Plants of japonica rice (Oryza sativa) cv. Chilbo were grown in plastic pots for 4 weeks in an incubator [14-h light/10-h dark, 28°C (day)/22°C (night), humidity 80%; Younghwa Science, Daegu, Korea]. On average, 300 g of dried soil was used to grow 10 plants in each pot. The effects of drought stress were measured at various time points, including 1, 2, 3, and 4 d after irrigation was withheld and then at 7 d after they were re-watered. Our mock treatment comprised a group of rice plants that continued to receive normal irrigation throughout the experimental period. To observe the physiological features between phyb mutants and wild type segregants, we used samples collected before water deficiency treatment (WD), 3 d after WD, and 7 d after re-watering. In three time points, we observed the root morphology of phyb mutants and wild type segregants.

We used the illumina platform to generate sequence reads (~26 GB) that comprised six transcriptome samples from the 2-d to 3-d drought-stressed plants plus the untreated control. In all, 100-bp paired-end reads were assessed with a FastQC toolkit (Andrews, 2010). Any adapter contaminations and low-quality reads (-phred33 and -q 20) were removed using both Cutadapt (Martin, 2011) and its wrapper tool, Trimgalore (Krueger, 2012). The resultant high-quality reads were taken for our TopHat pipeline, as described (Trapnell et al., 2013). On average, 94% of the filtered reads were mapped to the International Rice Genome Sequencing Project (IRGSP) 1.0 reference genome (Kawahara et al., 2013) and the gene features were estimated based on the gff3 annotation file provided in the Rice Genome Annotation Project (RGAP) database (http://rice.plantbiology.msu.edu/; Ouyang et al., 2007). Differentially expressed genes (DEGs) were evaluated by using Cuffdiff to compare between treatment conditions. Genes with p < 0.05 and log₂ fold-changes >1 (i.e., fold-change >2) were considered differentially expressed. Further screening among the initial DEGs was done based on fragments per kilo-base per million fragments mapped (FPKM) values (Trapnell et al., 2013). The GEO accession number is GSE92989.

To examine GUS expression patterns, we germinated seeds from two promoter trap lines in a Murashige and Skoog (MS) medium for 7 d. The resultant plantlets were then air-dried for 0.0, 0.5, 1.0, 2.0, or 4.0 h. Afterward, whole seedlings from all treatment groups were soaked for 30 min in a GUS-staining solution before their roots were photographed with a camera (Canon EOS 550D; Canon, Tokyo, Japan).

To identify any consensus cis-acting regulatory elements (CREs) in the promoters of our drought-inducible genes, we extracted 2-kb upstream sequences of ATG for LOC_Os04g52290 and LOC_Os07g02710 that had been validated through GUS assays in the current study plus those of LOC_Os09g35790 and LOC_Os02g04650, which have previously been reported as drought-inducible promoters based on the promoter-GUS system (Rerksiri et al., 2013; Jeong and Jung, 2015) from PLANTPAN (http://plantpan2.itps.ncku.edu.tw/; Chang et al., 2008). MEME searches were then performed with those sequences in the FASTA format via the Web server hosted by the National Biomedical Computation Resource (http://meme-suite.org/). We looked for up to four CREs with 10 maximum motif widths. Using the Motif Alignment and Search Tool (MAST) tool, we then searched DNA sequences for matches to the putative TF-binding site motifs found by the motif comparison tool (TOMTOM) within a set of promoter sequences (Bailey et al., 2006).

We employed the GO enrichment tool (Cao et al., 2012) to determine the biological roles of selected genes listed in the Rice Oligonucleotide Array Database (Jung et al., 2008b). This included any genes that were up-regulated during the 3 d of WD treatment. A fold enrichment value higher than standard (1) meant the selected GO term was over-represented. Terms with >2-fold enrichment values were also considered.

The rice MapMan classification systems covers 36 BINs, each of which can be extended in a hierarchical manner into subBINs (Usadel et al., 2005; Urbanczyk-Wochniak et al., 2006). Using diverse MapMan tools, a significant gene list selected from high-throughput data analysis can be integrated to diverse overviews. Here, we generated a dataset carrying locus IDs from RGAP in addition to average log₂ fold-change data for WD vs. mock treatment (control) conditions. For describing any genes up-regulated in response to 3 d of drought, we used four overviews: Metabolism, Regulation, Transcription, and Proteasome.

