A total of 541 accessions of rice (Oryza sativa L.) from major rice-growing areas worldwide, consisting of 328 indica, 162 japonica, 18 admix, 17 aus, and 16 basmati varieties (Supplementary Table S1), were obtained from the China Rice Data Center and the 3,010 Rice Genomes Project (Wang et al., 2018). All varieties were planted, and seeds were harvested in Lingshui, Hainan Province, China. A single nucleotide polymorphism (SNP) dataset was downloaded from the Rice SNP-Seek database (https://snpseek.irri.org/) (Mansueto et al., 2016). Non-sequenced varieties were sequenced on an Illumina HiSeq 2000 Sequencing Platform (Illumina Inc., San Diego, CA, USA), with an average sequencing depth of approximately 10×.

Fully mature seeds were harvested in field right after the grains turned to yellow, and dried in a 37°C oven for 7 days to break dormancy. Fifty rice varieties (Supplementary Table S2) were randomly chosen for germination test to determine the optimal salt stress conditions using 0.3%, 0.4%, 0.5%, 0.6%, 0.7%, 0.8%, 0.9% and 1.0% NaCl (w/v) dissolved in deionized water. Accordingly, the concentrations of 0.3% and 0.5% were subsequently used for GWAS. For each variety, 90 seeds were divided into three replicates and placed onto two layers of filter paper soaked with either 8 mL of deionized water, or 0.3% or 0.5% NaCl solutions, in 6-cm petri dishes without seal. All solutions were refreshed every two days. These dishes were cultured in an incubator at 28°C under a 12 h light/12 h dark photoperiod for 7 days. Germinated seeds, with the criterion of root lengths longer than the seed length, and shoot lengths longer than half of the seed length (Shi et al., 2017), were counted to calculate their germination rates on the 3^rd day (GR3), germination rates on the 4^th day (GR4), and germination indexes in 7 days (GI), with the calculation formula of GR3 = accumulated number of germinated seeds on the 3^rd day/the total number of seeds used × 100%; GR4 = accumulated number of germinated seeds on the 4^th day/the total number of seeds used × 100%; GI = ∑(G_t/T_t), where G_t is the number of germinated seeds on day t and T_t is the time in days corresponding to G_t (Shi et al., 2017). To obtain germination rate grades on the 3^rd (GRG3) and the 4^th day (GRG4), and germination index grades (GIG), the ratios of GR3, GR4 or GI for salt treatment to those in the control condition were calculated, and then assigned to germination grades from 1 to 11 (Supplementary Table S3). For seed germination in saline soil, pots with the size of 7 cm × 7 cm × 10 cm were filled with soil, and immersed in either deionized water or 0.5% NaCl solution for 24 hours to ensure saturation. Thirty seeds were sown into each pot and covered with a thin layer of soil. Subsequently, these pots were placed in an incubator at 28°C under 12 h light/12 h dark, watered with either deionized water or 0.5% NaCl solution in every two days. The germinated rates and shoot lengths were measured on the 10^th day after sowing.

The general linear model (GLM) with the first three principal components matrix calculated from genotypes was employed using Tassel 5.2.54 software (Bradbury et al., 2007) to perform GWAS for seed germination under salt stress. Across the whole panel of varieties, 302,900 high-quality SNPs were used for GWAS after removing SNPs with missing rates of more than 20% and minor allele frequencies of less than 5% (Ju et al., 2022; Mei et al., 2022). To identify significant SNPs, the Bonferroni correction method was used to calculate suggestive significance thresholds of associations, resulting in a p value of 1.65 × 10^-7 (0.05/N, with N = number of SNPs) (Shi et al., 2017). The qqman package in R3.4.1 software was used to generate Manhattan plots (Turner, 2014). Functional annotations for genes within genomic regions of interest were obtained in reference to the Nipponbare genome IRGSP 1.0 (Kawahara et al., 2013). Haplotype analyses were performed using SNPs in the coding sequences, with each haplotype possessing at least twenty rice varieties (Mei et al., 2022).

