A total of 60 barley genotypes archived at the Agricultural Research Center (ARC), Egypt, were selected to represent germplasm in Egypt (annually provided by ICARDA and ACSAD). The accessions divided according to their respective subspecies country of origin resulting and grain type-hulled and naked; 30 lines from each type (Table S1). The naked types were provided by The Egyptian/European Project on Sustainable Barley Production under Rain-fed in Egypt.

The collection was evaluated in field trials at three saline affected soil locations (Sakha; N31.3°, E30.9°, Elserw; N31.4°, E31.8°, and Elhosinia; N31°, E32.1° northwest, Egypt (Fig. 1) in 2012 and 2014). The genotypes (G) have been divided into four groups “sets”; 15 (G) per set. Each year (Y) and location (Loc.) was treated as a separate environment (E), 3Loc. × 2Y (E1 to E6). RCBD design with three replications (Rep.) was carried out for each set comprised of 12 experiments in total (4-exp./loc.) Each plot had four rows; it was 2 m long/30 cm among rows. Data were collected on: Days to Heading (DH) and Maturity (DM), Flag Leaf Area cm² (LA), Chlorophyll content (Chl as SPAD value), Plant Height (PH), Spike Length (SL), Peduncle length (Ped), Number of Tillers/ m² (NT), Number of Grains/spike (NG), thousand-Grain weigh (GW), Grain Yield GY t.h^–1, Biological Yield BY t.h^–1, Grain Filling Period (GFP), Grain Filling Rate (GFR) and Harvest Index (HI). The Mean, Mini/Maxi values, Standard Deviation (SD), Variance (S²), Coefficient of Variation (CV), heritability/broad sense (h_b) and R² coefficient of determination were calculated, and the genetic correlations over environment means were estimated between all pair traits. ANOVA was performed using the SAS statistical package for each set [14].

At the harvesting stage, the samples were taken from upper soil layer (0–30 cm) to carry out the chemical and physical analysis (Table S2).

A set of 32 SSR and 8 SNP specific markers were selected based on their distribution across the seven barley chromosomes and their response to abiotic stress [7,8] from Grain Genes database (Table 1). DNA was extracted and isolated from Leaves tissues (2-week-old) using CTAB method according to [15]. PCR was carried out in 10 μl reaction vessels containing 10 ng of genomic DNA, 0.3 μl/10 ng of each primer and 5 μl of Go-Taq (Promega, Japan). The amplification was for one cycle at 94 °C for 3 min, 30 cycles (1 min/94 °C, 1 min/55 °C and 1 min/72 °C finally 72 °C/5 min then at 4 °C). PCR products were separated on 2% agarose gel in a 0.1 × TBE buffer. The bands were scored using GelAnalyze software [16].

Number of allele, gene diversity and polymorphism information content (PIC) for each locus were calculated for markers using Power Marker 3.25 [17]. The genetic structure was analyzed by hierarchical AMOVA implemented in the software of GeneAlEx6.5 [18].

The population structure implied using the Bayesian model-based software program STRUCTURE was used to determine the number of clusters K = sub-populations [19]. We performed 10 runs; each run was a burn in length of 100,000 steps [20]. The PAST [21] and XLSTAT pieces of software were used to carry out UPGMA and PCA analyses, based on the Nei genetic distance matrix through binary data.

The association between the phenotypes and the markers was evaluated with a general linear model (GLM) in TASSEL.5 [11] during and over seasons. The phenotypic allele effect of SSR/SNP which was associated with 15 associated traits was estimated through comparison between the average phenotypic values over accessions with the specific allele. LD, r² values were calculated for each chromosome in the two sub-populations and plotted against genetic distance (cM) and (D’), [22] using TASSEL and PowerMarker pieces of software. The correlation of allele frequencies r² and standardized disequilibrium coefficient (D’) values estimated between all pair of loci were plotted against the genetic distance in cM to determine the map distance across them.

