The 563 soybean germplasms from the USDA Germplasm Collection were used for this study. These germplasms were originally collected from 26 countries, including Japan (159 germplasms), China (86), India (59), South Korea (46), North Korea (28), South Africa (26), Nepal (20), Indonesia (16), United States (16), Suriname (13), Thailand (10), and 15 other countries (39), plus 45 germplasms of unknown origin ( Supplementary Tables 1A, B ).

The phenotypic data for salt tolerance reactions in 563 soybean germplasms were downloaded from the USDA GRIN website: https://npgsweb.ars-grin.gov/gringlobal/descriptordetail?id=51054. The experiments were conducted in Illinois, United States by Randy Nelson at the USDA Soybean Collection in Urbana, IL. The salt reaction was scored as tolerant (T) and susceptible (S) for each accession. ‘1’ indicating tolerance and ‘9’ indicating susceptibility were used to perform GWAS to identify SNP markers associated with salt tolerance in this study ( Supplementary Table 1A ).

The germplasms were genotyped using Soy50K SNP Infinium Chips (Song et al., 2013). A total of 42,292 SNPs across 563 soybean germplasms were downloaded from SoyBase at https://soybase.org/snps/download.php (Song et al., 2015). For GWAS, 34,181 SNPs were selected after excluding those with more than 5% missing data, heterozygosity greater than 5%, and a minor allele frequency (MAF) less than 5%. These SNPs were distributed across all 20 chromosomes of the soybean genome ( Supplementary Figure 1 ).

In this study, 34,181 SNPs were included in the principal component analysis (PCA) and genetic diversity analysis. PCA and genetic diversity were analyzed using GAPIT 3 (Wang and Zhang, 2021), with PCA components set from 2 to 10 and NJ tree settings from 2 to 10. Phylogenetic trees were drawn using the neighbor-joining (NJ) method in GAPIT 3. Genetic diversity was assessed for all 563 tested germplasms and their salt tolerance using (1) 34,181 SNPs in GAPIT 3, and (2) 10,000 randomly selected SNPs in MEGA 7 (Kumar et al., 2016). The phylogenetic trees were drawn using MEGA 7 based on the Maximum Likelihood method with the parameters described in Shi et al. (2016, 2017).

GWAS was conducted using various models, including Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway (BLINK), Fixed and Random Model Circulating Probability Unification (FarmCPU), Generalized Linear Model (GLM), and Mixed Linear Model (MLM) in GAPIT 3 (Wang and Zhang, 2021). The analysis was performed on a panel of 563 germplasms using 34,181 SNPs. Multiple GAPIT models were utilized to identify robust and consistent SNP markers associated with salt tolerance in soybean. The significance threshold for germplasms was determined using Bonferroni correction of P-values with an α = 0.05 (0.05/SNP number). An LOD (logarithm of odds) value of 5.83 [Here, we use LOD instead of –log(P-value)] was used as the significance threshold based on the 34,181 SNPs.

Candidate genes associated with salt tolerance were sought within a 5 kb vicinity on both sides of the significant SNPs, following the methodology outlined by Zhang HY, et al., 2016. The candidate genes were extracted from the reference annotation of the soybean genome assembly, Wm82.a2.v1, available at https://phytozome-next.jgi.doe.gov/info/Gmax_Wm82_a2_v1.

In this investigation, ridge regression best linear unbiased prediction (RR-BLUP) from the rrBLUP package (Endelman, 2011) and Bayesian models, including Bayes A (BA), Bayes B (BB), Bayes LASSO (BL), and Bayes ridge regression (BRR), implemented in the BGLR package, were employed for predicting genomic estimated breeding values (GEBV) in GP. The analysis was carried out using R software version 4.3.1 (https://www.r-project.org/). Additionally, GEBV prediction was conducted using genomic best linear unbiased prediction (gBLUP), composite BLUP (cBLUP), marker-assisted BLUP (maBLUP), and settlement of MLM under progressively exclusive relationship (SUPER) BLUP (sBLUP) methods, implemented in the GAPIT package. The effectiveness of genomic prediction using these approaches has been documented in prior research studies (Shi et al., 2021; 2022; Jarquín et al., 2014; 2017; Zhang JP, et al., 2016).

