Positive interactions of major-effect QTLs with genetic background that enhances rice yield under drought

To improve the grain yield of the lowland-adapted popular rice variety Samba Mahsuri under reproductive-stage drought (RS) and to understand the interactions between drought QTLs, two mapping populations were developed using marker-assisted selection (MAS) and marker-assisted recurrent selection (MARS). The mean grain yield of pyramided lines (PLs) with qDTY2.2 + qDTY4.1 in MAS is significantly higher under RS and irrigated control than lines with single QTLs. Among MARS PLs, lines with four qDTYs (qDTY1.1 + qDTY2.1 + qDTY3.1 + qDTY11.1) and two QTLs (qDTY1.1 + qDTY11.1) yielded higher than PLs with other qDTY combinations. The selected PLs showed a yield advantage of 0.3–2.0 t ha−1 under RS. An allelic profile of MAS PLs having same qDTY combination but with different yields under drought was studied. Hierarchical clustering grouped together the selected lines with high yield under drought. Epistasis test showed the interaction of qDTY4.1 and qDTY9.1 loci with qDTY7.1 significantly increased yield under drought and all the lines with higher yield under drought possessed the conserved region of qDTY7.1 on chromosome 7. The positive interactions among QTLs, effectiveness of QTLs in different backgrounds, introgression of DTY QTLs together with resistance to biotic stresses shall help enhance grain yield under RS.

The increasing population, declining water availability with increasing demand, climate change and abiotic stresses are serious threats to world food security. Global rice consumption is reported to increase 8.8% from 2011 to 2020 and about 32.7% by 2050 1 . Drought is a recurring global, climatically induced calamity that affects rice production in arid, semi-arid, and humid areas as well. Tackling drought can provide excellent returns to rice research and development investments. In India, severe drought in 2002 and 2009 caused a 20% and 16% reduction in food grain, respectively, leading to higher prices and food security concerns 2 . In 2004, severe drought affected the crop on more than 2 million ha in Southeast Asia 2 . An effective breeding approach to develop drought-tolerant rice varieties could lead to food security under situations of accelerating food demand, depleting resources, and predicated climatic variability 3 .
Marker-assisted selection can generate an improved version of an existing elite genotype as it involves the transfer of favorable alleles from a donor to a recipient parent. Breeding approaches such as marker-assisted recurrent selection (MARS) involve taking advantage of elite alleles coming from two or more parental lines can be a viable alternative. MARS, a practice of improvement of low heritability traits 24,25 , involves the consideration of selection intensity to increase the frequency of favorable superior genes 21 and genetic drift to maintain diversity to carry on further improvement 26 . MARS has the potential to expand the gene pool of present cultivars and to expedite the development of new varieties. MARS involves selecting genotypes based on their favorable allele combinations and intermating them to produce the next generation [27][28][29][30] . Repeated intermating of heterozygous populations helps in successfully maintaining and escalating the genetic gain and variability [31][32][33][34] resulting from the optimum complement from both parents. Validation of the effects of QTLs that showed large and consistent effects in one background will promote their extensive use across different genetic backgrounds for increasing yield under drought. In the present study, two mapping populations were developed following marker-assisted breeding and marker-assisted recurrent selection using rice variety Samba Mahsuri as the recipient parent. The objectives of the study were to (1) evaluate the effects of qDTY 2.2 and qDTY 4.1 in a Samba Mahsuri background using the MAB approach, which had earlier shown effects in an IR64 background; (2) understand the interactions of drought grain yield QTLs qDTY 1 13 were used to genotype the population. Foreground and recombinant genotyping was used to identify true F 1 in WS 2011 (WS: wet season). The grain type of the selected marker assisted lines is provided in Supplementary Fig. S1. The complete scheme for the development of Samba Mahsuri pyramided lines and number of plants selected based on genotyping, phenotyping involving plant type, visual yield under drought stress and non-stress, and bacterial blight resistance is provided in Supplementary  Fig. S2.
