Genome-wide association analysis for rice salt tolerance at seedling and reproductive stages

Background Salinity is one of the main adverse environmental factors severely inhibiting rice growth and decreasing grain productivity. Developing rice varieties with salt tolerance (ST) is one of the most economical approaches to cope with salinity stress. However, the valuable resources for rice breeding towards ST remain to be identied. In this study, the salt tolerance of 220 rice accessions from RDP1, representing ve subpopulations, were evaluated at both seedling and reproductive stages. An apparent inconsistency was found for ST between the two stages. Through a gene-based / tightly linked genome-wide association study, a total of 214 single nucleotide polymorphisms (SNPs) related to 251 genes, signicantly associated with 16 ST-related indices, were detected at both stages. Eighty-two SNPs with low frequency favorable (LFF) alleles in the population were proposed to hold high breeding potential in improving rice ST. Fifty-four rice accessions collectively containing all these LFF alleles were identied as donors of these alleles. Through the integration of meta-QTL for ST and the response patterns of differential expression genes to salt stress, thirty-one candidate genes were suggested to be involved in the regulation of rice ST.

Introduction marker assisted selection (MAS), the SKC1 had been incorporated into some rice varieties with apparently enhanced ST in the seedling stage. It has been well recognized that ST at seedling and reproductive stages are different (Moradi et al., 2003;Liu et al., 2019;Lei et al., 2020;Chen et al., 2020); while, only few studies have reported ST QTLs/genes at the reproductive phase so far (Hossain et al., 2015;Liu et al., 2019;Lekklar et al., 2019;Ganie et al., 2019;Theerawitaya et al., 2020). Through the linkage mapping method, Hossain et al. (2015) reported 16 QTLs affecting 6 agronomic traits plus Na + content and the K + /Na + ratio in the ag leaf under stress conditions at the reproductive stage. In recent years, genome-wide association study (GWAS) has been developed in crops to enable large-scale and high-precision identi cation of elite alleles and their variants widely distributed in natural rice varieties Liu et al., 2019;Lekklar et al., 2019). With respect to salt stress, Kumar et al. (2015) rstly conducted GWAS with high density SNP (single nucleotide polymorphism) markers and identi ed 64 SNPs signi cantly associated with the K + /Na + ratio and some agronomic traits under salt stress at rice reproductive stage. Lekklar et al. (2019) investigated photosynthetic and yield-related traits of 190 Thai and Asian rice accessions exposed to salt stress at the owering stage, and identi ed 448 SNPs associated with ST by GWAS. By using 708 rice accessions and high-density SNPs within genes, Liu et al. (2019) detected 2255 SNPs that were signi cantly associated with ST-related traits at both seedling stage in greenhouse and reproductive stage at saline elds. In total, although a lot of ST related QTLs/genes have been reported, few of them, especially for those at reproductive stage, were characterized and successfully used in breeding Theerawitaya et al., 2020;Nayyeripasand et al., 2021).
Besides high-density SNPs generally used in GWAS, the other advantage of GWAS is that once a natural variety population is genotyped, it can be widely used in different studies towards diverse traits. For instance, to the best of our knowledge, the rice diversity panel l (RDP1), comprised of around 420 varieties from 82 countries/regions and genotyped by high-density SNP markers, has been widely employed in studying agronomic traits, biotic and abiotic stresses Famoso et al., 2011;Norton et al., 2014;Kang et al., 2016;McCouch et al., 2016;Chen et al., 2019;Feng et al., 2019). In the present study, the ST performance of 220 rice accessions from RDP1, representing ve subpopulations (tropic japonica/TRJ, temperate japonica/TEJ, indica/IND, aus /AUS, aromatic/ARO and admixture/ADMIX), has been evaluated at both seedling and reproductive stages. By using a gene-based / tightly linked GWAS analysis as reported in Liu et al. (2019), we identi ed 214 SNPs related to 251 genes that are signi cantly associated with 16 STrelated indices at both stages. Subsequently, fty-four rice accessions were proposed as donors with most valuable alleles in breeding and 31 promising candidate ST genes were recommended. These results provide useful information for further characterizing ST-related genes and for breeding ST varieties through MAS.

