Rice is one of the most important staple foods and plays an important role to ensure food safety. However, rice production consumes copious amounts of fresh water. In China, 49% of all fresh water resources are used for rice production (Zhang, 2007). Due to environmental damage and anomalous climatic variation, drought has become one of the main limiting factors for global grain production (Pennisi, 2008). Throughout Asia, 20% of all rice production areas are affected by drought per year (Gowda et al., 2011). This strengthens the disparity between supply and demand of rice production. Therefore, enhancing the drought resistance of rice is vital to reduce the effect of drought on rice production.

Drought, which is a soil water deficit, can result in insufficient moisture for a plant to adequately grow and complete its life cycle. Water shortage will seriously affect rice growth and lead to a series of physiological and biochemical changes. Many studies focused on plant drought resistance. These studies reported that drought usually leads to a delayed flowering time, restrained plant growth, and ultimately loss of yield in rice (Liu et al., 2005; Yue et al., 2005; Zou et al., 2007). The reduction in yield is due to a reduced spikelet number or reduced fertility of spikelet caused by drought stress (Ekanayake et al., 1989; Liu et al., 2010). In addition, drought can cause the change of antioxidant system and osmotic regulatory system in rice, accumulating antioxidants, and osmotic regulation substances (Ouyang et al., 2010).

With the development of molecular genetic technologies, more and more studies focused on the discovery of drought resistant genes and their function. These results are useful for breeders to develop drought resistant cultivars. QTL (quantitative trait loci) mapping used to be the main method to discover drought resistance loci. For example, many drought resistance-related QTLs were identified by using a recombinant inbred line (RIL) population derived from a cross of lowland indica rice Zhenshan 97 and upland japonica rice IRAT109 in previous studies (Yue et al., 2006; Liu et al., 2010). Based on the above drought resistance-related QTLs, 17 near-isogenic lines (NILs) were constructed and phenotypic variations of these NILs were investigated under drought and normal conditions, among them, qFSR4 was fine mapped for spikelet number, flag leaf width and root volume, further analysis showed that NARROW LEAF 1 (NAL1) regulating leaf width was located in the qFSR4 interval (Qi et al., 2008; Ding et al., 2011). Some studies have shown that NARROW LEAF 1 can affect root development and drought resistance in rice (Fujita et al., 2013; Cho et al., 2014). In addition, root traits are very important traits for drought resistance in plant. Dro1, a major QTL involved in deep rooting of rice was cloned by fine-mapping, it can increase rice yield under drought conditions (Uga et al., 2011, 2013). And qRT9, a QTL controlling root thickness and root length was mapped an 11.5 kb interval in upland rice (Li et al., 2015). Although linkage analysis for drought resistance has made some achievements in rice, many QTLs/genes of drought resistance remain hidden. Due to a rapid development of sequencing techniques, genome-wide association studies (GWAS) became a new method for gene mining of target traits. Compared to QTL analysis, GWAS is based on natural populations, and can detect multiple alleles at the same site (Flint-Garcia et al., 2003). GWAS have widely been used in human genetic studies as well as in plants, such as Arabidopsis, maize, sorghum, and rice (Atwell et al., 2010; Huang et al., 2011; Tian et al., 2011; Li et al., 2012; Riedelsheimer et al., 2012; Morris et al., 2013; Chen et al., 2014; Wen et al., 2014; Yang et al., 2014, 2015; Wang et al., 2015; Yano et al., 2016). Identification of the allelic variation underpinning the phenotypic diversity in rice will result in enormous practical implications for rice breeding.

Genome-wide association study (GWAS) has been widely used for the genetic analysis of rice. Huang et al. (2010) constructed the first high-density haplotype map in rice via re-sequencing of 517 landraces and cultivars, and they identified 37 strong association loci for 14 agronomic traits via GWAS. Their study was considered as the coming of age for GWAS in rice (Clark, 2010). The genetic architecture of rice metabolism was dissected via GWAS, and five candidate genes were identified or annotated (Chen et al., 2014). Recently, a high-throughput phenotyping platform was developed for GWAS, achieving good results and further promoting GWAS for the dissection of genetic mechanisms for other traits (Yang et al., 2014).

