Introduction

Rice (Oryza sativa L.) is essential for food security for more than half of the world's population (^{Jumaa et al., 2019}). Since rice has an evolutionary peculiarity of semi-aquatic, flooded rice paddies have become the major form of cultivation, growing in irrigated and rainfed lowland conditions, equivalent to 75 % and 19 % of the global production area, respectively (^{Kikuta et al., 2016}). Increasing grain yield of irrigated areas is not enough to supply future demand for rice; furthermore, expansion of production areas is restricted, due to the water scarcity (^{Parthasarathi et al., 2012}). Upland rice represents 4 % of the global rice production and is grown less than 9 % of total rice acreage in Asia, 46 % in Latin America, and 47 % in West Africa (^{Kikuta et al., 2016}). According to ^{Singh et al. (2014)}, upland rice accounts for 84 % of the total area in Sub-Saharan Africa and it is cultivated mostly by smallholder farms with an average area smaller than 0.5 ha. On the other hand, in Latin America, upland rice is cultivated in large-areas of mechanized harvesting (^{Bernier et al., 2008}).

Drought is one of the most severe abiotic stresses limiting rice yield worldwide and poses a serious threat to rice sustainability in rainfed agriculture (^{Wu and Cheng, 2014}). According to ^{Heinemann et al. (2015)}, rice yield in upland cultivation (tropical regions mainly) has its yield potential reduced by up to 35 % due to drought-stress conditions.

Reduction in water availability for plants results in a complex response characterized by a decrease in the water potential of its tissues, leading to several changes in different plant processes (^{Rosales et al., 2012}). Some processes reported for upland rice are (a) appropriate phenological patterns that combine crop growth and the amount of water available in the soil (water environment), (b) deep root system, (c) thick stems and reduced number of stomata, (d) osmotic adjustment to maintain cell homeostasis and, consequently, avoid a rapid decrease in leaf-water potential, and (e) senescence delay, also known as stay-green trait, which allows the maintenance of the photosynthetic capacity and the photoassimilate remobilization for a longer time period (^{Fukai and Cooper, 1995}; ^{Boonjung and Fukai, 1996}).

Thus, the establishment of sustainable crop systems of upland rice requires better understand of changes in the morphophysiological mechanisms, contributing to drought tolerance and yield effects. This study aimed to (a) identify a series of morphophysiological and agronomic traits related to drought tolerance in upland rice genotypes of Embrapa Core Collection under greenhouse cultivation, and (b) characterize morphophysiological components to be used as indicators for drought tolerance for plant breeding processes.

Materials and Methods

Results and Discussion

In crops, such as upland rice, where seeds are the product of interest, the main criteria for selecting agronomical tolerance to drought are the traits that lead to higher grain yield. In this study, the analysis of grain yield showed a significant difference (p < 0.05) for all variation sources. For 2015 and 2017, the genotypes accounted for 41 % and 50 % of the total sum of squares, while the environment (water level) accounted for 41 % and 44 % and the genotype versus environment interaction accounted for 18 % and 7 %, respectively. The agronomic performance (grain yield) of genotypes cultivated under two water treatments in 2015 and 2017 is shown in Figures 1A and 1B. In 2015, Bico Ganga, BRS Esmeralda, BRSGO Serra Dourada, BRSMG Curinga, Casca Branca, Guarani, Rabo de Burro, Rio Doce, and Três Meses Branco showed better yield under drought (average grain yield 70.9 g column⁻¹) and irrigated (average grain yield 119.8 g column⁻¹) conditions. In 2017, Agulhão, Bico Ganga, BRS Esmeralda, BRS Primavera, BRS Soberana, BRSMG Curinga, Guarani, and Rabo de Burro were more productive under drought (average grain yield 40.90 g column⁻¹) and irrigated (average grain yield 63.98 g column⁻¹) conditions. Among the upland rice genotypes evaluated, Bico Ganga, BRS Esmeralda, BRSMG Curinga, Guarani, and Rabo de Burro showed better agronomic performance at both water levels in both two years of trials and were then ranked as top genotypes. These genotypes probably present favorable alleles of drought tolerance that may be useful in breeding programs of upland rice. Two of these genotypes are modern cultivars (BRS Esmeralda and BRSMG Curinga), which could be qualified as parents in breeding programs of upland rice. For yield components, the average number of filled-grains and empty-grains was 138 and 46 in 2015, and 266 and 82 in 2017, respectively, in rice cultivated under irrigated condition (Table 2). For plants under drought, the total number of filled-grains and empty-grains were 99 and 37 in 2015, and 184 and 73 in 2017, respectively. The average value of 100-grain weight was 3.03 g under irrigated condition and 2.62 g under stress, determined only in 2015. IRAT 112 (41 %) and Douradão (52 %) presented the highest percentage of spikelet sterility under irrigated condition, and Moroberekan (94 %) and Branquinho 90 Dias (73 %) under drought condition in 2015 and 2017, respectively. In both years of trials, environmental conditions of phenotyping platform SITIS were severe. Particularly in 2017, in addition to artificially imposed water stress, the maximum temperature of 44.7 °C was 6.7 °C higher than the conditions of the 2015 trial, during the water deficit period. In addition, the minimum relative humidity of 26 % was 42 % lower than that of the 2015 trial (Table 3). According to ^{Choudhary et al. (2018)}, drought commonly occurs combined with other environmental stresses, such as excessive light incidence, heat, and low relative humidity, and characterizes multiplicity of stresses in the tropics. For rice, along with drought, high temperature (up to 33.5 °C) contributed to yield reduction due to the shortening of the vegetative period and high spikelet sterility (^{Peng et al., 1995}; ^{Matsui et al., 1997}; ^{Shah et al., 2011}).

