Genetic Dissection of Alkalinity Tolerance at the Seedling Stage in Rice (Oryza sativa) Using a High-Resolution Linkage Map

Although both salinity and alkalinity result from accumulation of soluble salts in soil, high pH and ionic imbalance make alkaline stress more harmful to plants. This study aimed to provide molecular insights into the alkalinity tolerance using a recombinant inbred line (RIL) population developed from a cross between Cocodrie and Dular with contrasting response to alkalinity stress. Forty-six additive QTLs for nine morpho-physiological traits were mapped on to a linkage map of 4679 SNPs under alkalinity stress at the seedling stage and seven major-effect QTLs were for alkalinity tolerance scoring, Na+ and K+ concentrations and Na+:K+ ratio. The candidate genes were identified based on the comparison of the impacts of variants of genes present in five QTL intervals using the whole genome sequences of both parents. Differential expression of no apical meristem protein, cysteine protease precursor, retrotransposon protein, OsWAK28, MYB transcription factor, protein kinase, ubiquitin-carboxyl protein, and NAD binding protein genes in parents indicated their role in response to alkali stress. Our study suggests that the genetic basis of tolerance to alkalinity stress is most likely different from that of salinity stress. Introgression and validation of the QTLs and genes can be useful for improving alkalinity tolerance in rice at the seedling stage and advancing understanding of the molecular genetic basis of alkalinity stress adaptation.


Introduction
Abiotic stresses are major challenges for sustainable crop production worldwide. The ongoing climate change is exacerbating the frequency and severity of abiotic stresses. Alkaline stress suppresses plant growth, development, and productivity due to exposure to osmotic stress, nutritional deficiency, and ionic imbalance [1][2][3]. High pH and high concentration of salt (Na 2 CO 3 or NaHCO 3 ) impedes plant performance by interfering with the water absorption capacity caused by ionic and osmotic stresses [4][5][6][7]. Furthermore, precipitation of iron and phosphorus due to high pH around the rhizosphere results in nutritional deficiency in plants [2,8]. Alkaline stress causes severe damage to the root cells due to increased injury to membranes, increased malondialdehyde content, and upregulation of cell death-associated genes [9,10].
Rice (Oryza sativa L.) is one of the most important cereal crops worldwide. However, most cultivated rice varieties are susceptible to saline and alkaline stresses. Both stresses affect all stages of plant development starting from germination to the reproductive stage [11]. The intensity of plants' injury depends on several factors including timing of stress and growth stages [12]. Alkaline stress is damaging to rice plants at both the seedling and reproductive stages [1,10,11]. Previous studies showed that there is either no or weak correlation between seedling and reproductive stage salinity tolerance [13,14] and therefore it has been suggested to screen the lines for salinity tolerance at the seedling stage followed by evaluation at the reproductive stage preferably under field conditions [15].
In saline-alkaline soil, soluble salts mostly consist of cations such as Na + , Ca 2+ , Mg 2+ , and K + and anions like CO 3 2− , HCO 3 − , Cl − , SO 4 2− , and NO 3 − . Plants are more susceptible to alkalinity due to high pH and the presence of CO 3 2− and HCO 3 − [16]. However, the role of CO 3 2− and HCO 3 − in the severity of alkalinity is rarely emphasized. Moreover, osmotic stress, ionic stress, and high pH under alkaline stress reduce the iron (Fe) solubility [17]. As Fe in the soil exists in oxidized form (Fe 3+ ), it becomes highly insoluble under aerobic conditions, especially at high pH [18], and plants show wilting and Fe deficiency symptoms [19]. Alkaline stress tolerance is a complex and polygenic trait [9,20,21], which makes developing alkalinity tolerant crop varieties challenging. Therefore, understanding the molecular basis of alkalinity tolerance is essential to make genetic progress.
The molecular genetic basis of salinity tolerance in rice has been investigated extensively compared to alkalinity tolerance [22][23][24]. The impacts of salinity have been evaluated at the morphological, physiological, and biochemical levels and genes contributing toward tolerance such as SKC1 [25] and DST [26] have been identified. However, progress in understanding the molecular genetic basis of alkalinity tolerance is lagging behind. As both alkaline and saline stresses involve a high concentration of Na + salt around the rhizosphere and in shoots, it has been assumed that there might be overlapping tolerance mechanisms for both stresses. Several studies showed that alkaline stress is more detrimental to crops compared with saline stress due to high pH and nutrient deficiency in addition to salt stress caused by Na 2 CO 3 [9,27,28]. Apart from tolerance to ionic imbalance and osmotic stress, plants have evolved mechanisms for tolerating high pH and nutrient deficiency [29,30]. Various mechanisms such as ion exclusion at root level and its restricted transport to shoots and increased K + uptake under stress environments have been reported to confer salinity tolerance in rice [31]. Few studies revealed that alkalinity tolerant cultivars possess genes for efficient acquisition of Fe and K under high pH [2,5,32]. Research efforts are needed to elucidate the mechanisms of the Na:K homeostasis, and Fe acquisition to enhance alkalinity tolerance in rice at the seedling stage.
The rapid development of genomics and molecular biology tools has enabled the dissection of the genetic basis of stress tolerance by identifying the major QTLs and genes to develop stress-tolerant rice cultivars using marker-assisted selection (MAS). A major alkaline tolerant QTL, qARL11, and two candidate genes from the F-box gene family, were identified using QTL mapping and a genome-wide association study [33]. Another major QTL qSNC3 for alkaline stress tolerance explained 21% of phenotypic variation [30]. Few other studies reported QTLs at the germination and seedling stage of rice under alkaline stress [34][35][36]. Rice plants overexpressing the LSD1-like-type zinc finger protein OsLOL5 and nucleus-encoded thylakoid protein OsY3IP1 improved alkalinity tolerance [37,38]. Similarly, Osppa6 was shown to play an important regulatory role in conferring alkalinity tolerance in rice [39]. However, there is no progress in the introgression of these QTLs/genes for rice improvement. In this study, we evaluated a recombinant inbred line (RIL) population and identified genomic regions for various morpho-physiological traits associated with alkaline tolerance in rice at the seedling stage. Since the next-generation sequencing has been used for genetic dissection of abiotic stress tolerance [22,40], we integrated results from both QTL mapping and whole genome sequencing (WGS) to identify candidate genes associated with alkaline tolerance in rice.

