area, may be useful. Since Populations 51 and 52 showed high levels of diversity and more private alleles than other populations, natural reserve areas should be established in these regions. In contrast, the NE populations maintain relatively high levels of genetic diversity and densities of individuals within populations, but low genetic differentiation among populations, implying that the species should be able to recover without assistance (if further disturbance is avoided) in this region. Selection of elite trees and the development of varieties for specific purposes are likely to be more important than provenance selection for conserving A. mono in Northeast China. To obtain the thorough overview required to develop an effective conservation strategy for this species further studies are required, including evaluation of the extent of its adaptive genetic variation.