Figure 1 Grain yield of upland rice (Oryza sativa L.) cultivated in SITIS Platform in years 2015 (A) and 2017 (B). The dotted lines define the average value of grain yield under irrigated and drought conditions. Genotypes that showed higher yield under irrigated and drought conditions were identified in the upper right-hand quadrant in Figures A and B. Bico Ganga, BRS Esmeralda, BRSMG Curinga, Guarani, and Rabo de Burro stood out in both years of trials. 

According to ^{Bernier et al. (2008)}, practices based on the assessment of agronomic performance of crop species require a long procedure, which limits breeding efficiency. Thus, a better understand of mechanisms of drought tolerance is necessary, since the association between main (grain yield and its components) and secondary (morphophysiology) traits could provide greater selection efficiency. For this, identifying morphophysiological traits related to drought tolerance is relevant to assist in the identification of mechanisms underlying these adaptation processes and thus in the selection of tolerant genotypes. In this study, upland rice plants reacted to drought stress by slowing down their growth (Table 4). In 2015, most genotypes (Agulhão, Aimoré, Amarelão, Arroz 4 meses, Bico Ganga, BRS Esmeralda, Primavera, BRSMG Curinga, Carajás, Casca Branca, Cirad 392, Douradão, Guarani, Moroberekan, Rabo de Burro, Rio Doce, Tangará, and Três Meses Branco) showed reduced PH (14 % on average) under drought condition. While, in 2017, only six genotypes (Agulhão, Bico Ganga, BRS Serra Dourada, Carajás, Rabo de Burro, and Rio Doce) presented an average reduction of 12 %. Overall, there was no difference between plants grown under both water levels for SDMB, TN, and PN, in both years of trials. Parameters LA and PL, taken in 2015, showed reductions of 13 % and 3 %. LA and PH were the main morphological traits affected by drought stress in top genotypes Bico Ganga, BRS Esmeralda, BRSMG Curinga, Guarani, and Rabo de Burro, in both years of trials. According to ^{Fischer et al. (2003)} and ^{Chaves et al. (2009)}, reduction of leaf growth and stem elongation in rice plants are the first processes affected by drought and could be considered as a tolerance mechanism, since they reduce the transpiration capacity, and consequently, plant demand for water. Furthermore, slowed growth (due to reduction of stomatal conductance, CO₂ assimilation and, consequently, photoassimilates production and accumulation) has been suggested as an adaptive trait for plant survival under stress. This trait allows plants to divert assimilates and energy into protective molecules to deal with stress (^{Zhu, 2002}) and/or keep root growth by increasing water acquisition (^{Chaves et al., 2003}; ^{Pandey and Shukla, 2015}).

The effect of the drought treatment was also evaluated by characterization of the root system, an important organ to increase rice yield under water stress (^{Pandey and Shukla, 2015}; ^{Kundur et al., 2015}). According to ^{Kato et al. (2006)}, rice root is complex, combining various root morphologies and showing considerable genotypic variation, also subjected to environmental effects. Thereby, a deep root system could improve adaptation of upland rice during drought by increasing capacity of extraction water, keeping high leaf water status with better crop performance under drought conditions (^{Kamoshita et al., 2004}; ^{Mishra and Salokhe, 2011}). In this study, the analysis of the root system of upland rice showed a significant difference (p < 0.05) for most variation sources. At depth 1 (5 - 25 cm), the genotypes accounted for 14 % of the total sum of squares, the environment (water level) accounted for 54 %, and the double interaction, genotype versus water lever, accounted for 32 %. At depth 2 (25 - 45 cm), the genotypes accounted for 62 % of the total sum of the square, the environment (water level) accounted for 20 %, and the double interaction, genotype versus water level, accounted for 17 %.