Marker-assisted recurrent selection for qDTY 1 with Samba Mahsuri grain type were selected and advanced in WS 2011. The lines with different QTL combinations and the highest phenotypic similarity to the recipient parent and good grain yield under drought stress and non-stress were selected in the BC 1 F 6 population and intercrossed in DS 2012. In WS 2012, 15641 F 1 seeds were produced. In each intercross F 2 generation, selection was done based on grain type similar to the recipient type, blast resistance (Magnaporthe oryzae) and bacterial blight (Xanthomonas oryzae pv. Oryzae) resistance, and foreground and recombinant selection were practiced using the peak and other foreground markers RM212 and RM486 (chromosome 1), RM525 and RM221 (chromosome 2), RM16 and RM520 (chromosome 3), and RM287 (chromosome 11) to select plants segregating for the respective introgressed drought QTLs in DS 2013. Plants fixed for different combinations of QTLs, grain type ( Supplementary Fig. S1), plant type and visual yield were selected in each generation and advanced to F 8 generations. The scheme for the development of Samba Mahsuri PLs using partial marker-assisted recurrent selection is provided in Supplementary Fig. S3.
Phenotypic evaluation of populations. The average days to flowering (DTF) of PLs in the marker-assisted backcross experiment varied from 79 to 88 days in the NS trials and from 82 to 98 days in the reproductive-stage (RS) drought stress trials (Table 1). At maturity, plant height (PHT) varied from 62 to 71 cm in the reproductive-stage drought stress trials and from 81 to 105 cm in the NS trials (Table 1). Plant height and DTF were severely affected by drought stress as reflected by the reduction of 19 to 34 cm in height and delayed flowering by 3 to 10 days. Grain yield (GY) varied from 114 to 1366 kg ha −1 in the RS drought stress experiments and from 2376 to 8133 kg ha −1 in the NS experiments ( Table 1). The grain yield reduction of 83% to 95% under drought stress compared with NS (control) indicated the severity of reproductive-stage drought stress faced by PLs. In the marker-assisted recurrent selection experiment, the average DTF of PLs varied from 80 to 97 days in the NS trials and from 85 to 116 days in the RS trials (Table 1). Plant height varied from 99 to 113 cm in the NS trials and from 73 to 85 cm in the RS drought stress trials (Table 1). Plant height decreased by 26 to 28 cm and flowering was delayed by 5 to 19 days. Grain yield ranged from 126 to 1751 kg ha −1 in the RS experiments and from 3131 to 5824 kg ha −1 in the NS experiments ( Table 1). The PLs IR 102818-10-266-3-2-2-6, IR 102818-10-276-1-2-2-9, IR 102818-10-227-1-2-1-9, IR 102818-10-227-1-2-1-6, IR 99734:1-33-69-1-39-6, IR 99734:1-33-69-1-12-8, and IR 99734:1-33-69-1-12-9 showed resistance to blast (Magnaporthe oryzae). The  Table 2, Supplementary Fig. S4 (A,B) and Table 3, Supplementary Fig. S4 (C,D) respectively.
The mean grain yield of PLs with qDTY 2.2 + qDTY 4.1 (class C) is significantly higher than in lines with a single QTL in DS 2016 ( Effect of QTL pyramiding on agronomic traits. The Samba Mahsuri PLs with either single or multiple qDTYs produced higher yield than the recipient parent in both marker-assisted backcross and marker-assisted recurrent selection experiments even under severe drought stress ( Supplementary Fig. S5). The severity of drought stress can be assessed by the water table level 8,9 ( Supplementary Fig. S6). The performance of the most promising drought-tolerant PLs in the marker-assisted backcross experiments at IRRI (Philippines), Hyderabad (India) and in the marker-assisted recurrent experiment at IRRI (Philippines) is presented in  Table 4). The selected promising PLs in the marker-assisted recurrent selection experiment flowered earlier than the recipient parent but the variability of PLs ranged from 53 to 100 days in NS and from 72 to 119 days under RS ( Table 5). The PHT of promising PLs in the marker-assisted recurrent selection experiment ranged from 86 to 122 cm and from 64 to 95 cm under NS and RS, respectively ( Table 5). The performance of PLs is dependent on season and level of stress (Tables 4 and 5). The grain yield advantage of selected promising PLs over Samba Mahsuri in MAB and MARS experiments is shown in Supplementary Tables S1 and S2, respectively. The grain yield advantage ranged from 297 to 4232 kg ha    Table S4). Visualization analysis of molecular marker scores was done using GGT 2.0 software 36 and colored chromosome bar segments representing allelic distribution are shown in Fig. 2. The PLs in Groups 1 and 5 that had shown better performance across seasons were reported to have a conserved region near RM21510 on chromosome 7. This region is also present in the IR64 introgressed line with qDTY 2.2 and qDTY 4.1 , IR 87707-445-B-B-B (Fig. 2). The PLs with a conserved genetic region on chromosome 7 had shown a yield advantage over PLs without the conserved genetic region under NS and severe reproductive-stage drought stress (Supplementary Table S5). Supplementary Fig. S6 validates the severity of reproductive-stage drought stress in WS 2015 and DS 2016.