Phenotypic Characterization under Alkaline Stress
After three weeks of exposure to alkaline stress, the RILs and parents showed a varying level of tolerance based on observation of several traits such as alkalinity tolerance score (AKT), chlorophyll content (CHL), shoot length (SHL), root length (RTL), root to shoot ratio (RSR), shoot dry weight (DW), shoot Na + concentration (SNC), shoot K + concentrations (SKC), and Na + /K + ratio (SNK) ( Table 1). Except RSR, Cocodrie and Dular showed a  Figure 1). AKT scores of Cocodrie and Dular were 3.7 and 7.7, respectively and both cultivars were scored 1 under control condition (Table S1 and Figure 2). All traits except CHL and SHL were normally distributed according to Shapiro-Wilk's test and showed transgressive segregants on both sides of the distribution. The log transformed CHL (LCHL) and log transformed SHL (LSHL) data under alkaline stress were used for QTL mapping. On the contrary, all traits under the control condition were normally distributed ( Figure S1). Cocodrie means were lower for AKT, SNC, and SNK but higher for SKC, LCHL, LSHL, RTL, DW, and RSR compared to Dular. Analysis of variance revealed significant differences among RILs for all traits except SNK ( Table 1). The RIL mean values for all the traits were in between the mean values of Cocodrie and Dular except LSHL ( Figure 2). The traits AKT, DW, RTL, RSR, SNC, and SKC showed moderate to high heritability, whereas SNK, LCHL, and LSHL exhibited low heritability. Heritability was significant and higher for all traits under control compared with the stress environment (Table S1). There was no difference for any morphological and physiological traits between Cocodrie and Dular under control condition ( Figure 3).

Phenotypic Characterization under Alkaline Stress
After three weeks of exposure to alkaline stress, the RILs and parents showed a varying level of tolerance based on observation of several traits such as alkalinity tolerance score (AKT), chlorophyll content (CHL), shoot length (SHL), root length (RTL), root to shoot ratio (RSR), shoot dry weight (DW), shoot Na + concentration (SNC), shoot K + concentrations (SKC), and Na + /K + ratio (SNK) ( Table 1). Except RSR, Cocodrie and Dular showed a significantly contrasting response for all other traits (Table 1 and Figure 1). AKT scores of Cocodrie and Dular were 3.7 and 7.7, respectively and both cultivars were scored 1 under control condition (Table S1 and Figure 2). All traits except CHL and SHL were normally distributed according to Shapiro-Wilk's test and showed transgressive segregants on both sides of the distribution. The log transformed CHL (LCHL) and log transformed SHL (LSHL) data under alkaline stress were used for QTL mapping. On the contrary, all traits under the control condition were normally distributed ( Figure S1). Cocodrie means were lower for AKT, SNC, and SNK but higher for SKC, LCHL, LSHL, RTL, DW, and RSR compared to Dular. Analysis of variance revealed significant differences among RILs for all traits except SNK ( Table 1). The RIL mean values for all the traits were in between the mean values of Cocodrie and Dular except LSHL ( Figure 2). The traits AKT, DW, RTL, RSR, SNC, and SKC showed moderate to high heritability, whereas SNK, LCHL, and LSHL exhibited low heritability. Heritability was significant and higher for all traits under control compared with the stress environment (Table S1). There was no difference for any morphological and physiological traits between Cocodrie and Dular under control condition (Figure 3).   RSR, root to shoot ratio; SNC, shoot Na + concentration; SKC, shoot K + concentration; SNK, shoot Na + :K + ratio. § t-test between Cocodrie and Dular; $ Genotypic difference between RILs based on ANOVA). *, **, *** Significant differences between the means Cocodrie and Dular at 0.05, 0.01, and <0.001 level of probability, respectively. ns Non-significant.

Correlations among Traits in the RIL Population
The correlations among all the traits were presented in Table 2. There was a significant negative correlation between AKT and alkaline stress responsive traits such as RTL, DW, RSR, LSHL, LCHL, SKC, and SNK. However, it showed a significant positive correlation with SNC. SNC was only significantly negatively correlated with RTL and RSR. SKC had a significant negative correlation with AKT and a significant positive correlation with SNK, LCHL, RTL, and DW. RTL was significantly and positively correlated to SKC and RSR whereas it was negatively correlated with SNC and AKT. LSHL showed a significant positive correlation to RTL and DW, but a significant negative correlation with AKT and RSR.

