Molecular dissection of maize seedling salt tolerance using a genome‐wide association analysis method

Summary Salt stress is a major devastating abiotic factor that affects the yield and quality of maize. However, knowledge of the molecular mechanisms of the responses to salt stress in maize is limited. To elucidate the genetic basis of salt tolerance traits, a genome‐wide association study was performed on 348 maize inbred lines under normal and salt stress conditions using 557 894 single nucleotide polymorphisms (SNPs). The phenotypic data for 27 traits revealed coefficients of variation of >25%. In total, 149 significant SNPs explaining 6.6%–11.2% of the phenotypic variation for each SNP were identified. Of the 104 identified quantitative trait loci (QTLs), 83 were related to salt tolerance and 21 to normal traits. Additionally, 13 QTLs were associated with two to five traits. Eleven and six QTLs controlling salt tolerance traits and normal root growth, respectively, co‐localized with QTL intervals reported previously. Based on functional annotations, 13 candidate genes were predicted. Expression levels analysis of 12 candidate genes revealed that they were all responsive to salt stress. The CRISPR/Cas9 technology targeting three sites was applied in maize, and its editing efficiency reached 70%. By comparing the biomass of three CRISPR/Cas9 mutants of ZmCLCg and one zmpmp3 EMS mutant with their wild‐type plants under salt stress, the salt tolerance function of candidate genes ZmCLCg and ZmPMP3 were confirmed. Chloride content analysis revealed that ZmCLCg regulated chloride transport under sodium chloride stress. These results help to explain genetic variations in salt tolerance and provide novel loci for generating salt‐tolerant maize lines.


Introduction
Soil salinization is a globally devastating environmental problem resulting from both natural and artificial processes, such as mineral weathering and irrigation. Salinity affects more than 800 million hectares of land, accounting for more than 6% of the world's land area (Munns and Tester, 2008). Various types of ions are responsible for salination, including sodium, potassium, calcium, magnesium and chloride (Zhu, 2016). Sodium chloride (NaCl) is the most abundant salt in soil because of its high solubility and ubiquitous distribution (Munns and Tester, 2008). Soil salinity has many negative effects on plant growth and development, including inhibition of seed germination, reduced root growth, plant height and fruiting levels, which ultimately decreases crop yield and quality (Sandhu et al., 2020). More seriously, sensitive crops can be killed by even slight salinity in soils (Luo et al., 2019b). The bases of these phenomena have been dissected using molecular genetics analyses. Such analyses have shown that various physiological and metabolic functions in plants are impaired by osmotic, ionic and oxidative stresses under saline conditions (Muchate et al., 2016).
Because of the harmful effects of salinity stress on plant growth and crop yields, it is necessary to exploit the genetic basis of variability in salt tolerance to improve plants' resistance to salt toxicity. Quantitative trait locus (QTL) mapping has revealed many genomic regions that affect important traits. In maize, QTL analyses of the germination rate, salt tolerance ranking, shoot fresh and dry weights, shoot K + /Na + ratio and Na + and K + concentrations in shoots were conducted on seedlings of 161 F 2:5 lines under hydroponic culture and saline field conditions. In total, 38 salt tolerance-related QTLs were detected on chromosomes 1, 3 and 5, with eight being major QTLs that individually explained more than 20% of the phenotypic variation (Cui et al., 2015). Using plant height in a saline field and a plant height-based salt tolerance index (plant height in a saline field/plant height in a normal field) as salt tolerance indicators, QTLs were detected in mature field-grown maize plants of 240 double-haploid lines. A major QTL, qSPH1, which was responsible for the two salt tolerance-related traits, was identified on chromosome 1 and explained 25.9%-31.2% of the phenotypic variation (Luo et al., 2017b). Some QTLs associated with salt tolerance have also been identified in other crops, such as rice (Zeng et al., 2021), wheat (Luo et al., 2021), soybean (Guan et al., 2014) and tomato (Frary et al., 2010). The results of those QTL mapping studies have provided insights into the genetic mechanisms of plant salt tolerance, which may be useful for breeding salt-tolerant crops.
