The PGRC wild barley collection currently maintains 3,782 accessions representing 16 countries from Morocco in the western region to Japan in the eastern region (Table S1). It was established since the 1970s, largely by several collection expeditions in the Near East and further west to Morocco during the 1980s–90s by the Canadian scientists Drs. Sati Jana, Bernard Baum, John Martens and George Fedak (e.g., see [33]), duplicating part of the National Small Grain Collection USDA-ARS, United States [21], and continued minor acquisitions, particularly from China and Ethiopia. In spite of these efforts, the collection has only two accessions from the Tibetan Plateau and is completely lacking representation from several other countries such as Libya, Russia, Pakistan, Azerbaijan and Uzbekistan. Based on the existing collection, a core subset of 269 accessions representing 16 countries were developed through a random sampling stratified with respect to country of origin and with the size roughly equal to the natural logarithm frequency of accessions for a country [42]. The detailed information on the selected accessions, including geographic records for 49 accessions, is given in supplemental materials Table S1. To infer regional structure, the accessions were grouped based on the distinct distribution of wild barley described in [43,44]. The eastern region represents wild barley from the Zagros Mountains and further east, while the western region includes wild barley from the Fertile Crescent and further west.

About 10 seeds were randomly selected from each accession and grown in the greenhouse at the Saskatoon Research Centre. Leaf tissue from individual two week old greenhouse grown seedlings was collected and freeze dried for 24 h. Genomic DNA was extracted from leaf tissue of one individual seedling randomly selected from an accession using a DNEasy Plant Mini kit (Qiagen, Mississauga, ON, Canada), quantified using a Thermo Scientific Nanodrop 8,000 spectrometer, and adjusted to a final concentration of 15 ng·µL⁻¹ with water.

Twenty-five informative barley SSR primers were selected based on the published marker information and genomic coverage from a consensus map for cultivated barley (Table 1; [45]) and these sequence specific primers were synthesized by IDT (Coralville, Iowa, U.S.A.). A three primer PCR method [46] was applied: a unique reverse primer, a unique forward primer with a T7 universal tail (T7 = TAATACGACTCACTATAGGG), and a third fluorescently labelled T7 primer containing one of four tags: FAM, VIC, PET, or NED. Fluorescently labelled primers were synthesized by Applied Biosystems-Life Technologies (Burlington, Ontario, Canada). Primers were re-suspended in water at 50 pmol·μL⁻¹. The PCR reactions were set up as: 1x New England Biolabs (NEB, Pickering, Ontario, Canada) Standard Buffer containing 1.5 mM MgCl₂, 0.1 mM each dNTP (Promega/Fisher Scientific) Nepean, Ontario, Canada), 188 fmol·μL⁻¹ reverse primer, 19 fmol·μL⁻¹ forward primer with T7 universal tail, 63 fmol μL⁻¹ fluorescently labelled T7 universal primer, 1 U Taq polymerase (NEB), and 30 ng template genomic DNA in a final volume of 20 µL. Samples were denatured at 94 °C, 3 min, selectively amplified with 22 cycles of 94 °C 30 s, 56 °C 30 s, and 72 °C 45 s, and then followed immediately in the same tube by fluorescent labelling with 22 additional cycles of 94 °C 10 s, 47 °C 20 s, and 72 °C 60 s with a final extension of 72 °C for 10 min.

Samples were diluted by combining 1 µL from each of four different reactions labelled with four different fluorescent labels and adjusting to 50 µL with water. Size standard was prepared by adding 24 µL of GeneScan 600 LIZ size standard to 890 µL of Hi-Di formamide (Applied Biosystems-Life Technologies, Burlington, Ontario, Canada). One microliter of diluted samples was added to 9 µL of size standard-formamide mixture. Samples were analyzed on an Applied Biosystems 3130xl genetic analyzer using POP-7 polymer (Applied Biosystems-Life Technologies, Burlington, ON, Canada) and a 36 cm capillary array. Data was collected using GeneMapper Software Version 4.1 with bin size set at 2bp. To minimize technique-born and scoring errors, six checks were repeated on each of three plates. Resulting data was manually proofread.

The scored alleles were first assessed for consistency of the checks and alleles of a primer pair with 5% or higher error rates over the three plates were discarded. The SSR data were analyzed for the level of polymorphism with respect to each primer, the accession country of origin, and the proposed accession region by counting the number of alleles, unique alleles, and rare alleles (with a frequency of 0.05 or smaller), generating the summary statistics on the allelic frequencies, and calculating the polymorphic information content (PIC) following the method described in [47]. This was done using a SAS program written in SAS IML [48]. To visualize the variation pattern, the numbers of alleles detected by all primer pairs were plotted against their frequencies of occurrence in all assayed accessions. To assess the impact of the variable number of accessions for each country on the polymorphism observed, a linear regression was made using SAS PROC REG [48] on the number of accessions (as independent variable) over each of the five independent variables: the number of alleles, the number of unique alleles, the number of rare alleles, the mean allelic frequency, and the group-specific proportion of variation (or Fst) obtained from the below analysis of molecular variance (AMOVA; [49]). The differences in allelic richness among accessions of pairwise countries or two regions were further assessed following the random permutation procedure described in [50]. The random permutation procedure allows for a significance test of the difference in allelic counts between groups of variable accession numbers.