Linkage Map and QTL Mapping
A total of 4679 SNP markers, generated by genotyping by sequencing, were used for linkage map construction. The linkage map covered 367 Mb of rice genome with a total genetic length of 1584 cM ( Table 3). The average chromosome length was 132 cM with 3 SNPs per cM. The average number of gaps with more than 5 cM was 3.2 per chromosome.

QTLs under Alkalinity Stress
Forty-six QTLs were identified for various morphological and physiological traits under alkaline stress using inclusive composite interval mapping (ICIM) ( Table 4 and Figure 4). Three QTLs were detected for AKT by ICIM and only qAKT8.27 was a major effect QTL with a contribution of 11% toward phenotypic variation (PV). The Dular allele increased the AKT value for all the QTLs. Two QTLs were identified for LCHL with desirable alleles from Cocodrie in both cases. The qLCHL1.38 was a major effect QTL with 27% of PV, while a minor effect QTL qLCHL8.17 explained 6% of PV. A total of 4 QTLs were detected for LSHL. The large-effect QTL (qLSHL2.31) had Dular allele responsible for increased mean. All other QTLs were minor effect QTLs and the desirable allele for these QTLs was contributed by Cocodrie. For root length, four QTLs were identified. All QTLs except qRTL5.01 were minor effect QTLs and had desirable alleles from Cocodrie, while qRTL5.01 had desirable allele from Dular with a total PV of 10%.
There were three QTLs for RSR on chromosome 9, while two QTLs on chromosome 6 and one each on chromosome 5 and 4 under alkaline stress. Both qRSR5.01 and qRSR6.28 were large effect QTLs with a PV of 14% and 13%, respectively. The allele for increasing the mean at qRSR5.01 was contributed by Cocodrie, while it was Dular allele for qRSR6.28. All other QTLs for RSR were small effect QTLs. Five QTLs were detected for shoot dry weight. All QTLs except qDW12.27 were small effect QTLs. In case of three QTLs, Dular allele was responsible for the increased mean, while Cocodrie allele was responsible for the remaining two QTLs. Na + and K + are the key factors determining alkalinity tolerance. For SNC, three large-effect QTLs and seven minor-effect QTLs were identified. All QTLs had alleles for increased mean contributed by Dular. The three major effect QTLs, qSNC3.15, qSNC3.32, and qSNC10.16, had a PV range of 10-13%. A total of seven QTLs were identified for SKC. Two major effect QTLs, qSKC9.22 and qSKC10.18, explained 10% and 15% of PV, respectively. All other QTLs were small effect QTLs with two each detected on chromosomes 2 and 4 and one on chromosome 1. The alleles for increased mean in case of all QTLs were contributed by Cocodrie. For SNK, there were four QTLs and only qSNK8.01 explained 10% of phenotypic variation whereas the remaining were minor effect QTLs with a PV value ranging from 4-9%. The desirable alleles for all QTLs were contributed by Dular. AKT, alkalinity tolerance scoring; LCHL, log chlorophyll content; LSHL, log shoot length; RTL, root length; RSR, root to shoot ratio; SNC, shoot Na + concentration; SKC, shoot K + concentration; SNK, shoot Na + :K + ratio. § qAKT, qLCHL, qLSHL, qRTL, qRSR, qDW, qSNC, qSKC, and qSNK are QTLs for alkalinity tolerance, log chlorophyll content, log shoot length, root length, root to shoot ratio, dry weight, shoot Na + concentration, shoot K + concentration, and shoot Na + :K + ratio, respectively. The number before the decimal represents the chromosome number and the number after the decimal indicates the physical position of the QTLs in mega base pair. β LOD logarithm of odds. ψ PVE (%) percentage phenotypic variance explained by the QTL.
Two major effect QTLs, qSKC9.22 and qSKC10.18, explained 10% and 15% of PV, respectively. All other QTLs were small effect QTLs with two each detected on chromosomes 2 and 4 and one on chromosome 1. The alleles for increased mean in case of all QTLs were contributed by Cocodrie. For SNK, there were four QTLs and only qSNK8.01 explained 10% of phenotypic variation whereas the remaining were minor effect QTLs with a PV value ranging from 4-9%. The desirable alleles for all QTLs were contributed by Dular.

QTLs under Control Environment
Twenty-six QTLs were identified for various morpho-physiological traits under the control environment by ICIM (Table S2). There was no QTL detected for AKT. Two and six QTLs were identified for CHL and SHL, respectively. The large-effect QTLs were qCHL7.01, qSHL1.03, and qSHL9.18 with PV of more than 10%. Two QTLs each were detected for RTL and RSR. One QTL on chromosome 6, qRSR6.13, had a PV of 38%. For SNC and SKC, five QTLs for each trait were identified. All QTLs for SNC and SKC except qSKC2.03 were small effect QTLs. The qSKC2.03 explained 21% of PV. Four QTLs for SNK were detected and qSNK7.05 was a large effect QTL. Incase of ten QTLs, Dular alleles were responsible for increased means, while Cocodrie alleles were contributed toward the increased means in the remaining QTLs.

