Epistatic interactions of major effect drought QTLs with genetic background loci determine grain yield of rice under drought stress

Epistatic interactions of QTLs with the genetic background and QTL-QTL interaction plays an important role in the phenotypic performance of introgression lines developed through genomic-assisted breeding (GAB). In this context, NIL pairs developed with various drought QTL (qDTY) combinations in the genetic background of IR64, TDK1-Sub1 and Savitri backgrounds were utilized to study the interactions. Multi-season phenotyping of NIL pairs harboring similar qDTY combinations provided contrasting performance for grain yield under drought (RS) (classified as high and low yielding NILs) but nearly similar performance under non-stress(NS) conditions. Genome wide genotyping data revealed a total of 16, 5 and 6 digenic interactions were detected under RS conditions in low yielding NILs of IR64, TDK1-Sub1 and Savitri respectively while no significant interaction was found in high yielding NILs under RS and NS conditions in any of the genetic backgrounds used in this study. It is evident from this study that existence of epistatic interactions between QTLs with genetic background, QTL-QTL interaction and interactions among background markers loci itself on different chromosomes influences the expression of a complex trait such as grain yield under drought. The generated information will be useful in all the GAB program of across the crops for precise breeding.

Drought is one of the most severe climatic threat to rainfed rice production affecting on more than 23 million hectares of South and Southeast Asia 1 . A further increase in drought intensity has anticipated due to ongoing global climatic changes. Development of climate-adapted crop varieties is an urgent need to counter the effect of more than 50% yield losses due to abiotic stresses 2,3 . Widely grown high yielding rice varieties are highly sensitive to drought. The recent advances in molecular marker and genomics technologies have played a major role in selection of drought-tolerant traits. Rice drought breeding programme at IRRI has successfully identified and validated some of the major and consistent effect grain yield QTLs under reproductive stage drought (qDTYs) such as qDTY 1 , qDTY 6. 1 8 and qDTY 12.1 9 using molecular markers. A considerable effort has been made to improve drought susceptible mega varieties of rice through marker-assisted introgression of qDTYs [10][11][12][13] . Assembling multiple alleles/QTLs/genes in a genetic background using marker-assisted introgression may be an appropriate approach to achieve the expected phenotypic variance in enhancing rice grain yield under drought stress 7,12,14 , submergence tolerance 15 , salt tolerance 16 , cold tolerance 17 , resistance to blast [18][19][20] , bacterial blight [21][22][23] , brown planthopper 24,25 . The introgression of major effect QTLs/genes and its combinations do not always result in desirable improvement for the targeted trait(s). In some cases, the effect of identified genetic loci in a QTL/gene pyramiding program is not sufficient to fully explain the desired phenotypic variation due to various interactions occurring for a complex quantitative trait. The possible interactions between the genetic make-up, epistasis, pleiotropy and linkage among/between the introgressed loci, with the genetic background and environment could have negative/positive effects on the expression of the traits that should never be underestimated [26][27][28] . QTL pyramiding approach has enormous potential to improve our understanding of interactions among QTLs and also help to develop new strategies to increase the efficiency of marker-assisted selection program. In this context, development of NILs using marker assisted selection (MAS) is a way forward to verify the effects of one or few major effect QTLs as well as increase the recovery of recurrent parent genome at a faster rate 29 . Rice NILs for complex traits such as drought 7,13 , blast 19,30 and brown planthopper 25,31 were developed and evaluated in different genetic backgrounds of popular rice varieties.
