We used 428 accessions from the 3 K-RG to evaluate the alkali tolerance of rice at the germination stage. The accessions contained 125 Geng accessions, 278 Xian accessions, 13 Aus, 6 Basmati and 6 admix accessions (Supplementary Table 1). A total of 120 seeds of each accession were dried at 50°C for 3 days to make them dry completely and break their dormancy. To disinfect the seeds, they were soaked in 3% NaClO solution for 30 min, and then the seeds were washed three times with sterile distilled water before testing. For each rice accession, 20 seeds were placed into a 90-mm-diameter Petri dish with filter paper. Three replicates were conducted under control (distilled water) and alkali stress conditions (0.15% Na₂CO₃ solution), respectively. As a control, 10 ml of distilled water was added to each Petri dish, and 10 ml of 0.15% Na₂CO₃ solution was added to each Petri dish as alkali stress. After that, the seeds were grown in a growth chamber under a 14-h light/10-h dark photoperiod (28°C/26°C) with 70% relative humidity for 7 days. Throughout the test, the solution was changed regularly every day.

Seeds were considered germinated when the root length was equal to the seed length and the shoot length was equal to half of the seed length. The germinated seeds were counted each day to calculate the germination rate (GR), germination energy (GE), and germination index (GI). We used the seed germination rates on day 3 and day 7 as GE and GR (GE or GR = N_t/N₀, where N_t represents the number of germinated seeds at day t and N₀ represents the total number of experimental seeds), respectively. The GI was calculated as follows: GI = ∑(G_t/T_t), where G_t is the accumulated number of the germinated seeds on day t and T_t is the time corresponding to G_t in days (Wang et al., 2010). The mean germination time (MGT) was calculated by using the formula: MGT = ∑T_tN_t/∑N_t, where N_t is the number of newly germinated seeds on day t (Alvarado et al., 1987). On day 7, the root length (RL) and shoot length (SL) of eight seeds randomly selected from each replicate were measured in both alkali stress and control conditions. Vigor index (VI) = (mean SL) × GI. We also used the ratios of the seven germination-related traits under alkali stress to the control for evaluating the response of rice accessions to alkali stress. The average of three replicates for each trait was used for data analyses.

The 3K-RG 4.8 M SNP dataset was downloaded from the Rice SNP-Seek Database (Alexandrov et al., 2015).¹ SNPs with the missing rate < 20% and minor allele frequencies >5% were filtered with PLINK (Purcell et al., 2007). Finally, a total of 2,942,166, 2,091,233, and 1,218,609 SNPs were used for GWAS in the whole, Xian and Geng panel populations, respectively. The GWAS was performed with EMMAX (Kang et al., 2010) based on a mixed linear model (MLM) to detect the associations between SNP and the alkali tolerance-related traits. The kinship matrix was generated using an identical-by-state matrix (with the “emmax-kin -v -h -d 10” parameter in EMMAX) based on the pruned subset of genome-wide SNP data (with the “indep-pairwise 50 10 0.1” parameter in PLINK) to account for the relatedness among accessions. For controlling population structure, the eigenvectors of the kinship matrix were calculated using GCTA (with the “-make-grm” parameter; Yang et al., 2011) and the first three principal components were used as covariates. The effective number of independent markers (N) was calculated with the GEC software (Li et al., 2012) and suggestive significance thresholds of association by the Bonferroni correction method (1/N) were calculated for claiming significant SNPs for the whole population (p = 2.15E-06), Xian (p = 2.76E-06) and Geng (p = 8.31E-06) subpopulations, respectively. Manhattan and quantile–quantile (Q–Q) plots of GWAS were created by the R package “qqman” (Turner, 2014). Based on the previously reported linkage disequilibrium (LD) decay in the 3 K-RG (Wang et al., 2018), the significant SNPs within a 300-kb region were considered a locus. A lead SNP in a locus was defined as the SNP with the lowest value of p, and the other significant SNPs within 150 kb on either side of the lead SNP were merged.