A number of functional genes and transcriptional factors related to salt tolerance have been reported. Many of them are involved in ion homeostasis maintenance, osmotic protection, antioxidant regulation, hormonal regulation and Ca 2+ signalling pathways (Muchate et al., 2016). For example, in the model plant Arabidopsis, the salt overly sensitive (SOS) signalling pathway (comprising SOS1, SOS2 and SOS3) involved in salt tolerance has been well characterized (Zhu, 2002;Zhu, 2016). In this pathway, SOS3 senses the Ca 2+ signal elicited by ion stress, and then interacts with SOS2. The activated SOS2 phosphorylates the plasma membrane Na + /H + antiporter SOS1, and the activated SOS1 extrudes Na + from the cytosol (Zhu, 2016). In rice, the SKC1 gene, which encodes a high affinity K + transporter (HKT), has been successfully isolated through map-based cloning. Overexpression of SKC1 was shown to reduce the Na + content and increase the K + content in the shoots of salt-stressed rice (Ren et al., 2005). In maize, ZmHKT1, which also encodes an HKT-type transporter, has been isolated from the major maize salt tolerance QTL, ZmNC1 (logarithm of odds, LOD = 12.51). A gene knockout analysis verified that ZmHKT1 is important for Na + homeostasis and salt tolerance in maize . However, despite the dissection of numerous genetic loci, the molecular basis of salt tolerance in plants is still far from being completely understood.
A genome-wide association study (GWAS) is a powerful method to study genetic variations associated with complex traits at the genome-wide level (Huang et al., 2010;Luo et al., 2019b;Xie et al., 2019). This method takes into account the historical recombination found in broad panels of diverse germplasm and population-wide linkage disequilibrium (LD) among single nucleotide polymorphisms (SNPs) and QTLs. Thus, a GWAS can circumvent the limiting low recombination rate of bi-parental populations to identify a wider range of genetic variations and provide complementary information (Lu et al., 2018). Recently, GWAS technology has been used to dissect multiple complex trait-related mechanisms in various plant species. In rice, a metabolic GWAS identified 36 candidate genes that modulate the levels of physiologically and nutritionally important metabolites . In soybean, a GWAS led to the identification of 11 candidate genes related to days to flowering, maturity and plant height . Additionally, using a GWAS, the natural variation of ZmVPP1 encoding one vacuolar-type H + pyrophosphatase was identified, which contributed significantly to drought tolerance of maize (Wang et al., 2016). Four stable QTLs were identified in maize, which played critical roles in controlling arsenic accumulation .
In the present study, a high-density SNP-based GWAS analysis was performed under normal and salt stress conditions to detect natural variations in alleles related to the salt stress response during the maize germination stage. The main purpose of this study was to investigate significant alleles and potential candidate genes associated with shoot and root traits related to salt tolerance. The findings of this study shed light on the molecular mechanisms associated with variations in salt tolerance among maize inbred lines and may assist in the development of molecular markers for the improvement of salt tolerance in maize.

Phenotypic variation analysis
The maize diversity panel consisting of 348 accessions collected from the USA, China, and CIMMYT has been used for several GWAS analyses Liu et al., 2017;Yang et al., 2011b) (Table S1). In this study, phenotypic data for nine important growth-related traits [shoot length (SL), root length (RL), full length of seedling (FL), shoot fresh weight (SF), root fresh weight (RF), full fresh weight of seedling (FF), shoot dry weight (SD), root dry weight (RD) and full dry weight of seedling (FD)], were collected from the association population under normal and salt-stressed conditions. The salt tolerance indexes of these nine traits were calculated by dividing the values measured under salt stress conditions by the values measured under control conditions. The phenotypic frequencies of all 27 traits (nine traits under normal conditions and salt-stressed conditions, and the salt tolerance indexes of these traits) exhibited normal or near-normal distributions ( Figure S1). The average values of SL, RL, FL, SF, RF, FF, SD, RD and FD were 4.03 cm, 5.83 cm, 9.87 cm, 0.12 g, 0.18 g, 0.30 g, 0.018 g, 0.02 g and 0.04 g, respectively, under salt stress conditions, compared with 11.10 cm, 11.27 cm, 22.37 cm, 0.35 g, 0.33 g, 0.67 g, 0.04 g, 0.03 g and 0.07 g, respectively, under control conditions. The average values of the salt tolerance indexes (R) of the nine traits were as follows: SL (SLR = 0.38), RL (RLR = 0.57), FL (FLR = 0.47), SF (SFR = 0.36), RF (RFR = 0.56), FF (FFR = 0.46), SD (SDR = 0.46), RD (RDR = 0.64) and FD (FDR = 0.54). All 27 traits showed a wide range of phenotypic variation, and their coefficients of variation were all greater than 25%. The repeatability for all traits was high (74.31%-97.26%) ( Table 1).