Genomic prediction (r-value) for salt tolerance was conducted across various soybean panels and scenarios. Firstly, GP was estimated using a training set to predict salt tolerance in the panel of 563 soybean germplasms. Predictions were estimated using four models: maBLUP, gBLUP, cBLUP, and sBLUP, utilizing all 34,181 SNPs in GAPIT3. Secondly, GP was executed using ten SNP sets: eight randomly selected SNP sets ranging from 10 to 10,000 SNPs, plus a GWAS-derived SNP marker set containing 10 markers (m10). These predictions were estimated using five GP models: BA, BB, BL, BRR, and rrBLUP. The prediction accuracy for salt tolerance was assessed using the average Pearson’s correlation coefficient (r) between the GEBVs and observed values in the validation set. Training and validation sets were randomly created 100 times, and the r-value was estimated for each iteration. The average r-value across the 100 iterations was then calculated for salt tolerance. In the GP scenarios, a higher r-value indicates greater prediction accuracy and better selection efficiency in GS, reflecting the reliability of the GP for salt tolerance.

The 563 soybean germplasms were divided into two groups, where 150 germplasms were salt tolerant and 413 were susceptible ( Figure 1 ; Supplementary Tables 1A, C ). These 150 salt-tolerant germplasms can be used as parents in soybean breeding programs to develop salt-tolerant lines.

Using GAPIT 3, the 563 soybean germplasms were divided into four distinct clusters (subpopulations), labeled Q1 to Q4, based on the analysis of 34,181 SNPs ( Figure 2 ; Supplementary Figure 2 ). The clustering was derived from the following analyses: (1) a 3D graphical plot of the principal component analysis (PCA) ( Supplementary Figure 2 , left), (2) a PCA eigenvalue plot ( Supplementary Figure 2 , right), and (3) phylogenetic trees constructed using the neighbor-joining (NJ) method ( Figure 2A , ring – left and Figure 2B , no-root - right). Additionally, the kinship plot confirmed the existence of these four groups among the 563 germplasms ( Supplementary Figure 3 ). Each was assigned to one of the four clusters (Q1 to Q4) ( Supplementary Table 1A ), and the resulting Q-matrix with four clusters was subsequently applied to the GWAS analysis.

Based on the analysis using four models (GLM, MLM, FarmCPU, and BLINK) in GAPIT 3, the multiple QQ plots showed a significant deviation from the expected distribution ( Figure 3 , right half; Supplementary Figure 4 , right), indicating the presence of SNPs associated with salt tolerance. The multiple Manhattan plots, covering all 34,181 tested SNPs, revealed several SNPs with LOD values greater than 5.83, primarily located on chromosomes 1, 2, 3, 7, and 16, suggesting the SNPs were associated with salt tolerance in the panel ( Figure 3 , left half; Supplementary Figure 4 , left).

Ten SNPs with LOD values greater than 5.83 were detected by at least one model (GLM or FarmCPU) in GAPIT 3 for salt tolerance ( Table 1 ). Three SNP markers associated with salt tolerance, Gm01_4306329_ss715579436, Gm01_4312808_ss715579441, and Gm01_4336306_ss715579451, are in a region from 4,306,329 bp to 4,336,306 bp, with an interval of 30 kb on chromosome 1. The LODs of these SNPs were greater than 6.5 in the GLM model ( Table 1 ). The other three SNP markers associated with salt tolerance, Gm02_36878905_ss715582154, Gm02_36940321_ss715582156, and Gm02_36991983_ss715582157, are located in a region from 36,878,905 bp to 36,991,983 bp on chromosome 2. These SNPs had an LOD greater than 6.7 in the GLM model and greater than 4.1 in a t-test ( Table 1 ), indicating the presence of a QTL in this region. The two SNP markers, Gm03_43213208_ss715586397 and Gm03_43220331_ss715586399, are located at 43,213,208 bp and 43,220,331 bp, respectively, on chromosome 3. These SNPs showed an LOD greater than 5.83 in either the GLM or FarmCPU model and greater than 8.9 in t-tests ( Table 2 ), suggesting a QTL in this region of chromosome 3 for salt tolerance. The SNP marker Gm07_41776639_ss715598058, located at 41,776,639 bp on chromosome 7, is associated with salt tolerance with an LOD of 6.44 in the GLM model and 5.94 in the t-test ( Table 2 ). The SNP marker Gm16_7722217_ss715625494, located at 7,722,217 bp on chromosome 16, is associated with salt tolerance with an LOD of 5.95 in the GLM model and 2.75 in the t-test ( Table 2 ).

Eleven genes are located within 5 kb upstream and downstream of 7 of the 10 SNP markers associated with salt tolerance ( Table 1 ; Supplementary Table 2 ). Gene information is based on the soybean reference genome Glycine max Wm82.a2.v1 (https://phytozome-next.jgi.doe.gov/info/Gmax_Wm82_a2_v1).