Results
Phenotypic variations at seedling and reproductive stages At the seedling stage, we found that most accessions began to wither and turn yellow on the 4 th day after salt treatment and died on the 6 th day after salt treatment, while a few accessions still showed acceptable growing status on the 6 th days after salt treatment (Fig. 1A, Supplementary Figure S1). According to the phenotypic distribution, we found that sRV-Chl6 and sRV-DR6 presented an apparently skewed distribution, while the rest four STrelated indices displayed a more scattered distribution (Supplementary Figure S1), demonstrating a significant variance of ST among accessions tested at the seedling stage. At the reproductive stage, ten ST-related indices were scored (Supplementary Table S1).
Na + content and Na + /K + ratio, which are commonly important indices in evaluating crop ST, clearly showed scattered distributions in the population, demonstrating varying responses to salt stress among accessions (Supplementary Figure S2). Similar trends were observed in 6 indices related to chlorophyll contents and Fv/Fm. Some accessions were severely affected by salt treatment, showing no panicle emergence, while some others exhibited apparently reduced panicle lengths (Fig. 1A). Therefore, the panicle length was measured as one of agronomical indices. We found that compared with the control, only a few accessions produced similar panicle lengths salt treatment conditions, and most had significantly shortened panicles including some without panicles (Supplementary Figure S2). The other agronomic index is seed-setting ratio (rRSD-SR), which showed a clearly scattered distribution in the population. Together, these data indicate that the accessions in this study display significant differences in the ST at both seedling and reproductive development stages. Correlation analysis showed that almost no significant correlations were found between ST indices from seedling stage and reproductive stage (Fig. 1B). This is consistent with previous studies, in which very few ST-related QTLs were reported conferring ST at both seedling and reproductive stages Hossain et al., 2015). In addition, we also conducted a multi-comparison of each ST index among 5 subpopulations, and only observed three traits that showed significant differences among some subpopulations (Supplementary Figure S3). One index was sRV-FW that significantly differed between ADMIX and TRJ subpopulations; and the other two were rRSD-SR and rRV-Fv/Fm7, which showed significant differences between AUS and TEJ subpopulations, and between AUS subpopulation and IND and TRJ subpopulations, respectively. This implies that the five subpopulations have almost no differentiation on ST.
In order to better portray the ST performance of the accessions, we conducted PCA for all measured ST-related traits. For the seedling stage, the PCA results showed that PC1 accounted for more than 45.2% of the total variance while PC2 was responsible for 18.8% (Fig. 1C). According to the length of each index line on PCA plot, the sRV-PH presented the strongest correlation with ST, followed by sRV-FW, sRV-Chl6, sRV-DR6, sRV-DR4 and sRV-Chl4. Negative correlations between photosynthesis and survival rates under stress have been well documented (Moradi and Ismail, 2007), consistent with our data that sRV-Chl4 and sRV-Chl6 were significantly negatively correlated with the sRV-DR4 and sRV-DR6, respectively, reflected by their opposite directions on the PCA plot. For the reproductive stage, the PCA results showed that PC1 and PC2 accounted for 56.5 % and 13.5% of the total variance, respectively (Fig. 1D). Comparatively, the two agronomic traits, rRSD-SR and rRSD-PL, directly associated with grain yield, showed strong correlations with ST, while other indices presented similarly moderate contributions to ST, according to the PCA analysis. The Na + content and Na + /K + ratio, commonly used for evaluating ST in seedlings (Zelm et al, 2020), did not appear to be strongly correlated with ST at the reproductive stage. Except for the correlation between rRSD-SR and rRV-Fv/Fm7, all other correlations among 10 indices reached a statistically significant level (Fig. 1B). Both Na + content and Na + /K + ratio showed apparently negative correlations with the rRV-Chl and rRV-Fv/Fm indices, while positive correlations were identified with the two agronomic traits, rRSD-PL and rRSD-SR, indicating that a higher content of Na + leads to a more severe hurt to the plants (Fig. 1B). Together, the PCA data indicate that different indices have divergent contributions to rice ST at different stages.

Genome-wide association study of ST-related indices
To elucidate the genetic variance and identify the potential gene(s) with ST at both seedling and reproductive stages, we conducted a gene-based / tightly linked GWAS for 16 ST-related indices. A total of 201332 SNPs located within genes and their flanking regions were identified, which related to 27923 unique genes (Supplementary Table S2). The average interval distance of these SNPs on genome is 1892 bp and ranges from 1543 bp to 2225 bp on different chromosomes (Chrs), showing a relatively even distribution on rice genome (Supplementary Figure S4). By using two statistical analysis models, CMLM and BLINK, we obtained a total of 252 and 508 SNPs significantly associated with ST, respectively, which covered 15 ST-related indices except for sRV-DR4 on which no significant SNP was detected (Supplementary Table S3-18). Comparatively, a total of 224 SNPs were consistently detected by both methods (Supplementary Table S19). After removing 10 SNPs that were repeatedly detected in different ST indices, 214 SNPs in total were chosen for further analysis.