In previous study, GWAS has been used to dissect drought resistance in maize (Lu et al., 2010; Wang et al., 2016; Zhang et al., 2016). However, a systematic examination of drought resistance in rice via GWAS was still called for. In this study, we conduct GWAS on a natural rice population under well watered and drought stress conditions, exploring the drought resistance of rice. The collection contains a wide range of genetic variation (Wu et al., 2015), highlighting its suitability for mapping studies aimed at detecting genetic variation and segregation in a diverse set of rice varieties. Using this strategy, 18, 5, and 6 loci could be associated with plant height, grain yield per plant, and drought resistant coefficient. Nine known genes were identified for three traits. The function of these genes is either directly or indirectly related to the three target traits. We also discovered new loci that contribute to these three traits and that were missed by previous studies. Among these, OsRLK5 was preliminarily confirmed to positively regulate yield-related traits under drought condition.

Drought is one of the main abiotic stresses affecting the production of most field crops. Drought stress can occur at any crop growth stage and can affect productivity to variable degrees depending on the onset time, duration, and intensity of drought. Drought stress typically suppresses plant growth at the vegetable stage and has been reported to directly reduce production at the reproductive stage (Yue et al., 2005, 2006). Agronomists and breeders have made significant progress in improving the drought resistance of crops, however, the genetic and molecular basis for drought resistance in crops remains largely unclear. In this study, drought resistance of the rice germplasm was evaluated using contrasting soil moisture regimes. Combining high density SNPs and phenotypes of this germplasm, we conducted GWAS for plant height, grain yield per plant, and drought resistant coefficient. Five, two, and two known genes for the same three traits were found within or close to the associated loci, respectively. These associated loci and candidate genes could enhance the knowledge for understanding the mechanisms underlying rice drought resistance and could be used in improvement of drought resistance of rice via molecular breeding. In addition, many drought resistant germplasm accessions were obtained in this study that should be valuable in rice breeding program facing water scarcity.

Compared to linkage analysis, GWAS greatly accelerated both speed and accuracy of locating QTL or candidate genes. For example, OsSPL13 has been quickly identified based on genome-wide association study for the grain size and expression profiling in panicles between small-grain and large-grain varieties. Further analysis indicated that OsSPL13 positively regulates cell size in the grain hull, resulting in enhanced grain length and yield in rice (Si et al., 2016). In this study, nine reported functional genes were found to be near the significant GWAS loci (Table 4). The function of these genes is consistent with their associated traits. The discovery of the novel drought resistant gene OsRLK5 and the new functions of OsGAN1 in this study further support GWAS as a viable approach to quickly identify new genes that are contributing to the complex traits.

Numerous factors affect plant height of rice. Gibberellin (GA) is one of the most important determinants (Monna et al., 2002; Sasaki et al., 2002; Spielmeyer et al., 2002). In this study, several candidate genes (SD-1, OsGA2ox3, and OsDOG), controlling rice plant height were involved in synthesis or signal transduction of GA. For example, SD-1 encodes a gibberellin 20-oxidase, a key enzyme in the biosynthesis of gibberellin catalyzing (GA53 → GA44 → GA19 → GA20) (Monna et al., 2002; Sasaki et al., 2002; Spielmeyer et al., 2002). Interestingly, the GWAS approach also found that SD-1 is important for plant height under drought stress condition (Figure 1), suggesting that SD-1 may be involved in drought resistance. OsGA2ox3 was also involved in the GA metabolic pathway, and it has been reported that OsGA2ox3 regulates plant height and tiller number (Lo et al., 2008). Another candidate gene for rice height, OsDOG, has been reported to act as a new dynamic equilibrium regulator for the gibberellin metabolism. It has furthermore been reported that OsDOG negatively regulates cell elongation and plant height in rice (Liu et al., 2011). OsSAP11 was an A20/AN1 zinc-finger gene and contained stress-associated proteins. Over-expression of OsSAP11 can enhance the tolerance to water deficit and salt stress in transgenic Arabidopsis plants (Giri et al., 2011). OsDOG/OsSAP11 was identified within the important locus for plant height, especially as a significant signal with lower p-value under drought stress conditions.

Auxin plays an important role in plant height. In this study, OsGH3-2 was identified at the qPH1.2 interval, encoding an enzyme catalyzing the IAA conjunction to amino acids. OsGH3-2 has been reported to be involved in the regulation of plant height, root and stomatal development, as well as in the modulation of ABA level and drought resistance (Du et al., 2012). Two more candidate genes (LOC_Os09g08130 and LOC_Os10g06710) that were detected for plant height are also related to the auxin metabolism. LOC_Os09g08130 was annotated as indole-3-glycerol phosphate synthase and was located upstream of the indole alkaloid biosynthesis pathway. LOC_Os10g06710 was annotated as amidase, involved in the metabolic pathway of indole acetic acid. Indole acetic acid is an important phytohormone and plays a crucial role in cell division, differentiation, and elongation, in root development and plant height regulation (Petersson et al., 2009; Du et al., 2012; Lu et al., 2015).