We shortlisted potential candidate genes for alkali tolerance when they met at least one of the following criteria: (1) genes with significant SNPs and functional annotation related to abiotic stress based on the Nipponbare reference genome IRGSP 1.0 (Kawahara et al., 2013) and the funRiceGenes database (Yao et al., 2018); (2) genes harboring SNPs significantly associated with more than three traits; (3) genes containing SNPs of the most significant association (p < 0.05/effective SNP number) with each trait. In the case of an SNP located in the intergenic region, the downstream gene was chosen. LDBlockshow (Dong et al., 2021) was used to estimate the local LD block region containing each candidate gene. The haplotype analysis was performed on all candidate genes using all SNPs within the gene coding sequence region, in which synonymous SNPs were ignored (merged in one haplotype; Zhang et al., 2021). There were at least 10 rice accessions in each haplotype. Duncan’s multiple range post-hoc tests were completed with the “agricolae” package (de Mendiburu, 2021) in R to compare phenotypic differences among the haplotypes (n ≥ 10 rice accessions). The expression profiles of candidate genes in Nipponbare were obtained from the RiceXPro database (Yutaka et al., 2013).

For detecting expression levels of candidate genes, we screened two representative accessions (one alkali-tolerant and one alkali-sensitive) from each subspecies. After 24 h of alkali stress with 0.15% Na₂CO₃, 100 seed embryos were sampled under alkali stress and control conditions, respectively. Total RNA was extracted from germinating seeds using plant RNA extraction kit (Tiangen Biotechnology) and reverse transcribed using reverse transcription kit (Invitrogen). Real-time qRT-PCR analyses were conducted with Taq Pro Universal SYBR qPCR Master Mix (Vazyme, Q712-02). All primers used for qRT-PCR are listed in Supplementary Table 2. The relative expression levels were calculated using the 2^-∆∆CT method (Livak and Schmittgen, 2002).

We used one-way ANOVA to test the differences between subpopulations in the traits related to alkali tolerance by the “agricolae” (de Mendiburu, 2021) package in R. Spearman’s correlation coefficients between traits were calculated using the “corrplot” package in R. Box plots for displaying the trait distribution and bar charts for illustrating haplotype frequency were generated using the “ggplot2” (Ginestet, 2011) package in R.

Seven traits including GR, GI, MGT, VI, RL, SL, and GE were measured for 428 rice accessions under alkali stress with 0.15% Na₂CO₃ (named “TraitName+S”) and control condition with distilled water (named “TraitName+C”) at the germination stage (Table 1). Furthermore, we calculated the ratios of these traits under alkali stress and control conditions (named “R + TraitName”) in order to determine the degree of alkali damage (Supplementary Table 1). Under alkali stress and control conditions, the germination-related traits were distributed continuously, suggesting that multiple loci may be involved in controlling these traits (Supplementary Figure 1). We observed 12 accessions with GRS of 1.00, including eight Xian accessions, three Geng accessions, and an Aus accession (Supplementary Table 1). Among the four accessions with the shortest MGTS, all were Xian accessions. In addition, we found some accessions exhibited extreme phenotypes in other germination-related traits, such as “Lumaozhan” with the highest GES (0.90) and GIS (1.04), “Heimangdao” with the longest SLS (6.03 cm), “ARC 15091” with the longest RLS (4.53 cm), and “NCS 766” with the highest VIS (6.17).

In comparing the germination-related traits under alkali stress and control conditions, we found that alkali stress significantly impaired the seed germination and growth, extending germination time and reducing GE, GR, GI, VI, RL, and SL (Figure 1A). GE was the most strongly affected by alkali stress among the seven traits. On average, MGT was delayed by 14.9% under alkali stress, and GR, GI, VI, RL, SL, and GE decreased by 32.2%, 42.6%, 55.9%, 73.6%, 26.7%, and 78.9%, respectively. We used the ratio of the trait value under alkali stress to control to determine the extent of damage caused by alkali. Accordingly, alkali damage on these germination-related traits followed the following order: GE > RL > VI > GI > GR > SL > MGT (Figure 1B). Furthermore, the Xian subpopulation (n = 278) exhibited significantly higher RVI, RRL, RGE, but significantly shorter RMGT than the Geng subpopulation (n = 125; Figure 1C), suggesting that Xian accessions are generally more alkaline tolerant than Geng accessions.

According to the correlation analysis among the germination-related traits in the whole population, MGTS and RMGT were negatively correlated with the other traits, whereas GRS, RGR, GIS, RGI, VIS, and RVI demonstrated a strong positive correlation with one another (Figure 1D). In addition, correlation analysis results in the subpopulations Xian and Geng were similar to the results in the whole population (Supplementary Figure 2). These findings indicate that it is valuable to identify the loci associated with these germination traits and corresponding favorable alleles in molecular breeding for rice alkali tolerance.