To evaluate the functional significance of our candidate genes, we compared our gene list with the Overview of functionally characterized Genes in Rice Online database (OGRO, http://qtaro.abr.affrc.go.jp/ogro), which summarize rice genes with known functions (Yamamoto et al., 2012).

Using the Rice Interactions Viewer tool (http://bar.utoronto.ca/interactions/cgi-bin/rice_interactions_viewer.cgi; Chandran and Jung, 2014), we generated a hypothetical protein–protein interaction network involving transcription factors (TFs), kinases, transporters, and functionally characterized genes. The network was edited with the Cytoscape tool (3.2.0 version; Shannon et al., 2003).

Four-week-old seedlings of phyb mutants and wild type (WT) segregants (control) were subjected to drought stress for 3 d (drought-stressed) and then grown for 7 d after they were re-watered (recovered). Recovered, drought-stressed and control root samples were collected, and immediately frozen in liquid nitrogen and stored at −80°C. H₂O₂ measurement, activity of ascorbate peroxidases (APX) and catalase (CAT) enzymes were performed as described previously (Garg et al., 2012). All experimental data were recorded as the means of three independent experiments.

Our quantitative real-time PCR (qRT-PCR) analysis was conducted as follows. Roots were sampled from WT and phyb mutant plants at 4 weeks old and immediately frozen in liquid nitrogen. After total RNAs were isolated using RNAiso kits (Takara Bio, Shiga, Japan), first-strand cDNA was synthesized with MMLV Reverse Transcriptase (Promega, WI, USA) and the oligo(dT) 15 primer. Synthesized cDNAs were amplified using a SYBR Premix Ex Taq (TaKaRa) before qRT-PCR was performed on a Rotor-Gene Q instrument system (Qiagen, Hiden, Germany). For normalizing the amplified transcripts, we used a primer pair for rice ubiquitin 5 (OsUbi5/Os01g22490; Jain et al., 2006). All primers for these analyses are summarized in Table S1.

Four-week-old rice seedlings grown in an incubator were exposed to a water deficit for 1, 2, 3, or 4 d (Figures 1A–D). After then, we rewatered rice seedlings in all stressed conditions and had grown them for 7 d. Most of the rice plants that were drought-treated for 2 days (2 WD) were not recovered but those treated for 3 days were mostly recovered (Figures 1E–H). To identify the DEGs in root samples, we compared between the WD treatment (1–4 d) and mock treatment (control) plants. The expression patterns were then examined for two genes that had been identified as molecular markers of the drought-stress response, i.e., OsDREB2b (LOC_Os05g27930) and OsbZIP23 (LOC_Os02g52780) (Xiang et al., 2008; Matsukura et al., 2010). As expected, stressed roots (from Days 3 to 4) showed increased expression of those genes (Figures 1I,J). After then, we observed the morphological features and measured the dry weight (Figures 2A–E). As a result, the dry weights of the root samples under WD were not changed, while that of untreated control is increasing. In addition, we identified that dry weight after recovery of root samples exposed to WD stress for 2 d increased, whereas that of root samples exposed to WD stress for 3 d decreased and finally the rice plants were mostly died due to the damage augmented by the stress (Figures 2F–J). This result indicates that WD stress for 3 d is a critical time point in our experimental condition to determine the survival under WD stress. Thus, it will be very important to know the mechanism how plant roots overcome unfavorable growth condition caused by long duration expose to WD stress.

Transcriptome analysis was conducted with RNA-Seq technology using roots from control (C) and WD plants (2 or 3 d of water stress). In the 2-d treatment, 716 genes were up-regulated and 787 genes were down-regulated when compared with the control (Figure 3A, Table S2). In the 3-d trial, 1,098 genes were up-regulated and 865 genes were down-regulated (Figure 3B, Table S3). In all, 585 up-regulated genes were common to both test periods (Figure S1). For closer examination, we focused on genes that were up-regulated after 3 d of drought treatment because we had determined that this length of time is riskier for rice plants, based on their survival rates. A heat-map was constructed with data for log₂ fold-change values between WD and control roots, and log₂ intensities in all replicates of 1,963 DEGs, including upregulation and downregulation (Figure 3B, Table S3).