NILs harboring QTLs/genes for abiotic/biotic stresses in the elite backgrounds with superior performances are useful material for immediate release to farmer's field. NILs are also useful source to study the stable epistatic interactions between the introgressed donor segment and the rest of the recurrent parent genome which gives an important clue about differences in performance of the same set of NILs under variable conditions. The expressions of QTLs/genes for complex traits are strongly affected by the genetic backgrounds 32,33 . The estimation of genetic backgrounds or recurrent parent genome (RPG) recovery with different types of markers and number of markers also affect the estimation of RPG recovery as a good number of markers with good coverage estimates better information 19 . Better estimation of RPG recovery at early generation of selection can be useful to eliminate linkage drags which can cause unexpected traits in the MAS breeding products because of neighboring genes of the target QTLs. Also, the favorable allele at one locus in a particular background may become unfavorable in another genetic background due to the epistatic interaction 34 . Role of epistasis phenomena in plant breeding has been discussed for a long time, but the extent of the expression level of quantitative traits is due to this complex genetic phenomena are poorly understood 35 . Epistatic or non-allelic interaction is a genetic factor and may change the magnitude of phenotypic expression of QTLs by suppressing/enhancing the loci effect associated with a trait to the other loci of genetic background 34,36 . The epistatic effect is a deviation from the sum of independent effects of QTLs/genes therefore biased estimation of QTL effects is possible in genetic models assuming no interaction in QTL mapping studies 37,38 . The epistatic interaction of QTLs and background or modifying loci has been recently reported for quantitative traits in various crops [38][39][40][41] .
Genomic interactions play an important role in deciding the performance of MAS lines for grain yield under drought and various significant QTL x QTL interactions in pyramided lines have been reported for drought-related traits 12,28,42,43 . Most of these studies highlighted capture of such interactions for increased performance of QTLs pyramided lines under drought. But none of these studies reported reasons for a lower yield of NILs with same combinations of pyramided QTLs in the same or different recipient backgrounds. Keeping in this view, the present study is an effort (i) to evaluate the performance of drought NILs of IR64, TDK1-Sub1 and Savitri backgrounds under reproductive stage moderate and severe drought stress and non-stress conditions (ii) to identify promising drought tolerant NILs with higher recovery of recipient genome (iii) to study epistatic interaction of major effect grain yield drought QTLs and with background marker loci.

Results
Phenotypic performance of drought NILs possessing qDTY under RS and NS conditions. Significant phenotypic differences under RS and NS were observed among the parents and NILs in IR64, TDK1-Sub1 and Savitri backgrounds. The results of ANOVA, means, LSD, heritability of IR64 NILs with a qDTY and qDTY combinations under RS (SS and MS) and NS are summarized in Table 1. Among IR64 NILs, the average yield reduction was found 87% under SS and 74% under MS as compared to NS conditions. Trial mean of NILs in IR64 background for GY was 690 kg ha −1 under SS in 2015 DS while it was 1358 kg ha −1 under MS in 2017 DS. Broad-sense heritability (H) of GY under RS ranged from 0.78 to 0.89 in the years of 2015 and 2017 DS. The NIL pairs harboring same qDTY or qDTY combinations with on par background recovery, provided significantly different grain yield (GY) under similar level of drought stress were classified into high and low yielding NILs (Table 1). For instance, nine IR64 NILs possessing qDTY 12 Mean performances of measured traits of TDK1-Sub1 NILs, heritability and LSD is presented in Table 2. Grain yield under severe stress (GYSS) ranged from 0 kg ha −1 to 479 kg ha −1 with trial mean of 327 kg ha −1 . The recurrent parent TDK1-Sub1 yielded 14.4 kg ha −1 while the donor parent IR55419-04 performed well with GY of 881 kg ha −1 under SS situation. GY mean of MS trial (GYMS) was 889 kg ha −1 , with range of 300 kg ha −1 to 1555 kg ha −1 . The average yield reduction was more than 90% under SS while 84% less grain yield was recorded under MS among TDK1-Sub1 NILs. Heritability for GY was recorded 0.89 to 0.91 in RS and 0.58 in NS. The pyramided three NILs with qDTY 6.1 + qDTY 6.2 combination yielded 0 to 479 kg ha −1 under SS situation. However, the performance of both lines was 300 and 1555 kg ha −1 respectively in MS. Similarly, NIL pair consisting of qDTY 3.1 + qDTY 6 www.nature.com/scientificreports www.nature.com/scientificreports/ The mean PH under SS, MS and NS were 64 cm, 69 cm and 92 cm, respectively. Significant differences in DTF with trial mean of 94 days and flowering differences of 7 days between entries was observed under RS. Earliest line flowered in 80 days while most of the lines were of late maturing type and it flowered in 96 days. As observed in the case of IR64 NILs, TDK1-Sub1 NILs was also produced nearly similar grain yield under NS situation.
