Rice is arguably the most important staple food that feeds more than half of the world population (Dhakarey et al., 2017). Modern semi dwarf rice varieties, a result of the green revolution, were selected primarily for high yield and resistance to biotic stresses, especially for irrigated ecosystems where available water is considered abundant for irrigation (Sandhu and Kumar, 2017). These varieties lacked resistance to abiotic stresses, such as water stress (drought, flooding, and submergence), problematic soil (salinity, alkalinity etc.), temperature extremes (heat, chilling, and freezing) and nutrient stress (deficiency and/or toxicity).

Rice is a profligate user of water, and it alone receives about 35% of the global surface water irrigation (Bouman et al., 2007). Erratic rainfall patterns due to the current and imminent environmental instabilities will increase the scarcity of water in arid and semi-arid regions and also are a great threat to the quality of water, where available, for crop use. Drought situation around the globe is apparent. In traditionally irrigated areas, such as in the U.S., drought is becoming an increasing concern, not only due to the climate change but also because of the demand for water for competing uses, which can lead to ground water paucity. In U.S. specifically, despite the present flooding situation in Houston 2017 and Louisiana in August 2016, overwhelming drought situations in the Southern U.S. (2010–2013), mid-western U.S. (2012–2013), in California (2014) are testimony to the fact that climate is changing and drought is inexorable. In addition, recurrent hurricanes in the coastal U.S. lead to the inundation of sea water into the cultivated area, thus making the irrigation water saline. The immediate effect of salinity is osmotic stress that makes the crops experience a physiological drought even with abundance of water available for irrigation. Thus, there is a need to develop rice varieties with drought resistance and acceptable yield.

Drought resistance is a complex phenomenon, which is the manifestation of both drought tolerance (tissue tolerance, maintenance of photosystem, etc.) and drought avoidance (deep root, leaf rolling, etc.) traits that are governed by multiple genes. Although several secondary drought resistance traits, such as green leaf area, canopy temperature, relative water content, etc. (Atlin and Lafitte, 2002; Richards et al., 2010) were identified to improve selection efficiency for yield under drought, complexity of the methods to phenotype the secondary traits limited their inclusion as a selection criterion in breeding. Also, expression of most loci governing these traits is influenced by both environmental conditions and genetic background, thus making difficult the assessment of their contribution to grain yield. Therefore, direct selection for yield was considered the most successful approach to improve drought resistance in rice (Kumar et al., 2008). Major criteria to evaluate the performance of genotypes against drought are drought score, grain yield and spikelet fertility (Swain et al., 2014). However, data accumulated in the past suggest that drought tolerance for yield components is largely associated with genetic and physiological factors independent from those determining the traits per se (Lang et al., 2013).

Genetic mapping of quantitative trait loci (QTLs) to identify genomic regions controlling grain yield under drought and subsequent utilization of markers linked to the QTLs in marker-assisted pyramiding of desirable alleles is an effective alternative to breeding high yielding drought resistant rice (Ashraf, 2010). Despite limited success of using QTLs in molecular breeding due to the repeatability of QTLs (Bernier et al., 2008), significant progress has been made toward mapping QTLs for drought resistance traits, such as morphological, physiological and yield related traits in rice (Swamy et al., 2017) by the integration of molecular markers and sequencing technologies in rice breeding studies (Thomson et al., 2010; Swamy et al., 2014).

