MICROCHECKER found no evidence of scoring errors, but some samples were detected to have null alleles. We failed to amply these alleles despite two to three more genotyping attempts. Mutations in the flanking region may prevent the primer from annealing to template DNA during amplification of microsatellite loci by PCR [28], but we still kept these loci for further analyses because the frequency was relatively small (<5%). All populations deviated significantly from Hardy-Weinberg equilibrium (p < 0.05), with the observed heterozygosity being lower than expected (mean observed 0.031, range 0.000–0.205 vs. expected 0.426, range 0.018–0.797).

All nSSR loci were polymorphic in all populations. The mean allele richness (A_R) and Shannon’s information index (I) were 1.9 (1–3.1) and 0.793 (0.034–1.744), respectively. The fixation index was high (mean 0.913, range 0.202–1). The outcrossing rate was low (mean 8.1%, range 0%–66.4%), but three populations (SY, J2, and K2) showed atypically high outcrossing rate (>48%). The region of MDRY had the highest genetic diversity (A_R = 14.0, and I = 2.349). Observed heterozygosity (H_O = 0.063) and expected heterozygosity (H_E = 0.881) of this region were also higher than other regions (Table 1).

All cpSSR loci showed relatively low diversity, the mean allele richness (A_R) and Shannon’s information index (I) for cpSSRs were 1.6 (1–3.2) and 0.793 (0–0.908). CpSSRs indicated that Korea has the highest allelic richness (A_R = 4.6) and Shannon’s information index (I = 0.932) among all regions.

For nSSRs, CONTRIB revealed no difference in regional contribution to total diversity. However, for allelic richness, the highest contribution was made by the SWC region, followed by the regions MDRY and NEC, mainly due to their high own diversity. The lowest contributions came from regions NWC and Japan. For cpSSRs, the SWC region made the greatest contribution to total diversity and allelic richness due to both diversity and differentiation. Besides, the regions of NEC, Korea and Japan made high contributions to allelic richness due to differentiation (Figure 2).

MCMC structure reconstruction of nSSRs showed moderate genetic structure. When Evanno’s [29] ad hoc estimator of the actual number of clusters was used, ΔK indicated modes at K = 2 (Figure 3a). The average percentages of membership for eight geographical regions of individuals in each of the two clusters were calculated. Most samples (>66%) of the group SC, MDRY, SWC, NEC, Japan and Korea were assigned to cluster 1, and most individuals of group NEC (79.4%) and NC (86.5%) to cluster 2 (Table 2, Figure 3b). No geographic structure was detected for cpSSRs. The UPGMA dendrogram of both nSSRs and cpSSRs divided the eight regions into the same two geographical clusters (Figure 4).

Analysis of nSSRs by AMOVA revealed that 6.0% of genetic variation was due to the genetic distance between the two clusters, 46.7% among populations within clusters and 47.3% between individuals within populations. Similar results were obtained from cpSSRs (6.8%, 57.0% and 36.25%, respectively) (Table 3). A mantel test indicated a significant isolation by distance for cpSSRs (r² = 0.021, p = 0.002), but not for nSSRs (r² = 0.004, p = 0.074).

The allele size permutation test rendered non-significant differences between F_ST and R_ST estimates (p = 0.004 for nSSRs and p = 0.01 for cpSSRs; 10,000 iterations), indicating R_ST estimates were more appropriate than R_ST for our data. We found high population genetic differentiation (R_ST) (cpSSRs: 0.499 and nSSRs: 0.622). For cpSSRs, the overall level of inferred gene flow (Nm) was 0.502 individuals per generation among the populations; and for nSSRs, the gene flow (Nm) was 0.251.

Standardized differences test and Wilcoxon sign-rank test based on both SMM and TPM model showed recent reduction in seven populations: Chengkou (CK), Wuchang (WC), Tongbai (TB), Keshan (KS), Jizhou (JZ), Japan1 (J1) and Korea5 (K5). A recent bottleneck effect was also detected in three additional populations of Wuqing (WQ), Jiaohe (JH) and Japan 5 (J5) using TPM model by the Wilcoxon sign-rank test (Table 4). The mode-shift test in allele frequency attributed L-shaped distribution to all populations, which was consistent with normal frequency class distribution ranges (p > 0.05).