The same traits were also observed in the Savitri NILs ( www.nature.com/scientificreports www.nature.com/scientificreports/ Mean PH of Savitri NILs was significantly reduced under drought stress conditions ranged from 61 to 85 cm in SS, 58 to 75 cm in MS and from 78 to 102 cm in NS conditions. PH was reduced by 17 to 20 cm in drought stress. The mean days to flowering (DTF) in Savitri NILs ranged from 72 to 81 days in SS, 76 to 82 days in MS and from 74 to 81 days in NS conditions. DTF was severely affected under drought for recipient parent Savitri varied from 103 days in NS while 113 and 96 days was recorded in SS and MS conditions respectively. The heritability (H) was high for DTF (0.84,0.85,0.92) and medium to high for PH (0.72,0.80,0.88) and GY (0.74,0.89,0.58) under SS, MS and NS conditions for Savitri NILs (Table 2).
Promising drought NIL lines with the maximum genetic background coverage of recipient parents. Recurrent parent genome recovery (RPG) were estimated using 111, 107 and 89 polymorphic SSR markers among NILs carrying qDTYs in IR64, TDK-Sub1 and Savitri backgrounds, respectively. The background recovery among the NILs of multiple genetic backgrounds ranged from 85 to 93% in the case of IR64 NILs, 86 to 90% in the case of TDK-Sub1 NILs and 80 to 89% in the case of Savitri NILs. Most of the IR64 NIL lines showed more than 90% RPG recovery followed by TDK-Sub1 and least recovered lines belong to Savitri background. Graphical representation of the selected pyramided NILs achieving higher genome recovery of respective backgrounds is depicted in Supplementary Table 3. These superior NILs with excellent drought tolerance could be used a drought tolerant line/variety to be released after multi-location evaluation in national and provincial coordinated trials in target countries.

QTL-QTL and QTL-background interaction in drought NILs. A total of 16 digenic interactions
between marker loci were detected in low yielding NILs of IR64 background under SS and MS situation. Out of 16 digenic interactions, we identified 6 interactions between drought grain yield QTLs (qDTY 2.2 , qDTY 2.3 and qDTY 3.2 ) and background markers loci while 9 interactions were found among background markers loci of different chromosome. A Significant negative QTL-QTL interaction was also found between introgressed qDTY 1.2 (RM543-RM212) located on chromosome 1 with qDTY 2.3 (RM263-RM573) identified on chromosome 2. In context of QTL and background marker loci interactions, qDTY 2.2 (RM154-RM236) interacted negatively with two genetic loci (RM7-RM332) on chromosome 3 and chromosome 4 (RM142-RM119), while qDTY 2.3 (RM263-RM573) was interacted negatively with genetic loci (RM108-RM215) on chromosome 9 and chromosome 10 (RM271-RM269) under SS situation (  Fig. 1b). Out of 9 digenic background markers interaction detected in present study, two negative interactions with marker interval (RM7-RM232; chromosome 3) was found at single loci position at 65 cM with background markers located on chromosomes (9 and10). Similarly, two negative interactions with marker interval (RM246-RM543; chromosome 1) were found at single loci position at 141 cM with background markers located on chromosomes (9 and 11). The PVE% of epistatic QTLs ranged from 3.47 to 25.46% under SS to MS. Surprisingly, none of high yielding NILs expressed any interactions in SS and MS conditions (Fig. 1c,d). Also, no significant interactions were also observed under NS conditions for IR64 NILs including both high yielding as well as low yielding NILs (Fig. 1e). A heat map according to LOD score along the chromosomes reflecting all the significant interactions among low yielding IR64 NILs under SS and MS situations is presented in Supplementary Fig S4. Low yielding NILs of TDK1-Sub1 background had shown 5 significant epistatic interactions and all these interactions were detected between introgressed QTL qDTY 6.1 (RM587-510) and background markers loci located across chromosomes 2, 3, 5, 6 and 11. Two negative interactions were found between QTL qDTY 6.1 (RM587-510) and marker pairs RM115-RM275 and RM473-RM206 located on chromosomes 6 and 11 respectively while 3 interactions were found