The genetic structure of the assayed accessions was first inferred using prior defined grouping with respect to accession origin: country and region. The inferences were specifically done by estimating genetic distances among groups and clustering the prior defined groups using Arlequin version 3.5 [51]. The Arlequin program provides a partition of total SSR variation into within- and among- group variation components, and a measure of inter-group genetic distances as the proportion of total SSR variation residing between wild barley accessions of any two groups [49]. Significance of resulting variance components and inter-group genetic distances was tested with 10,100 random permutations. The neighbor joining clustering of the prior defined groups was made using NTSYS-PC 2.01 [52] that was based on the AMOVA-based estimates of genetic distances with respect to country origin. As the structural inference may be biased from variable accession numbers for these countries, an additional AMOVA was also made for countries with more than six accessions.

The genetic structure was also inferred using model-based Bayesian methods. The Bayesian method available in the BAPS software [53] was applied to estimate the number of clusters. Clustering of individual accessions was done using the model for non-linked markers and 20 replicate runs of the algorithm with the upper-bound values (K) for the number of clusters ranging between 2 and 10. For further verification, the other Bayesian method available in the program STRUCTURE version 2.2.3 [54,55] was used to detect population structure and to assign accessions to subpopulations. The STRUCTURE program was run 30 times for each subpopulation (K) value, ranging from 2 to 10, using the admixture model with 10,000 replicates for burn-in and 10,000 replicates during analysis. The final population subgroups were determined based on (1) likelihood plot of these models, (2) the change in the second derivative (∆K) of the relationship between K and the log-likelihood [56], and (3) stability of grouping patterns across 30 runs. For a given K with 30 runs, the run with the highest likelihood value was selected to assign the posterior membership coefficients to each accession. A graphical bar plot was then generated with the posterior membership coefficients.

The optimal genetic structure obtained from the BAPS program was further analyzed with respect to accession country and region to characterize its feature. The genetic divergence among inferred clusters was assessed based on Nei’s minimum distance generated from BAPS. The inferred genetic structure was further compared for consistency with the genetic relationships of individual accessions obtained from two commonly applied approaches. A principal component analysis (PCA) of the 269 accessions was performed using NTSYS-PC 2.01 [52] based on the similarity matrix of 359 SSR alleles, and plots of the first three resulting principal components were made to assess the accession associations. A neighbor-joining (NJ) analysis of all 269 accessions was also made using PAUP* [57] based on the original data of 359 SSR alleles and a radiation tree was displayed using MEGA 3.01 [58]. The resulting PCA plots and NJ trees were individually labeled for the inferred structures.

Additional AMOVA analyses were also made based on the optimal genetic structure derived from the two Bayesian analyses above. The average dissimilarity of each accession was estimated following the method described in [59]. This average dissimilarity measures the overall genetic difference between an accession of interest and the remaining 268 accessions assayed, reflecting its genetic distinctiveness within the core subset.

The assayed SSR primers detected a total of 359 alleles at 25 loci in the 269 wild barley accessions (Table 1). These alleles sampled both transcribed and non-transcribed regions of seven barley chromosomes. The number of loci per chromosome ranged from 2 to 5. The number of alleles detected per locus ranged from 4 to 35, and the number of rare alleles (with a frequency of 0.05 or smaller) per locus ranged from 0 to 25. The average allele frequency for all alleles at a locus ranged from 0.029 to 0.253. Overall, the observed allelic frequencies ranged from 0.004 to 0.708 with an average of 0.072. Specifically, 212 of the 359 alleles were rare alleles; 290 alleles had a frequency of 0.1 or smaller; and 20 alleles had a frequency of 0.3 or larger. The PIC value for each marker ranged from 0.488 to 0.947 and averaged 0.794 (Table 1).

The core subset was selected using a stratified sampling strategy from the wild barley collection with respect to country of origin. The number of accessions per country ranged from one to 36 (Table 2). A variable number of accessions per country were significantly associated with the variation in the estimates of five parameters per country (Table 2): the number of alleles (R² = 0.72, p < 0.0001), the number of unique alleles (R² = 0.63, p < 0.0004), the number of rare alleles (R² = 0.66, p < 0.0002), the mean allelic frequency (R² = 0.57, p < 0.001), and the country-specific Fst value (R² = 0.41, p < 0.01). These associations clearly demonstrated the effectiveness of the stratified sampling in capturing country-wise SSR variability from the PGRC wild barley collection. The detailed diversity estimates for the core subset with respect to country are shown in Table 2.