QTLs under Control Environment
Twenty-six QTLs were identified for various morpho-physiological traits under the control environment by ICIM (Table S2). There was no QTL detected for AKT. Two and six QTLs were identified for CHL and SHL, respectively. The large-effect QTLs were qCHL7.01, qSHL1.03, and qSHL9.18 with PV of more than 10%. Two QTLs each were detected for RTL and RSR. One QTL on chromosome 6, qRSR6.13, had a PV of 38%. For SNC and SKC, five QTLs for each trait were identified. All QTLs for SNC and SKC except qSKC2.03 were small effect QTLs. The qSKC2.03 explained 21% of PV. Four QTLs for SNK were detected and qSNK7.05 was a large effect QTL. Incase of ten QTLs, Dular alleles were responsible for increased means, while Cocodrie alleles were contributed toward the increased means in the remaining QTLs.

Candidate Genes Identification by Integrating Data from QTL Mapping and Whole Genome Sequencing
Using the flanking markers of QTLs, the genes present in the QTL intervals of the alkalinity tolerance traits were determined (Table S3). Five large effects QTLs (qSNC3.32, qSNK8.01, qAKT8.27, qSKC9.22, and qSKC10.18) were selected for detecting candidate genes based on polymorphic SNP and InDels present between parents. After removing SNPs and InDels in the intergenic, downstream, and upstream regions, all low, moderate, and high impact variants of genes present in the QTL intervals between Cocodrie and Dular were examined (Table S4). Only high-impact SNPs and InDels were considered and a total of 63 candidate genes were identified (Table S5). After removing the hypothetical genes and expressed proteins, the number of candidate genes was narrowed down to thirty (Table 7). There were twelve candidate genes within the interval of qSNC3.32 based on the frameshift mutation, stop lost, stop gained, and splice donor variants differentiating Cocodrie and Dular. These genes were retrotransposon protein (LOC_Os03g56560 and LOC_Os03g57380), ad-003 (LOC_Os03g56830), calreticulin family protein (LOC_Os03g59264), no apical meristem protein (LOC_Os03g59730), cys-rich domain-containing protein (LOC_Os03g61430), small heat shock protein (LOC_Os03g61940), hydrolase (LOC_Os03g62070), WD40-like Beta Propeller Repeat family protein (LOC_Os03g62370), MYB family transcription factor (LOC_Os03g62379), pentatricopeptide (LOC_Os03g62400) and OsWAK28-OsWAK receptor-like protein kinase (LOC_Os03g62430). In the case of qSNK8.01, seven genes were identified based on several stops gain, splice acceptor and donor, and frameshift variants and these were Cytochrome P450 (LOC_Os08g01470), NBS-LRR disease resistance protein (LOC_Os08g01580), Rf1, mitochondrial precursor (LOC_Os08g01640), dehydrogenase (LOC_Os08g01760), matrix attachment region binding protein (LOC_Os08g01810), CR4L subfamily gene (LOC_Os03g01830), and protein kinase family protein (LOC_Os08g02050).

Validation of Expression of Alkalinity Tolerance Related Genes
Ten genes were selected from the high-impact variants to determine their expression pattern under alkalinity stress compared to control (no stress) in rice by real-time reverse transcription PCR (qRT-PCR) ( Table S6). The expression of LOC_Os03g59730 (no apical meristem protein), LOC_Os03g62370 (WD40-like beta propeller repeat family protein), LOC_Os09g39100 (cysteine protease EP-B 1 precursor), and LOC_Os10g34650 (retrotransposon protein) were downregulated in Cocodrie under alkaline stress whereas the expression level of these genes sharply increased in Dular 6 h after exposure to alkaline stress ( Figure 5). In contrast, the expression of LOC_Os03g62430 (OsWAK28-OsWAK receptor-like protein kinase); LOC_Os03g62379 (MYB family transcription factor), LOC_Os08g02050 (protein kinase family protein), LOC_Os08g44690 (transposon protein), LOC_Os10g34490 (ubiquitin-carboxyl extension) and LOC_Os10g35170 (semialdehyde dehydrogenase, NAD binding domain-containing protein) were upregulated in Cocodrie compared to Dular under alkaline stress.   ( Figure 5). In contrast, the expression of LOC_Os03g62430 (OsWAK28-OsWAK receptorlike protein kinase); LOC_Os03g62379 (MYB family transcription factor), LOC_Os08g02050 (protein kinase family protein), LOC_Os08g44690 (transposon protein), LOC_Os10g34490 (ubiquitin-carboxyl extension) and LOC_Os10g35170 (semialdehyde dehydrogenase, NAD binding domain-containing protein) were upregulated in Cocodrie compared to Dular under alkaline stress.