The phenotypes of three significantly salt tolerant (CI7, CIMBL115 and GEMS37) and three significantly salt sensitive (CIMBL127, CIMBL157 and BY807) maize inbred lines were shown in Figure 1a. In general, the nine measured traits (SL, RL, FL, SF, RF, FF, SD, RD and FD) were significantly lower (35.6%-63.7% decrease, P < 0.001) under salt treatment conditions than under control conditions. The SL, SF and SD showed greater decreases than RL, RF and RD in response to salt stress, consistent with previous reports (Luo et al., 2017a;Luo et al., 2018) (Figure 1b-e). The nine measured traits of seedlings were significantly and positively correlated with each other (P < 0.001) under salt stress and control conditions. As expected, the salt tolerance indexes of the nine traits were negatively correlated with the nine traits under control conditions. The correlation coefficients of the traits under the same conditions were relatively higher than the correlation coefficients of the traits between different conditions. Furthermore, the correlation coefficients between traits under salt and control conditions were the greatest in all correlation coefficients among traits under salt treatment, under control treatment and their salt tolerance indexes ( Figure S2).

GWAS mapping
In total, more than 1.06 million high-quality SNPs were obtained from an RNA-seq project and the MaizeSNP50 BeadChip (Fu  1937-1951Ganal et al., 2011). The SNPs were filtered with a minor allele frequency <0.05, and the remaining 557 894 SNPs were used for the association analysis. To determine the optimal model for the association analysis, three models (K, Q, Q + K) were compared using quantile-quantile plots ( Figure S3). The Q and Q + K models resulted in a greater control of false-negative errors, and the Q + K model was more reliable than the Q model. Moreover, by combining the two covariates of population structure (Q) and K, the Q + K method can effectively control type I errors (false positives) (Lu et al., 2018;Zhang et al., 2016). Therefore, all subsequent GWAS analyses were performed using the Q + K model. Manhattan plots for all traits are shown in Figure S4. The quantile-quantile (QQ) plots for all traits were shown in Figure S5.
A total of 183 SNP-trait associations with P < 1.79 × 10 −6 were identified (Table S2), and they involved 149 unique SNPs. According to the LD decay distance of this maize population, a 200-kb region (AE100 kb) around each significant SNP was defined as a QTL (Deng et al., 2017;Wang et al., 2019). The QTLs with overlapping intervals for the same traits were merged. In this way, we identified 83 QTLs related to salt stress and 21 QTLs in the control, with an average of 4.6 and 3.0 loci for each trait, respectively (Table S3; Figure 2a-c). Briefly, 36 QTLs (42 significant SNPs) were identified for the nine traits under salt treatment conditions, and the proportion of phenotypic variation (R 2 ) explained by each locus ranged from 6.9% to 11.2%, with a mean of 7.6%. In the control, 21 QTLs (29 significant SNPs) were identified for seven traits, and the proportion of phenotypic variation explained by each locus ranged from 6.7%-9.7%, with an average of 7.6%. Based on the salt tolerance indexes of the nine traits, 47 QTLs (78 significant SNPs) were identified, with R 2 values ranging from 6.6%-10.5% (average, 7.2%).
The co-localization of QTLs identified from multiple traits was summarized ( Figure 2; Table S4). Overall, 11 QTL intervals were simultaneously detected by multiple salt-related traits (Figure 2a

Candidate gene prediction and expression profiling
The physical positions of the 83 salt-related QTLs from the maizeGDB database (www.maizegdb.org, B73 RefGen_v2) were searched to identify the genes present within these QTL regions. A total of 420 genes were obtained. Based on their annotations in the maizeGDB, Gramene and TAIR databases, 16 genes were found to be involved in tolerance to salinity or water deficiency. By analysing the LD of significant SNPs associated with candidate genes, we found that three genes were outside the LD regions of their significant SNPs (r 2 < 0.1). Therefore, they were excluded from the candidate genes. The remaining 13 genes were considered as candidate genes (Table 2).
GRMZM2G477325, which encodes a plasma membrane protein 3 (PMP3) on chromosome 7, was within the QTL simultaneously detected by five salt-associated traits (Table S4). This gene is known to be involved in ion homeostasis in yeast mutant under salt stress . Another gene, GRMZM2G071119 (ZmCLCg), located on chromosome 2, encodes an unknown protein. However, its Arabidopsis ortholog is a voltage-gated chloride channel that participates in chloride transmembrane transport (Nguyen et al., 2016), indicating that it is a likely candidate gene for chloride transport in maize. Four candidate genes (GRMZM2G102754, GRMZM2G176085, GRMZM2G027351 and GRMZM2G166049) located within the hot spot on chromosome 5 (4.56-19.35 Mb) ( Figure 2b). The Arabidopsis orthologs of GRMZM2G102754 and GRMZM2G176085 play roles in ethylene signal transduction  and oxidative stress responses (Valério et al., 2004) under salt stress, respectively. GRMZM2G027351 and GRMZM2G166049 encode a calcium-dependent protein kinase 18 (cdpk18) (Mittal et al., 2017) and a btb/poz domain protein (mab17, math-btb17) (Juranić et al., 2012), respectively. The Arabidopsis ortholog of GRMZM2G166049 is involved in homeobox-leucine zipper transcription factor degradation (Lechner et al., 2011).