between background markers loci (RM300-RM475 on chromosome 2; RM411-RM319 on chromosome 3; RM592-RM437 on chromosome 5) with a common loci (RM587-510) associated with qDTY 6.1 (Table 4, Fig. 2a). The PVE% of interacting QTLs ranged from 12.04 to 19.95%. Low yielding NILs under MS (Fig. 2b) and high yielding NILs under SS (Fig. 2c) and MS (Fig. 2d) have not shown any significant interactions. Also, no significant interactions were observed under well-watered (NS) conditions for TDK1-Sub1 NILs including both high yielding and low yielding NILs (Fig. 2e). A heat map according to LOD score along the chromosomes reflecting the significant digenic interactions found among low yielding TDK1-Sub1 NILs under SS situations is presented in Supplementary Fig S5. In total, 6 epistatic interactions were identified among background markers loci under SS and MS in low yielding lines of Savitri background (Table 4, Fig. 3a,b). The interaction between introgressed QTLs (qDTY 3.2 , qDTY 12 (Fig. 3c,d). Also, no significant interactions were found under NS condition for Savitri NILs including both high yielding as well as low yielding NILs (Fig. 3e). A heat map according to LOD score along the chromosomes reflecting the significant digenic interactions found among low yielding Savitri NILs under SS and MS situations is presented in Supplementary Fig S6.

Discussion
Epistasis is a genetic phenomenon of interaction which may enhance or reduce the expression (depending on degree and direction) of interacting loci underlying QTLs associated with the complex trait 44,45 . QTL mapping and interaction studies in recent years in different crops suggested the possible presence of epistasis such as (1) interactions between introgressed QTLs (2) interactions between introgressed QTLs and 'background' (modifying) loci and (3)   www.nature.com/scientificreports www.nature.com/scientificreports/ through MAS is most desirable step for generation of drought-tolerant lines/varieties. In this study, contrasting performance of a pair of NILs possessed similar qDTY and qDTY combinations were analyzed to identify the possible epistatic interactions of introgressed qDTYs QTLs and background marker loci under different level of stress and non -stress conditions. Genetic backgrounds IR64, TDK1-Sub1 and Savitri, used in this study to develop NILs which produces higher grain yield under drought stress condition, widely cultivated and accepted by farmers in India, Lao PDR and Nepal, respectively. Foreground selection using the flanking markers associated with qDTYs QTLs ensured successful incorporation of the target QTLs in NILs of respective backgrounds (Supplementary Table S1). Marker-assisted background selection using uniformly distributed polymorphic markers have been extensively used to track recovery of background genome 19,47,48 . Rice drought breeding has witnessed some of the successful MAS introgression of qDTYs QTLs in elite backgrounds 7  www.nature.com/scientificreports www.nature.com/scientificreports/ moderate to high heritability for grain yield under drought stress for QTL identification 4,8 . In present study, 600 genome-wide SSR markers were used for estimating the similarity between backgrounds and their derived NILs. Promising NILs have captured 89-93% recovery of respective backgrounds (Table 3) while 44 NILs used in this study possessed 85-93% genome recovery. Phenotypic selection combined with marker-assisted background selection approach after qDTY introgression was followed, that could be one of the reasons not to achieve the expected recovery in some of the NILs. Similar concern was discussed earlier by Dixit et al. 49 in his study of marker-assisted breeding to combine submergence and drought tolerance in rice. However, in future drought NILs advancement, donor segment can be further minimized by using the recombinant selection markers and generation of whole genome profiling data through SNP-chips 19 of respective DTY QTLs. A similar approach was also explained by Jena et al. 25 in context to rice NILs developed for brown planthopper (BPH) resistance having low recovery (82-89.5%) of recurrent parent genome IR24 even after three back crosses.