The data related to 15 agronomic-traits of the 60 genotypes is summarized in Table S3. Over the six environments, GY varied from 1.16 to 2.64 t.h^–1 and the mean GY was 2.02 ± 0.31 t.h^–1; E6 recorded the highest mean values for most traits, compared to the others. The CV% concerning yield and agronomic traits (Table S3) varied from 4.74 in HD to 17.92 in LA, which indicates that there was considerable variation for these traits. ANOVA identified significant differences among G, E and G × E for all traits, showing highly significant P values ≤ 0.01 (Table S4) for the main effects. The heritability (h_b) of the 15 traits ranged from 59% for BY to 88% for MD over locations; by using data pooled over the two years and all locations, the highest significant positive correlation was found between BG vs. GY (r = 0.64, P < 0.01), (Table S5).

The allele number, gene diversity and PIC were calculated to estimate the genetic diversity in a barley panel. A total of 40 markers (32 SSR and 8 SNP) were used to measure the genetic diversity of the population. Out of them, 27 SSR and 6 SNP showed polymorphism, with a total of 98 alleles; the gene diversity and the PIC values of the 33 polymorphic loci ranged from 2 to 6, 0.07 to 0.80 and 0.06 to 0.77, with an average of 2.76, 0.43, and 0.36, respectively (Table 1). Seven SSR loci (Bmac0030, EBmac0501, Bmac0316, EBmac0701, Bmag0125, Bmag0120 and GMS006); two SNP loci (scssr09398 and scssr07970); SNP showed high PIC and gene-diversity values ≥ 0.51 (Fig. S1).

STRUCTURE analysis results showed that the statistic Δk was used to determine a suitable value for K. Here, the Δk value was much higher for the model parameter K = 2 (Fig. 2), suggesting that the total panel could be divided into two major sub-populations. Sub.P1 contained hulled accessions from ARC, ICARDA and ACSAD, and sub.P2 contained naked types (European and five Egyptian cultivars). Both sub-populations (based on grain type and geographic region) contained 30 accessions; the genotypes with partial membership may exhibit distinctive identities.

AMOVA was performed to analyse the distribution of genetic diversity between and within the populations (Table S6). Highly significant (P ≤ 0.001) genetic variance was observed within and among sub-populations, recording a total 77 and 23% of total variance, respectively, with the main genetic variation occurring within sub-populations. For allele frequency, the revealed FST value was 0.235 among the sub-populations. UPGMA and PCA results are presented in Figs. S2 and S3, respectively. The results showed a spatial pattern that corresponds to the geographic region and the grain types. The hulled genotypes from ARC, ICARDA, and ACSAD were grouped into the same cluster, while the naked types (25 European landrace and five Egyptian cultivars) were in the second group.

Pairwise LD was estimated using the squared-allele frequency correlations (r²) and was found to decay rapidly with the genetic distance; a significant LD was detected across the genome of the 60 barley genotypes and extent of LD varied with different chromosomes (Fig. S4). The proportion of marker pairs exhibiting significant r² with P values ≤ 0.01 is reported in (Fig. 3A) as a function of the inter-marker distance. The scatter plots of the LD values of r² and D’ values as a function of the inter-marker distance cM for the entire samples are summarised in Fig. 3B and C. Overall, D’ values are high across inter-marker distances, ranging from −1.00 to 1.00 with an average of −0.70. The significant intra-chromosomal r² values (P < 0.01) ranged from 0.00 to 0.45 with an average of 0.04 for the whole panel (data not shown). The plot of r² versus the sum of PIC and GD values for intra-chromosomal loci pairs with r² > 0.05 are displayed in Fig. 3D.

Marker–trait associations (MTA) for 15 salt tolerance-related traits were performed with the GLM model (Table S7). A total of 33 marker (27 SSR and 6 SNP) loci with the R² range of 3–40, significant QTLs were identified on –log₁₀(P) threshold P ≤ 1.9 × 10^–5, and were associated with salt tolerance-related traits and differed from individual to individual over the two years. The association mapping approach in the first year was less than QTLs in the second year. A total of 107, 237 and 187 significant MTA (P ≤ 1.9·10⁻⁵) were identified among 1350 marker × trait combinations for the first and second years as well as for the mean over all the years, respectively. The strongest QTL for salt tolerance on 4H–log₁₀(P) was consistently found and co-localises with QTLs for other growth-related traits such as MD, Chl, PH, Ped, NP, BY, GY, and GW, located around 52.10 cM on 4H, with −log₁₀(P) scores close to 11 (MD/2nd year). Another important region was found at 35.39 cM on 7H with highly significant QTL−log₁₀(P) scores (9.57) for NG-related traits (first year). In addition, the region on 1H 87.83 cM had highly significant QTL–log₁₀(P) scores (10.32) for MD during the second year (Fig. 4).