The expression levels of ZmCLCg (GRMZM2G071119) and ZmPMP3 (GRMZM2G477325) in shoot of B73 under salt stress were analysed. Results showed that ZmCLCg and ZmPMP3 were responsive to salt stress in B73. In shoot, the expression of ZmCLCg was up-regulated at 8 h ( Figure 3b) while ZmPMP3 was up-regulated at 1 h post salt treatment (Figure 3c).

Functional verification of candidate genes ZmCLCg and ZmPMP3
After culturing in control or saline water for 10 days, growth parameters of B104, B73, the significantly salt-tolerant maize inbred line GEMS37, and the significantly salt sensitive maize inbred line CIMBL157 were compared ( Figure S6a-e). The RL, RF, SL and SF of GEMS37, B104 and B73 were significantly higher than those of CIMBL157. The salt tolerance indexes of RL, RF, SL and SF in GEMS37 were 59.8%, 41.8%, 41.6% and 49.3%, respectively. In B104, they were 60.9%, 57.7%, 48.1% and 46.3%, respectively. In B73, they were 53%, 49.0%, 43.9% and 35.6%, respectively. In CIMBL157, they were 32.3%, 23.6%, 10.0% and 11.4%, respectively. The growth parameters under salt stress and the salt tolerance indexes of B104 and B73 were close to that of salt tolerant maize inbred line GEMS37 ( Figure S6a-e). These results suggested that both B104 and B73 were salt tolerant maize inbred lines.
To validate the function of candidate gene ZmCLCg in salt tolerance, the maize inbred line B104 with ZmCLCg gene knock out was generated by CRISPR/Cas9 technology. The CRISPR/Cas9 vector used for ZmCLCg editing contains a synthetic polycistronic gene with three tandem tRNA-gRNA structures, which has the advantage of producing multiple mature gRNAs through endogenous tRNA-processing system (Xie et al., 2015) (Figure 4a). Three 20-bp sequences in the first, third and fourth exons of ZmCLCg were chosen as Cas9-gRNA cleavage sites (Figure 4a,b). PCR and sequencing analysis identified one mutant plant in the third exon (Mutation efficiency = 10%) and 7 mutant plants in the fourth exon (Mutation efficiency = 70%) among 10 independent T0 transgenic lines. These mutants were self-pollinated to obtain homozygous mutant maize plants. Three zmclcg knock-out mutants and the wild type were used to investigate their biomass parameters under salt stress. The zmclcg-1 has single base insertion mutation at the target site of exon 4, the zmclcg-2 has single base deletion at exon 3 and single base insertion at exon 4, and the zmclcg-3 has 28-bp deletion mutation at the target site of exon 4 (Figure 4c). These mutations will lead to frame shift mutations in ZmCLCg gene. All three zmclcg mutants showed a greater reduction in root length, root fresh weight, shoot length and shoot fresh weight compared with that of the wild type under 100 mM NaCl treatment (Figure 4d-h). These results suggested that ZmCLCg conferred salt tolerance in maize. Under 100 mM NaCl treatment, the chloride content in shoots of three zmclcg mutants was significantly higher than that of wild type  (Figure 4i), indicating that the salt tolerance function of ZmCLCg was associated with chloride transport.
To verify the salt tolerance function of candidate gene ZmPMP3, one B73 EMS mutant (Mut_Sample: EMS4-0a0498) with termination mutation in the second exon of ZmPMP3 was obtained from maize EMS mutant library (http://www.elabcaas.c n/memd/) (Figure 5a,b). The biomass of the zmpmp3 mutant and its wild type was determined under control and salt treatment. We observed that the zmpmp3 mutant showed significantly decreased root length, root fresh weight, shoot length and shoot fresh weight than the wild type under 100 mM NaCl condition (Figure 5c-g). These results indicated that ZmPMP3 played a role in maize salt tolerance.