Promoter traps employing the GUS reporter gene system have been used to identify promoters involved in regulating tissue-specific and stress-responsive expression patterns (Jung et al., 2005, 2006, 2015). Our RNA-Seq data analysis revealed 98 genes showing >3 (log₂ scale)-fold upregulation by stress when compared with the control (Table S3). We then searched the potential promoter trap lines of 38 genes and examined GUS expression patterns in 7-d-old seedlings. Among these, the promoter trap lines of two genes (PFG 3A-03417 for LOC_Os04g52290 and PFG 3A-13738 for LOC_Os07g02710) displayed GUS expression in the roots after plants were exposed to a water deficit for 0–4 h (Figures 4A,C, Figure S2). We noted with interest that longer exposure to stress (i.e., 2–4 h) was associated with stronger GUS expression. This drought-related expression was verified by qRT-PCR (Figures 4B,D). However, the other lines did not show GUS activity, indicating that they were not real promoter trap lines. The efficiency (7.2%) showing GUS activity in the promoter trap candidate lines is a little higher than previous reports (Jeon et al., 2000; Jeong et al., 2002). Our findings demonstrated that the promoter trap system, when combined with qualified genome-wide transcriptome data, is a very effective way to identify the activity of an endogenous promoter. This also enables researchers to develop novel promoters.

To identify the CREs involved in the response to a water deficit, we used two promoters from this study plus two that have shown both WD-inducible expression patterns using GUS reporter systems (from previous studies) as well as significant upregulation in our RNA-Seq data (Rerksiri et al., 2013; Jeong and Jung, 2015). In total, the promoters of four genes displaying significant upregulation in stressed roots were analyzed using the MEME tool (Bailey et al., 2006). From these, we found an AACA motif, a target of early activation tagged (EAT) 2 (TOE2), dehydration response element binding factor 3 (DREB3), and dehydration-responsive element-binding 2C factor (DREB2C) elements (Figure 4E, Table 1). Because they were common to all of our selected promoters, we propose that they will be crucial in future efforts to manipulate responses against water deficiencies in rice roots.

To determine the functions of 1,098 genes up-regulated by drought in rice roots, we studied the GO terms of those genes within the “biological process” category. In all, 19 terms were highly over-represented in our gene list, with p < 0.05 and fold-enrichment values of >2 (log₂)-fold, as we have also previously reported (Jung et al., 2008b; Table S4).

Of these, “valyl-tRNA aminoacylation” (14.0-fold enrichment) was the most significantly enriched by drought stress. Metabolic profiling of Pisum sativum L. (Charlton et al., 2008) has shown that valine along with threonine, leucine, and isoleucine are important metabolites that are most highly accumulated in response to a water deficit. Moreover, the levels of branched chain amino acids, such as valine, leucine, and isoleucine are greatly increased under water stress in leaf tissues from cultivars of wheat (Triticum aestivum; Bowne et al., 2012). All of these results suggest that aminoacylation of valyl-tRNA might be activated during drought periods.

Genes for “phospholipid biosynthetic process” (11.5), “thiamin biosynthetic process” (7.7), “cysteine biosynthetic process from serine” (7.4), “glycogen biosynthetic process” (6.3), and “cellular amino acid biosynthetic process” (4.7) also have important roles in the abiotic-stress response (Figure 5). For example, Phospholipase C (PLC) plays a role in catalyzing hydrolysis and AtPLC1 gene is known to be induced by drought and salt stresses (Hirayama et al., 1995). Furthermore, thiamine (vitamin B1) is up-regulated under salinity or osmotic-stress conditions by abscisic acid (ABA) in Arabidopsis thaliana (Rapala-Kozik et al., 2012). Transgenic Arabidopsis plants over-expressing T. aestivum cysteine protease exhibit higher drought tolerance and greater cysteine protease activity under water-stress conditions than do WT plants (Zang et al., 2010). The effects of lipids on signaling, intracellular trafficking, and cytoskeletal organization also have important roles in responses to drought and salinity (Wang et al., 2007). We also identified other significant GO terms such as “catabolic process,” “Transport,” and “etc.” that are significant in the drought response (Figure 5). Therefore, the biological processes identified here as being closely associated with WD might prove to be novel resources for improving our understanding about the molecular mechanism and components involved in conferring plant tolerance to moisture stress.

The MapMan program is very effective for visualizing diverse overviews associated with high-throughput transcriptome data (Jung and An, 2012). We uploaded fold-change data and Locus IDs for 1,098 upregulated genes (Table S3) to various overviews installed in that program.