Pyramiding effects of low-frequency favorable SNP-carrying genes
To identify favorable alleles/haplotypes with potential effects on ST improvement, we conducted a haplotype analysis for each SNP locus (Supplementary Table S19). According to the result of t test between the two genotypes in each SNP locus, we found that 172 SNPs related to 201 genes exhibited significant differences on corresponding ST index value, and the frequency for the favorable alleles ranged from 2.80% to 97.18% in the population (Supplementary Table S19). In particular, we are interested in those favorable alleles less than 50% in the population and consider them as low-frequency favorable (LFF) alleles, because they are not utilized in most varieties and so have a relatively high breeding value in ST improvement for most varieties. Certainly, those high-frequency favorable (HFF) alleles (>50% in the population) are also helpful to improve some varieties without them. After screening, a total of 82 LFF alleles (refer to 96 genes) with the frequency ranging from 2.80% to 50% were shortlisted, which included 57 and 25 (67 and 29 genes) at the seedling and the reproductive stages (Supplementary Table S20), respectively.
To assess whether these LFF alleles have pyramiding effects on ST improvement, we evaluated the relations between the ST index value and the number of LFF alleles. As shown in Supplementary Figure S5 and S6, an apparently linear relationship was observed between each of 16 ST index values and the number of LFF alleles, although the degrees of slopes varied among each other that reflect the magnitude of effects.
These additive effects inform that the adaptability of rice to salinity can be strengthened by combining LFF alleles that dominate ST-related indices. Based on this consideration, we then calculated the number of stage-specific LFF alleles accumulated in each accession. At the seedling stage, we found that no varieties contained more than half of the total LFF alleles and the highest number of LFF alleles accumulated in a variety was 26 (Fig. 3A). Most varieties harbored less than 10 LFF alleles and around 46% (number of 102) varieties possessed no more than 4 LFF alleles. At the reproductive stage, one variety carried the highest number of 13 LFF alleles, and up to 63% germplasms (139) contained less than 4 LFF alleles (Fig. 3B).
Considering the fact that LFF alleles are distributed in different germplasms, we selected the accessions that possessed top 15% stage-specific LFF alleles as promising germplasms at each developmental stage. At the seedling stage, 30 accessions carrying LFF alleles with the numbers ranging from 14 to 26 were identified as potential ST breed donors ( Fig. 3A), which covered all 57 LFF alleles (Supplementary Table S21). Among them, most accessions contained 14 to 18 LFF alleles and 7 accessions possessed more than 20 LFF alleles. At the reproductive stage, 25 accessions carrying at least 9 LFF alleles were identified as potential ST breed donors with 3 of them possessing more than 12 LFF alleles (Fig. 3B).
In order to better present the ST performance of these potential donors, we projected them on the PCA plot  (Singh and Flowers, 2010). Yet, we found an accession (NSFTV72 / IR8) that featured a longer life span in ST although it did not display the best ST at each of the two stages ( Fig. 3C, D). This variety accumulated 14 seedling LFF and 11 reproductive LFF alleles ( Fig. 3A, B), suggesting its potential value as a donor for simultaneously transferring multiple LFF alleles from both stages in future breeding programs.