The genetic diversity of wild soybean was studied previously using SSRs [21,25,26,30,31]. However, this is the first time a study uses both nuclear and plastid SSRs to analyze the extent and structure of genetic variation across the whole species range. Wild soybean showed a relatively high population diversity (H_E = 0.426), which is similar to the result from previous studies [31,32], Considering life form and breeding system have a highly significant influence on genetic diversity [33], we compared genetic diversity of wild soybean with other predominantly self-pollinated wild species, such as wild emmer (Triticum turgidum ssp. dicoccoides) (H_E = 0.19) [34], wild barley (Hordeum spontaneum) (H_E = 0.138) [35], and officinal wild rice (Oryza officinalis) (H_E = 0.22) [36]. This may be caused by the special seed dispersion of the wild soybean, the pod dehiscence could discharge the mature seeds to a distance of 0–5 m (up to 6.5 m) [22]. High outcrossing rate for certain populations maybe another reason for high genetic diversity in wild soybean.

The nSSRs showed that MDRY region has the highest diversity, which is consistent with several previous studies. For example, Shimamoto et al. [13] reported the highest diversity in the Yangtze River region using RFLP markers. Southern China (including regions of MDRY, SWC and SC) was proposed as the wild soybean center of genetic diversity in a study by Wen et al. [37] using a combination of SSRs and morphological traits. The same region was also pointed as origin and center of diversity using SSR markers and nucleotide sequences in a study by Guo et al. [25]. Compared with previous results, our study applied more detailed regional division, and the center of diversity was similar, but not as obviously different than in previous studies.

Compared with nSSRs (A_R = 1.9; I = 1.794), the cpSSRs showed less diversity (A_R = 1.6; I = 0.932), which is congruent with Powell et al. [38], who used both nSSRs and cpSSRs of wild soybean samples from a germplasm bank. Similar results have been observed in other studies using both types of SSR markers in other species [39–41]. This is consistent with low substitution rate of plant chloroplast cpDNA sequences compared with nDNA [42]. The cpSSRs could offer unique insights into ecological and evolutionary processes in wild plant species in some situation [43], Differing from that of nSSRs, cpSSRs revealed that Korea has the highest wild soybean genetic variation.

Breeding system, life form, effective population size, genetic drift and gene flow are the major evolutionary effects on population genetic structure, with the effect of breeding system being the predominant one [44,45]. Populations of self-fertilizing species are expected to have lower allelic diversity, lower levels of heterozygosity, and high differentiation among populations than populations from outbreeding species [45]. Here, both nSSRs and cpSSRs showed high inter-population genetic differentiation and low gene flow, as expected in the predominantly selfing wild soybean, combined with low seed and pollen dispersal ability. The seed dispersal distance of wild soybean is short, and 95%, 99%, and 99.9% of the produced seeds disperse within 3.5, 5.0, and 6.5 m, respectively after natural pod dehiscence [22], and nearly 81.4% of the loci were found to be positively correlated in the first two distance classes (0–10 m) [6]. Low pollen dispersal ability can be surmised from the estimates of outcrossing rate in wild soybean, which varied from 2.3% (range 2.4%–3.0%) [46] to 13% (range 9.3%–19%) [47] using allozymes and 3.4% (range 0%–37.4%) applying nSSRs [21]. We found a higher mean outcrossing rate (8.1%), with extremely high values in some populations (G5_YT: 21.1%; G6_SY: 66.4%; G7_J2:52.7%; G8_K2: 48.7%). Despite high selfing rates, occasional outcrossing rate can be subsequent. Occasional high outcrossing was detected in other predominant self-pollinated species such as wild barley (t = 25.1%) [48]. The high outcrossing rate in some populations of a predominantly selfing species can be a consequence of rare or sporadically occurring specific environmental conditions (temperature, humidity, wind, insect pollination, etc.) [48]. In this study, the populations with high outcrossing rate were found in different habitats from all eight eco-regions, and could not be ascribed to a particular abiotic environmental factor, which can suggest an importance of some biotic factor such as high pollinator visiting activity [6], more studies should be carried out to fully resolve this issue.