We conducted a GWAS based on the mixed linear model for the traits related to alkali tolerance (Supplementary Figures 3–8). Using a Bonferroni correction based on the effective numbers, the genome-wide significant value of p thresholds were set at 2.15E-06, 2.76E-06, and 8.31E-06 for the whole population, Xian subpopulation, and Geng subpopulation, respectively. Consequently, 271, 51, and 503 SNPs significantly associated with alkali tolerance at the germination stage were identified in the whole population, Xian subpopulation and Geng subpopulation, respectively. A total of 18 significantly associated SNPs were detected both in the whole population and Xian subpopulation (Supplementary Figure 9A). However, there were no overlaps between the identified SNPs in Xian subpopulation and those in Geng subpopulation. By analyzing the genes in which the significantly associated SNPs were detected, it was found that there was also no overlap between the genes identified in Xian subpopulation and those in Geng subpopulation (Supplementary Figure 9B). The results suggested that the genetic mechanism of alkali tolerance in subpopulations Xian and Geng may be different. We merged the adjacent SNPs around 300 kb as a locus to reduce the redundancy in association signals of different traits. Consequently, a total of 90 loci involved in 124 associations between 106 lead SNPs and 14 traits were identified in the whole population or at least one of the two subpopulations (Supplementary Table 3).

According to the annotations of the genes with identified SNPs, no reported genes for alkaline tolerance were identified. However, we found two known salt tolerance genes (OsRLCK253 and OsNCED5) harboring lead SNPs (rs8_17550068 with p = 4.92E-07 and rs12_26284167 with p = 6.1E-06) associated with RSL in Geng subpopulation (Figure 2A). Moreover, a WRKY transcription factor gene OsWRKY76 was significantly associated with RRL in the whole population (Figure 2B). In addition, we identified several new loci with strong associations with alkali tolerance (Figures 2C–E; Supplementary Table 3). For example, a lead SNP rs5_26741198 was significantly associated with SLS (p = 8.68E-11) in the whole population. The lead SNP of rs3_34261591 showed the strongest association signal (p = 2.15E-8) with MGTS in Xian subpopulation. The lead SNP rs8_24108555 with the most significant association (p = 9.42E-9) for GES was identified in Geng subpopulation (Figures 2C–E).

A comprehensive analysis was carried out concerning the 90 alkali tolerance loci, each of which was associated with one to five traits (Figure 2F; Supplementary Table 3). For example, locus 4 was associated with GRS, RLS, RGR, and RGI in the whole population and with RLS in Geng subpopulation. Locus 7 was associated with GRS, RGR, and RVI in the whole population. Locus 44 was associated with GES, RGE and RRL in the whole population and with RRL in Xian subpopulation. Locus 59 was associated with GRS and RGR both in the whole population and Xian subpopulation. Locus 88 was associated with GIS, VIS, SLS, and RVI in the whole population and with RSL in Geng subpopulation.

We selected 17 genes as candidate genes at 15 loci (Supplementary Table 4; see Methods for detail), and conducted a gene-CDS-haplotype analysis to determine whether there were significant differences in alkali tolerance between different haplotypes of each gene. Finally, eight genes with significant phenotypic differences between haplotypes were identified as important candidate genes for alkali tolerance, including LOC_Os09g25060, LOC_Os03g60240, LOC_Os03g08960, LOC_Os01g12000, LOC_Os04g41410, LOC_Os11g35350, LOC_Os12g09350, and LOC_Os12g13300. Based on qRT-PCR analysis, the expression levels of LOC_Os09g25060 and LOC_Os03g60240 were both higher in the alkali-sensitive Xian accession than in the alkali-tolerant Xian accession. On the contrary, LOC_Os11g35350 was expressed more strongly in the alkali-tolerant Xian accession compared to the alkali-sensitive Xian accession. There were no significant differences in the expression levels of the other five genes between the two Xian accessions (Figure 3A). For the two Geng accessions, the expression level of LOC_Os09g25060 in alkali-sensitive accession was also significantly higher than that in alkali-tolerant accession, while LOC_Os01g12000 was expressed more strongly in the alkali-tolerant accession than in alkali-sensitive accession (Figure 3B).