Discussion
Saline and alkaline stress adversely affect the growth and development of rice plants resulting in reduced grain yield [35]. Compared with saline stress, alkalinity not only disrupts the ionic balance but also causes deficiency of essential nutrients and damages to cellular machinery due to high pH and the presence of carbonate cations [29,40]. Alkaline stress is more complicated than saline stress and deciphering the mechanisms of alkalinity tolerance at the seedling stage of rice is critical for improving rice production. Although several QTLs and candidate genes for alkalinity tolerance have been identified in rice [6,30,[33][34][35][36]41,42], the molecular mechanism of alkalinity tolerance at the seedling stage in rice is not yet elucidated in greater detail. Thus, there is a need to unravel the molecular mechanisms and candidate genes associated with alkalinity tolerance in rice.
Alkaline stress reduced the growth and performance of all RILs. However, there was significant variation in the extent of damage among RILs for all traits except SNK (Table 1). Some RILs performed better than the parents under alkaline stress, suggesting the presence of transgressive segregants (Figure 2), which could be due to the complementary gene action of additive alleles dispersed in the parental lines [43]. The comparison of the performance of both parents under control and alkaline stress environment clearly showed that the impact of alkaline stress was more damaging to Dular than Cocodrie. Furthermore, there was a reduction in heritability estimates for all traits under alkaline stress compared to control suggesting the influence of alkaline stress on the expression of the traits as observed in earlier studies [44,45]. The traits with high to moderate heritability such as AKT and SNC could be exploited in breeding program for improving alkalinity tolerance in rice.
Correlation analysis is crucial to investigate the relationship among the physiological and morphological traits under stress conditions. Association of high AKT with high SNC in RILs suggesting inefficient exclusion of Na + from roots and shoots was in agreement with earlier studies [46,47]. In susceptible lines, salt stress depolarizes the plasma membrane and reduces the K + uptake by affecting the expression of K + acquisition genes [48]. A negative and significant correlation between AKT and SKC in our study (Table 2) led to the same conclusion that tolerant RILs accumulated more potassium than susceptible lines. Tolerance was reflected by low AKT, low SNC, and high SKC. We observed a significant negative correlation between AKT and traits such as RTL, RSR, DW, LSHL, and LCHL. The association between root length and shoot length, chlorophyll, and dry weight were positive under the stress environment. An increase in root length under alkaline stress increased SHL, CHL, and DW. This suggests that a deep root system and high chlorophyll content under alkaline stress can improve alkalinity tolerance due to the ability to overcome osmotic stress and the nutrient deficiency in rice. Longer and deeper roots help to cope with high salt concentration and high pH near the surface and supply nutrients to the plants. A study showed that an approximately 20% increase in root surface under stress environment could mitigate stress by the instantaneous response of plants to uptake more nutrients and water [49]. Similarly, the transgenic rice overexpressing OsMADS25 enhanced salt tolerance by developing deeper roots, while knock-down of OsMADS25 increased salt sensitivity due to reduced root growth [50]. In this study, a significant negative association was found between SNC and SKC. Heritability for these physiological traits were low to medium in this study. In contrast to this study, a positive correlation between Na + and K + and high heritability of these traits were reported under saline stress [51]. However, results comparable to our study were reported in rice under alkaline stress [30]. Furthermore, Cocodrie was tolerant to alkali stress in contrast to its susceptibility to saline stress [52]. These findings suggested that alkaline tolerance mechanisms might be different from those for salinity tolerance to cope with the ionic imbalance and Fe deficiency under high pH.
We detected twenty-one QTLs for SNC, SKC, and SNK. Cocodrie alleles were responsible for increasing means at all SKC QTLs whereas Dular alleles increased means at all SNC and SNK QTLs. These findings signified the desirability of the Cocodrie allele to improve alkalinity tolerance. However, same tolerant parental allele contributed toward increased trait means at both Na + and K + QTLs under salt stress in an earlier study [53]. Therefore, it will be interesting to elucidate the underlying genes for Na + and K + accumulation in shoots under alkaline stress. Accumulation of K + and maintenance of low Na + :K + ratio and/or compartmentation of Na + by Cocodrie could be the reason for alkalinity tolerance as reported in a previous study [5]. Most of the SNC QTLs of this study (qSNC2.22, qSNC2.32, qSNC3.15, qSNC3.32, qSNC5.06, and qSNC10.16) co-localized with the QTLs detected in previous alkaline stress studies ( Table 5). The qSNC3.32 co-localized with qRKC3.32 identified in our earlier study [36]. Some SNC QTLs co-localized with the QTLs detected for morphological traits. These findings confirmed the association between physiological and morphological traits for stress tolerance. The qSNC3.32 with a contribution of 13% of PV detected in this study co-localized with qSNC3 [30] and two candidate genes (LOC_Os03g62500 and LOC_Os03g62620) in this QTL region showed differences in expression level between the parents under alkaline stress. It is possible that other genes within this genomic region might be responsible for alkalinity tolerance. This possibility was supported by our sequence analysis of qSNC3.32 region which led to identification of more candidate genes downstream of the detected genes [30]. Furthermore, QTLs for Na + concentration in the same region were detected under saline stress [54,55]. We further narrowed down the number of candidate genes to twelve within the qSNC3.32 interval (Table 7). Receptor-like kinases play a significant role in plant signal transduction pathways. The upregulation of receptor kinase gene OsWAK28 in Cocodrie compared to Dular ( Figure 5) suggested the role of this gene in response to alkaline stress. In contrast to this study, the knockdown of OsWAK112 improved rice salt tolerance which could be due to a different role in signal transduction process [56]. WD40 family protein (LOC_Os03g62370) was downregulated in Cocodrie compared to Dular ( Figure 5). Contrary to our study, a rice WD40 protein gene OsABT was upregulated in rice under salt stress [57]. These results further corroborated our hypothesis of differences in tolerance mechanisms for alkalinity and salinity in rice. Transcription factors serve as a control switch for plant responses under stress environments [58]. The transcription factor genes, LOC_Os03g62379 and LOC_Os03g59730, were upregulated and downregulated, respectively in Cocodrie ( Figure 5). In earlier transcriptomic studies, OsMYB2P-1 [59] and OsNAC52 [28] were differentially expressed between tolerant and sensitive cultivars under alkaline stress. These results suggest that MYB and NAC transcription factors play a crucial role in alkaline stress response and both LOC_Os03g62379 and LOC_Os03g59730 may be involved in regulating processes that affect tolerance to alkaline stress in rice, particularly at the seedling stage.
In case of SKC, qSKC1.13, qSKC2.19, qSKC4.16, qSKC4.31 and qSKC10.18, co-localized with the QTLs identified under alkaline stress in earlier studies (Table 5). Both qSKC9.22 and qSKC10.18 were large effect QTLs. The qSKC10.18 overlapped with an earlier reported QTL [36], while qSKC9.22 was a novel QTL for alkalinity tolerance. In case of qSKC9.22, two potential candidate genes, LOC_Os09g39090 (vignain precursor) and LOC_Os09g39100 (cysteine protease EP-B 1 precursor) with high impact variants differentiated Cocodrie and Dular. Similarly, eight candidate genes were detected based on high-impact DNA polymorphisms between Cocodrie and Dular in case of qSKC10.18 ( Table 7). None of our SNK QTLs were same as earlier identified QTLs. The qSNK8.01 was a large effect QTL and qSNK2.03, qSNK7.05 and qSNK11.27 were minor effect QTLs. Of particular interest is the qSNK8.01 and sequence analysis narrowed the candidate genes number to seven within the QTL interval ( Table 7). Most of the genes detected in qSKC10.18 and qSNK8.01 interval were not previously identified under alkaline stress. The LOC_Os09g39100 encoded a cysteine protease, related to cell apoptosis, which regulates cell death across the plant genome [60]. Damage to roots under alkaline stress in rice was associated with the upregulation of cell-death-related genes [10]. Cysteine protease was downregulated in Cocodrie and upregulated in Dular ( Figure 5). Alkalinity tolerance of Cocodrie could be due to the suppression of cell death-related gene (LOC_Os09g39100) under high pH conditions. The expression of the ubiquitin-carboxyl gene (LOC_Os10g34990), semialdehyde dehydrogenase, NAD binding domain-containing protein (LOC_Os10g35170), and protein kinase family protein (LOC_Os08g02050) were upregulated in Cocodrie, while the expression of retrotransposon protein (LOC_Os10g34650) was reduced in Cocodrie under stress environment ( Figure 5). Many studies demonstrated relationship between salinity tolerance and the protein ubiquitination process [61,62]. Similarly, semialdehyde dehydrogenase family genes, NAD binding protein, and retrotransposons have been reported to be responsive to abiotic stresses [63][64][65][66][67]. Since the role of these genes in alkaline stress response is unclear, these novel genes should be investigated in future for their potential in improving alkalinity tolerance in rice.
A total of twenty-five additive QTLs were detected for AKT, LCHL, LSHL, RTL, RSR, and DW. Except for LCHL and AKT, QTLs identified for each trait had varying contributions from Cocodrie and Dular. These results suggested contribution of both Cocodrie and Dular alleles to the expression of these traits. These observations reinforced the complex nature of the alkaline tolerance in rice and thus pyramiding of QTLs for different traits will be required to mitigate the impacts of alkalinity stress. The colocalization of qAKT8.27 and qDW8.27 in the same region as in the previous study [36] could be due to pleiotropic effect or tightly linked genes [68]. Similar results were observed in the case of qDW1.38, qLCHL1.38, qLCHL1.38, and qDW1.38 which were congruent to qSHL1.38 detected in our earlier study [36]. Selection of these QTLs could be helpful in breeding for alkalinity tolerance as selection for one trait will indirectly help select the other desirable traits.
The alkalinity tolerance scores reflect the overall performance of plants under alkaline stress. Lines sensitive to alkaline stress showed a reduction in the shoot, root, dry weight, and chlorophyll content. The correlation analysis confirmed the negative association of AKT with these traits. In this study, Dular alleles were responsible for increasing the mean effects for all three QTLs (qAKT8.02, qAKT8. 27,qAKT12.27). Therefore, a low AKT score and corresponding QTL allele from Cocodrie are desirable from the breeding perspective. None of these QTLs colocalized with the QTLs from earlier alkaline stress tolerance studies. This showed that these additive QTLs might be novel alkaline stress tolerance loci. In case of the major effect QTL qAKT8.27, sequence analysis delimited this region to only one transposon protein gene (LOC_Os08g44690). The transposons have been implicated in various plant stress responses [69][70][71][72]. The upregulation of LOC_Os08g44690 in our expression study suggested its role in phenotypic plasticity and stress adaptation.