The UPGMA and Neighbor-joining dendrogram based on Nei’s genetic distance and assignment test revealed two clusters of wild soybean in both nSSRs and cpSSRs. One cluster was formed by the NC and NWC regions, and the other one was formed by six geographic regions including NEC, SWC, SC, MDRY, Korea and Japan. The absence of differentiation among East China, Southern Japan and the Korean Peninsula (CJK region) is surprising. Fluctuations in sea level among the CJK region throughout the Quaternary (or even in the mid-late Tertiary) provided abundant opportunities for population fragmentation and allopatric speciation at the CJK region. Applying nDNA and cpDNA sequences, the previous phylogeographic studies on Croomia japonica [49], Kirengeshoma palmata [50], and Platycrater arguta [51] suggested deep allopatric-vicariant differentiation of disjunct lineages in the CJK region [52]. Wild soybean might have seen a continuous distribution throughout the CJK region through the exposed East China Sea (ECS) basin when the sea level fell by 85-130/140 meters during Last Glacial Maximum (LGM; 24,000–18,000 years before present) [53,54], the disjunct distribution among this region formed following the submergence of ECS land bridge, and there may be insufficient time for lineage sorting and differentiation. Wild soybean has salt resistance [55], and could grow easily in the salty conditions of a sea shore hence they have more chance to migrate along the land bridge among the CJK regions during glacial periods. We could not totally exclude the possibility of exchange of wild soybean among the CJK region via long distance dispersal due to disappear of the ECS land bridge. However, it is just a speculation and will need further studies.

In this study, a bottleneck effect was detected in seven populations: Chengkou (CK), Wuchang (WC), Tongbai (TB), Keshan (KS), Jizhou (JZ), Japan 1 (J1) and Korea 5 (K5). The CK and JZ populations are from undisturbed habitats with very small population sizes, while another five populations are situated in disturbed habitats: populations WC, J1 and K5 are along roadsides; population AF is beside an abandoned railway; populations KS and TB are along the ridge of some fields. Population KS is a relic from a larger population predating farming reclamation, and only limited individuals are left. In brief, the five populations were significantly affected by anthropological activities. Wild soybean can adapt to a wide variety of habitats with adequate water. However population size of wild soybean will rapidly decrease in the habitat, with subsequent degradation of genetic diversity and allelic richness. Conversation of wild soybean is therefore a priority, and should focus on regions already affected genetically.

When selecting conservation sites one must also consider a population’s contribution to total diversity and allelic richness. The SWC region was inferred to have greatest contribution to total diversity and allelic richness with both nSSRs and cpSSRs. Wild soybean in this region shows an unusually small population size, combined with a fragmented distribution: several populations from Ninglang county of Yunnan province and Chayu county of Xizang province are separated from the main populations by as much as 400 km. Furthermore, both previous ex situ and in situ conservation initiatives have paid little attention to this region, and only dozens (from a total of 6172) of wild soybean seed accessions from this region have been collected and stored in the Chinese Crop Germplasm Resources databank (http://icgr.caas.net.cn/cgrisintroduction.html). This region deserves high conservation priority.

A total of 604 wild soybean individual leaf samples were obtained from 43 populations across most of the species distribution (Figure 1). Five populations represented two countries, Korea and Japan, and 5 to 6 populations represented each of six regions of China (Table 5). Total genomic DNA was extracted from silica gel-dried leaves using the CTAB method of Doyle and Doyle [56]. The extracted DNA was resuspended in 0.1× TE buffer (10 mmol/L Tris-HCl, PH 8.0, 1 mmol/L EDTA) to a final concentration of 50–100 ng/μL.

Genotyping was performed using 20 nSSRs representing all 20 wild soybean linkage groups corresponding to the 20 chromosomes, and five cpSSRs from intergenic regions. All the 25 loci are polymorphic and have been used in previous studies [15,21,57] (Table 6). PCR reactions were performed in 15 μL reactions containing 30–50 ng genomic DNA, 0.6 μM of each primer, 7.5 μL 2× Taq PCR MasterMix (Tiangen Biotech, Beijing, China). PCR amplifications were conducted under the following conditions: 94 °C for 2 min; 35 cycles at 94 °C for 30 s, 50 °C for 40 s, and 72 °C for 1 min; followed by a final extension step at 72 °C for 7 min. Primers are shown in Table 6. All the SSR markers were polymorphic based on electrophoresis performed on an ABI 3730 DNA sequencer (Applied Biosystems, Foster City, CA, USA). Fragment length sizes were scored automatically using the program GeneMapper (Applied Biosystems).