Role of digenic interactions between DTY QTLs affecting the performance of one QTL over the other drought QTL has been well documented in various QTL mapping and marker-assisted drought breeding programme 12,42,43 . Positive interaction of qDTY 12.1 with qDTY 2.3 and qDTY 3.2 have been reported by Dixit et al. 42 which increases grain yield under drought significantly while Shamsudin et al. 12  Interaction between the target QTL/gene with the genetic background has been discussed earlier for various complex traits in rice 18,50,51 . However, there have been limited attempts to identify particular background markers loci interacting with introgressed QTLs and influencing the overall expression of target trait. There is an urgent need to identify such epistatic gene interactions which complicate the genotype-phenotype relationship of complex traits such as drought. In present study, we have detected negative interactions between drought QTLs www.nature.com/scientificreports www.nature.com/scientificreports/ (qDTY 2.2 , qDTY 2.3 and qDTY 3.2 ) with IR64 background markers loci located on chromosomes (2, 3, 4, 10 and 11), among background markers loci located on different chromosomes and between drought grain yield QTL (qDTY 1.2 ) with qDTY 2.3 , which might be one of the reason of poor performance of low yielding NILs carrying DTY QTLs (Table 1). The influence of epistatic drought QTL interactions explained PVE% ranging from 3.47% to 25.46% which could be very crucial in determining the grain yield under drought. Five digenic interactions were found in low yielding NILs of TDK1-Sub1 background, where qDTY 6.1 (RM587-RM510) had shown negative interaction with background marker located on five different chromosomes. Savitri background NILs had shown 6 epistatic interactions between background markers loci in low yielding lines. Many of such epistatic interactions are background specific. However, in our study, the high effect of many of the qDTY QTLs interaction with the genetic background marker loci indicate that these are biological interactions and not just statistical. The QTLs qDTY 2.2 , qDTY 2.3 , qDTY 3.2 and qDTY 6.1 have shown interactions with the background markers on the same chromosome in three other genetic backgrounds (data unpublished).Some earlier studies 32,52 highlighted the importance of epistatic QTL effect over the additive QTLs are influencing the expression of introgressed trait in crops like Arabidopsis and rice. Thomson et al. 53 reported a loss of Saltol QTL effect in developed NILs and suggested an interaction between Saltol and other background loci. Similarly, Babu et al. 54 observed the difference in level of seedling stage salt tolerance among Saltol introgressed advanced lines and assumed the existence of a possible interaction between Saltol QTL and genetic background but these reports did not report on the particular background markers loci interacting negatively with Saltol QTL, resulting in variable expression of Saltol introgressed lines. Revealing such epistatic interactions in present study, between introgressed QTL and background markers loci sheds more light in understanding the differences in phenotypic expression of MAS introgressed/pyramided lines for quantitative traits that can be useful in future MAS programs.

Conclusions
Performance of NILs carrying same major effect drought grain yield QTLs/QTLs combinations are strongly affected by interactions with the genetic background loci. This is the first report on in-depth analysis of loss of introgressed QTL effects in SS, MS due to negative interactions between drought QTLs and background markers loci, drought QTL-QTL interaction and among interacting background markers loci on various chromosomes in SS, MS in low yielding NILs but not under NS. For both low as well as high yielding NILs, no interaction was detected under NS. Absence of such negative interactions in high yielding NILs under MS, SS indicate future MAS program to carefully select against such interactions to increase grain yield under drought. Promising NILs in background of popular varieties (IR64, TDK1-Sub1 and Savitri) with absence of negative interactions and www.nature.com/scientificreports www.nature.com/scientificreports/ performing well in drought stress and non-stress conditions might be a very useful lines for use in future MAS/ variety to be released for drought-prone ecosystems.