The root system properties (length, surface area, volume, and diameter) of upland rice plants during the drought period are shown in Figure 2. Under irrigated condition, the genotypes that stood out mostly in terms of length, area, volume, and root diameter were IRAT 112, Agulhão, BRSMG Curinga (top genotype), Comum, Rabo de Burro (top genotype), and Saia Velha, at both depths. Under drought condition, the highlight was BRSMG Curinga followed by Agulhão, Comum, Rabo de Burro, and Saia Velha. Therefore, among top genotypes, BRSMG Curinga, and Rabo de Burro presented greater robustness of the root system, mainly at depth 2 (25 - 45 cm), irrespective of the water level applied. This is in accordance with ^{Pandey and Shukla (2015)}, which describe that under water deficit, root growth is usually kept, while shoot growth is inhibited. Conversely, ^{Ji et al. (2012)} found a more extensive deeper root growth in a tolerant rice cultivar, IRAT109, after 20 days of irrigation cut-off. The findings of our study indicate a mechanism at the molecular level underlying a constitutive root growth for the root traits evaluated. Water deficit is an important environmental constraint and influences all physiological processes in plant growth, affecting gas exchange mechanisms (^{Ma et al., 2018}).

Figure 2 Root length (cm), Root area (cm²), Root volume (cm³), and Root diameter (mm), at first (soil layer of 5 - 25 cm) and second (soil layer of 25 - 45 cm) depths of the soil cultivated with upland rice (Oryza sativa L.). Plants were grown under irrigated and drought conditions. Capital letters compare genotypes within each water regime and small letters compare water regimes within each genotype. Means followed by the same letter do not differ by the Tukey test 5 % error probability. Parameters were evaluated in 2017. 

The stress effects on A, E, gs, Ci, WUE, and CE in upland rice plants are shown in Table 5. During phase I, where control and stress columns were in similar conditions of soil water availability, there was genetic variability among rice accessions, implying a contrast for the gas exchange traits evaluated, in both years of trials. In 2015, A ranged from 8.34 to 25.31 µmol CO₂ m⁻² s⁻¹, E ranged from 2.60 to 9.34 mmol H₂O m⁻² s⁻¹, and gs (number and activity of stomata) ranged from 0.16 to 0.52 mol H₂O m⁻² s⁻¹. In 2017, 7.95 to 25.91 µmol CO₂ m⁻² s⁻¹, 4.16 to 11.65 mmol H₂O m⁻² s⁻¹, and 0.13 to 0.47 mol H₂O m⁻² s⁻¹, respectively. In phase II, 5^th day after irrigation cut-off, we observed mechanisms, such as leaf-rolling and stomatal closure. These events soften the solar radiation incidence and transpiration rate, respectively, increasing water conservation and delaying water deficit. Low values of A, E and gs were observed in both years of trials. In 2015, A ranged from 1.27 to 15.67 µmol CO₂ m⁻² s⁻¹, E ranged from 0.97 to 4.14 mmol H₂O m⁻² s⁻¹, and gs ranged from 0.04 to 0.47 mol H₂O m⁻² s⁻¹. In 2017, A ranged from 0.82 to 19.21 µmol CO₂ m⁻² s⁻¹, E ranged from 1.07 to 6.27 mmol H₂O m⁻² s⁻¹, and gs ranged from 0.02 to 0.14 mol H₂O m⁻² s⁻¹. Four out of five top genotypes (Bico Ganga, BRS Esmeralda, BRSMG Curinga, and Rabo de Burro) showed average reduction of 84 %, 72 %, and 81 % in A, E, and gs, respectively, in plants cultivated under drought. In phase III, after plants under stress received 50 % of water at the column base for 10 days, only three genotypes, Três Meses Branco (2015), Branquinho 90 Dias, and Rabo de Burro (2017) restored the functioning of the photosynthetic machinery, since stressed plants showed values of photosynthetic rate similar to those of irrigated plants. Conversely, for the other genotypes including Bico Ganga, BRS Esmeralda, BRSMG Curinga, and Guarani (top genotypes), recovery of A, E, and g_s was 40 %, 46 %, and 30 %, respectively, in stressed plants.

When water deficits start to increase, leaf stomatal conductance usually decreases faster than carbon assimilation, leading to increased WUE. The WUE reflects the multiple environmental stimuli perceived and the capacity of a particular genotype to sense the onset of changes in moisture availability and therefore to fine-tune its water status in response to the environment (^{Wilkinson, 2004}; ^{Blankenagel et al., 2018}).

However, despite the negative impact of water deficit on gas exchange, in both years of trials, Bico Ganga, BRS Esmeralda, BRSMG Curinga, and Guarani (top genotypes) improved their WUE (44 %) when compared with optimal irrigation conditions. This was most probably due to higher stomatal control efficiency, keeping approximately 40 % of the photosynthetic process and drastically reducing stomatal conductance (70 %) by closing the stomata process. Although Rabo de Burro did not show increase in WUE, it presented a recovery of the gas exchange apparatus compared to irrigated plants, which can be justified partly by its vigorous root system.