Discussion
Maize is sensitive to salt stress, but is often planted on saltcontaminated land because most farmland is salinized. Therefore, elucidating the genetic architecture of salt tolerance in maize is instrumental for improving its salt tolerance. In this study, a GWAS analysis based on SNP markers was used to dissect the genetic basis of salt-related traits, including seedling length, fresh weight, dry weight and salt tolerance indexes. Multiple QTLs were identified for each trait, with some QTLs being simultaneously detected from several salt-related traits. Moreover, candidate genes involved in salt tolerance were predicted and analysed, which may serve as potential targets in studies on the molecular mechanisms underlying salt tolerance in maize.
All of the traits were seriously inhibited under salt stress (Figure 1), consistent with previous reports (Luo et al., 2019a;Luo et al., 2017a;Luo et al., 2018). The pairwise correlations among nine growth parameters in both salt-stressed and control conditions were significantly positive (r = 0.22-0.96), demonstrating strong genetic correlations among them. The salt tolerance indexes of some traits were weakly correlated with the growth parameters in control and salt stress conditions ( Figure S2), indicating that different genetic mechanisms may control these traits.
In previous studies, GWAS analyses have successfully identified genomic regions associated with tolerance to various abiotic stresses Zhao et al., 2018). In the present study, the 2300-Mb whole maize genome was covered by 557 894 high-density SNPs, with an average interval of 4.1 kb between SNPs, allowing for fine-resolution QTL mapping. Population structure is a main limitation in GWAS studies because it induces false-positive associations. Several statistical methods were evaluated to reduce false positives (Lu et al., 2018;Zhang et al., 2016), and the results suggested that the Q + K mixed linear model had a greater ability to correct both false-positive and false-negative associations than the K and Q models. The QQ plot for SLS generated by three models we compared in this study ( Figure S3). Consistent with previous studies (Lu et al., 2018;Zhang et al., 2016), results indicated that the Q as well as the Q + K model could better control the false-negative errors than the K model, and the Q + K model was more reliable than the Q model. Therefore, we used K + Q model for GWAS analysis in this study.
Through the association mapping of root and shoot traits under control conditions, 21 QTLs were identified across all of the chromosomes, and 16 of the QTLs were related to root growth. Six QTLs associated with root growth matched to previously reported QTL regions (Burton et al., 2014;Cai et al., 2012;Li et al., 2015). Four QTLs controlling root length (located in 45.42-45.62-Mb and 37.56-37.76-Mb regions on chromosome 1, 171.53-171.73-Mb region on chromosome 4, and 149.72-149.92-Mb region on chromosome 6) co-localized with qSolPriLen1, qARL21-1, qTRL14-1 and qARL26-1, respectively, which were identified from linkage populations in previous studies (Burton et al., 2014;Cai et al., 2012). Another QTL for RF, which was located at the 105.95-106.15-Mb region on chromosome 10, co-localized with qRDW110-1 for RD, which was identified from a BC 4 F 3 maize population (Cai et al., 2012). A QTL localized in the 136.76-136.97 Mb region on chromosome 7 co-localized with qRDW7 for RD, which was identified from a recombinant inbred line population . Two novel QTLs were detected by multiple traits (Table S4). Traits that shared the same QTLs were closely and significantly correlated with each other (r = 0.72-0.92), consistent with a previous study (Lu et al., 2018). Thus, the association analysis data appeared to be reliable. The novel QTLs for maize seedling growth identified in this study might enhance our understanding of the genetic basis for maize growth and development.
Only a few studies have tried to genetically map salt tolerance in maize (Cui et al., 2015;Luo et al., 2019b;Luo et al., 2019a;Luo et al., 2017b;Sandhu et al., 2020;Xie et al., 2019). The GWAS analysis in this study builds on the results of those studies and allowed us to identify salt tolerance-related SNP markers throughout the genome. Based on 18 traits, 83 QTLs distributed across all 10 maize chromosomes were significantly associated with salt tolerance. The proportion of phenotypic variation explained by individual significant SNPs was less than 11.2%, implying that salt tolerance is a complex minor-effect quantitative trait. Among the QTLs, 11 matched to those reported in previous studies (Cui et al., 2015;Luo et al., 2019b;Luo et al., 2019a;Luo et al., 2017b). For example, one QTL covering the region from 147.06 to 147.26 Mb on chromosome 1 was located within qSPH1 (Luo et al., 2017b), QStr1 (Cui et al., 2015), QTwc1 (Cui et al., 2015) and QSkcs|skcn1 (Cui et al., 2015), which were detected in linkage analyses, implying that our results were reliable. The significant SNPs of chr3.S_3201547 and chr3.S_3206938 for survival rate identified by Luo et al. (2019b) had strong LD with the significant SNP of chr3.S_3208836 for root length under salt stress identified in this study (r 2 > 0.2). In addition, 11 QTLs were identified by multiple traits (Table S4), suggesting that they had pleiotropic effects on salt tolerance. As expected, traits that shared the same QTLs had significant positive correlations with one another (r = 0.426-0.962).