Candidate genes were investigated by quantitative real-time PCR (qRT-PCR) using RNA extracted from shoots and roots of 7-day-old Zhonghua11 (a japonica rice variety, ZH11) seedlings grown under 0.3% or 0.5% NaCl, or control (deionized water) conditions. To examine expression levels of candidate genes in varieties with different germination grades, shoots from 7-day-old seedlings of ST (L_363, L_417 and L_505, with GRG3 and GRG4 ≤ 2) and SS varieties (L_065, L_274 and L_352, with GRG3 and GRG4 ≥9) grown under 0.5% NaCl or control conditions were collected and analyzed by qRT-PCR. To examine expression patterns of OsTMF in time-course salt stress treatments, germinated seeds under 0.5% NaCl or control conditions for 1, 2, or 3 days, and seedlings grown in Hoagland solution treated with or without 0.5% NaCl for 1, 3, or 7 days, were sampled. Total RNA was extracted using TRIzol™ reagent (Thermo Fisher, USA), cDNA was synthesized using the PrimeScript™ RT reagent kit (Takara, Japan), and qRT-PCR was conducted using TB Green^® Premix Ex Taq™ II (Takara, Japan) on LightCycler^® 96 (Roche Life Science, Switzerland). Primers for qRT-PCR were designed using Primer3 (https://primer3.ut.ee) based on coding sequences of genes (Supplementary Table S4). OsActin1 was used as an internal control for normalization, and the 2^−△△CT method was used to calculate relative expression levels.

Clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated nuclease 9 (CRISPR/Cas9) constructs were made according to Cheng et al. (2021), using the guide sequences for candidate genes designed by the web tool CRISPR-GE (http://cbi.hzau.edu.cn/cgi-bin/CRISPR) (Xie et al., 2017; Supplementary Table S5), and transformed into ZH11 via Agrobacterium-mediated transformation (Nishimura et al., 2007). Sanger sequencing (Sangon Biotech, Beijing) was performed to identify homozygous mutants. PCR products amplified with HPT-F (5′-TGCCGTCAACCAAGCTCTGA-3′) and HPT-R (5′-GCTTCGATGTAGGAGGGCGT-3′) primers were used to select transgene-free lines.

Correlation analyses and significant differences were assessed between pairs of evaluated traits using SAS software (Shi et al., 2017). Differences between varieties, subpopulations, treatments, or haplotypes were examined using a one-way ANOVA method based on a significance level of 0.01. Figures were generated using GraphPad Prism 8.0 software (https://www.graphpad.com).

To define optimal conditions for assessing rice seed germination under salt stress, seeds of 50 varieties randomly selected from 541 varieties collected from different regions of the world (Supplementary Table S2) were germinated in either deionized water (control) or different concentrations of NaCl solutions (see Materials and Methods). Well-germinated seeds, according to the criterion reported (Shi et al., 2017) were counted every day over a 7-day period. GRG3, GRG4 and GIG were calculated, with germination grade 1 for the highest, and grade 11 for the lowest salt tolerances (see Materials and Methods). Results showed that, under 0.3% and 0.5% NaCl treatments, all three traits were consistently distributed (Supplementary Figure S1), therefore, these two treatments were used for subsequently experiments.

