The generation and basic molecular properties of OsNRT2.1/OsNRT2.2 over-expression transgenic lines (OE1, OE2, OE3) was previously described in Chen et al. (2016). The genetic background of OsNRT2.1 over-expression transgenic lines was cv. Wuyunjing7 (WYJ). The generation and basic molecular properties of OsNAR2.1 RNAi transgenic lines (r1, r2) was previously described in Yan et al. (2011). The genetic background of OsNAR2.1 RNAi transgenic lines was cv. Nipponbare (WT).

We amplified the OsNAR2.1 or OsNRT2.3a open reading frame sequence from cDNA isolated from Oryza sativa L. ssp. Japonica cv. Nipponbare using the primers listed in Supplementary Table S1. The PCR products were cloned into the pMD19-T vector (TaKaRa) and confirmed by restriction enzyme digestion and DNA sequencing. The restriction sites (AscI and KpnI) or (SpeI and SpeI) were incorporated into primers to facilitate cloning into the pUbi:OsNAR2.1 vector or the p35S:OsNRT2.3a vector, respectively. These constructs were introduced into Agrobacterium tumefaciens strain EHA105 by electroporation and then transformed into cv. Nipponbare or cv. Wuyunjing7 as described previously (Chen et al., 2016) (Supplementary Figure S1).

To determine seedling survival under drought stress, the seedlings were grown for 21 days under highly irrigated conditions by maintaining 10 equal-sized seedlings per pot. Irrigation was withheld for 12 days, followed by re-watering for 7 days. Photographs were taken before the stress (0 days), at the end of the stress period, and a week after recovery. Five biological replicates were maintained among all lines.

For hydroponic drought stress experiments, rice seedlings were grown with the normal IRRI solution for 2 weeks and then transferred to nutrient solution supplemented with 10% (w/v) PEG6000 for 2 weeks. The hydroponic experiments were carried out in a growth room with a 14 h light (30°C) (8:00 – 22:00)/10 h dark (22°C) (22:00 – 8:00) photoperiod and 60% relative humidity. In all treatments, nutrient solutions were replaced every 2 days. At the end of the drought stress experiments, measurements of chlorophyll content, relative water content, photosynthesis rate and water use efficiency were recorded.

The rice plants for the limited water supply experiment were cultivated in plots at the experimental site of Nanjing Agricultural University, Nanjing, Jiangsu, China. The plots size was 1 m × 1 m and the seedlings were planted in a 5 × 5 array. Urea was used as nitrogen fertilizer, and 180 kg N/ha of fertilizer was applied on each plot. All plants were normally irrigated for 2 weeks after transplanting. Then limited water supply (60% field capacity) was applied to the experimental group and fully irrigated conditions (100% field capacity) was applied to controls until completion of the life cycle, and its agronomic characteristics were analyzed.

RNA was extracted from the root or shoot of seedlings under well-watered and drought stress conditions. Total RNA was extracted using TRIzol reagent (Vazyme Biotech Co., Ltd., China). DNase I-treated total RNAs were subjected to reverse transcription (RT) with HiScript II Q Select RT SuperMix for qPCR (+gDNA wiper) kit (Vazyme Biotech Co.). Triplicate quantitative assays were performed using the 2 × T5 Fast qPCR Mix (SYBRGreenI) kit (TsingKe Co, Ltd., China). The primers for qRT-PCR are shown in Supplementary Table S2.

Genomic DNA was extracted from leaves of WT and transgenic plants using the SDS method (Chen et al., 2016). 60 μg of DNA was digested with EcoRI and HindIII restriction enzymes 10 h at 37°C. The digested DNA was separated on a 1% (w/v) agarose gel and transferred to a Hybond-N⁺ nylon membrane, and hybridized with the coding sequence of hygromycin-resistance gene as hybridization probe (Chen et al., 2016).

The degree of relative chlorophyll content in the fully expanded last leaf was determined using a SPAD-502 Chlorophyll Meter (Minolta Co.). Relative water content was measured as described (Ramegowda et al., 2014) in the leaves for chlorophyll content measurements. Relative water content = (fresh weight - dry weight)/(turgid weight - dry weight) × 100%.