Methods
Plant materials. A total of 44 drought NILs derived from three backgrounds (33 NILs of IR64, 5 NILs of TDK1-Sub1 and 6 NILs of Savitri) were used in this study (Supplementary Table S2). We initiated the process of NILs development with a total of 130 lines (70 lines from IR64, 32 lines from TDK1-Sub1 and 28 lines from Savitri) developed through marker-assisted breeding for the three genetic backgrounds for the present study. However, before including all the lines we attempted foreground selection with QTLs linked markers as well as background marker selection of all the lines with 600 genome wide background SSR markers and those lines which follows the minimum criteria of RPG recovery (>80%) were taken for the selection of NILs for further evaluation. As a result, finally we have selected a total of 44 drought NILs derived from three backgrounds (33 NILs of IR64, 5 NILs of TDK1-Sub1 and 6 NILs of Savitri) were used in this study. A total of eight sets of IR64 NILs (20 high yielding and 13 low yielding) possessing different QTL and QTL combinations was developed through crossing of IR64 with different drought tolerant donors contributing different grain yield QTLs. Four different drought tolerant donors were utilized for development of the selected NILs, for instance IR77298-14-1-2-17 (qDTY 2.2 and qDTY 4.1 ), IR 86918-B-315 (qDTY 1.1 and qDTY 1.2 ), Vandana (qDTY 2.3 and qDTY 3.2 ) and Way Rarem (qDTY 12.1 ) were crossed with IR64 and through marker assisted backcross breeding the above mentioned NILs were developed 7 . The pyramided version of NILs was developed through intercrossing of NILs of different gene combinations 7 . Two sets of TDK1-Sub1 NILs (2 high yielding and 3 low yielding NILs) possessing different QTL combinations were utilized in the present study. The pyramided version of NILs were developed from the cross between IR55419-04/2*TDK1. After confirming the presence of three QTLs (qDTY 3.1 , qDTY 6.1 and qDTY 6.2 ) in pyramided line, it was utilized for backcrossing with submergence tolerant parent TDK1-Sub1 to develop NILs in TDK1-Sub1 background 8,49 . Similarly, two sets of Savitri NILs (3 high yielding and 3 low yielding NILs) were developed from intercrossing of two BC 1 mapping population (IR 77298-5-6-18/2*Savitri IR 74371-46-1-1/2*Savitri) to develop lines carrying qDTY 3.2 and qDTY 12.1 QTLs in Savitri background 10,11,13 . All of the NILs used in the present study in all the three background were re-confirmed with earlier reported markers for the presence of QTL and QTL combinations. www.nature.com/scientificreports www.nature.com/scientificreports/ Evaluation of NILs under reproductive-stage drought stress and non-stress conditions. To study the grain yield variability of NILs pyramided with same QTL and QTL combinations, a total of 33 NILs derived from IR64, 5 NILs from TDK1-Sub1 and 6 NILs from Savitri backgrounds were evaluated during the dry seasons (DS) of 2015 and 2017 in a transplanted lowland ecosystem under reproductive-stage drought stress (RS) and non-stress (NS) conditions at the Ziegler Experiment Station of IRRI, Los Baños, Laguna, Philippines (14 °30′N longitude 121 °15′E latitude). Seedlings were raised in a wet-bed nursery for 21 days before transplanting into the main experimental field. The field lay out was generated using the statistical software Plant Breeding Tools (PBTools v1.4) developed at IRRI. The experimental design was alpha lattice in 2 replications in two-row plots of 5 m length with row to row and plant to plant spacing of 20 × 20 cm. Continuous supply of irrigation at 5 cm standing water was provided till the crop maturity in non-stress experiment. Fertilizer nitrogen, phosphorus, and potassium (NPK) were applied at recommended dose of 120:30:30 kg ha −1 . In RS trials, two doses of nitrogen fertilizer (basal and first split) were applied before initiating stress and a third dose was applied in adjustment with life-saving irrigation.