In addition to increased relative stomatal limitation, drought stress is responsible for reducing maximum Rubisco carboxylation activity and electron transport and therefore ribulose bisphosphate (RuBP) regeneration (^{Perdomo et al., 2017}). The carboxylation efficiency could be considered an estimate of the Rubisco activity, illustrating its limitations under stress conditions (^{Niinemets et al., 2009}). In our study, all upland rice genotypes showed a poor capacity to overcome limitation in CO₂ diffusion by stomata and mesophyll and effective CO₂ fixation (70 % of CE reduction) during phase II for both years of trials. After replenishing 50 % of water at the column base for 10 days, recovery of 55 % and 64 % in the carboxylation efficiency was observed in 2015 and 2017, respectively. Considerable loss of Rubisco activity during stress conditions were also reported for sugarcane subjected to water deficit (^{Saliendra et al., 1996}; ^{Vu and Allen Jr., 2009}). Overall, a response pattern was not observed among genotypes with greater yield performance under water deficit, since they showed divergent physiological responses of gas exchange.

Furthermore, remobilization of photoassimilates from vegetative into reproductive structures may have a significant effect on grain yield, although this component was not evaluated in our study. As demonstrated for cereals (^{Blum et al., 1994}) and legumes (^{Chaves et al., 2002}), nutrient pre-anthesis reserves are used for grain filling in addition to current assimilates. In rice, drought-induced leaf senescence also promotes assimilate allocation to grains under development, shortening grain filling and increasing the grain filling rate (^{Sehgal et al., 2018}). Moreover, senescence and reserve mobilization are integral components of plant development and basic strategies in stress mitigation (^{Lemoine et al., 2013}).

Water stress effects on Ψw, RWC, Ψs, and OA, evaluated only in the 2015 trial, are shown in Table 6. Among top genotypes, BRS Esmeralda, BRSMG Curinga, Guarani, and Rabo de Burro showed a more pronounced gradient of ψ_w and probably enhanced water absorption capacity. Besides, advance of the most severe internal damage may have reduced in the reproductive organs under the drought period. Conversely, Bico Ganga kept high water potential during the water deficit period, which may be associated to a more robust root system in the second soil layer and thus higher panicle water potential, which probably contributes to increased grain yield. According to ^{Guimarães et al. (2016)}, plants that prevent dehydration presented higher water potential and earliness in flowering, lower height, lower leaf area or lower tillering. Regarding the trait RWC, which is directly related to the plant water status, values ranged from ∼ 82 % in leaves under irrigated condition to 75 % for stressed plants. On the other hand, BRSMG Curinga presented significant RWC reduction due to the stress imposed. This divergent responses regarding leaf water status suggest greater capacity of top genotypes to save water during drought and stimulate an adjustment of the photosynthetic capacity to tolerate changes in water availability (^{Silva et al., 2007}; ^{Rodrigues et al., 2009}; ^{Graça et al., 2010}). The mechanism of osmotic adjustment (OA), usually accomplished by accumulation of compatible solutes (ψs) and maintenance of RWC, although significant for all genotypes, was numerically higher for BRS Esmeralda, followed by Bico Ganga and Guarani, compared to BRSMG Curinga and Rabo de Burro (top genotypes). This mechanism in upland rice plants during the reproductive phase allows maintenance of adequate physiological state, in which the leaves remain green and cool for a longer time, besides allowing the establishment and retention of spikelet and, consequently, grain yield sustenance (^{Fischer et al., 2003}).

This study describes important aspects of drought-induced effect on upland rice, providing a better understanding of morphophysiological changes under water deficit. Top genotypes showed distinct strategies by activating different physiological responses: higher ability to save water on leaves (Bico Ganga, BRS Esmeralda, BRSMG Curinga and Rabo de Burro), lower leaf water potential (Bico Ganga, BRS Esmeralda, BRSMG Curinga and Guarani), higher ability to reduce vegetative structures (Bico Ganga, BRSMG Curinga and Rabo de Burro), higher efficiency in the use of water (Bico Ganga, BRS Esmeralda, BRSMG Curinga and Guarani), higher photosynthetic capacity (Guarani), and improved ability to absorb water from drying soil, either by osmotic adjustment (Bico Ganga, BRS Esmeralda and Guarani) or additional investment in the root system (BRSMG Curinga and Rabo de Burro). Therefore, different mechanisms, such as vegetative morphology, gas exchange, water status, and root system could be explored simultaneously to support the development of drought-tolerant rice cultivars by breeding programs.