The same leaf is used to measure photosynthetic rate. Photosynthesis rates were measured in rice seedlings using a Li-COR6400 portable photosynthesis system equipped with a LED leaf cuvette (Li-COR, Lincoln, NE, United States). Water use efficiency was calculated using photosynthesis measurements and the transpiration rate (Ramegowda et al., 2014). Water use efficiency = photosynthesis/transpiration rate.

Rice seedlings of WT and transgenic plants were grown in IRRI nutrient solution containing 1 mM NH₄⁺ for 2 weeks and then nitrogen starved for 5 days. The plants were transferred first to 0.1 mM CaSO₄ for 1 min, then to a complete nutrient solution containing 1.25 mM ¹⁵NH₄NO₃ or 1.25 mM NH₄¹⁵NO₃ (atom % ¹⁵N: ¹⁵NO₃₋, 99%; ¹⁵NH₄⁺, 99%) for 5 min, and finally to 0.1 mM CaSO₄ for 1 min. And the ¹⁵N influx rate was calculated following the method in Chen J. et al. (2017).

All data were analyzed by ANOVA using the statistical SPSS 10 program (SPSS Inc., Chicago, IL, United States). The different letters indicate a significant difference between the transgenic line and the WT (P < 0.05, one-way ANOVA).

The expression of OsNAR2.1 and OsNRT2s under drought stress was analyzed by qRT-PCR. The results showed that the expression of OsNAR2.1, OsNRT2.1, OsNRT2.2, and OsNRT2.3a was affected by 10% PEG6000 treatment and OsNRT2.4 did not show any response to drought stress (Figure 1). Expression of OsNAR2.1 and OsNRT2.3a was increased by 10% PEG6000. As a result, the expression of OsNRT2.1 and OsNRT2.2 was inhibited. Among them, OsNAR2.1 expression showed the most drastic change under drought stress, the expression increased by nearly 4 times that of fully irrigated conditions (Figure 1).

Results showed that the expression of OsNAR2.1 was significantly enhanced in both shoots and roots of plants treated with 10% PEG6000 (Figure 2). Overall, OsNAR2.1 expression in roots was significantly higher than in shoots, which was consistent with published work by Feng et al. (2011). In roots, OsNAR2.1 expression increased after 10% PEG6000 application and reached maximal expression levels after 6 h, followed by a gradual decline with prolonged treatment (12–72 h) to a 3 to 5-fold increase (Figure 2A). In shoots, the expression of OsNAR2.1 reached the highest level after 12 h of treatment, with a 3.5-fold increase (Figure 2B).

The OsNRT2.1 over-expression transgenic lines were designated as OE1, OE2 and OE3 and OsNAR2.1 RNAi transgenic lines as r1 and r2 previously used in Chen et al. (2016) and Yan et al. (2011). We also performed molecular verification to confirm these lines. qRT-PCR analysis showed that the expression of OsNRT2.1 in OE1, OE2 and OE3 increased significantly as compared with WYJ, (Supplementary Figure S2A). Southern blot analysis showed that OsNAR2.1 RNAi transgenic lines r1 and r2 are two independent lines (Supplementary Figure S3A) and, compared with WT, the expression of OsNAR2.1 in root and shoot of r1 and r2 decreased significantly (Supplementary Figure S3C).

We used the Agrobacterium tumefaciens-mediated method to generate transgenic lines. The p35S:OsNRT2.3a expression vector was inserted into Wuyunjing7 (O. sativa L. ssp. Japonica cv., WYJ) and the pUbi:OsNAR2.1 expression vector was inserted into Nipponbare (O. sativa L. ssp. Japonica cv., WT). On the basis of Southern blot analysis and RNA expression level, we selected three independent lines of p35S:OsNRT2.3a designated as WA1, WA2, and WA3 (Supplementary Figures S2B,C), two independent lines of pUbi:OsNAR2.1 designated as Ox1 and Ox2 (Supplementary Figures S3B,D).