Candidate genes associated with ST-related indices
Since all the SNPs applied in the above GWAS analysis are located in or very close to the annotated genes, we would like to mine some candidate genes for ST. Out of 214 significant SNPs, we found that 39 SNPs were located in the coding regions (Supplementary Table S23). These included 17 synonymous SNPs and 22 non-synonymous SNPs resulting in either amino acid substitutions or premature stop codons. The rest 175 SNPs were in non-coding regions, which can be further subdivided into three groups: 127 in intergenic regions, 5 in 5' UTR, 7 in 3' UTR and 36 in intron (Supplementary Table S23). Altogether, these 214 significant SNPs cover 251 genes, in which 222 are annotated and 29 encoded hypothetical proteins of unknown functions (Supplementary Table S19).
To prioritize GWAS-derived SNPs/genes, we conducted a co-localization analysis with STrelated QTLs published previously. Since most of these reported QTLs data came from independent studies that were not entirely consistent with each other, we performed a metaanalysis to identify consensus QTLs with high reliability. A sum of 375 ST-related QTLs from 18 independent studies were collected for meta-analysis, which included 308 in the seedling stage, 55 in the reproductive stage and 12 in vegetative and reproductive stages (Supplementary Table S24). In meta-analysis, QTLs in vegetative and reproductive stages were all considered as reproductive ST QTL. As a result, we obtained 41 seedling meta-QTL for ST (hereafter named as sMqST) and 17 reproductive meta-QTL for ST (rMqST), which were scattered on all 12 chromosomes (Supplementary Table S25, Fig. 4A). The interval length of these sMqST and rMqST ranged from 0.054 Mb to 4.734 Mb and from 0.139 Mb to 5.614 Mb, respectively (Supplementary Table S25). The average length of sMqST was 0.789 Mb and apparently less than 2.375 Mb of rMqST (Fig. 4B). Through comparison, we found a total of 45 SNPs located in these MqST intervals (Fig. 4A).
To screen candidate genes for ST, we consulted publicly available data on differentially expressed genes (DEGs) of rice under salt stress at diverse growth stages, which offered a list of 131 DEGs that related to 116 out of 214 SNPs (Supplementary Table S26). To the best of our knowledge, only one of them, OsPP1 (LOC_Os03g16110), encoding a protein phosphatase, has been reported as an ST regulation gene (Liao et al., 2016). Rice plants overexpressing OsPP1a can recover from salinity-induced oxidative damage via boosting antioxidant enzyme systems to maintain a relative redox homeostasis. SNP-3.8889557, associated with rRSD-SR on chromosome 3, was found to be located in the 3' UTR of OsPP1, which accounted for 14.48% of the total variation of rRSD-SR (Supplementary Table S23). Twelve accessions out of 208 (5.77%) carried the 'A' allele at this position with an average rRSD-SR of 56%, while the accessions possessing the alternative 'G' allele were more sensitive to salinity treatment, with an average rRSD-SR of 84% (Fig. 4C). In comparison, the favorable 'A' allele of OsPP1 was able to rescue 28% losses on seed setting rate due to salt stress during the reproductive stage. This implies that mining DEG data could be a feasible strategy to screen candidate ST genes.
By comparison among 45 SNPs located within MqST intervals, we identified 31 DEGs related to 26 SNPs (Table 1). The chromosome 3 had most co-located DEGs with MqST, and no co-located DEGs and MqST were found on chromosome 5, 6 and 11 (Fig. 4A). According to the gene annotation, we found that except for 3 genes belonging to hypothetical protein or predicted protein, most of the rest 28 DEGs presented potential function in various kinds of stress resistance, such as OsPRI1, OsSPL2, OsULT1, OsERF60, and so on (Table 1). It is worth further validating the function of these candidates in ST. In addition, fifteen of these candidates were from LFF alleles, whose contributions on ST improvement required testing in practice (Supplementary Figure S7). One of the candidate genes at each stage was chosen for further analysis below. LOC_Os08g04390, known as OsPRI1, was associated with SNP-8.2146875 in the interval of sMqST8-1. In rice, OsPRI1 is a linker between the iron-binding sensor OsHRZ1 and the Fedeficiency-responsive gene OsIRO2 . Under Fe deficiency, this HLH transcription factor positively regulates OsIRO2 expression via directly binding to its promotor. Through global bioinformatics analysis, the HLH transcription factor was found to participate in salt responsive gene regulatory networks . Overexpression of OsIRO2 results in an elevation of Fe uptake and translocation in rice (Ogo et al., 2007). Being an essential mineral element for plant growth and development, Fe has also been proposed as a defensor against salinity since it can boost the production of antioxidative enzymes to mitigate salinity-induced oxidative damage Sharma et al., 2012;Ghasemia et al., 2014). In addition, salt stress would impose deleterious effects on Fe acquisition and distribution in plants . Consequently, efficient absorption of Fe with the help of OsIRO2 and its upstream regulator OsPRI1 would have the potential to confer greater tolerance to saline-alkaline stress in rice. The accessions containing 'A' allele in the regulatory of OsPRI1 displayed a lower sRV-FW of 52%, while only 13 accessions carried the 'G' allele that showed a better sRV-FW value (Supplementary Figure S7). This demonstrated that introducing 'G' allele of OsPRI1 from the 13 accessions could further improve rice ST.
Another candidate gene is OsTOGR1 (LOC_Os03g46610), encoding a DEAD-box RNA helicase co-located with rMqST3-2. At higher temperature, the intrinsic helicase activity of OsTOGR1 increases, thereby promoting stabilization of pre-rRNA homeostasis . Although it is uncertain whether OsTOGR1 directly participates in ST through regulation of RNA surveillance, its two family members, OsPDH45 and OsSUV3, have been demonstrated to function in ST (Amin et al., 2012;Macovei and Tuteja, 2012;Tuteja et al., 2013). In particular, overexpression of PDH45 confers salt tolerance to rice at both seedling and reproductive stages. SNP-3.26377803, located in the CDS region of OsTOGR1, contains two kinds of variations 'G' and 'T'. Thirteen accessions containing the favorable 'T' allele significantly reduced the rRSD-SR of around 25% (Supplementary Figure S7), implying its breeding potential in ST breeding program.