To investigate the significant metabolic pathways involved in the response to water-deficiency stress, we analyzed the Metabolism overview associated with 1,098 genes (Figure 6). Amino acid metabolism (19 elements), lipids (16), secondary metabolism (14), cell walls (14), nucleotides (5), mitochondrial electron transport (5), photosystems (4), and the Calvin Cycle (3) were clearly related to this stress (Figure 6A, Tables S5, S6). These results implied that rice roots might trigger the above metabolic pathways to increase drought tolerance. Our Regulation overview of 1,098 genes demonstrated that 104 TFs, 29 genes related to protein modification, and 50 genes associated with protein degradation were expressed in rice roots during periods of water stress (Figure 6B). Of these, TFs were the most abundant, meaning that they are largely involved in regulating the response and tolerance of rice to drought conditions. Therefore, those genes are considered potential candidates for further study.

Accordingly, we found nine myeloblastosis (MYB) oncogenes, six Apetala2/Ethylene Responsive Element Binding Proteins (AP2/EREBPs), five Basic Helix-Loop-Helix (bHLH) genes, four homeobox genes, four Chromatin Remodeling Factors, four SET-domains, three WRKY domains, two Cys2His2 (C2H2) zinc fingers, two C3H zinc fingers, two MADS box genes, two members of the TEOSINTE BRANCHED1, CYCLOIDEA, AND PCF FAMILIES, and two basic leucine zipper (bZIP) TFs for water-deficiency stress (Figure 6C, Tables S5, S6). Dehydration-responsive element-binding protein (DREB) is part of a subfamily of the AP2/EREBP and plays a significant role in the abiotic-stress response by specifically binding to the dehydration-responsive element/C-repeat (DRE/CRT) cis-acting element (Mizoi et al., 2012). For example, OsDREB1C, OsDREB1F, and OsDREB2A are constitutively expressed under drought conditions (Dubouzet et al., 2003; Wang et al., 2008), and overexpression of OsDREB1F, OsDREB1G, and OsDREB2A improves tolerance to moisture stress by transgenic rice plants (Chen et al., 2008; Wang et al., 2008; Cui et al., 2011). Likewise, bZIP, MYB, and WRKY genes have been explored for their ability to improve drought tolerance in rice. OsbZIP16, OsbZIP23, OsbZIP46, OsbZIP71, and OsbZIP72 have critical functions in ABA signal transduction and they positively regulate drought tolerance (Xiang et al., 2008; Lu et al., 2009; Chen et al., 2012; Tang et al., 2012; Liu et al., 2014). Rice gene OsMYB4 has a positive role in conferring drought tolerance when expressed in transgenic plants of Malus domestica and Solanum lycopersicum (Vannini et al., 2007; Pasquali et al., 2008). As an R2R3-type MYB gene, OsMYB2 over-expressed in plants makes them more tolerant to salt, cold, and dehydration while overexpression of OsMYB48-1, another MYB-related TF, increases drought and salinity tolerance in rice (Yang et al., 2012; Xiong et al., 2014). Increased expression of OsWRKY11 under the control of HSP101 promoter also leads to enhanced drought tolerance (Wu et al., 2009), as does overexpression of OsWRKY45 (Tao et al., 2011). In addition, various TFs, e.g., NAC, bHLH, Zinc fingers, and homeobox genes, are involved in the plant response to water deficits (Zhang et al., 2016).

We found that the responses of RING-finger E3 ligases (18 elements), Ubiquitin E3 F-Box (5), and ubiquitin protease (2) were clearly related to drought stress (Figure 6D, Tables S5, S6). For example, OsDIS1 is a C3HC4 RING finger E3 ligase, and the RNAi transgenic plants are tolerant to drought stress (Ning et al., 2011). The RING-finger containing E3 ligase O. sativa SALT-AND DROUGHT-INDUCED RING FINGER 1 (OsSDIR1) is a candidate gene for engineering drought tolerance in various crop plants. Its expression is also up-regulated by excess NaCl (Gao et al., 2011). The expression patterns of 47 OsRFP genes in response to abiotic stresses have been monitored via semi-quantitative reverse transcription PCR and in silico analysis (Lim et al., 2013). Our results also suggest a strong correlation between drought and activity of RING finger E3 ligase.