To examine variations in seed germinations under salt stress in 541 rice varieties, seeds of these varieties were placed on plates with water, 0.3% or 0.5% NaCl, and germinations were counted daily over a 7-day period. Variations of seed germinations under salt stress were observed in different varieties (Supplementary Table S1). As showed for two representative varieties, L_541 with lower salt tolerance, and L_022 with higher salt tolerance in germination, numbers of germinated seeds differed on the 3^rd and the 4^th day, with more germinated seeds for the ST variety (Figure 1; above dash lines). All three traits, GRG3, GRG4 and GIG, showed large variations (Figure 2), with the highest coefficient of variation observed for GRG4 (73.53%) under 0.3% NaCl condition, followed by GRG4 under 0.5% NaCl (57.31%), and GRG3 under 0.3% NaCl (53.43%) (Table 1). Phenotypic variations in subpopulations were also examined, and results showed that, under treatments of either 0.3% or 0.5% NaCl, indica varieties displayed lower GRG3 (Supplementary Figure S2A) and GRG4 (Supplementary Figure S2B) when compared with other 4 subpopulations that showed no significant differences, suggesting a higher salt tolerance of indica varieties. To be noted, no significant difference was observed for GIG among these subpopulations (Supplementary Figure S2C). In addition, on average, GRG3 was higher than GRG4 under both 0.3% and 0.5% NaCl (Table 1), suggesting that the germination of rice seeds on the 3^rd day was more vulnerable to salt stress than that on the 4^th day. Further analyses revealed that GRG3 and GRG4 had moderately positive correlations at these two NaCl concentrations (Supplementary Table S6).

To dissect the genetic basis of rice seed germination under salt stress, GWAS was conducted based on the GLM model using 302,900 SNPs from public database and in-house re-sequencing project. These SNPs were evenly distributed across all 12 chromosomes, with a density of about 800 SNPs/Mb (Supplementary Table S7). The significance threshold for GWAS was set to 6.78 (−log₁₀(0.05/N), N = effective SNP number) after Bonferroni correction. Manhattan plots were generated, showing that 1,494 SNPs were significantly correlated with rice seed germination under salt stress, in particular, GRG3 (Figure 3A; Supplementary Table S8) and GRG4 (Figure 3B; Supplementary Table S8). These SNPs explained 4.54% to 10.06% of phenotypic variations. No SNP association was identified for GIG (Figure 3C).

To identify putative genomic regions associated with salt tolerance, we screened for co-located and clustered SNPs detected under both 0.3% and 0.5% NaCl treatments. Three QTLs, qGRG3-1 (chr.3; 16,666,490 − 17,013,993) and qGRG3-2 (chr.5; 27,858,716 − 28,081,822) associated with GRG3 (Figure 3A), and qGRG4 (chr.4; 19,891,019 − 20,069,206) associated with GRG4 (Figure 3B), were identified. Notably, the qGRG4 locus contained a high-affinity potassium transporter OsHAK1, harboring a significant SNP (S04_19891019, −log₁₀(p) = 7.19 under 0.3% and −log₁₀(p) = 7.58 under 0.5% NaCl). Since OsHAK1 is known to regulate K⁺/Na⁺ homeostasis in rice seedlings under salt stress (Chen et al., 2015), our subsequent work was focused on two other loci, qGRG3-1 and qGRG3-2.

To identify candidate genes that are responsive to salt stress in qGRG3-1 and qGRG3-2 loci, the roots and shoots of 7-day-old ZH11 seedlings grown under 0.3% or 0.5% NaCl treatments were collected, and expression analyses were performed using qRT-PCR to analyze expressions of genes located in these two loci. No consistent responses to salt stress were observed in any of the 19 genes in the qGRG3-1 locus (Figure 4A). In contrast, 8 of 21 genes examined in the qGRG3-2 locus showed responsive expressions under the salt stress (Figure 4B). In particular, LOC_Os05g48620, LOC_Os05g48730, and LOC_Os05g48900 were up-regulated in roots and shoots of seedlings exposed to either 0.3% or 0.5% NaCl, while expressions of LOC_Os05g48650 and LOC_Os05g48670 were only up-regulated under 0.5% NaCl treatment (Figure 4B). Three other genes, LOC_Os05g48600, LOC_Os05g48760 and LOC_Os05g48930, were down-regulated in roots upon exposure to 0.3% or 0.5% NaCl (Figure 4B, right). We selected these genes as candidates for further analysis.