Discussion
Identi cation of genetic resources for rice breeding towards salt tolerance Rice is relatively tolerant to salt during germination, active tillering, and maturity but is very sensitive at the early seedling stage and reproductive stage (Singh and Flowers, 2010;Ganie et al., 2019). To mitigate the adverse effects of salinity on rice, a prospective practice is to breed varieties with in-built salt tolerance at both sensitive stages and develop new rice varieties with enhanced resilience to salt stress environments. In general, ST at these two sensitive stages is not correlated to each other. Nor is there a statistically signi cant correlation between the ST-related indices measured at the seedling stage and the grain yield at the reproductive stage under salt stress (Moradi et al., 2003). Therefore, a comprehensive assessment of ST in rice must be based on determination of certain ST-related indices through the whole growth period, and combination of multiple ST QTLs/genes is suggested in the breeding practice. Considerable efforts have been devoted to the identi cation of ST QTLs/genes at each sensitive stage independently, such as the characterization of the Salto at the seedling stage (Thomson et al., 2010;Zeng et al., 2002;Mohammadi et al., 2013;Bimpong et al., 2014;Hossain et al., 2015). However, only a few attempts have been made to discover loci associated with ST across developmental stages and especially at the reproductive stage In this study, we evaluated the ST of 220 rice accessions representing ve subpopulations at both the seedling and the reproductive stages by assessing a combination of morphological and physiological parameters. Our results indicate that the degree of ST varies among this rice panel at both stages, while no correlation was found between ST-related indices measured at the seedling stage and the reproduce stage (Fig. 1B). Also, only few ST indices showed signi cant differences among subpopulations (Supplementary Figure S3). These allow us to infer that rice ST does not experience an arti cial or natural selection, although apparent difference on ST related indices can be observed among varieties. This could account for the fact that not a single dominating index could directly regulate rice ST especially among various development stages. As a result, we employed PCA to determine the contributions of different indices on ST, and found that the two morphological indices (rRSD-PL and rRSD-SR) displayed apparently stronger correlations with ST than those of the other 8 indices, including Na + and Na + /K + ratio at the reproductive stage. The rRSD-SR has been universally employed for evaluating rice ST in previous studies, while the rRSD-PL was seldom used. In future studies, the rRSD-PL could be emphasized in both ST-related genetic and breeding researches at the reproductive stage. To readily pinpoint the genes responsible for ST, a GWAS analysis was undertaken using 201k within-gene SNPs (Supplementary Table S2). A total of 117 and 97 SNPs were identi ed to be signi cantly associated with different ST-related indices at the seedling and reproductive stages (Fig. 2), respectively. However, we did not detect any overlapping SNPs at both stages, highlighting the independence of ST between these two stages and the genetic complexity.
The ultimate goal of these signi cant SNPs discovery is to deploy them in the breeding program to develop new salt-tolerant rice varieties or germplasm adaptable to salinity across the whole growth stages. A prospective practice is to combine diverse ST favorable alleles at both sensitive stages of rice. Comparatively, LFF alleles were on the priority list in view of their great potential in genetic improvement.
Among the 214 SNPs in total, we identi ed 57 and 25 LFF alleles at the seedling and the reproductive stages (Supplementary Table S20), respectively. Importantly, our preliminary results indeed show that rice accessions accumulating more LFF SNPs typically display relatively superior ST to those comprising fewer LFF SNPs (Supplementary Figure S5 and S7). This means that increasing the frequency of these LFF alleles in rice population could be a feasible approach to further improve rice ST. In order to accelerate the utilization of these LFF alleles, 30 and 25 potential donors were selected at seedling and reproductive stages, respectively, which collectively covered all 82 LFF alleles and displayed a good performance on ST at both stages (Fig. 3). We also noted some deviations where accessions carrying LFF alleles similar to those of other salt-tolerant lines exhibit phenotypic salt-sensitivity. This is probably due to the complex interactions among different genes or between genes and genetic backgrounds, which results in various phenotypes although these accessions harbored similar LFF alleles. Undoubtedly, pyramiding these LFF alleles using the potential donors discovered in this study with previously identi ed ST loci/genes will facilitate the development of highly tolerant varieties against salinity. In addition, these identi ed LFF SNPs could also be converted into breeder-friendly and cost-effective markers, like Kompetitive Allele Speci c Polymerase Chain Reaction (KASP) marker, for future molecular breeding.