3.6.1 Growth traits

3.6.2 Yield component traits

The important economic characteristics are generally quantitative in nature and exhibit a considerable degree of interaction with the environment. In this study, most traits were significantly affected by changes in E, G and G × E; the highly significant environment effect and its high variance component could be attributed to the differences among the test locations in soil characters [23]. The heritability, h_b%, exceeded 86% in combined and phenotypic means, reflecting a broad variation in the panel, by providing information on the magnitude of the inheritance of quantitative traits [24]. Although the h_b% for many traits might be influenced by environmental conditions, it is mainly affected by main effects and G × E effects during the growth cycle [25,26]. This indicates that the phenotypic expression of one genotype might be superior to another genotype in one environment, but inferior in a different environment; the current study has shown that this also holds true for salt stress tolerance in barley [9,27]. In contrast to plant tolerance to abiotic stresses, which is mostly dependent on monogenic traits, the adaptive mechanisms of the plants exhibit more complexity in response to salt stress, and are thus more difficult to control and engineer [2,28]. This can be explained by the influence of the G, E and G × E, as suggested by the results of the QTL analysis for most agronomic traits [29,30].

Molecular marker studies provide ample evidence that barley represents distinct gene pools. In the present study, STRUCTURE was applied to gain insight into the stratification of the germplasm collection [19]. Two sub-populations were detected among the accessions with significant divergence (P > 0.001). Within this structure, several patterns were quite prominent (e.g., the differentiation between ARC, ICARD, ACSAD and European accession sources or hulled and naked, respectively). AMOVA results revealed high genetic diversity within the 77% sub-population, while the low genetic differentiation among sub-populations of 23% may be due to gene flow that resulted from developing by crossing and their nature. However, the choice of marker plays a role in the structure detected, as shown by groups that appeared clearly using SSR/SNP markers [31]. We presented the landraces with an inferred ancestry corresponding to a different origin; for example, (S_1) represents freshly introduced populations in the region, which can result from seed exchange among the farmers, while the European and the Egyptian naked cultivars (S_2) had a high genetic similarity value, and are also the same grain type [20]. The set of markers (SSR/SNP) employed in this study demonstrates the usefulness of markers for the identification of barley genotypes. However, we need more markers and genotypes [32].

The extent of pairwise LD significantly varied among the barley chromosomes, suggesting that this factor may affect the resolution of association mapping for QTL located on different chromosomes [29,33]. Running the association analysis with a so-called partitioning trait, i.e. a trait showing a high correlation with the assessed stratification, can best assess the efficacy of a correction for population structure. Several molecular markers influenced more than two examined traits [9,32].

In this study, we have identified QTLs affecting salt tolerance and related traits using an association-mapping approach, while correcting for population structure and relatedness (Fig. 4). Numerous significant maker-trait associations were found in different regions of the barley genome controlling salt tolerance and related traits; among them, 45, 52 and 46 QTLs were detected during the first and second years as well as over the years for 14 associated traits. The results showed a strong QTL on 1H, 2H, 4H and 7H independently of population structure controlling salt tolerance that co-localised with QTL for other traits such as Chl, MD, PH, GFP, BY, GY, HI and GW, which are important factors for salt tolerance improvement in barley. Although only a small sample and limited markers were used, the results are credible because many of the loci that were identified were associated with traits that were common in previous reports of linkage or association mapping [1,9]. These results can be well understood considering the polygenic nature of the investigated traits. Tolerance to abiotic stresses (e.g., drought and salinity) is inherited quantitatively [34].

Association mapping has proven to be a powerful approach to dissect the complexity of quantitative traits in plants [12]. From the breeding perspective, stress tolerance can be described as the ability to maintain stable yields under stress environments. Abiotic stress tolerance is a quantitative trait and is controlled by genetic factors, and many genes for abiotic tolerance have been lost during domestication [9]. Association mapping provides valuable information on genomic regions that control quantitative traits, but it also has limitations due to poor sampling of the allelic variation present in the barley gene pool for each one of the loci affecting salt tolerance [30].