To evaluate the significance of our candidate genes, we searched the literature to determine if functions for them have been reported previously. This was accomplished with the online OGRO database, which provides a thorough summary of rice genes that have been characterized through molecular and genetic techniques (Yamamoto et al., 2012). Of the 41 genes found in that database (Table 2), 18 have been linked to various abiotic-stress responses in rice. They include PhyB (Liu J. et al., 2012), OsbZIP52/RISBZ5 (Liu C. et al., 2012), OsETOL1 (Du et al., 2014), trehalose-6-phosphate synthase 1 (OsTPS1; (Li H. W. et al., 2011), OsTZF1 (Jan et al., 2013), and sHSP17.7 (Sato and Yokoya, 2008) for drought; OsCIPK15 (Xiang et al., 2007), OsGAPC3 (Zhang et al., 2011), glyoxalase I (OsGLYI-11.2; Mustafiz et al., 2014), Programmed cell death 5 (OsPDCD5; Yang et al., 2013), OsPLDα1 (Shen et al., 2011) OsTPS1 (Li H. W. et al., 2011), and OsTZF1 (Jan et al., 2013), for salinity; OsbZIP52/RISBZ5 (Liu C. et al., 2012) OsLti6b (Kim et al., 2007), OsTPS1 (Li H. W. et al., 2011), and v1 (Kusumi et al., 2011) for chilling; and ETHYLENE OVERPRODUCER 1-like (OsETOL1) for tolerance to submergence (Du et al., 2014; Table 2). These findings indicate that our candidate genes are potentially involved in plant responses to abiotic stress (including drought) and suggest the possibility of crosstalk among related pathways.

Regulatory genes are primary targets when investigating diverse stress responses and developmental processes. Among the 1,098 genes that were up-regulated by 3 d of drought treatment, we identified 104 TFs, 44 kinases, 27 transporters, and 40 functionally characterized genes (Figure 7, Table 2, Table S6). Understanding the regulatory relationships among them can improve our ability to develop novel strategies for enhancing plant tolerance to various stresses. To increase our knowledge, we utilized the Rice Interactions Viewer to generate a hypothetical protein–protein interaction network associated with those up-regulated genes mentioned above (Ho et al., 2012). Four categories were used as query in the refined network: 10 functionally characterized genes (red lettering in large circles of Figure 7), 12 TFs (large brown circles), four kinases (large yellow circles), and three transporters (large sky-blue circles). We also highlighted elements with multiple interactions among these regulatory and functionally characterized genes. Details about locus IDs, gene names, and putative functions are provided in Table S7.

Of the 41 genes with known functions, alleles with T-DNA insertions were found for mutants of phytochrome B (phyb)/LOC_Os03g19590 (Figure 7, number 10). As we have reported previously (Liu J. et al., 2012), a drought-tolerant phenotype has been confirmed in 30-d-old greenhouse-grown seedlings of phyb mutant Line 4A-02226, which has a T-DNA insertion in the 3rd intron (Figures 8A–F). This study informed that drought tolerance in phyb mutant is caused by reduction of stomatal density but not by root morphology. However, in our study, the comparison of dry weight between WT and phyb mutant roots exposed to WD stress for 3 d revealed that phyb mutant roots have 48.5% less dry weight than that of WT. More interestingly, after recovery process, WT did not survive and the dry weight of the root was reduced, while phyb mutant survived and the dry weight of roots even increased due to the continuous growth of the roots (Figures 8G–J). This result indicates that root development is very important for the drought stress tolerance and also we are interested in the mechanism how rice roots overcome damage caused by prolonged WD stress. It was known that processing of reactive oxygen species (ROS) play important roles in root development of other plant species (Causin et al., 2012; Zhao et al., 2016).

Several studies have reported a correlation between major active enzyme activity and drought (Huang et al., 2010; Huda et al., 2013; Xu et al., 2016). The enzyme-catalyzed reactions are the main mechanisms for the removal of superoxide and H₂O₂. Especially, ascorbate peroxidases (APXs) and catalases (CATs) are known to be important enzymes involved in H₂O₂ removal under drought stress (Noctor et al., 2014). Therefore, we analyzed the expression patterns of CAT and APX antioxidant enzyme families using real-time PCR analyses for the samples under normal condition, drought stress condition for 3 d, and recovery after drought stress. As a result, we found that expressions of five of the eight APX genes and one of the three CAT genes were upregulated in phyB mutant under drought stress, suggesting the major contribution of APX to remove ROSs produced by drought stress in rice roots (Figure 9A). Hydrogen peroxide (H₂O₂) contents were measured to confirm the accumulation of ROS in the roots of phyb mutant. Subseuently, phyb mutants always maintained high H₂O₂ contents compared to wild type and in addition showed the highest H₂O₂ contents under drought stress (Figure 9B). Measurement of APX enzyme activity between phyb mutants and wild type segregants demonstrates that phyb mutants have the better performance of APX activity at drought stress for 3 d than that of wild type. In addition, the CAT activity of the phyb mutants was consistently higher than that of WT, although it decreased during drought stress rather than normal condition (Figure 9C). Therefore, we estimate that phyb mutant might retain the tolerance to drought stress by effective ROS scavenging in root besides controlling stomatal density in leaf.