Discussion
In the present study, bacterial blight-/blast-tolerant, lowland-adapted but drought-susceptible, high grain and cooking quality line, Samba Mahsuri was used as a recipient parent to improve its yield under drought.    As shown earlier in IR64 background (Swamy et al. 13 ), in the current study also, qDTY 2.2 , qDTY 4.1 combination showed higher yield advantage under drought over single QTLs in Samba Mahsuri background indicating the effectiveness as well as positive interactions between these two QTLs in multiple genetic backgrounds. The release of IR64 PLs with qDTY 2.2 + qDTY 4.1 in India, Nepal, and Myanmar, validate the effect of these QTLs in reducing yield loss under reproductive-stage drought stress in variable environments (Sandhu et al. 38 ).
This is the first study in rice comparing the yield advantage under severe reproductive stage drought achieved through MAS and MARS strategies. Severe and cyclic drought stress exposure to the population in the present study assisted in identifying true drought-tolerant lines 39 with different growth duration 40 . The grain yield advantage in selected promising PLs over the recipient parent is high in MARS compared to MAS. The superiority of MARS lines may be because of accumulation of higher proportion of drought tolerant alleles during recurrent selection process as compared to the lines developed through MAS. MARS in sweet corn 41,42 , soybean and sunflower 43 , maize 18,21,[43][44][45][46][47] , wheat 48 , and cucumber 24,25,49 has proven to be effective in increasing the frequency of favorable alleles with improvement in grain yield, grain yield-related traits, and drought tolerance.
Drought stress in general occurs together with a high prevalence of biotic stresses such as bacterial blight, blast, and brown spot. The development of high-yielding drought-tolerant rice varieties with tolerance of biotic stress could considerably help to control heavy yield losses. However, very few molecular breeding studies have been undertaken to study the combined effect of abiotic and biotic stress tolerance simultaneously in mapping populations. The present study reports an integrated strategy of QTL pyramiding to develop PLs with high grain yield under reproductive-stage drought stress together with tolerance of biotic stress. We observed that early generation-systematic screening of large F 2 population for biotic stress tolerance may provide opportunity to select lines for the second targeted trait (high grain yield under RS) in reduced time span. The developed PLs having tolerance to both biotic and abiotic stresses may help to identify, exploit and understand the mechanism of potential QTLs/genes providing tolerance to multiple stresses. It may also serve as useful resource for crop improvement program directed toward improving agronomic traits and multiple stress resistance.
The GY advantage of the PLs with qDTY 1.1 and qDTY 11.1 either single or in combination with other QTLs under RS and NS indicates the superiority of qDTY 1.1 and qDTY 11.1 in marker-assisted introgression programs over other QTLs. qDTY 1.1 has been reported to be associated with increased yield under multiple conditions (dry direct seeded, drought, non-stress) and multiple backgrounds (Swarna, IR64, MTU1010) 6,12,50 . Dixit et al. 17 reported the performance of qDTY 11.1 in a Swarna background. In this study, across the PLs with two, three, four QTLs combinations, qDTY 1.1 showed positive interaction with qDTY 11.1 as lines possessing these two combinations of QTLs outyielded lines with other QTLs combinations. The study indicates that the breeding programs targeting MAS of qDTY 1.1 should also include qDTY 11.1 for higher yield advantage under both NS and RS.
The use of donors having qDTY 9.1 (qDTY 9.1 ; IR 77298-5-6-B-18) 13 supported the presence of qDTY 9.1 in the Samba Mahsuri background in the current MAS QTL introgression program. The presence of qDTY 9.1 in Samba Mahsuri (Fig. 3) and IR64 background 13 under reproductive-stage drought stress conditions indicates the effectiveness of the genetic region in different backgrounds. The stable performance of selected PLs in the Philippines and Hyderabad, India (Supplementary Table S1), indicates the effectiveness of qDTY 9.1 in different backgrounds (Samba Mashuri, IR64) and environments.