The most promising candidate genes were identified based on their functional annotations and their homologs as screening references. Finally, we identified 13 candidate genes located in 12 corresponding QTLs (Table 2). The phenotypic differences reached significant levels (P < 0.001 or 0.01) between the two alleles of each of the strongest trait-associated SNPs ( Figure S8). The ZmPMP3 (GRMZM2G477325) was reported to regulate ion homeostasis in yeast mutant under salt stress . It was located within a QTL containing three leading SNPs (chr7.S_168404086, chr7.S_168404089 and chr7.S_168405013)  (Table S4). The biomass of the zmpmp3 mutant decreased significantly under 100 mM NaCl condition compared to its wild type, which verified that ZmPMP3 conferred maize salt tolerance ( Figure 5). Another candidate gene, GRMZM2G071119 (ZmCLCg) was located in a QTL (5.74-5.95 Mb, chromosome 2) harbouring 11 peak SNPs ( Figure S8a). The AT5G33280 encodes a voltage-gated chloride channel responsible for chloride transmembrane transport in Arabidopsis (Nguyen et al., 2016), and ZmCLCg is homologous to AT5G33280, indicating that it is a promising target gene for salt tolerance. Maize zmclcg knock-out mutants were generated using the CRISPR/Cas9 technology, and we found that the biomass of three zmclcg mutants were significantly lower than that of the wild type under salt stress, which confirmed the role of ZmCLCg in maize salt tolerance (Figure 4). Thus, these results verified the accuracy of our GWAS results and indicated that GWAS is an effective strategy to uncover DNA regions related to salt tolerance in maize.
Two genes encoding transcription factors, myb73 (GRMZM2G121570) and ereb137 (GRMZM2G028386), were located within two loci that explained more than 10% of phenotypic variation (Table 2). Moreover, these two regions were detected simultaneously by three traits ( Table 2). The peak SNP of chr1.S_201957459 was located within GRMZM2G028386. Thus, these two transcription factors may play important roles in salt tolerance.
The hot spot on chromosome 5 contained four candidate genes: GRMZM2G102754, GRMZM2G176085, GRMZM2G027351 and GRMZM2G166049. GRMZM2G102754 is a homolog of Arabidopsis ethylene insensitive 2, which is involved in ethylene signal transduction and leaf senescence regulation under salt stress . GRMZM2G102754 harboured the peak SNP of chr5.S_12267893. GRMZM2G176085 is a homolog of a peroxidase superfamily protein in Arabidopsis that responds to oxidative stress (Valério et al., 2004). Both cdpk18 (GRMZM2G027351) and mab17 (GRMZM2G166049) were located in the same QTL interval identified by two traits, and encode proteins that may be involved in signal transduction and transcriptional regulation under salt stress conditions (Mittal et al., 2017).
The transcript levels of 12 candidate genes fluctuated from 0.5-168 h after salt treatment and were up-regulated at certain time points (Figure 3), indicating that all these genes were responsive to salt stress in maize. Consistent with the results of Fu et al. (2012), the expression of ZmPMP3 showed a fluctuating trend, which was up-regulated at some time points and down regulated at some time points .
Overall, 13 candidate genes which may be related to maize salt tolerance were identified by GWAS mapping in this study. Two candidate genes of ZmCLCg and ZmPMP3 were selected for functional verification. The zmclcg mutants were obtained by target site editing with an efficient CRISPR/Cas9 vector carrying three tandemly arrayed tRNA-target-gRNA. Three zmclcg mutants were obtained, with one mutant had mutations in both exons 3 and exon 4. The biomass of three zmclcg mutants and one zmpmp3 EMS mutant were compared with their wild-type plants under 100 mM NaCl treatment, respectively, and the results showed that the root length, root fresh weight, shoot length, shoot fresh weight of all mutants were significantly lower than that of the wild type, which verified the salt tolerance function of both ZmPMP3 and ZmCLCg. Chloride content analysis further indicated that ZmCLCg was associated with chloride transport in maize.