Candidate genes involved in regulation of rice salt tolerance
Page 14/31 The majority of the 214 SNP-carrying genes have not been studied in the context of ST. By integration of DEG data and Meta-QTL results, we obtained 31 candidate genes that could play a crucial role in tolerance against salinity (Table 1). Besides the two candidates (OsPRI1 and OsTOGR1) described in the previous section, at least 4 more genes have already been reported with diverse biological functions although their contributions to ST have not been elucidated so far. Based on the prior literature, the current understanding of these 4 genes is discussed next with respect to their potential roles in ST.
LOC_Os09g34100 (SNP-9.20121667.), known as OsGLYII3, associated with sRV-DR6 at the seedling stage encodes a Glyoxalase II involved in the glyoxalase pathway (Yadav et al., 2007;Musta z et al., 2011) which is required for glutathione-based detoxi cation of methylglyoxal. Earlier studies have already revealed the potential role of this pathway in conferring ST of tobacco (Singla-Pareek et a l., 2003;2006): overexpression of glyI from Brassica juncea in tobacco imparts tolerance to the transgenic plants during salt stress (Veena et al., 1999); overexpression of glyI and glyII together can enhance the ST of transgenic plants to a higher extent (Singla-Pareek et a l., 2003;2006). In rice, OsGLYII3, together with OsGLYI1, were identi ed as salt responsive gene (Veena et al., 1999;Yadav et al., 2007), with increased transcription levels upon salt stress.
Leaf segments from OsGLYII3-overexpressing lines could retain a larger amount of chlorophyll in the presence of salt as compared to the control (Singla-Pareek et a l., 2008). Consequently, OsGLYII3 is prominent causative gene for ST.
LOC_Os02g45660 (SNP-2.27769525.), known as OsIspF, is associated with sRV-PH at the seedling stage and encodes a 2-C-methyl-D-erythritol 2,4-cyclodiphosphate synthase, which is the fth enzyme in the MEP pathway (Huang et al., 2018). Located in chloroplast, OsIspF plays a pivotal role in chloroplast development. Rice plants carrying mutated forms of OsIspF display yellow-green leaf phenotype throughout development, with signi cantly reduced contents of Chlorophyll and carotenoid. Relationships between photosynthesis, chloroplast and survival rates of rice under salinity have been well documented (Moradi and Ismail, 2007;Nan et al., 2020); therefore, OsIspF could be a prominent gene for ST.
LOC_Os09g29584 (SNP-9.17990633), also named OsWAK84, is associated with sRV-DR6 at the seedling stage and encodes a wall-associated kinase. Wall-associated kinases in rice have been proposed to be involved in plant immunity (Li et al., 2009;Delteil et al., 2016;Harkenrider et al., 2016;Hu et al., 2017) and response to mineral toxicities (Xia et al., 2018). In contrast, there has not been a single report that reveals their relationships with rice ST. However, it was reported that an Arabidopsis WAK-like gene 4 (AtWAKL4) responded notably to salt stress, with a 5-fold increase in AtWAKL4 transcripts after salt treatment of the seedlings, while destruction of its promoter via T-DNA insertion extenuated the adaptability of Arabidopsis to salt stress (Hou et al., 2005).
LOC_Os03g44660 (SNP-3.25139625), known as OsGRL7, is associated with rRV-Chl7 at the reproductive stage and encodes a Glutaredoxins (GRXs) family protein of GRL-type (Garg et al., 2010). Though the precise functions of GRL-type proteins in ST have not been investigated, overexpression of a CPYC-type GRX-OsGRX20 was shown to result in enhanced tolerance to salt stress (Ning et al., 2018). It was proposed that the capacity of GRXs to reduce oxidized disul des is a decisive factor to maintain redox homeostasis under high salinity (Rouhier et al., 2005;Ning et al., 2018).
The other candidate ST genes jointly supported by DEG data and Meta-QTL results have been hardly studied in the previous literature. In summary, our ndings are anticipated to add potential novel gene resource to the list of ST contributing factors. Further functional tests will be needed to validate their roles in ST.