We examined expressions of these eight candidate genes in three ST (L_363, L_417, and L_505) and three SS varieties (L_065, L_274, and L_352), selected as described in Materials and Methods. Results showed that LOC_Os05g48730 (Supplementary Figure S3A) and LOC_Os05g48930 (Supplementary Figure S3B) exhibited higher levels of expression in ST varieties than those in SS varieties, under both the control and salt stress conditions. Three other genes, LOC_Os05g48620 (Supplementary Figure S3C), LOC_Os05g48670 (Supplementary Figure S3D) and LOC_Os05g48760 (Supplementary Figure S3E), showed elevated expressions under salt stress in ST varieties, but not in SS varieties. For LOC_Os05g48600, elevated expressions were observed in L_363 (ST) and L_417 (ST) under salt stress, but no significantly changed expressions in L_505 (ST), and three SS varieties (L_065, L_274 and L_352; Supplementary Figure S3F). For LOC_Os05g48650 and LOC_Os05g48900, although no consistent pattern was observed between ST and SS varieties, elevated expressions were detected for LOC_Os05g48650 under salt stress in all varieties except L_065 (Supplementary Figure S3G), whereas no salt-induced expressions were observed for LOC_Os05g48900 in neither ST nor SS varieties (Supplementary Figure S3H).

To further characterize the eight candidate genes in qGRG3-2 locus, knockout mutants were generated using CRISPR/Cas9-mediated gene editing in ZH11. For each gene, at least two independent homozygous transgene-free lines with frameshifts in translation were identified in either T1 or T2 generations, confirmed by PCR amplifications and sequencing (Supplementary Table S5). We tested the germination rates of mature seeds from mutant lines under the control, 0.3% and 0.5% NaCl treatments. Among lines assayed, ostmf-1 and ostmf-2 (Figures 5A, B), carrying mutations in LOC_Os05g48620 that encodes OsTMF (a rice homolog of human TATA modulatory factor, TMF), exhibited delayed germination under 0.3% NaCl treatments when compared to ZH11 (Figures 5C, D). OsTMF was previously reported to attenuate cold tolerance in rice by modulating expressions of cell wall biosynthesis-related genes (Xu et al., 2020). With the increased NaCl concentration (0.5%), further delays of germination were observed (Figures 5C (below dashed lines), E).

To examine the germinations of ostmf-1 and ostmf-2 in saline soil, we sowed their seeds in pots, filled with soil, and soaked in deionized water or 0.5% NaCl. On the 10^th day, no evident phenotypic difference was observed in ostmf-1 and ostmf-2 under the control conditions, when compared to ZH11 (Figure 5F). However, in soil soaked with 0.5% NaCl, both ostmf-1 and ostmf-2 showed lower germination rates (Figure 5G) and shorter shoot lengths (Figure 5H).

To investigate the expression of OsTMF in response to salt stress, we conducted qRT-PCR analyses in seeds and seedlings using RNAs collected at different days after growing in 0.5% NaCl and deionized water. Results showed that the OsTMF expression was increased in seeds after growing in 0.5% NaCl solution for 1 day, and then decreased gradually, whereas it maintained stable in those seeds grown in water (Figure 5I). In the shoots and roots of seedlings, salt stress enhanced OsTMF expressions from day 1 to 7, whereas in the control conditions, OsTMF expression remained at stable and low levels throughout the time course. These findings suggest that OsTMF may positively regulate seed germination under salt stress.

To examine the relationship between OsTMF haplotypes and seed germination under salt stress, haplotype analysis was conducted in 541 rice varieties. In the coding region of OsTMF, six SNPs were detected (Figure 6A), with which we defined four major haplotypes, based on their presences in at least 20 varieties. We observed that Hap1, Hap2, and Hap4 are predominant present in the indica subpopulation, while Hap3 is predominant in the japonica subpopulation (Figures 6A, B). GRG3 of varieties harboring Hap2 were 3.710 and 5.794 under 0.3% and 0.5% NaCl treatments, respectively, which were lower than those varieties carrying Hap1, Hap3, or Hap4 (Figure 6C), indicating that Hap2 is more tolerant to salt stress. These results together suggest that Hap2 is a potential haplotype to be used for improving rice seed germination under salt stress.