To test the involvement of OsNRT2.1, OsNAR2.1, and OsNRT2.3a in drought tolerance, the drought stress response of their transgenic lines was tested at the vegetative stage. Irrigation was withheld for 12 days, followed by re-watering for 7 days (Figure 3A). The survival rate of WYJ, OsNRT2.1 and OsNRT2.3a over-expression transgenic lines was not significantly different and was about 27% (Figure 3B).

However, in OsNAR2.1 transgenic lines (Figure 4A), the change in survival rate was significant. The recovery rate in WT was approximately 43.6%, in OsNAR2.1 RNAi transgenic lines was 21.0%, and in OsNAR2.1 over-expression transgenic lines was 86.3% (Figure 4B).

To further assess the effect of OsNAR2.1 on growth response to drought stress, the seedlings of OsNAR2.1 transgenic lines were grown in normal IRRI solution for 14 days and then transferred into the nutrient solution containing 10% PEG6000 for 14 days, nutrient solution without PEG6000 was used as a control (Supplementary Figures S4A,B). Under drought conditions, wilting and foliar chlorosis was observed in WT and OsNAR2.1 RNAi lines but not in OsNAR2.1 over-expression lines (Supplementary Figure S4C).

Compared to control, drought stress led to the suppression of root and shoot growth in all plants (Figure 5A,B and Supplementary Figure S4D). As compared with WT, the biomass per plant of OsNAR2.1 RNAi lines was reduced by 34.2% and that of OsNAR2.1 over-expression lines was increased by 35.7% in normal solution (Supplementary Figure S4D). However, under drought stress, the biomass of OsNAR2.1 RNAi lines was reduced by 60.3% and in OsNAR2.1 over-expression lines biomass was increased by 68.2% as compared with WT (Supplementary Figure S4D). Compared with WT, the roots biomass of OsNAR2.1 RNAi lines was reduced by 40.7% in normal solution and by 76.1% under drought stress (Figure 5A). The shoots biomass of OsNAR2.1 RNAi lines was reduced by 35.1% in normal solution and by 66.9% under drought stress (Figure 5B). Similarly, compared with WT, the roots biomass of OsNAR2.1 over-expression lines was increased by 36.8% in normal solution and by 62.8% under drought stress (Figure 5A). Moreover, the shoots biomass of OsNAR2.1 over-expression lines increased by 33.8% in normal solution and by 59.9% under drought stress (Figure 5B).

There was no significant difference in chlorophyll content, relative water content, photosynthesis rate or water use efficiency between OsNAR2.1 RNAi lines and OsNAR2.1 over-expression lines under normal solution (Figure 5C–F). However, after 14 days of drought stress as compared with WT, the chlorophyll content of OsNAR2.1 RNAi lines was reduced by 15.6% and the OsNAR2.1 over-expression lines was increased by 14.5% (Figure 5C). Under drought stress, the OsNAR2.1 RNAi lines showed a reduction in relative water content of about 13.1% compared with WT, but OsNAR2.1 over-expression lines showed an increase of 15.5% (Figure 5D). Compared with WT, the photosynthesis rate of OsNAR2.1 RNAi lines was reduced by 26.1% and the OsNAR2.1 over-expression lines was improved by 20.1% following drought stress (Figure 5E). Water use efficiency of WT in normal conditions was significantly lower than that under drought conditions (Figure 5F). The OsNAR2.1 RNAi lines showed a reduction in water use efficiency by 19.9% as compared with drought-stressed WT and OsNAR2.1 over-expression lines showed an increase of 17.5% (Figure 5F).

To understand the role of OsNAR2.1 in drought stress responses, we analyzed the expression of several known genes involved in abiotic and biotic stress in OsNAR2.1 transgenic lines, including OsMYB2 (Yang et al., 2012), OsNAC10 (Jeong et al., 2010), OsSNAC1 (Hu et al., 2006), OsDREB2a (Cui et al., 2011; Mallikarjuna et al., 2011), OsAP37 (Oh et al., 2009; Ramegowda et al., 2014), OsP5CS1 (Sripinyowanich et al., 2013) and OsRAB16C (Chen et al., 2015). Under normal conditions as compared with WT, the expression of all of these genes in the roots of OsNAR2.1 RNAi lines was decreased and in OsNAR2.1 over-expression lines was increased (Figure 6A). Moreover, in shoots, only OsMYB2 and OsRAB16C expression responded to OsNAR2.1 manipulation, showing a decrease in OsNAR2.1 RNAi lines and an increase in OsNAR2.1 over-expression lines. Conversely, no significant difference was found in the expression of non-stress related genes in transgenic lines (Figure 6C). Under drought stress, the expression of all of these genes was significantly decreased in the roots or shoots of OsNAR2.1 RNAi lines and significantly increased in OsNAR2.1 over-expression lines (Figure 6B,D).