Most of the yield related QTLs reported in the literature came from studies that focused on different components associated with yield rather that grain yield under drought stress from studies involving drought resistant local landraces and ecotypes (Ikeda et al., 2013; Sun et al., 2013; Yan and Bao, 2014). The QTL, qDTY12.1, located on chromosome 12, was the first QTL reported for rice grain yield in upland reproductive-stage drought stress conditions (Bernier et al., 2007). It's contribution to improve grain yield under drought tolerance is significant in terms of the genetic variance explained (43 to 51%) and its consistent effects in multiple environments (Bernier et al., 2007, 2009a,b; Boopathi et al., 2013). qDTY12.1 is the only locus utilized to generate drought resistant lines with increased grain yield for highly diverse upland and lowland rice ecosystems (Bernier et al., 2007, 2009a; Mishra et al., 2013); however, map-based cloning of qDTY12.1 to identify causal gene(s) is not yet available. QTL, qDTY1.1 on chromosome 1, with 8.55–30% additive effects, is one of the most recent examples of major QTLs discovered for grain yield under drought identified in different genetic backgrounds and from different donors, with the positive allele coming from the drought resistant varieties N22 (Vikram et al., 2011) and Dhagaddeshi (Ghimire et al., 2012). Two other QTLs, qDTY2.3 on chromosome 2 and qDTY3.2 on chromosome 3 (Tang et al., 2013) derived from Vandana cultivar, interact with qDTY12.1 to enhance yield and harvest index of qDTY12.1-positive lines under severe upland and lowland drought conditions (Dixit et al., 2012). qDTY2.3 had a large effect, accounting for 18–31 % (R²) improved grain yield under severe lowland stress (Yadaw et al., 2013; Saikumar et al., 2014). qDTY3.2 also accounted for variation in canopy temperature during flowering and seedling shoot dry weight under stress and drought recovery (Saikumar et al., 2014). Likewise, Zhang et al. (2014) reported five stably expressed QTLs (QSnp1b, QGyp2a, QSnp3a, QSf8, and QSnp11) from a set of 20-30 QTLs for drought resistance-related traits in a cross between Lemont (drought sensitive) and Teqing (moderately drought tolerant). QTL QSnp1b and QGyp2a mapped with qDTY1.1 (Vikram et al., 2011) and qDTY2.1(Venuprasad et al., 2009), respectively, whereas other QTLs localized in chromosomal regions harboring genes for abiotic stress response and hormone signaling (Ye et al., 2009; Jeong et al., 2013). Five QTLs on chromosomes 1, 6, 8, 10, and 12 with 2.47–13.89% effect on grain yield under drought were identified (Lang et al., 2013) with the largest effect QTL on chromosome 12 coincided with a minor QTL for grain thickness (Zhang et al., 2013). Sixteen drought grain yield QTLs distributed over all chromosomes except 5, 7 and 8 were reported in rice at IRRI (Kumar et al., 2014). Meta-QTL analyses of 15 populations identified 14 MQTLs on seven chromosomes (four on chromosome 1, two each on chromosomes 2, 3, 8, and 10, and one each on chromosomes 4 and 12), with 4–28% effect on drought grain yield (Swamy et al., 2011). This study showed that the qDTY12.1 was present in 85% of the lines, whereas qDTY1.1, qDTY1.2, qDTY3.2, qDTY4.2, and qDTY8.1 were present in >50% of the lines.

In the southern U.S., rice breeding programs have been focused to improve grain yield and quality with resistance against pathogen and insect pests. Except for a few reports (De Leon et al., 2015, 2016), limited attention has been paid to understanding the genetics toward breeding for abiotic stress tolerance. Evaluation of the impact of drought on grain yield under field conditions is an ideal situation to dissect genes/QTLs associated with drought tolerance, but it requires a dry season, a factor that is beyond researcher's control. Therefore, the present study was undertaken under controlled greenhouse. This is the first study, to our knowledge, that reports identification of genetic factors controlling grain yield under drought using the southern U.S. cultivar with an objective to identify molecular markers associated with QTLs for rice grain yield under drought conditions.

One hundred seventy three out of 330 SSRs selected from the Gramene database showed polymorphism between the parents, Cocodrie and Vandana. However, 24 SSRs produced complex banding pattern, similar-sized alleles or skewed segregation among the F₂ individuals, and thus were excluded from further analysis. Only 5 out of 33 Indels generated distinct polymorphic bands between the parents. Of the 37 genic SSRs, 22 SSRs were confirmed being polymorphic with good resolution between the parents. Similarly, 37 out of 120 SNPs were polymorphic without cross-amplification between the parents. Thus, a total of 213 markers showing distinct, reproducible polymorphism between the parents were used for mapping.

The present study targeted grain yield as it is the main focus of current rice breeding efforts in the development of improved drought resistant rice varieties (Venuprasad et al., 2012a,b; Dixit et al., 2014; Saikumar et al., 2014). By using grain yield as the selection criterion, we detected six QTLs that primarily control mechanisms toward higher yield under drought using F_{2 : 3} lines of two parental cultivars, Cocodrie and Vandana with contrasting drought response. Involvement of other genetic mechanisms, such as those underlying drought avoidance and plant recovery following stress, contributing to the yield cannot be overruled as leaf rolling was observed among the progeny. Grain yield (weight) is the result of the sum of carbohydrate accumulated before heading and translocated/loaded after heading (Swain et al., 2014), and thus is primarily determined by highly correlated traits, such as panicle number per plant, spikelet number per panicle, and spikelet fertility. It is not surprising that loci associated with these traits have been reported co-localized under both irrigated and drought stressed conditions (Venuprasad et al., 2012b; Mishra et al., 2013; Saikumar et al., 2014). Therefore, grain yield was the only trait that was measured under drought in the present study.