The presence of the conserved allelic region on chromosome 7 (marker RM21510) in selected promising PLs showed higher yield advantage and its absence showed lower yield advantage under RS even if the other qDTYs were present (Fig. 2, Supplementary Table S3). The introgressed QTLs do not explain the entire phenotypic variation of the PLs for GY, this potentially indicates the importance of capturing the positive digenic interaction of qDTY 7.1 with qDTY 4.1 and qDTY 9.1 (Fig. 3, Supplementary Table S6). The epistasis interactions of these loci suggest their importance in elucidating the genetic basis of GY in high yielding PLs under RS. Understanding the genetic composition of these identified loci and their effect on GY may allow us to use these loci to achieve maximum GY advantage under RS. Earlier, Dixit et al. 22 reported the positive interaction of qDTY 2.3 and qDTY 3.1 with qDTY 12.1 and a significant effect on grain yield increase under RS drought stress. Unlike Dixit et al. 22 where in interactions between two identified QTLs were reported, in the present study, even if no QTLs for grain yield under drought near RM10 on chromosome 7 has been detected, the effect of the interaction of this region with qDTY 4.1 and qDTY 9.1 has significantly enhanced the yield under drought. In an earlier study, genetic loci for grain yield under drought stress, plasticity for root dry weight and total water uptake, and total root length were identified 49 located near qDTY 7.1 and qDTY 9.1 , respectively. The broad understanding of the interactions identified in the present study and introgression of such specific combinations of a few QTLs may be more effective in increasing yield than random combinations of many QTLs. The complexity of effect and interaction among the most favorable alleles or haplotypes controlling the trait of interest, differential expression of expressed regions under variable conditions, and increase in frequency of positive alleles over negative contributing alleles may be responsible for the differential behavior of PLs having the same QTLs. " Previously identified QTLs and genes in qDTY 4.1, qDTY 7.1 and qDTY 9.1 may provide insights as to why the QTLs identified in the present study contribute to grain yield advantage under RS. The QTLs associated with germination 51 , panicle length (sp2(t)) 52 , tiller number (tp7(b)) 53 , days to flowering (dth7.1) 54 and rice stripe virus resistance (Rurm1) 55 were identified in qDTY 7.1 region. Various genes related to stress-signaling, stress-responses and signaling, growth and development processes, and hormonal regulation and transcription factors were reported within qDTY 4.1 and qDTY 9.1 by Swamy et al. 13 . The upstream region of qDTY 7.1 (15603452-18640879 bp) found to be associated with the gene responsive to phosphate uptake efficiency (Li et  be a probability that the genomic region (15414191-18878266) on chromosome 7 may be involved in providing tolerance to multiple stresses involving increasing nutrient uptake to increase yield under drought and biotic stress resistance. Further detailed studies on targeted genomic region on chromosome 7 may reveal additional information.

Conclusions
The results reported in the present study indicate the effectiveness of introgressed QTLs, interaction of QTLs with other loci, to further enhance GY under reproductive-stage drought stress while following MAS and MARS approaches. The GY advantage achieved is more in MARS than in MAS. Drought-tolerant Samba Mahsuri PLs with a grain yield advantage of 0.5-1.0 t ha −1 under reproductive-stage drought stress were developed. The PLs showing similar grain type and plant type as Samba Mahsuri together with the tolerance to biotic and abiotic stress may act as a candidate to replace the variety Samba Mahsuri. The positive interaction of introgressed QTLs with other QTL/genetic region and genetic background as reported in the present study could be one of the possible reason for the variable effect of introgressed QTLs in PLs. Identification of these positive interactions, allele mining and complete sequencing of promising PLs could reduce to a certain extent QTL x genetic background interactions often observed under RS.