Detailed analysis of drought response in OsNAR2.1 transgenic lines was performed by maintaining plants at 60% field capacity (Supplementary Figure S5), the yield of transgenic lines was measured by comparing the results with the fully irrigated field which we designated as a normal field. The seed setting rate and the grain number per panicle of WT under limited irrigation were 16.5 and 13.1%, respectively, lower than those of fully irrigated field. As a result, grain weight was reduced by 23.3% and grain yield was reduced by 26% (Figure 7B–D and Supplementary Figure S6D). Moreover, the yield of OsNAR2.1 over-expression lines under limited irrigation was not significantly different from the WT under full irrigation (Figure 7D).

As compared with WT, the effective tiller number and grain number showed no difference in transgenic lines in the normal field. However, in the drought field, OsNAR2.1 RNAi lines showed a decrease of 30.6 and 29.8%, respectively and OsNAR2.1 over-expression lines increased by 24.4 and 13%, respectively (Figure 7A,B). Compared with WT, the seed setting rate and grain yield of RNAi lines was decreased by 14.3 and 18.2%, respectively in the normal field. In addition, as compared with WT, the seed setting rate and grain yield were reduced by 45 and 65.2% in RNAi lines under drought stress. However, the seed setting rate and grain yield in OsNAR2.1 over-expression lines were increased by 5.6 and 9.1% in the normal field as compared to WT. Under drought conditions the OsNAR2.1 over-expression lines, seed setting rate and grain yield was increased by 16.3 and 26.6%, respectively (Figure 7C and Supplementary Figure S6D).

We also analyzed other agronomic traits amongst OsNAR2.1 transgenic lines. When there was plentiful irrigation available, the total tiller number per plant and 1000-grains weight of transgenic lines of OsNAR2.1 showed no difference. However, in limited water supply a 13.2% decline and 11.5% increase were observed in OsNAR2.1 RNAi lines and OsNAR2.1 over-expression lines, respectively, for tiller number and a similar pattern was observed in 1000-grain weight as compared to WT (Supplementary Figures S6A,C). When full irrigation was applied, panicle length declined in OsNAR2.1 RNAi lines and increased in OsNAR2.1 over-expression lines by about 19.0 and 25.8%, respectively, for grain weight it was 15.8 and 8.6%, respectively, for dry weight it was 11.1 and 8.4%, respectively as compared to WT. In contrast, when water supply was limited panicle length declined in OsNAR2.1 RNAi lines and increased in OsNAR2.1 over-expression lines by 39.1 and 25.8%, respectively, for grain weight it was 50.4% decrease and 31.5% increase, respectively, for dry weight it was 50.0 and 17.1%, respectively. Furthermore, the grain/straw ratio in OsNAR2.1 RNAi lines decreased by 43.5% and in OsNAR2.1 over-expression lines increased by 17.4% under drought conditions (Supplementary Figure S6).

We analyzed short-term NO₃⁻ and NH₄⁺ uptake in same-size seedlings of the transgenic lines by exposing the plants to 1.25 mM NH₄¹⁵NO₃ or 1.25 mM ¹⁵NH₄NO₃ for 5 min to determine the effect of OsNRT2.1, OsNRT2.3a or OsNAR2.1 expression on root NO₃⁻ and NH₄⁺ influx into intact plants. Compared with WYJ, overexpression of OsNRT2.1 increased NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ influx by 29.4% (Supplementary Figure S7A) and 12.6% (Supplementary Figure S7B), the ratio of ¹⁵NH₄⁺ to ¹⁵NO₃⁻ influx decreased by 12.9% (Supplementary Figure S7C). Compared with WYJ, overexpression of OsNRT2.3a did not significantly affect the influx of NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ by rice roots at seedling stage (Supplementary Figures S7A,B). As compared with WT, the NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ influx of OsNAR2.1 RNAi lines decreased by 43.0% and 46.0%, that of OsNAR2.1 over-expression lines increased by 32.8% and 30%, there was no significant difference in the ratio of ¹⁵NH₄⁺ to ¹⁵NO₃⁻ influx (Supplementary Figures S7D–F).