The four genes, Glyma.01G039600 (leucine-rich repeat receptor-like protein kinase family protein), Glyma.01G039700 (Vps51/Vps67 family protein, components of vesicular transport), Glyma.01G039800 (galactosyltransferase family protein), and Glyma.01G040000 (glutathione S-transferase TAU 18), are physically close to the three SNP markers—Gm01_4306329_ss715579436, Gm01_4312808_ss715579441, and Gm01_4336306_ss715579451— ( Tables 1 , 2 ).

Similarly, candidate genes Glyma.02G195400 (syntaxin of plants 121), located at 36,872,648 bp to 36,875,320 bp on chromosome 2, Glyma.03G230400 (invertase H), Glyma.03G230500 (plus-3 domain-containing protein), Glyma.03G230600 (protein of unknown function, DUF538), and Glyma.03G230700 (importin alpha isoform 4) on chromosome 3 and Glyma.16G076200 (pyrimidin 4) and Glyma.16G076300 (long-chain fatty alcohol dehydrogenase family protein on chromosome 16 were within physically close to the significant SNPs on those chromosomes ( Tables 1 , 2 ).

GP in the reference: The GP analysis yielded moderate to high r-values of 0.46, 0.60, and 0.44 for the maBLUP, gBLUP, and sBLUP models, respectively. These estimates were obtained using a training set to predict salt tolerance in a panel of 563 soybean germplasms genotyped with 34,181 SNPs in ( Supplementary Figure 5 ). These results indicate that GS is effective for salt tolerance selection.

GP in cross-prediction using randomly selected SNP markers: GP using randomly selected SNP markers for cross-prediction yielded the following average r-values: 0.19 (ranging from 0.14 to 0.20) for the 10-SNP set (r10); 0.22 (ranging from 0.17 to 0.24) for the 100-SNP set (r100); 0.25 (ranging from 0.24 to 0.26) for the 200-SNP set (r200); 0.29 (ranging from 0.28 to 0.30) for the 500-SNP set (r500); 0.30 (ranging from 0.29 to 0.31) for the 1,000-SNP set (r1000); 0.31 (ranging from 0.29 to 0.33) for the 2,000-SNP set (r2000); 0.30 (ranging from 0.28 to 0.32) for the 5,000-SNP set (r5000); and 0.28 (ranging from 0.27 to 0.30) for the 10,000-SNP set (r10000) (Table 2; Figure 4 ). These results demonstrate that the r-value increased with the number of randomly selected SNPs, with an average r-value rising from 0.19 in the 10-SNP set to 0.30 in the 10,000-SNP set. This suggests that a randomly selected SNP set consisting of at least 1,000 SNPs (r = 0.30) should be used in GS for selecting salt tolerance.

GWAS-derived SNP marker set: The average r-value was 0.38, ranging from 0.36 to 0.40, for the 10-marker set (m10) ( Table 2 ; Figure 4 ). These results indicate that the r-value was moderately high, exceeding 0.35 and surpassing those of SNP sets randomly selected from 10 SNPs to 10,000 SNPs. This suggests that GWAS-derived SNP markers can be effectively used for GP and for selecting salt tolerance in soybean breeding through MAS and GS.

GP Model: All five GP models—BA, BB, BL, BRR, and rrBLUP—exhibited similar r-values, indicating that each model is effective for selecting salt tolerance in GS.

The phylogenetic analysis showed that the 150 salt-tolerant germplasms were distributed throughout the tree of 563 germplasms and did not separate into distinct groups of susceptible and tolerant germplasms ( Supplementary Figure 6 ). This indicates that the 150 salt-tolerant germplasms have broad genetic backgrounds. Further analysis revealed that these 150 salt-tolerant germplasms can be divided into three distinct groups ( Supplementary Figures 7 , 8 ), confirming that they possess different genetic backgrounds.

Among the 150 salt-tolerant germplasms, six countries contributed a total of 104 germplasms: Japan (46), China (31), India (12), Suriname (7), Indonesia (6), and Nepal (5) ( Supplementary Table 1C ). Phylogenetic analysis of these 104 germplasms showed that germplasms from the same country generally clustered together ( Supplementary Figure 9 ), suggesting that germplasms from the same country share similarities in genetic backgrounds. Specifically, germplasms from Nepal are closer to those from Japan, followed by China; germplasms from Indonesia are closer to those from Suriname; while germplasms from India are more distinct ( Figure 5 ). This clustering suggests that geographic factors influence the distribution of salt-tolerant germplasms.