QTLs for grain yield under drought in managed stress environments, such as controlled greenhouse conditions may not be translatable to target field environments where the timing and severity of drought vary over years (Prince et al., 2015), and such conditions will change the responses of plant's traits involved in drought-resistance mechanisms (Kamoshita et al., 2008). However, under circumstances where a drought experiment under field conditions with unpredictable rainfall pattern and lack of a rainout shelter is a distant reality, simulated drought under greenhouse studies could provide important primary information on the heritability of the trait and chromosomal regions associated with grain yield by near-exact manipulation of the water regime. In addition, greenhouse conditions will eliminate the edaphic variability (soil texture, soil temperature, etc.) and other environmental confounding factors (humidity, biotic stresses) that hamper precise phenotyping under natural field conditions and detection of causative QTLs. In the present study, the broad sense heritability was moderate (28%), which is similar to the results obtained under controlled conditions by Manickavelu et al. (2006).

Using 59, not previously reported markers (22 SSRs and 37 SNPs), along with 159 SSRs and five indels for genetic mapping, we identified six QTLs for grain yield in chromosome 1, 5, 8, and 9 under drought, whereas no QTL for drought grain yield was found on chromosomes 5 and 8 out of the 16 QTLs reported by Kumar et al. (2014) and 7 meta-QTLs identified by Swamy et al. (2011). Markers RD0107, RM594, and OSR27_RM128 associated with grain yield under drought QTLs, qGYD1.1 and 1GYD1.2, in our study were located at 12.9 Mb, 15.15 Mb and 30.7 Mb of chromosome 1, respectively and were far from qDTY1.1 (Vikram et al., 2011). However, qGYD1.1 and qGYD1.2 co-localized with minor effect QTLs for drought tolerance and drought avoidance traits, such as measuring fitness, yield, and the root system identified from experiments carried out in PVC pipes-grown plants with Zheshan 97/IRAT109 progeny (Yue et al., 2006). Marker RM 128 flanking qGYD1.2 in our study co-located with metaQTL, mqMRL_1-2 at 32.5 Mb for maximum root length, and other QTLs for root thickness, penetrated root thickness and penetrated root length (Courtois et al., 2009). Similarly, marker RM128 was also found to flank a QTL, ph1.4 for plant height with RM226 (Kaladhar et al., 2008), which suggested pleiotropic effect of a chromosomal region affecting grain yield and plant anatomical features. The effect of qGYD1.1 (Table 1) appeared to be dependent on genetic background, as the effect was derived from the drought sensitive Cocodrie cultivar.

Although a region on chromosome 5 between SNP markers S4134205–S7643153 was found harboring four QTLs for grain yield and related traits in two genetic backgrounds and different environments (Wang et al., 2014), this segment did not overlap with the QTL, qGYD5.1 (RM17710-cvssr21) identified in our study. On the other hand, a minor, inconsistent QTL for 100 grain weight under drought was reported on chromosome 5 (RM509-RM430) in the Zhenshan 97/IRAT109 population (Yue et al., 2006).

A QTL on chromosome 8 (qtl 8.1) with the marker interval RM337-RM3664 and peak at RM8020 was found to be associated with grain yield under well-watered upland conditions (Bernier et al., 2007) that overlapped with metaQTL MQTL8.1 (RM337-RM902) for drought grain yield (Swamy et al., 2011) around RM25 reported in Apo/IR72 and Vandana/IR64ApoIR (Venuprasad et al., 2012a). QTLs associated with yield, spikelet fertility and root traits of drought plants relative to control plants were identified in chromosome 8 using Zhenshan 97/IRAT109 (Yue et al., 2006). The QTL, qGY8.1 for grain yield under aerobic conditions (Sandhu et al., 2013) and qDTY8.1 under drought stress (Vikram et al., 2012) co-localized between RM339 and RM210, close to regions harboring QTLs for other yield related traits and root length (Sandhu et al., 2013). Genome-wide association studies also identified the marker RM6070 of chromosome 8 to be significantly associated (P < 0.01) with both plant height (R² = 3.96) and percentage seed set (R² = 12.85%) (Zhou et al., 2012). The markers, RD0806_4 and RM23017, linked to qGYD8.1 identified to be associated with grain yield under drought (R² = 6.4) from our study overlapped with the MQTL8.2 (Swamy et al., 2011). A major QTL located on chromosome 8 explaining 54% of the phenotypic variance for grain yield was reported in Swarna × O. nivara population (Kaladhar et al., 2008). All these reports indicate that chromosome 8 could be a hot spot for alleles with positive effect on yield and drought recovery traits.