To compare genetic diversity among countries of variable samples sizes, the allelic differences among wild barley accessions of pairwise countries were estimated and tested using a random permutation procedure. Significant allelic differences were found largely among accessions from countries with more than six accessions (Table 3). The accessions from Lebanon and Morocco had the most significant allelic differences from the others, and the accessions from the countries in the eastern region had the least significant allelic differences. The largest significant allelic differences were found between accessions from Lebanon and Israel or Jordon. The country-specific Fst estimates the country-specific contribution to the overall SSR variation and provides some measure of the within-country variation. When the countries with more than six accessions were considered (see Table 2), Israel and Jordan had accessions with the smallest Fst value (0.231 or the most within-country variation), followed by Syria (0.236), Turkey (0.237), and Iran (0.237). The country with the largest Fst value (or the least within-country variation) was Lebanon (0.274), followed by Morocco (0.257).

Similarly, SSR variation was also estimated with respect to region (Table 2). The numbers of accessions assayed for the western and eastern regions were 216 and 53, respectively. More alleles, including unique and rare alleles, were observed for the accessions from the western region than from the eastern region and such allelic difference was statistically significant (p < 0.0001), based on the random permutation test. However, the mean allelic frequency (0.075) observed for the western region was slightly lower than that (0.123) for the eastern region. Similarly, the accessions from the western region had slightly more SSR variation with the region-specific Fst value of 0.0741 than that from the eastern region (0.0752).

The average dissimilarity (AD) of each accession was estimated. The AD estimates ranged from 0.103 to 0.141 with a mean of 0.114 (Table S1). The accession with the least AD estimate was from Cyprus (CN48879), while the three accessions from Israel (CN80241, CN79043 and CN78947) displayed the largest AD estimates. Overall, more individual accessions from Israel and Jordan displayed distinct genetic backgrounds.

The dominant genetic structure detected in the core set was at the country level, followed by the proposed region (Table 4, Figure 1). The percent of total SSR variation residing among the accessions of various countries was 24.23, followed by the regions (7.43) (Table 4). Excluding the countries with less than seven accessions did not significantly reduce the proportion of the total SSR variation (24.16%) present among the assayed countries. Clustering 268 accessions from 15 countries revealed no distinct separation between accessions representing two regions and genetically distinct clusters from Lebanon, Greece and Cyprus (Figure 1A). These patterns of variation were consistent with those PCA-inferred genetic relationships of 269 accessions (Figure 2A). The eastern accessions were not clearly separated from those western accessions. The accessions from Lebanon and Greece were genetically most differentiated and separately from those from the western region. Similarly, the neighbor-joining analysis of individual accessions revealed the same patterns of variation as those shown in the PCA plot (Figure 2B). Several clusters were observed, but they were not distantly diverged. Interestingly, seven accessions from Japan were not closely associated with those accessions from the eastern region, rather with some accessions from the western region. Two accessions from China (CN28546 collected from Tibet; CN28547 from Daofu, Sichuan) were clustered together and associated with those from the western region.

However, when the countries with less than seven accessions were not considered, the accessions from four countries in the eastern region (Afghanistan, Japan, Iran and Turkmenistan) were closely grouped together and genetically more related to the accessions from Turkey and Syria (Figure 1B). These results not only confirmed the expected effects of variable sample sizes on the structural inferences at the country or regional level, but also provided some support for the distinction in wild barley accessions between the eastern and western regions [43,44]. Also, the accessions from Lebanon and Israel displayed the most genetic differentiations (Figure 1B). This pattern of variation is consistent with the findings of allelic differences reported above and our current understanding of wild barley genetic diversity mentioned earlier.

The BAPS analysis revealed five optimal clusters in this core subset with the highest log likelihood of −19,454.3 (Figure 3) and a large partition (20.3%) of the total SSR variation (Table 4). Nine (3.3%) accessions had multiple memberships across the five clusters and largely resided between clusters 2 and 4 (Figure 3A). The clusters ranged in size from 15 to 104. The cluster composition varied with respect to accession country and region. The accessions from Lebanon and Greece contributed to clusters 3 and 1. The cluster 4 consisted mainly of the accessions from Cyprus and Syria. The cluster 2 represents the western accessions and the cluster 5 dominantly consisted of the eastern accessions, while both clusters 2 and 5 had accessions mixed from both regions. Clearly the clusters 3 and 1 displayed substantial divergence from the other clusters (Figure 3B). The detailed memberships of each cluster for the assayed accessions were given in Table S1.

The STRUCTURE analysis was also applied to infer genetic structure in the core subset, considered K = 2 to 10 clusters, and revealed four optimal clusters (Figure 4A), with the highest log likelihood of −16,724.3 (Figure 4B) and a large partition (16.3%) of the total SSR variation (Table 4). The optimal clusters were supported from the rate of change in the second derivative of the log likelihoods over various Ks analyzed, giving K = 4 as an optimal (Figure 4C). Clearly, each of the four optimal clusters has a considerable proportion of mixed memberships sharing between clusters.