Conclusions
The present study demonstrated the power of integrating QTL mapping with nextgeneration sequencing to elucidate the complex mechanisms associated with alkalinity tolerance. In addition to identification of the genomic regions using a high-resolution genetic linkage map, the SNPs and InDels identified from the whole genome sequence analysis helped to select putative candidate genes associated with alkalinity tolerance and the differential expression of some of these genes demonstrated their role in response to alkaline stress. Despite the susceptibility of Cocodrie to salinity [52], it showed a high degree of tolerance to alkaline stress, suggesting different genetic mechanisms controlling tolerance to both stresses. The congruency of QTLs with earlier salt tolerance QTLs could be due to commonality in morpho-physiological response to salinity and alkalinity. However, analysis of the genome, transcriptome, and metabolome of genotypes with contrasting stress response at a global scale will provide insights into the mechanisms differentiating adaptation to salt and alkali stresses in the future. The QTLs and genes identified in this study can be useful for improving alkalinity tolerance in rice at the seedling stage and advancing understanding of the molecular genetic basis of alkalinity stress adaptation.

Choice of Parents
Cocodrie and Dular were used as parents to develop a recombinant inbred line (RIL) population for this study. Cocodrie, a long grain rice cultivar released by the Louisiana State University Agricultural Center [73], is tolerant to alkaline stress [36]. Dular is an upland-adapted well-known donor for drought tolerance from India [74], but susceptible to alkaline stress. The individual plants of the F 2 population developed from the cross between Cocodrie and Dular were advanced to F 8 by the single seed descent method to develop a RIL population.

Seedling Stage Screening for Alkalinity Tolerance
This experiment was set up in the LSU AgCenter greenhouse ( Figure S2). A total of 189 RILs along with parents were screened for alkalinity tolerance at the seedling stage. A randomized complete block design (RCBD) for both control and stress experiments with three replications was used. Seeds were exposed to 50 • C for five days to break the dormancy. Fifteen seeds per replication were sown in the sand-filled "4 × 4" square pots. Plants were allowed to grow in the nutrient solution with pH 5.6 until two leaf stage. Nutrient solution was prepared by dissolving 1 g/L of Jack's professional fertilizer (20-20-20) (J.R. Peters Inc., Allentown, PA, USA) and 300 mg/L ferrous sulfate. A protocol described by Singh et al. [36] was used for alkalinity tolerance screening. At the twoleaf stage, plants were exposed to alkaline stress of 0.20% and 0.40% sodium carbonate (Na 2 CO 3 ) solution with pH 10 for the first two and third week, respectively. Plants in the control experiment were allowed to grow under normal conditions.
Only five uniform seedlings were used for morphological and physiological observations. Chlorophyll content (CHL) was measured using SPAD-50 chlorophyll meter (Spectrum Technologies, Inc., Aurora, IL, USA) 10 days after exposure to stress. When the sensitive parent Dular showed susceptibility symptoms related to alkaline stress, plants were scored visually for alkalinity tolerance (AKT) on a scale of 1 to 9 following Singh et al. [36]. Shoot length (SHL) was measured from the base of the tip of the seedling to the longest leaf. Root length (RTL) was recorded from the base of the culm to the tip of the root. Rootto-shoot ratio (RSR) was computed by dividing root length by shoot length. Five plants were oven-dried at 65 • C for five days and then shoot dry weight (DW) was obtained. For Na + and K + , shoot samples were oven-dried at 65 • C for 10 days and 0.5 g of the ground sample from each line per replication was digested in 5 mL of HNO 3 and 3 mL of H 2 O 2 at 152-155 • C for 3 h [75]. The concentrations of shoot Na + (SNC) and shoot K + (SKC) from the digested samples were measured using a flame photometer (model PFP7, Bibby Scientific Ltd., Staffordshire, UK). Standard curves for Na + and K + were derived from the standard solutions of different dilutions. Then, the final concentration of Na + and K + were computed using standard curve readings. Shoot Na + :K + ratio (SNK) was calculated by dividing the shoot Na + concentration by shoot K + concentration.

Statistical Analysis
R (version 2.2.1) and SAS software (version 9.4) were used for statistical analysis [76,77]. Shapiro-Wilk test was performed in R for each trait to assess the normality of the data and log transformation was used to transform the data for the traits that were not normally distributed. Descriptive statistics and histograms for each trait were obtained using the R. Aov function in R was used for computing the analysis of variance (ANOVA). To determine the relationship between morphological and physiological traits under alkaline stress, Pearson correlation coefficients were calculated in R. SAS was used to estimate broad sense heritability following Holland et al. [78].

Genotyping-by-Sequencing of the Mapping Population and SNP Identification
Leaf tissues were collected from 189 RILs and parents (Cocodrie and Dular) grown in the control environment. The genomic DNA was extracted using a modified CTAB method [79]. Extracted DNA was purified by the genomic DNA clean and concentrator kit (Zymo Research Corp. Irvine, CA, USA). The DNA concentration in each sample was estimated by Nanodrop ND-100 Spectrophotometer (Thermo Fisher Scientific, Wilmington, USA). The concentration of DNA was adjusted to 30-60 ng/µL for library construction. The protocol reported by Elshire et al. [80] was used for library preparation using the ApeKI restriction enzyme followed by single-end sequencing at the Genomic Diversity Facility of Cornell University.
The TASSEL 3 GBS pipeline was used to process the raw sequence data [81]. Raw sequences without barcode were removed using the TASSEL plugin and good barcode reads were aligned to the Nipponbare reference genome using Bowtie 2 [82]. SNP calling and filtering were done using the TASSEL pipeline. The SNPs with high genotyping error, low coverage, and high heterozygosity were purged. Duplicate SNPs were merged. The remaining SNPs were manually filtered. SNPs markers with more than 10% missing alleles and SNPs with monomorphic alleles between both parents were discarded. The heterozygous SNPs were considered as missing data.