For locus 71 on chromosome 9, the LD block region was predicted from 14.94 to 14.99 Mb (48.93 kb) and included 491 SNPs (Figure 4A). LOC_Os09g25060 (OsWRKY76) had a lead SNP rs9_14985169 (p = 2.09E-06) for RRL in the whole subpopulation, which is a member of the rice WRKY transcription factor gene family (Ross et al., 2007). Three major haplotypes of LOC_Os09g25060 were detected based on five SNPs in the coding region shared by at least 10 accessions (Figure 4B). Hap1 is predominant in Xian subpopulation and Hap2 is prevalent in Geng subpopulation (Figure 4D). In comparison of the RRL across the three haplotypes, Hap1 and Hap2 had a significantly higher RRLs of 0.28 and 0.26 than Hap3, which had a mean RRL of 0.19 in the whole population (Figure 4C).

The LD block region for locus 32 on chromosome 3 was predicted from 34.256 to 34.267 Mb (11.16 kb) and contained 37 SNPs (Figure 5A), including the alkali tolerance candidate gene LOC_Os03g60240 that had the most significant SNP rs3_34257992 (p = 7.96E-08) for MGTS in Xian subpopulation. Using 22 SNPs (including five non-synonymous SNPs) in the coding region of LOC_Os03g60240, six haplotypes were detected, each of which was shared by at least 10 accessions (Figure 5B). In the whole population, Hap4 had the lowest mean MGTS (3.98). Thus, we defined Hap4 as the favorable haplotypes of alkali tolerance, which were carried only by Xian accessions (Figures 5C,D).

For locus 7 on chromosome 1, the most significant association peak for RGR, GRS, and RVI in the whole population, was predicted from 6.47 to 6.56 Mb (85.5 kb) and included 730 SNPs based on local LD analysis (Figure 6A). One of these SNPs, rs1_6533274, located in the intergenic region of two genes (LOC_Os01g11990 and LOC_Os01g12000), was significantly associated with RGR, GRS, and RVI. We identified four haplotypes (n ≥ 10 accessions) based on eight SNPs in the coding region of LOC_Os01g12000, the downstream gene of rs1_6533274 (Figure 6B). Among the haplotypes, Hap1 and Hap3 were both enriched in Xian subpopulation, while Hap3 had significantly higher GRS, RGR, and RVI than the other haplotypes (Figures 6C,D).

The LD block region of locus 82 was estimated to be from 20.71 to 20.75 (38.1 kb) on chromosome 11 including 438 SNPs. The SNP rs11_20724735, located in the candidate gene LOC_Os11g35350, was significantly associated with RRL in the Xian subpopulation (Figure 7A). Using six SNPs within the coding region of LOC_Os11g35350, four major haplotypes (n ≥ 10 accessions) were identified (Figure 7B). Hap3 had significantly higher RRL, which were mainly present in the Xian accessions compared to the other haplotypes (Figures 7C,D).

Since RGE was most sensitive to alkali stress, it was used to identify the favorable haplotype of each candidate gene. As a result, Hap1 of LOC_Os09g25060, Hap3 of LOC_Os01g12000, Hap5 of LOC_Os12g13300, Hap4 of LOC_Os03g60240, Hap3 of LOC_Os04g41410, Hap6 of LOC_Os12g09350, Hap4 of LOC_Os11g35350, and Hap1 of LOC_Os03g08960 were detected as alkali-tolerant haplotypes. We used a combined haplotype analysis in order to determine the breeding potential of candidate genes for alkali tolerance. There remained five groups comprising three candidate genes after the removal of rare haplotype combinations (n < 15 accessions). The Groups I, II, III, and IV were enriched in the Xian subpopulation, whereas Group V was mainly found in the Geng subpopulation, which showed clear Xian-Geng differentiation (Figure 8A). Compared to the other four groups, Group I containing the alkali-tolerant haplotypes at LOC_Os03g60240, LOC_Os09g25060, and LOC_Os03g08960 had a significantly higher RGE (Figure 8B). Furthermore, the more favorable haplotypes of these candidate genes were pyramided in rice, the stronger the alkali tolerance. The results suggest that developing a pyramid of multiple favorable haplotypes/alleles for alkali tolerance may be an effective strategy for improving rice alkali tolerance at the germination stage.