Plant materials
A total of 220 rice accessions including 5 subpopulations from RDP1 were selected, and comprised 53 TRJ,  Table S1). Twenty germinated seeds of each variety were placed in a 96-well PCR plate cut at the bottom and put into a container for hydroponic culture in greenhouse. At two-leaf stage, the seedlings were transferred to the IRRI nutrient solution with pH 5.5 (Xia et al., 2015). The solution was changed every three days till four-leaf stage, and then the 0.8% NaCl (136.8 mmol/L) was used for salt treatment. The nutrient solution without NaCl was used as the control.
The same rice germplasm as the seedling stage was soaked and seeded in a 32-well seedling tray. At fourleaf stage, the seedlings were transplanted into 15 L plastic buckets with soil in greenhouse. Each plastic bucket contained 5 plants, and at jointing stage 4 plants with the same growth status were retained. Then at booting stage, each barrel was treated with 0.5% NaCl (85.5 mmol/L). Each variety treated by clean water was used as the control.

Measurement of ST related traits or indices at seedling and reproductive stages
At the seedling stage, on the 2 nd , 4 th and 6 th day after salt treatment, the second leaf from the top of the seedling was selected and the chlorophyll content of the leaf was measured by SPAD 502 chlorophyll meter.
At the reproductive stage, the chlorophyll content of rice ag leaf was determined on the 0, 7 th , 14 th and 21 st days of salt treatment. On the 6 th day after salt treatment at the seedling stage, the plant height, root length and fresh weight were measured. On the 4 th and 6 th days after salt treatment, the number of dead seedlings under salt stress was counted, and then the death rate was calculated. 5 to 8 seedlings with the same growth status were selected as a replicate, each treatment along with the control was repeated 3 times.
At the reproductive stage, the ag leaves of rice were selected on the 7 th , 14 th and 21 st day after salt treatment to determine the chlorophyll uorescence parameters, respectively. After dark adaptation to 30 min, the initial uorescence (Fo) was measured with weak measuring light, and then the maximum uorescence (Fm) was measured with a strong ash (5000 μmol·m -2 ·s -1 , pulse time 0.7 s). The variable uorescence (Fv) was calculated as Fv = Fm-Fo, as well as the maximum photochemical e ciency of photosystem II (PSII) (Fv/Fm), and the potential photochemical e ciency of PS II (Fv/Fo). After 14 days of salt treatment, the third leaf from the top of each plant was sampled and dried. The dried sample was crushed by a grinder, and 0.1 g of sample was placed in a test tube with a lid, then 10 mL of 100 mM acetic acid was added and bathed at 90 ℃ for 2 h. After cooling and 5000 rpm centrifugation for 15 min, the supernatant was transferred to the 10 mL centrifuge tube. After proper dilution, the contents of Na and K were determined by atomic absorption spectrophotometer (Pin AAcle 900F, Perkin Elemer Co., Ltd). The panicle length and the number of grains per panicle of each variety were measured at the end of grain lling, and the seed setting rate was calculated. Four plants per bucket were set as a replicate. Each treatment along with the control was repeated 3 times.
According to the method of Qi et al. (2005), Excel 2016 and R 3.6.3 were used for data processing and correlation analysis.
Manhattan and Q-Q plots were generated with the "CMplot" package in R environment (Yin et al, 2020).

Differential gene expression analysis
A total of 25 rice salt-tolerant transcriptome datasets from 7 independent studies were collected from GEO database (https://www.ncbi.nlm.nih.gov/geo/). R package "limma" was used to mine differential expression genes (Ritchie et al, 2015). Transcripts or genes with P-values < 0.05 and absolute value of log2 (fold change) > 1 were assigned as differentially expressed genes (DEGs).