Linkage Map Construction and QTL Mapping
IciMapping software v.4.1 was used for the construction of linkage map and QTL mapping [83]. Genotypic data were encoded as '2', '0', and '−1' to represent Cocodrie, Dular, and missing alleles, respectively. The SNP markers were grouped based on their physical location on the chromosome. Recombination distance between the markers was calculated using the Kosambi mapping function [84] and was used for the ordering of markers within each chromosome.
The mean value of three replications for each of the nine morphological and physiological traits was used for QTL analysis. Except for CHL and SHL, the data was normally distributed for the remaining seven traits and was used directly for QTL mapping. The data for CHL and SHL were log-transformed to improve the normality before QTL mapping and the traits were labeled as LCHL and LSHL, respectively. Inclusive composite interval mapping (ICIM) was used for detecting additive QTLs for alkalinity tolerance. To declare the significant additive QTLs, a scanning window size of every 1 cM with a 2.0 LOD score was used. The additive effect and phenotypic variation explained by each QTL were estimated. The positions of the right and left flanking markers were used to define the QTL intervals. The QTLs were named using trait name, followed by chromosome location and Mb position of the QTL.

Whole Genome Resequencing of Parents and Detection of SNPs and InDels
The raw whole genome sequencing data of Dular (Accession number: PRJNA284427) was downloaded from National Center for Biotechnology Information (NCBI). Cocodrie was sequenced earlier in our lab at the School of Plant, Environment and Soil Sciences, LSU Agricultural Center and the whole genome sequencing data was submitted to NCBI (Accession number: PRJNA632686). The NGS QC toolkit (v2.3.3) was used for removing adapters and low-quality reads from the FASTQ files of Cocodrie and Dular. Reads with a Phred quality score of more than 30 were used for mapping using Burrows-Wheeler Alignment (BWA v0.7.17) [85,86]. SAMtools (v1.12) was then used to sort and convert the SAM files to BAM file [87]. Before variant calling, the BAM files were processed by Picard tools from the Genome Analysis Toolkit (GATK v4.0) [88]. The genomic variants present between two parents were identified using the Haplotype Caller tool in GATK [89]. VariantFiltration tool was used for stringent filtering and variants with quality depth (QD) below 2, strand odds ratio (SOR) above 3, fisher strand (FS) above 60, mapping quality (MQ) below 40, mapping quality tank sum (MQRankSum) below −12.5, and read position rank sum (ReadPosRankSum) below −8 for SNPs and QD below 2.0, FS above 200, and ReadPosRankSum below −20 for InDels, were removed. After the filtering, SnpEff v5.0e was used for annotation of the variants by selecting polymorphic alleles between Cocodrie and Dular, while sites with synonymous, upstream, downstream, intron, and intergenic variant effects were removed [90]. The physical locations of SNP markers flanking the major effect QTLs were used against the MSU Rice Genome Annotation release 7.0 to make an inventory of genes present within the QTL intervals.

Expression Analysis of Selected Genes by Real-Time Quantitative Reverse Transcription PCR (qRT-PCR)
The seeds of both parents, Cocodrie and Dular, were incubated at 50 • C to break the dormancy and then germinated in Petri plates. After germination, twenty uniform seedlings per replication were transferred to a hydroponics setup using nutrient solution (1 g/L of 20-20-20 Jack's professional fertilizer) at the LSU AgCenter greenhouse. The experiment was conducted with three replication for control and alkalinity stress. The seedlings were allowed to grow under normal conditions. The nutrient solutions were changed after every two days. At the two-leaf stage, plants in the control experiment were allowed to grow in normal conditions and seedlings in the stress experiment were exposed to alkaline stress (0.5% Na 2 CO 3 with pH 10). Leaf samples were collected from both sets of experiments at 0 and 6 h of stress and immediately frozen in liquid N. Leaf samples were then stored at −80 • C until RNA extraction. Three biological replicates per treatment were used to extract total RNA using Trizol reagent. The quality and quantity of RNA were assessed in 1.2% agarose gel and a ND-1000 Spectrophotometer (Thermofisher Scientific, Waltham, MA, USA), respectively. The samples were then treated with PerfeCTa DNase 1 (Quantabio, Beverly, MA, USA) and iScipt™ first strand cDNA synthesis kit (Bio-Rad Laboratories, Hercules, CA, USA) was used to synthesize the first-strand cDNA following the manufacturer's instructions. The genes for the expression analysis were selected from the whole genome sequence analysis. The primers for the qRT-PCR reaction were designed using PrimerQuest (Integrated DNA Technologies, Inc., Coralville, IA, USA) (Table S6). EF1α (LOC_Os03g08010) was used as an internal standard for expression normalization. The reactions were run in three technical replicates from the pooled cDNA samples of the biological replicates on an Applied Biosystems QuantStudio 3 Real-Time PCR System (Thermofisher Scientific, Waltham, MA, USA) using iTaq™ Universal SYBR Green Supermix (Bio-Rad Laboratories, Hercules, CA, USA) [91]. The expression level of genes was determined using 2 −∆∆CT method [92].