Materials and Methods
Plant material, phenotyping, and management. The study was conducted at the International Rice Research Institute (IRRI), Los Baños, Laguna, Philippines. To evaluate the effect of QTLs that had earlier shown an effect in the IR64 background under drought, two approaches were used. The popular rice variety Samba Mahsuri (occupying 3.3% of rice growing area in India), was used as a recipient to develop mapping populations through marker-assisted breeding and marker-assisted recurrent selection approaches. Samba Mahsuri is a medium-tall (90-100 cm) variety with 140-145 days duration, superfine grain with excellent grain and cooking quality, hulling percentage of 70%, head rice recovery of 75%, kernel length of 5.45 mm, kernel breadth of 1.97 mm, L/B ratio of 2.7, elongation ratio of 1.85, alkali spreading value of 5.5, and amylose content of 24.8%. Samba Mahsuri is a very popular variety, preferred for its fine slender premium grain and excellent cooking quality. The drought-tolerant indica rice varieties/pre-breeding lines developed at IRRI, IR 87728-75-B-B possessing qDTY 2.2 and qDTY 4.1 and IR55419-04 possessing qDTY 1.1 , qDTY 2.1 , qDTY 3.1 , and qDTY 11.1 , were used as donors in marker-assisted backcross and marker-assisted recurrent selection approaches, respectively. The scheme of developing the Samba Mahsuri PLs and number of selected plants in each subsequent generation using MAB and partial MARS approaches is shown in Supplementary Figs S2 and S3, respectively. The lines were screened under lowland transplanted control and lowland reproductive-stage drought stress conditions. Screening of PLs was conducted using an α-lattice or randomized complete block design (RCBD) or augmented RCBD along with drought-tolerant and susceptible checks and donor/recipient parents in 1-4-row plots 3-5 m in length, with 0.20-0.25-m row-to-row spacing and 2.0-2.5 g seed per linear meter. For all the trials, the seeds were sown in a raised bed nursery and 21-25-day-old seedlings were transplanted to the main field. Inorganic fertilizers NPK (nitrogen, phosphorus, and potassium) were applied @ 120:30:30 kg ha −1 . Weeds, insect pests, and snails management was done as described by Venuprasad et al. 5 . For non-stress, the trials were conducted under irrigated, transplanted, flooded, puddled, and anaerobic conditions with no drought stress. The reproductive-stage drought stress experiments were carried out in an automated rainout shelter facility at IRRI.
For reproductive-stage stress, transplanted experiments were maintained as described by Sandhu et al. 58 . The drought stress was initiated at 32 days after transplanting. After the inception of the stress, the soil water potential was measured using tensiometers (30 cm depth) in DS 2013 and DS 2014 (only). The plots in the reproductive stage drought stress treatments were rewatered when the soil water potential dropped to −50 to −70 kPa (tensiometer). The decline in water table depth was measured on a daily basis with a meter scale inserted into a 1.1-m polyvinyl chloride pipe in the experimental fields at regular intervals in all RS treatments. The pipes were placed at 1.0-m depth with 10 cm of pipe remaining above the soil surface. The plots were rewatered when water table level reached 100 cm below the soil surface and most lines were wilted and exhibited severe leaf drying. This cyclic reproductive stage drought stress allows the effective screening of broad range growth duration PLs 40 .
Data on days to 50% flowering (days) were recorded when 50% of the plants in the plot started flowering. At maturity, when 80-85% of the panicles turned golden yellow, plant height (cm) was measured as the mean height of three plants per plot from the base to the tip of the tallest panicle. The grains were harvested from each plot, dried to a moisture content of 14%, and weighed. Grain yield data were normalized as per moisture content for yield computation (kg ha −1 ). Visual observation on grain type similar to that of the recipient parent was made in the field per plot during plot selection and per plant during panicle selection. Blast (caused by Magnaporthe oryzae) and bacterial blight (caused by Xanthomonas oryzae pv. Oryzae) screening was done twice first at F 2 and then at F 8 stage. Mixed inoculum for races present in Philippines was used for blast inoculation. Genes pik-s, pi2, pi5(t), pi9 showed resistance against the mixed inoculum. For bacterial blight inoculation race 1 (PXO61) and race 2 (PXO86) was used. Inoculation and scoring for bacterial blight were done at maximum tillering stage as described by Kauffman et al. 59 . Inoculation for blast screening was done 10 days after seeding in the blast nursery and, after an exposure of 22 days, scoring was done based on the SES scale (0: highly resistant, 1-2: resistant, 3-4: moderately resistant, 5-6: moderately susceptible, 7-8: susceptible, 9: highly susceptible) 60,61 . Phenotypic selection was used for biotic screening. Selection for resistance genes against blast and bacterial blight was not targeted in the present study.