We also analyzed OsNAR2.1 and OsNRT2s expression in OsNRT2.1, OsNRT2.3a or OsNAR2.1 transgenic lines. Compared with WYJ, the expression of OsNAR2.1 increased significantly in OsNRT2.1 over-expression lines, but there was no significant difference in the expression of OsNRT2.2, OsNRT2.3a and OsNRT2.4 (Supplementary Figure S8A). There was no significant difference in the expression of OsNAR2.1, OsNRT2.2 and OsNRT2.4 between OsNRT2.3a over-expression lines and WYJ (Supplementary Figure S8B). The expression of OsNRT2.1, OsNRT2.2 and OsNRT2.3a decreased in OsNAR2.1 RNAi lines, increased in OsNAR2.1 over-expression lines, and the expression of OsNRT2.4 is not affected by OsNAR2.1 (Supplementary Figure S8C).

Many studies have shown that nitrogen supply has an essential influence on the response of plants to drought stress (Chaves et al., 2002; Guo et al., 2007; Ding et al., 2018). OsNRT2.1 is a high affinity nitrate transporter, mainly expressed at the root and responsible for nitrate uptake (Katayama et al., 2009; Feng et al., 2011). Previous studies have shown that overexpression of OsNRT2.1 increases the expression of OsNAR2.1 and this may be responsible for increased nitrogen uptake (Chen et al., 2016; Luo et al., 2018). This is consistent with our data showing that compared with WYJ, overexpression of OsNRT2.1 increased NH₄¹⁵NO₃ influx by 29.4% (Supplementary Figure S7A) and the expression of OsNAR2.1 increased significantly in OsNRT2.1 over-expression lines (Supplementary Figure S8A). Increased nitrate uptake could promote ammonium uptake by rice (Duan et al., 2006; Li et al., 2006). However, although ¹⁵NH₄NO₃ influx of OsNRT2.1 overexpression lines increased by 12.6% as compared with WYJ (Supplementary Figure S7C), the ratio of ¹⁵NH₄⁺ to ¹⁵NO₃⁻ influx decreased by 12.9% (Supplementary Figure S7C).

Tang et al. (2012) reported that OsNRT2.3a is responsible for the transport of nitrate from root to shoot. Previous studies have shown that overexpression of OsNRT2.3a did not increase ¹⁵NH₄¹⁵NO₃ uptake in rice (Fan et al., 2016). This is consistent with our data showing compared with WYJ, overexpression of OsNRT2.3a did not significantly affect the influx of NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ by rice roots at seedling stage (Supplementary Figures S7A,B). This is likely due to the fact that overexpression of OsNRT2.3a did not lead to a significant increase in expression of OsNAR2.1 (Supplementary Figure S8B). Previous work has shown that overexpression of OsNAR2.1 increase nitrate and ammonium absorption in rice, and increase rice biomass under 1.25 mM NH₄NO₃ treatment (Chen J. et al., 2017). These results are consistent with our data showing as compared with WT, the NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ influx of OsNAR2.1 RNAi lines was reduced by 43.0 and 46.0%, that of OsNAR2.1 over-expression lines was increased by 32.8 and 30%, there was no significant difference in the ratio of ¹⁵NH₄⁺ to ¹⁵NO₃⁻ influx (Supplementary Figures S7D–F). Also, the expression of OsNRT2.1, OsNRT2.2 and OsNRT2.3a decreased in OsNAR2.1 RNAi lines and increased in OsNAR2.1 over-expression lines (Supplementary Figure S8C).