The alkali stress of 0.15% Na₂CO₃ can significantly inhibit rice germination (Figure 1A). Qi et al. (Qi et al., 2005, 2006) proposed the primary method to evaluate the alkali tolerance during rice germination is to measure GR, GI, and relative alkali damage rate, and they reported a wide variation range and reasonable distribution of GR, GE and their relative alkali damage rate under 0.10% and 0.20% Na₂CO₃ condition. In our study, GE was the most sensitive trait of the seven germination-related traits measured under alkali stress (Figure 1B), which is consistent with previous studies (Lv et al., 2013). Furthermore, the alkali tolerance levels of rice germplasms also varied considerably during germination in this study (Supplementary Figure 1). Xian accessions had shorter RMGT and higher RVI, RRL, and RGE than Geng accessions, indicating that Xian accessions had generally higher tolerance to alkali stress than Geng accessions at the germination stage (Figure 1C). This is contrary to the previous findings by Shi et al. regarding salt tolerance in rice during germination (Shi et al., 2017), inferring different mechanisms for salt and alkali tolerance in rice.

Two salt tolerance-related genes OsRLCK253 and OsNCED5 were associated with alkali tolerance at the germination stage in this study. OsRLCK253, which encodes a receptor-like cytoplasmic kinase, can improve the water-deficit and salt tolerance in transgenic Arabidopsis plants by affecting several common endogenous gene expressions (Giri et al., 2011). OsNCED5, encoding a 9-cis-epoxycarotenoid dioxygenase, positively regulates ABA level, enhances salt tolerance, and accelerates leaf senescence in rice (Xiong et al., 2014). In both OsRLCK253 and OsNCED5, however, there were no significant differences in alkali tolerance between different haplotypes. In addition, no cloned genes with large effects on salt tolerance were detected in this study, further supporting the perspective that different genetic mechanisms may regulate alkali tolerance and salt tolerance in rice (Sun et al., 2021).

The expression levels of candidate gene LOC_Os09g25060 were significantly higher in alkali-sensitive accessions compared with alkali-tolerant accessions in both Xian and Geng subpopulations, suggesting its negative regulatory role in alkali tolerance during rice germination. LOC_Os09g25060, encoding a WRKY transcription factor OsWRKY76, is downregulated in the roots of Dongdao-4 in response to saline–alkaline stress (Li et al., 2020a). Previous studies have shown that the genes associated with jasmonic acid (JA) biosynthesis and JA signaling are enhanced in the knockout plants of both OsWRKY62 and OsWRKY76 (Liang et al., 2017). Moreover, salt stress can trigger the activation of the JA signaling pathway and inhibit the cell elongation in the elongation zone of root (Valenzuela et al., 2016). Based on haplotype analysis of OsWRKY76, Hap1 and Hap2 had higher RRL than Hap3 (Figure 4C), suggesting that OsWRKY76 may affect root development by activating the JA signaling pathway, which ultimately affects rice alkali tolerance at the germination stage.

Rice can alleviate the damage of alkali stress through osmotic regulation, ion balance, antioxidant protection, and hormone regulation (Fang et al., 2021). The alkali tolerance of rice at the germination stage is a complex trait, which is controlled by multiple genes.

Among 90 loci for alkali tolerance in this study, only five loci (5.6%) were also detected in previous QTL mapping studies (Supplementary Table 3), primarily because most reported QTLs for alkali tolerance were identified at the seedling stage. In other words, the genetic mechanisms of alkali tolerance in rice may partially differ between the seedling and germination stages. Our results indicated that rice germplasms with more favorable haplotypes on candidate genes had better alkali tolerance (Figure 8), which suggests that developing an optimal haplotype combination through pyramiding multiple favorable haplotypes should be an effective strategy for improving alkali tolerance during rice germination. In addition, it is possible, through inter-subspecific crosses in rice breeding programs, to make introgression of favorable haplotypes specific in the Xian subpopulation into the Geng subpopulation to provide important donor resources for improving the alkali tolerance of the Geng subpopulation. For example, rice breeders may use marker-assisted selection for Hap4 of LOC_Os03g60240 in the progeny of inter-subspecific crosses to shorten germination time under alkali stress. Thus, the candidate genes associated with alkali tolerance and their favorable haplotypes will be useful to rice alkali-tolerant breeding.