Plant materials and treatments
The association mapping panel consisted of 348 maize inbred lines, with 141 from tropical and subtropical zones and 126 from the temperate zone Liu et al., 2017;Yang et al., 2011b). The ZmCLCg knock-out maize lines were obtained by CRISPR/Cas9 technology. The EMS mutant of ZmPMP3 (Mut_Sample: EMS4-0a0498) was obtained from the maize EMS mutant library (http://www.elabcaas.cn/memd/). The ZmPMP3 gene (GenBank number: MW113229), ZmCLCg gene (GenBank number: MW113230) sequences of B73 maize inbred line, and the ZmCLCg gene sequence of B104 (GenBank number: MW113231) maize inbred line had been submitted to GenBank database. Primers for PCR amplification of full length of the ZmCLCg gene are listed in the Table S6.
Maize seeds were sterilized with 1% v/v NaClO for 10 min, and then rinsed three times with sterile water. Subsequently, sterilized maize seeds were sown and hydroponically cultured in a maize seedling identifying apparatus (Chinese patent number: ZL200920177285.0) placed in a greenhouse at 26 AE 1°C and 60% relative humidity under a 12-h light/12-h dark (150--180 μmol m −2 s −1 ) photoperiod. Details of the apparatus's operation are as follows: Maize seeds were fixed between two plates (170 mm × 50 mm), and the two plates were inserted into grooves on the inside surface of a container (325 mm × 190 mm × 95 mm). Each container had 10 grooves on the inside surface. Filter paper sheets were placed between the seeds and the plates and were humidified with 16-mm deep nutrient solution (800 mL). All maize materials were laid out in a randomized complete-block design with three replications (10 plants per replication).

Phenotypic data collection
The SLs and RLs of maize seedlings were measured with a ruler. The fresh and dry weights of seedling shoots (SF and SD, respectively) and roots (RF and RD, respectively) were measured using an electronic analytical balance. For dry weight determination, fresh shoot and root samples were oven-dried at 80°C for 3 d and then weighed. Ten seedlings for each replicate and three biological replicates were analysed. After measurements, the salt tolerance index for each trait was calculated using the formula: salt tolerance index for each trait = measured value under salt stress/ measured value under normal condition. Phenotypic trait distributions, correlations and frequency distributions were determined using R version 3.4.4 (http://www.r-project.org/).
The chloride content was determined by ion chromatography. Shoot of seedlings were dried, grounded and was passed through 40-mesh sieve. Transfer 0.2 g sample into a 50 ml tube and add 10 ml solution containing 3.5 mM Na 2 CO 3 and 1.0 mM NaHCO 3 . After digesting at 80°C for 1 h, the digested solution was filtered through 0.45 μm membrane. Finally, the chloride content was determined using a Dionex ICS 600 ion chromatography (Thermo Fisher Scientific, Agawam, MA, USA) with a Dionex IonPac TM AS14 chromatographic column (Thermo Fisher Scientific). The solution containing 3.5 mM Na 2 CO 3 and 1.0 mM NaHCO 3 was used as eluent and the flow rate was 1.0 ml/min.

Genome-wide association analysis
More than 1.06 million high-quality SNPs obtained from an RNAseq project

Candidate gene expression profiles
The expression profiles of candidate genes were determined in a transcriptome analysis. The transcriptome library preparation and sequencing were performed by Annoroad Gene Technology Co., Ltd. (Beijing, China) (Yu et al., 2017). Total RNA was extracted from shoots and roots of maize seedlings using TRIZOL (Invitrogen, Carlsbad, CA), and the purity, integrity and concentration were checked by electrophoresis and the Bioanalyzer 2100 system (Agilent Technologies, Santa Clara, CA). The transcriptome libraries were generated using the NEBNext ® Ultra ™ II Directional RNA Library Prep Kit for Illumina ® (New England Biolabs, Ipswich, MA) according to manufacturer's instructions. The quality and quantity of the libraries were controlled with the Bioanalyzer 2100 system (Agilent Technologies) and a CFX96 Real-Time PCR Detection System (Bio-Rad, Hercules, CA). The resulting libraries were sequenced using a HiSeq X ten instrument (Illumina, San Diego, CA). The raw sequences were cleaned by removing low-quality reads and reads containing adaptors. The cleaned reads were mapped to the maize reference genome (www.maizegdb.org) to obtain expression data for the identified genes. The sequencing data had been deposited in the Sequence Read Archive database (Accession number: PRJNA670840).