Photosynthesis is the main source of biomass accumulation in all plants, it is also one of the most sensitive physiological processes during abiotic stress (DaMatta et al., 2002). Drought stress can inhibit stomatal opening and photosynthesis, and can lead to an increase in water use efficiency (Silva et al., 2010; Waraich et al., 2011; Allahverdiyev, 2016). Compared with WT, there were no significant differences in chlorophyll content, relative water content, photosynthesis rate and water use efficiency between OsNAR2.1 RNAi lines and OsNAR2.1 over-expression lines in well-watered conditions (Figure 5C–F). After 14 days of drought stress as compared with WT, the photosynthesis rate of OsNAR2.1 RNAi lines decreased by 26.1%, and of OsNAR2.1 over-expression lines increased by 20.1% (Figure 5E). OsNAR2.1 RNAi lines showed a 19.9% reduction in water use efficiency compared with drought-stressed WT and OsNAR2.1 over-expression lines increased efficiency by 17.5% (Figure 5E). Because of changes in nitrogen uptake, photosynthetic rate and water use efficiency, compared with WT, the biomass of OsNAR2.1 RNAi decreased by 34.2% under normal solution and 60.3% under drought stress, the biomass of OsNAR2.1 over-expression lines increased by 35.7% under normal solution and 68.2% under drought stress (Figure 5A,B and Supplementary Figure S4D).

There are two kinds of nitrogen applications for rice plants: ammonium and nitrate, with different nitrogen supply patterns and effects on plant growth. The root activity of rice with ammonium was significantly higher than that with nitrate, which might be related to the characteristics of aquaporins (Siemens and Zwiazek, 2003). Plants supplied with NH₄⁺ had a higher assimilation rate and stomatal conductance than those applied with NO₃⁻ (Guo et al., 2002). Compared with nitrate nutrition, ammonium nutrition increases photosynthesis rate under water stress at the early development stage of rice (Guo et al., 2007) and enhances the tolerance of rice seedlings to drought conditions (Li et al., 2009). Under drought stress, rice seedlings grow better in ammonium nutrition (Guo et al., 2002, 2007). Under PEG6000 treatment water stress, plants under NO₃⁻ treatment had significantly lower chloroplast content than those under NH₄⁺ treatment (Li et al., 2012; Xiong et al., 2017). Wilkinson et al. (2007) observed that the decrease in stomatal closure and leaf elongation rate was more sensitive to drought stress in maize plants supplied with high nitrate. Compared with nitrate supply, ammonium can significantly improve the drought resistance of rice. In the present study, irrigation was withheld for 12 days, followed by re-watering for 7 days (Figure 3A, 4A). The survival rate of WYJ, OsNRT2.1 transgenic lines and OsNRT2.3a transgenic lines was not significantly different, about 27% (Figure 3B). Approximately 43.6% of the WT recovered, 21.0% of the OsNAR2.1 RNAi transgenic lines recovered, and 86.3% of the OsNAR2.1 over-expression transgenic lines recovered (Figure 4B). The results showed that over-expression of OsNRT2.1 could increase ammonium uptake by rice, but at the same time increased more nitrate uptake (Supplementary Figure S7C), which might counteract the positive effect of ammonium on rice drought resistance. Overexpression of OsNRT2.3a did not increase nitrogen uptake or drought tolerance in rice. Compared with WT, the NH₄¹⁵NO₃ and ¹⁵NH₄NO₃ influx of OsNAR2.1 RNAi lines decreased significantly, and that of OsNAR2.1 over-expression lines increased significantly, there was no significant difference in the ratio of ¹⁵NH₄⁺ to ¹⁵NO₃⁻ influx (Supplementary Figures S7D–F). This ability to increase uptake of NH₄⁺ as well as NO₃⁻ may play a role in the mechanism of OsNAR2.1 induced drought tolerance in rice.