In particular, a significant SNP (sf0918825447) with p-value of 4.18E-07 for plant height is located at the promoter region of OsGNA1. The associated peak signal was detected with a lower p-value under drought stress condition compared to normal water condition, implying its contribution to drought resistance. A previous study demonstrated OsGNA1 to play an important role in root development. Compared to wild type rice plants, mutant gna1 plants exhibited reduced root elongation, shorter taproot, lateral root and root hairs (Jiang et al., 2005). Similarly, the mutant of OsGNA1 exhibited shorter root length and seedling height compared to wild type plants in this study. These results suggest that OsGNA1 can regulate plant height and drought resistance. A haplotype analysis revealed that Hap5 was likely to improve drought resistance in rice because Hap5 had the highest PH at the adult stage under drought stress and had the longest maximum root length at seedling stage. Moreover, Hap5 had the highest DRC (Table S3).

In this study, the candidate genes for drought responsive associated loci were discovered together with a set of RNA-Seq data for rice plants under contrasting moisture regimes (unpublished). A total of 162 genes showed significantly different expression levels in leaves between drought treated and normal water treated plants. Two functional genes, OsPYL2 and OsGA2ox9 were discovered in this study. OsGA2ox9 was involved in GA metabolic pathways, and a previous study reported it to control plant height, tiller number, and root length in rice (Lo et al., 2008). Roots play an important role for absorbing water and nutrients. Plants can benefit from longer roots for water uptake from deeper soil layers as drought stress occurs. These above results imply that OsGA2ox9 was likely affecting drought resistance via regulation of root growth in rice. OsPYL2 was reported to be an ABA receptor (Tian et al., 2015). ABA is involved in the adaptation of plants to abiotic stress. The combination of ABA and the ABA receptor can activate the ABA signal transduction and trigger the corresponding physiological reaction, including stomatal closure (Cutler et al., 2010; Kim et al., 2010). Thus, OsPYL2 is likely to regulate stomatal behavior to reduce water loss, then to enhance the drought resistance. Furthermore, our preliminary research verified the function of OsRLK5 to be positively related to the leaf water content and grain yield under drought stress (Figures 7, 8).

Two known genes (OsCYP51G3 and OsRRMh) were mapped for grain yield in this study, and the functions of these genes correlate well with the grain yield trait. OsCYP51G3 has been reported to mediate the biosynthesis of phytosterols and brassinosteroids, and to regulate seed setting rate (Xia et al., 2015). OsRRMh plays an important role during the transition from vegetable to reproductive phase in rice. Compared to the wild-type, the OsRRMh RNAi lines exhibit enlarged panicles while over-expression of OsRRMh lines lead to lower fertility rates and less number of spikelets per panicle (Liu and Cai, 2013).

Genome-wide association study (GWAS) based on large-scale re-sequencing provides a powerful tool to discover genetic variants that can be used for crop improvement, including improvement of drought resistance. GWAS has been successfully used at a high resolution to uncover associations that involve complex traits from inbred lines of field-grown maize and rice landraces and cultivars with genetic variants (Huang et al., 2010; Lu et al., 2010; Chen et al., 2014; Wen et al., 2014; Wu et al., 2015; Yano et al., 2016). Using high density SNPs at a whole-genome level, GWAS provides high-resolution genetic mapping that can narrow down the associated regions to candidate genes. Our study demonstrates that known genes as well as novel genes and loci can be dissected via GWAS analysis. Moreover, candidate genes discovered in this study provide useful information for further studies in drought resistance of crops. Functional identification of these candidate genes will improve our understanding of the mechanisms of crop responses to drought stress.

LL, HL, HM, and XM designed this study. XM, KX, SC, TL, and XL performed the experiments. XM, FF, and HW analyzed the data. XM, HL, and LL drafted the manuscript.

This work was supported by grants from the National Program for Basic Research of China (2012CB114305); the National High-Tech Research and Development Program of China (2014AA10A603); the Shanghai Municipal Commission of Science and Technology (15DZ2290700; 16ZR1431200); the Project of Subject Construction, Shanghai Academy of Agricultural Sciences, Grant No. SAAS-2016(02); the Youth Talent Development Project supported by the Shanghai Municipal Agriculture Commission (No. 2015-1-6).