Our candidate genes for conferring tolerance to water deficiencies were identified from RNA-Seq transcriptome analysis of root samples from 4-week-old soil-grown rice plants. The samples were well-qualified for transcriptome analysis based on time-course physiological responses of roots under water deficiency and recovery compared to controls as well as expression patterns analysis with marker genes. Our transcriptome data is useful for identifying novel targets for drought tolerance. When RNA-Seq data were compared with public microarray data under drought stress conditions, 32.05% of those genes (352 of 1,098) showed similar trends in their stress responses (Table S8). However, the others are unique in our experiments, suggesting novel candidate genes for future studies. In addition, the RNA-Seq approach can provide more candidates that cannot be discovered by microarray technology because the latter is able to detect the expression profiles of fixed-target transcripts. In parallel with this description, we did not identify Affymetrix probe IDs for 279 genes (i.e., 25.40% of the total) of DEGs after 3 d of drought treatment. By contrast, the most popular array system in rice, Affymetrix array platform, covers ~90% of the annotated genes in the rice genome (Jung et al., 2008a; Chandran and Jung, 2014). Therefore, unique DEGs under drought treatment in this study might be valuable for future studies and further applications to understand novel regulatory mechanism for drought tolerance responses.

MapMan analysis can reveal a global view of the metabolic pathways associated with the drought-stress response in roots. Many rice genes in pathways for carbohydrate, cell wall, lipid, and amino acid degradation are stimulated when the water supply is limited, thereby implying that plants use those metabolites and energy resulted from degradation pathways to tolerate stressful growing conditions. For example, a unique Ni²⁺-dependent and methylglyoxal-inducible GLY I functions in the response and adaptation of rice to abiotic stresses (Mustafiz et al., 2014). The phospholipase Dα (PLDα) mediates H(+)-ATPase activity and is involved in salt tolerance (Shen et al., 2011; see also Figure 6A, Table 2). Therefore, we propose that the ubiquitin-mediated pathway for protein degradation associated with our candidate genes is evidence that F-box E3 ligases or RING box E3 ligases have dominant roles in the process. For example, OsRZFP34 (LOC_Os01g52110), encoding a rice RING zinc-finger protein, is positively involved in the opening of stomata, even under ABA treatment, and might have a potential role in abiotic-stress tolerance (Hsu et al., 2014). Moreover, the D3 (LOC_Os06g06050) gene, encoding an F-box leucine-rich repeat protein orthologous to Arabidopsis MAX2/ORE9, is involved in tillering and shoot-branching (Ishikawa et al., 2005). Our transcriptome data also suggest additional roles in the drought response (Figure 6D, Table 2). Therefore, we believe that many genes within the degradation pathways of primary metabolites, as well as biosynthetic pathways for secondary metabolites, positively influence drought tolerance in rice roots. However, functional elucidation of major, yet uncharacterized, regulatory genes could lead to novel ways for improving the plant response to a water deficiency.

In our network analysis, we found that LOC_Os08g39140, encoding heat shock protein 90 (HSP90) in circle 71 of the network, potentially interacts with seven gene products. Among them, four have known functions (i.e., Receptor for Activated C-Kinase 1A, or RACK1A; PDCD5, TPS1, and Metal Tolerance Protein 1, or MTP1) while the others are TFs (Figure 7). Heat-shock proteins/chaperones play a crucial role in protecting plants through proteins and membranes stabilization and protein re-folding under stress conditions (Wang et al., 2004). Among the products that interact with HSP90, three are related to abiotic-stress responses, indicating significant roles for HSP90. The three include OsPDCD5, involved in salinity (Yang et al., 2013); OsTPS1, involved in drought, salinity, and cold (Li H. W. et al., 2011); and OsMTP1, which improves plant tolerance to excess heavy metals such as Zn and Cd (Yuan et al., 2012). In addition, RACK1A functions in the production of ROS and resistance to rice blast infection (Nakashima et al., 2008). One inevitable consequence of drought stress is that more ROS are generated in diverse cellular compartments, e.g., the chloroplasts, peroxisomes, and mitochondria. However, ROS accumulations in response to a water deficiency are tightly controlled by a versatile and cooperative antioxidant system that modulates intracellular ROS concentrations and sets the redox-status of a cell (Cruz de Carvalho, 2008). These activities suggest that RACK1A is involved in a ROS-mediating drought-stress pathway.