The qtl9.2 for biomass yield involving Vandana/Way Rarem was delimited by RM201 and RM205 (Bernier et al., 2007), which overlapped with qGYD9.1 (RM201-cvssr35) for grain yield under drought from our study. However, the positive allele was contributed by Cocodrie in our study as opposed to Vandana that contributed the positive allele in the study by Bernier et al. (2007). Using the advanced backcross lines at seedling stage, a QTL for increased root length and drought tolerance was mapped in the segment between RM201 and RM328 of chromosome 9 (Lang et al., 2013), which was close to the Dro1-KP allele for deep rooting (Uga et al., 2013). In addition, qDTY9.1 for grain yield under drought was flanked by RM105-RM434 (Swamy et al., 2013), which is the single QTL on chromosome 9 for enhanced yield under lowland conditions (Kumar et al., 2014). Chromosome 9 has been documented to harbor hot spots for QTLs for maximum root length and other root traits detected in multiple environments and in different genetic backgrounds (Courtois et al., 2009). The qGYD9.1 detected in our study co-localized with meta-QTLs at ~20 Mb for root traits, such as maximum root length, root thickness and deep root weight (Courtois et al., 2009). Drought avoidance and drought tolerance mechanisms appear to be the underlying effects of the QTLs on chromosome 9 toward improved grain yield, thus emphasizing the necessity to pyramid these traits.

Other QTLs, previously reported in populations involving Vandana as a parent were qDTY6.1 on chromosome 6 (Venuprasad et al., 2012a) under aerobic drought stress and irrigated lowland conditions, and qtl2.1 on chromosome 6 with significant effects on grain yield under non-stress conditions (Bernier et al., 2007). However, no QTL was found on chromosome 6 associated with drought grain yield in the mapping population in our study. Interestingly, our greenhouse drought study did not identify a QTL on chromosome 12, unlike qtl12.1 that was identified in F₃ generations of Vandana/Way Rarem (Bernier et al., 2007) with the positive allele for high yield under drought contributed by the drought sensitive Way Rarem. However, qGYC12.1, a QTL for grain yield under non-stressed control with ~50% additive effect (PVE 4.8%) was identified in our study (Supplementary Figure S1; Supplementary Data S3). QTLs identified for yield under unstressed control conditions did not overlap with any of the drought grain yield QTLs in our study. Since, the present work was conducted to dissect loci with effect on grain yield under controlled conditions, it is not surprising that QTLs reported here do not overlap/co-localize with the reported QTLs coming from studies carried out under upland and lowland drought situations. However, QTLs, such as DRO1 for drought tolerance or drought avoidance have been found in controlled conditions (Uga et al., 2011). In most studies involving phenotypic evaluations of drought tolerance under upland and lowland conditions across multiple environments, few grain yield QTLs have shown consistent effects. Drought tolerance QTLs detected for upland rice varieties may not be suitable for rainfed lowland conditions and vice versa (Mackill et al., 1996), mainly because of the unique hydrology of rainfed lowlands in which soil transitions from flooded and anaerobic to drought and aerobic (Wade et al., 1999).

Due to the complexity of the components involved in grain yield under drought, many QTLs for different yield-related traits co-localize in same chromosomal regions with pleoiotropic effects. Although, many genetic studies have revealed the location of QTLs of interest for grain yield, only a few reported the genes and their mode of action within the discovered QTLs (Ashikari et al., 2005). The QTLs/genes presented here are expected to provide further clues to identifying underlying mechanisms involved in improved grain yield under drought stress. Our on-going experiments are aimed at confirming the genomic regions and narrowing down of the number of genes reported within the QTLs in the present study through comprehensive studies involving high-resolution linkage mapping via high-throughput genotyping by sequencing of advanced generation progeny such as RILs/BILs and phenotyping under natural conditions with rain-out shelter facilities.

NB: conceptualized and designed the experiment; NB, VM, JS, ES, and SL: performed the experiments; JS, AG, and RB: analyzed the data; JS and NB: wrote the manuscript.