Generation of ZmCLCg CRISPR/Cas9 knock-out maize lines The pCambia2300-Spe vector containing the SpCas9n-pBE construct was used for plasmid construction (Wu et al., 2019). In the CRISPR/Cas9 vector, the sequences of U3 promoter, tRNA, esgRNA and the protein sequence of SpCas9 are according to previous reports (Cong et al., 2013;Xie et al., 2015). In addition, the coding sequence of SpCas9 was codon-optimized for high expression in maize and was synthesized by Nanjing Kingsley Biotechnology Co., Ltd. For CRISPR/Cas9 vector construction, the Cas9n-PmCDA1-UGI-t35s fragment in the SpCas9n-pBE construct was replaced with Cas9-t35s fragment using SnaBI and AscI, and the ZmUbi1-Hpt-t35s fragment was replaced with the ZmUbi1-bar-t35s fragment using AscI and AvrII. The CRISPR/Cas9 vector also contains a synthetic polycistronic gene that harbours three tandemly arrayed tRNA-target-gRNA (Xie et al., 2015). The three target sites located at the first, third and fourth exons of ZmCLCg were assembled with gRNA using BsaI. The resulting vector was transformed into the Agrobacterium tumefaciens strain EHA105 (Weidi Biotech, Shanghai, China). The Agrobacterium-mediated method was applied to transform immature embryos of maize inbred line B104. A total of 10 independent T0 transgenic plants were generated. The genomic fragments encompassing the target sites for each transgenic plant were sequenced by Sanger sequencing. No T0 plants with mutation at the target site of exon 1 were detected. One and seven T0 plant with mutation at the target site of exon 3 (Mutation efficiency = 10%) and exon 4 (Mutation efficiency = 70%) were identified, respectively. All T0 plants with mutation at the target sites were self-pollinated to get T1 progenies. PCR products covering the target sites of all T1 plants were sequenced to identify T1 plants with homozygous mutation in target sites. The confirmed homozygous mutant T1 plants and the wild-type plants were used for salt stress treatment and phenotypic determination. The primers for PCR amplification of target sites are listed in the Table S6.

Statistical analyses
Comparisons of phenotypic data and gene expression levels were conducted by unpaired two-tailed Student's t tests. Student's t tests and correlation analyses were conducted using the t test and correlation functions in GraphPad Prism 5 software (http://www. graphpad.com/), respectively. The means, standard deviations, coefficients of variation, kurtosis and skewness were calculated using the column statistics function in GraphPad Prism 5 software. Repeatability for each trait was determined using the R software (Luo et al., 2019). Repeatability was calculated with the formula: repeatability = σ 2 G =ðσ 2 G þ σ 2 e Þ, while σ 2 G and σ 2 e represent the genetic variance and the error variance, respectively.  Figure S2 Pearson's correlation coefficients (r) between 27 maize traits (nine each in the control and under salt stress conditions, and nine salt tolerance indexes). Correlation coefficients were calculated from mean values of three biological replicates for each trait of the maize association population. See Figure S1 caption for abbreviations.

Figure S3
Quantile-quantile (QQ) plots of genome-wide association study (GWAS) results using different association models for maize shoot length trait under salt treatment conditions. Horizontal dashed red line represents significance threshold (1.79 × 10 -6 ). Figure S4 Manhattan plots for all 27 maize traits using Q + K mixed linear model. See Figure S1 caption for abbreviations. Figure S5 QQ plots for all 27 maize traits using Q + K mixed linear model. See Figure S1 caption for abbreviations Figure S6 Growth status and growth parameters of B104, B73, the significantly salt-tolerant maize inbred line GEMS37, and the significantly salt sensitive maize inbred line CIMBL157 after culturing in control or saline water for 10 days. (a) Growth status; (b) Root length; (c) Roof fresh weight; (d) Shoot length; (e) Shoot fresh weight. Data are shown as the mean AE SE of three independent experiments. The P values were calculated by a twotailed Student's t test. Figure S7 KOG functional categories for maize genes within significant QTL regions Figure S8 Candidate regions associated with salt tolerance and phenotypic differences between two alleles of most significant trait-associated single nucleotide polymorphisms (SNPs). SNPs within candidate regions are highlighted in green. Peak SNPs are marked by arrows. (a-d), Candidate regions associated with RLS, SFS, RLS and RLS located on chromosome 2, 7, 7 and 10, respectively (left). Phenotypic differences in RLS, SFS, RLS and RLS between two alleles of peak SNPs (right). ***P < 0.001 Table S1 List of 348 maize lines used in this study Table S2 List of significant maize SNP-trait associations and detailed information identified by GWAS