Microsatellite data from each population was tested for amplification errors and null alleles, large allele dropout or stuttering using 1000 randomizations in MICROCHECKER v.2.2.3 [58]. Genepop v. 3.4 online [59] was used to check for deviation from Hardy-Weinberg expectations and between loci in each population using exact tests with 10,000 dememorizations, 100 batches and 1000 iterations. Significance level was adjusted using the sequential Bonferroni correction for multiple comparisons [60]. For nSSRs, the number of alleles per locus (A), the numbers of different alleles (N_a), the observed heterozygosities (H_O), expected heterozygosities (H_E), fixation index (F_IS) and Shannon’s information index (I) were calculated using GenALEx v. 6.4 [61]; allelic richness (A_R) was calculated by FSTAT 2.9.3.2 [62]; outcrossing rate (t) was calculated from the fixation index using the equation t = (1 − F_IS)/(1 + F_IS) [63]. For cpSSR, the number of alleles per locus (A), the numbers of different alleles (Na) and Shannon’s information index (I) were calculated using GenALEx v. 6.4 [61]. In each individual, genetic variants at all cpSSR and nSSR sites were combined into haplotypes. Then, each region was characterized for its plastid DNA diversity using the number of haplotypes detected and gene diversity estimated using the program CONTRIB [64]. Contribution of each region to total diversity (CT) and to total allelic richness (CTR) were calculated according to Petit et al. [65].

Genetic differentiation was investigated using the model based clustering method STRUCTURE 2.1 [66,67] for nSSRs. Burn-in time and replication number were set to 100,000 and 100,000 (further generation following the burn in) for each run, respectively. The number of populations (K) in the model was systematically varied from 1 to 10. In order to decrease the margin of error, the average value of 20 simulations performed for each K was used. We used the ΔK method [29] representing the highest median likelihood values to assign wild soybean accessions using the online tool Structure Harvester [68]. For the chosen K value, the run that had the highest likelihood estimate was adopted to assign individuals to clusters. The 10 runs with the lowest DI values for the selected K-value were retained, and their admixture estimates were averaged using CLUMPP v. 1.1.1 [69], applying the greedy algorithm with random input order and 1,000 permutations to align the runs and calculate G’ statistics. Results were visualized using DISTRUCT 1.1 [70].

Nei’s genetic distance (D) and Goldstein’s distance [(δμ)²] are commonly used for microsatellite. Considered Goldstein’s distance (δμ)² showed bias at small sample sizes and the bias was directly related to the number of alleles and range in allele size [71], a dendrogram based on Nei’s (1978) [72] genetic distance (D) between groups was constructed using the UPGMA method implemented in the PHYLIP v. 3.68 [73]. In order to make sure the results of UPGMA method, a neighbor joining tree also was constructed using PHYLIP v. 3.68. A hierarchical analysis of molecular variance (AMOVA) [74] implemented in Arlequin v. 3.11 [75] was used to partition the observed genetic variation into among clusters, among populations within a cluster and among individuals within a population.

Two commonly estimators of population differentiation are F_ST, based on allele identity, and R_ST, which incorporates microsatellite-specific mutation models. We used the allele size permutation test in SPAGeDI [76] to test whether allele sizes were informative in wild soybean microsatellite data set, which would indicate that mutatioin has contributed to differentiation. Because R_ST was shown to be most appropriate for our data, (see results), the global R_ST across all samples was calculated in Arlequin v. 3.11 [75]. Gene flow were quantified using the approach of transform estimates of R_ST into indirect estimates of the average number of migrants exchanged per generation among populations (Nm) [77]. Gene differentiation (R_ST) was calculated using AMOVA analyses based on population levels, gene flow was estimated from R_ST (nSSRs: Nm = 0.25(1 − R_ST)/R_ST; cpSSRs: Nm = 0.5(1 − R_ST)/R_ST).

We assessed demographic history based on microsatellite data using different and complementary methods. Heterozygosity excess test [78] and mode-shift test [79] from BOTTLENECK 1.2.02 [80] were used to detect the recent population bottleneck. This program conducts tests for recent (within the past 2N_e to 4N_e generations) population bottlenecks that severely reduce effective population size (N_e) and produce an excess in heterozygosity. Heterozygosity excess test was performed under two mutation models: stepwise mutation model (SMM) and two-phase mutation model (TPM). The model of TPM include both 95% single-step mutations and 5% multiple-step mutations, as recommended by Piry [80]. Heterozygosity excess was detected using the one-tailed Wilcoxon sigh-rank test and standardized differences test on 20 nSSR loci [80]. Significance was determined also by the standardized differences and Wilcoxon tests. Mode-shift test detects allele frequency to investigate whether allele frequency distort from the expected L-shaped distribution. During a bottleneck, the loss of rare alleles occurs more rapidly than the associated decrease in expected heterozygosity, as rare alleles do not contribute to H_E as much as common alleles, and thus distort the allele frequency distribution from its expected L-shaped distribution [78].