We analyzed the expression of several known genes involved in abiotic and biotic stress in OsNAR2.1 transgenic lines, including OsMYB2, OsNAC1, OsSNAC1, OsDREB2a, OsAP37, OsP5CS1, and OsRAB16C. Under normal conditions as compared with WT, the expression of all of these genes in the roots of OsNAR2.1 RNAi lines was decreased and in OsNAR2.1 over-expression lines was increased (Figure 6A). Conversely, no significant difference was found in the expression of other genes in transgenic lines (Figure 6C). In drought stress, the expression of all of these genes was significantly decreased in the roots or shoots of OsNAR2.1 RNAi lines and increased in OsNAR2.1 over-expression lines (Figure 6B,D). Previous studies have shown that overexpression of these genes increases drought tolerance in rice (Hu et al., 2006; Kim et al., 2009; Oh et al., 2009; Jeong et al., 2010; Cui et al., 2011; Mallikarjuna et al., 2011; Yang et al., 2012; Ramegowda et al., 2014; Chen et al., 2015). Overexpression of OsNAR2.1 increased ammonium uptake in rice, while OsNAR2.1 RNAi decreased ammonium uptake (Supplementary Figures S7E,F). Ammonium supply significantly increased abscisic acid (ABA) accumulation in plants (Zdunek and Lips, 2001; Rahayu et al., 2005; Fernández-Crespo et al., 2015; Ding et al., 2016). Under drought stress, ABA accumulation is essential to regulate the expression of drought-tolerant genes (Jang et al., 2004; Schraut et al., 2005; Lian et al., 2006; Mahdieh and Mostajeran, 2009; Parent et al., 2009), promote root water uptake and plant growth (Sharp, 2002; Zhang et al., 2006). OsNAR2.1 regulates the expression of these drought regulating genes in rice under stress (Figure 6B,D); suggesting that the OsNAR2.1 mediated increase in their expression is involved in rice drought tolerance.

Compared with fully irrigated conditions, the grain yield of WT decreased by 26% under limited water supply (Figure 7D). The seed setting rate and grain number per panicle of WT in limited water supply were 16.5 and 13.1% lower than those in full irrigation; resulting in a 23.3% reduction in grain weight (Figure 7B,C and Supplementary Figure S6D). This agrees with previous studies showing that drought conditions impair pollen germination, seed setting rate, and grain number per panicle, resulting in decreased grain yield (Ekanayake et al., 1989, 1990; Saini and Westgate, 1999; Kim et al., 2009; Guan et al., 2010).

Drought stress also reduced the tiller number of rice (Kang et al., 2017). Compared with WT, the effective tiller number of OsNAR2.1 RNAi lines and OsNAR2.1 over-expression lines had no significant difference under complete watering conditions. The effective tiller number of OsNAR2.1 RNAi lines decreased by 30.6% and OsNAR2.1 over-expression lines increased by 24.4% under limited water supply (Figure 7A). Compared with WT, there was no significant difference in the grain number per panicle between OsNAR2.1 RNAi lines and OsNAR2.1 over-expression lines under complete watering conditions, but under limited water supply, the grain number per panicle of OsNAR2.1 RNAi lines decreased by 29.8% and that of OsNAR2.1 over-expression lines increased by 13.0% (Figure 7B). Similarly, compared with WT, the seed setting rate of OsNAR2.1 RNAi lines decreased by 14.3% and that of OsNAR2.1 over-expression lines increased by 5.6% in fully irrigated field. Further, that of OsNAR2.1 RNAi lines was decreased by 45.0% and that of OsNAR2.1 over-expression lines was increased by 16.3% under limited water supply conditions (Figure 7C).

The grain yield of rice is the result of various factors (Kim et al., 2009; Guan et al., 2010). In our experiments, changes in photosynthetic rate, water use efficiency, tiller number, grain number per panicle and seed setting rate led to changes in rice yield. Finally, compared with WT, the grain yield of OsNAR2.1 RNAi lines decreased by 18.2% and that of OsNAR2.1 over-expression lines increased by 9.1% under full watering conditions (Figure 7D). Further, OsNAR2.1 RNAi lines decreased by 65.2% and that of OsNAR2.1 over-expression lines increased by 26.6% under limited water supply (Figure 7D). In conclusion, overexpression of OsNAR2.1 could improve the drought resistance without causing adverse growth phenotypes in rice.