The three other TFs with important roles in drought tolerance include, first, a zinc finger domain (circle 11 in our network). Overexpression of ZFP36 in rice elevates the activities of antioxidant enzymes and enhances plant tolerance to water and oxidative stresses (Zhang et al., 2014). Second, a loss-of-function mutant of a drought and salt tolerance gene (DST) stimulates stomatal closure and reduce stomatal density, resulting in greater drought and salt tolerance in rice (Huang et al., 2009). Finally, an upstream activation factor (UAF in circle 16) regulates the pace of transcriptional activity by binding to an upstream element of target genes in response to diverse signals (Carthew et al., 1985). Therefore, our network data that integrate this diverse functional information will be useful resources when designing more detailed molecular mechanisms for future applications that can enhance tolerance to a water deficiency.

Hydrogen peroxide (H₂O₂) is a signaling molecule involved in the regulation of specific biological and physiological processes and is a relatively stable non-radical ROS. However, abiotic stresses such as drought increase the production of H₂O₂ in plants, which can cause serious damage to biomolecules (Slesak et al., 2007). To solve this problem, plants are endowed with H₂O₂-metabolizing enzymes, among which CAT and APX play the most important role (Sofo et al., 2015; Li et al., 2016). For example, Panicum sumatrense with strong drought tolerance has been reported to increase the activity of antioxidant enzymes such as CAT and peroxidase during drought stress (Ajithkumar and Panneerselvam, 2014). Olive plants also increase CAT activity in the leaves when water deficit conditions, thereby limiting cellular damage due to ROS (Sofo et al., 2004). As the drought increased in soybeans, a clear increase in the amount and activity of APX was observed and APX2 overexpressing plants in rice showed increased APX activity and improved stress resistance (Kausar et al., 2012; Zhang et al., 2013). These facts support that CAT and APX are antioxidant enzymes that play a key role in recovery from drought stress (Faize et al., 2011; Pinheiro and Chaves, 2011). We confirmed an increase in APX activity in the phyb mutant roots during drought stress. In addition, we observed that phyb mutant had higher CAT activity than WT under any circumstances (Figure 9C).

Despite the higher level of APX and CAT activity, phyb mutants have higher H₂O₂ accumulation than wild type plants. One possible explanation on this phenamenum is that rice plants under drought stress might use H₂O₂ as a second messenger to trigger ROS removing pathway. It was demonstrated that H₂O₂ plays key roles in processing external biotic and abiotic stimuli thorugh signal transduction pathways in plant cells (Petrov and Van Breusegem, 2012). In addition, several reports in rice support the significance of increased H₂O₂ in tolerance responses against drought stress: overexpression of abscisic acid, stress, and ripening 5 (OsASR5) gene exhibits a drought resistant phenotype despite higher H₂O₂ content under drought stress compared to wild type (Li J. et al., 2011); drought and salt tolerance (dst) mutant has increased stomatal closure and reduced stomatal density due to accumulated H₂O₂, consequently resulting in enhanced drought tolerance in rice (Huang et al., 2009); rice similar to RCD one (SRO/OsSRO1c) promotes stomatal closure and H₂O₂ accumulation via stress-responsive NAC 1 (SNAC1) and DST regulators, resulting in tolerance responses against drought stress (You et al., 2013).

Recent studies indicate that phytochrome influences the growth and development of roots through the role of photoreceptors and their associated signaling mechanisms. For example, in Arabidopsis, light was efficiently conducted through the stems to the roots, resulting in the accumulation of ELOTATED HYPOCOTYL 5 (HY5) protein by photoactivated phyb (Lee et al., 2016). We therefore expect that OsPhyB acts as a negative regulator of APX and CAT activity by a transcription factor or phytochrome interacting factor (PIF) during drought stress in rice roots (Figure 10). Details of this model need to be elucidated through further studies.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.