Meta-analysis of rice ST QTLs
Information of ST QTLs at seedling and reproductive stages were collected from 18 references published from 2004 to 2020, which included QTL con dence interval, linkage molecular markers, LOD value, phenotypic variation rate (PVE%), and mapping parents (Supplementary Table S24). According to the positions of the molecular markers linked with these ST-responsive QTLs on rice reference genome of Nipponbare, an integrated physical map for all ST responsive QTLs and their linked molecular markers was obtained. Then, based on the ratio of 1 cM per 244 kb (Chen et al., 2002), this integrated physical map was further converted to a virtual genetic map, namely a consistent genetic map, which included 616 and 134 markers linked with ST QTLs at seedling and reproductive stages, respectively. This conversion is based on the assumption that the ratio of genetic to physical distance is invariant throughout the genome except that in the centromeric regions, which has been used in previous studies for meta-analysis of QTLs related to several traits (Courtois et al., 2009). According to the information of all ST responsive QTLs integrated in this consistent genetic map, the Meta-QTL analysis was performed by using the software BioMercator V4.2 (Sosnoswki et al., 2012;Veyrieras et al., 2007). The method relies on a clustering algorithm based on a Gaussian mixture model and enables the determination of the probable number of clusters considered as the "true" QTLs underlying the QTLs observed in a given region. The optimal number of clusters was chosen by means of an information-based criterion, and the position and con dence interval of the meta quantitative ST loci (MqST) was then estimated. rRV-Na:the RV of Na + content; rRV-Na/K:the RV of Na + /K + content; rRSD-SR:the RSD of seed-setting rate;

Abbreviations
rRSD-PL:the RSD of panicle length; rRV-Chl7:the RV of chlorophyll content on the 7th day; rRV-Chl14:the RV of chlorophyll content on the 14th day:rRV-Chl21:the RV of chlorophyll content on the 21st day; rRV-

Consent for Publication
Not applicable.

Availability of Data and Materials
All relevant data in this study are included in the article and its supplementary les.

Competing Interests
The authors declare that no competing interests exist.
Funding Figure 1 Phenotype and PCA analysis of indices under salt stress at seedling and reproductive stages A, Comparison of rice accessions with (+) and without (-) NaCl treatment. Pictures for seedling stages were taken at 4 days after treated with and without 0.8 % NaCl. Pictures for reproductive stages were taken at 14 days after treated with and without 0.5 % NaCl. Left and right pictures in each panel refer to different rice accessions with different tolerance to salt stress at seedling or reproductive stages, respectively. B, Pearson correlation coe cient matrix of 16 ST indices. The cross indicates that the correlations did not reach statistically signi cant level at P < 0.05. Red and blue refer to positive and negative correlation, respectively. C and D, PCA analysis of ST indices at seedling and reproductive stages, respectively. Total of 220 accessions were plotted on PCA plot with different shapes and colors. Each arrow represents an index and the length refers to its contribution on ST. The angles captained by any of the two arrows less than 90º imply the two indices having positive correlation, otherwise represent negative correlation between the two indices.

Figure 2
Distribution of signi cant SNPs associated with different ST indices on chromosomes Green and red dots indicate the corresponding SNP explained phenotypic variation less than 10% and more than 10%, respectively. The arrows indicate the SNPs explain more than 14% total phenotypic variation of the corresponding ST index. Triangles indicate the SNPs associated with two different ST indices. The length of chromosome was dimensionally drew based on the physical length of each chromosome (http://rice.plantbiology.msu.edu). Comparison of the signi cant association SNPs, Meta-QTL and differentially expressed genes A, Integrated map for association SNPs, Meta-QTL intervals and differentially expressed genes (DEGs). Circles and triangles on the left side of each chromosome represent LFF and not LFF SNPs, respectively. Green and pink colors indicate the association SNPs detected at seedling and reproductive stages, respectively. Squares on the right side of the chromosomes represent the DEGs. Red and blue vertical lines refer to meta-QTL intervals of seedling and reproductive stages, respectively. B, Distribution of interval length of meta-QTL in reproductive stage (rMqST) and in seedling stage (sMqST). C, Haplotype analysis of the SNP-3.8889557 associated differentially expressed gene OsPP1. The n=192 and n=12 indicate the number of the accessions harbored 'G' and 'A' haplotypes, respectively. The average rRSD-SR value of the accessions containing each of the two haplotypes reached statistically signi cant difference at p<0.05.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementalFigures47.pptx TableS126.xlsx