Genotyping. Molecular marker work was carried out in the Molecular Marker Application Laboratory of IRRI. Fresh young leaves from all lines were collected, freeze-dried, and the DNA extracted using the modified CTAB protocol 62 . A total of 112 SSR markers linked to two qDTY regions were tested for polymorphism in the marker-assisted breeding experiment. The markers linked to qDTY 2.2 (RM236, RM279, RM109) and qDTY 4.1 (RM335, RM551, RM518) were found to be polymorphic in the Samba Mahsuri background and were used for foreground selection. The lines having the same QTL combination were found to be segregating for yield components. To study the allelic pattern of these lines, a total of 650 SSR markers distributed across the rice genome were tested for polymorphism. A total of 99 polymorphic markers were used for the background study. In the marker-assisted recurrent selection experiment, a total of 200 markers linked to four qDTY regions were tested for polymorphism. Foreground and recombinant selection were carried out using RM212 and RM486 (chromosome 1), RM525 and RM221 (chromosome 2), RM16 and RM520 (chromosome 3), and RM287 (chromosome 11) polymorphic markers. Amplification was carried out using polymerase chain reaction (PCR) and PCR products were resolved in non-denaturing polyacrylamide gel electrophoresis (6% or 8%) depending on product size. Gels were stained with SYBR SafeTM DNA, viewed after 20 min, and DNA profiles were scored based on parent allelic profile. Stepwise selection involving phenotyping and genotyping was used to select and advance the desirable plants in every generation.

Statistical analysis.
Mean and standard error of difference of the experiments with alpha lattice were calculated using the linear mixed model of CROPSTAT version 7.2.3 considering replications and blocks within replication as random effects and lines as fixed effects.
where μ is the overall mean, g i is the effect of the i th genotype, r j is the effect of the j th replicate, b lj is the effect of the l th block within the j th replicate, and e ijk is the error. For the RCBD, the model used was: where μ is the overall mean, g i is the effect of the i th genotype, r j is the effect of the j th replicate, and e ijk is the error. For the augmented RCBD, the model used was: where μ is the overall mean, g i is the effect of the i th genotype, b l is the effect of the l th block, and e ilk is the error.
Class analysis for qDTY PLs. Class analysis for qDTY PLs was performed using SAS v9.2 (SAS Institute Inc. 2009), considering the effects of QTLs and genotypes within the QTL as fixed effects and replicate and blocks within replicate as random effects. The model used to see the performance y ijkl of the j th genotype nested within the i th QTL class in the l th block within the k th replicate is as follows: where μ is the population mean, r k is the effect of the k th replicate, b(r) kl + q i is the effect of the l th block within the k th replicate, q i is the effect of the i th QTL, g(q) ij is the effect of the j th genotype nested within the i th QTL, and e ijkl is the error 63 . ANOVA and F test using SAS v9.2 (SAS Institute Inc. 2009) were used to see whether the QTL classes differed significantly from each other.
Graphical representation of the genome. Graphical representation of molecular marker data was performed using the software Graphical Genotypes (GGT 2.0) 36 . The homozygous donor allele, homozygous recipient allele, heterozygous allele, and recombinant allele were scored as ' A' , 'B' , 'H' , and 'R' , respectively. The estimated proportion of the A, B, H, and R alleles in each PL was calculated using GGT 2.0.
Diversity studies of PLs segregating for grain yield. DARwin 6.0.013 software was used to compute a pairwise distance matrix by calculating a dissimilarity matrix 35 . An Unweighted Pair Group Method with Arithmetic Mean (UPGMA) followed by bootstrap analysis with 1000 permutations was used to construct a neighbor joining tree. Q x Q interactions. Detection of QTL x QTL interactions in the marker-assisted pyramided population was performed in QTL Network 2.1 based on a mapping methodology summarized by Yang et al. 64 . A two-dimensional genome scan with 1000 permutation was performed to identify additive x additive interaction. Determination of the QTLs intervals, detection of Q x Q interactions, and their additive x additive effect were considered as significant at P ≤ 0.01.