Reproductive stage drought stress experiment was conducted as described previously 4,6 . In brief for RS experiment normal irrigation was maintained in the field up to 30 days after transplanting, after which the water was drained out to initiate stress and continued till the harvesting. Water table depth was measured using PVC pipe with a plugged hole at the bottom and inserted in soil to 1 m depth and 15 cm above the soil surface in experimental stress field at regular interval. Life-saving irrigation was provided at level of severe stress when all the susceptible checks showed severe leaf rolling with minimum probability to recover upon watering and water table remained below 1 meter. Life-saving irrigation provided through flash flooding was drained out after 24 hrs to initiate further cycle of stress.
The stress imposed was classified from severe to moderate based on the relative yield reduction to non-stress (NS) experiments explained earlier by 42,55 . Experiments with yield reduction of <70% were characterized as severe stress (SS) and experiments with yield reduction of 31-70% were characterized as moderate stress (MS) as earlier reported by Kumar et al. 55 .
Observations recorded. Data on days to 50% flowering (DTF), plant height (PH), and grain yield (GY) were recorded in RS and NS experiments. DTF was observed as the number of days from seeding to the 50% the plants in a plot flower. The PH of three random plants from each NILs was measured before maturity from the soil surface to the tip of the main tiller and then averaged for mean analysis. Grains from individual plot were harvested at physiological maturity, oven-dried to a moisture content of 14% before weighing 6,9 . The grain yield in grams was further converted in kg ha −1 before entry means analysis in PBTools.  6.2 ) and Savitri backgrounds (qDTY 3.2 and qDTY 12.1 ) were confirmed in respective NILs using the earlier reported peak and flanking markers of respective grain yield drought QTLs 4,6-9 . Fresh and young leaves from 25 days old transplanted seedlings were collected in plot-wise bulk from each NIL-and their respective parents. Genomic DNA was extracted from collected leaves using a modified CTAB protocol developed by Murray and Thompson 56 , dissolved in 200 μl of TE (Tris-EDTA) buffer and stored at −20 °C. PCR amplification with the molecular markers was performed on a thermal cycler (G-Storm GS1, UK) at Genotyping service laboratory (GSL), IRRI, Philippines. Each PCR reaction mixture of 15 μl consisted of 10 ng of rice genomic DNA, 1 × PCR buffer, 100 μM dNTPs, 100 μM oligonucleotide primers and 1 unit of Taq polymerase. PCR products were resolved casting high-resolution 8% (v/v) polyacrylamide gel electrophoresis (PAGE) (CBS scientific, model MGV-202-33) and resolved in a 1x TBE buffer at 90 volts for 1-2 h, depending on the PCR product sizes. The separated DNA fragments after electrophoresis were stained with SYBER Safe ™ and visualized under UV trans-illuminator (AlphaImager ™ System).

NILs genotyping.
To study the genomic reconstitution of NILs with the similar QTL or QTLs combinations but showing variation in grain yield under drought, background genotyping was carried out on the selected NILs and parents. A total of 600 SSR markers equally distributed on all 12 chromosomes were selected from the Gramene database (http://www.gramene.org/) for identification of polymorphic markers between parents, which were further utilized for background selection in order to estimate recurrent parent genome (RPG) recovery.
Statistical analysis, recurrent parent genome recovery and epistatic interaction analysis. The phenotypic data collected from drought stress and non-stress experiments were statistically analyzed for the computation of trial means and standard error of difference (SED) using PBTools v1.4. Least significant difference (LSD) "at the 5% and 1% significant levels" were used to compare the means of the test entries and to infer the significant differences of the traits studied between parents and each NILs. Linear mixed model for analysis of variance was calculated with following equation: where, Yijk is measurement recorded in plot, µ is overall mean, Gi is effect of i th genotype, Rj is the effect of the j th replicate, BK (Rj) is the block effect of j th replicate and eijk is the error. The genotypes were considered as fixed and the replicates and block effects were random for estimating the entry means. The recurrent parent genome (RPG) recovery percentage of drought NILs of various background were estimated with the following formula 57 = + × N Recurrent parent genome (RPG) recoery % 2(B) (H) 2 100 www.nature.com/scientificreports www.nature.com/scientificreports/ Where, B=SSR marker loci homozygous for genetic background; H = no. of marker loci still in heterozygous; N = total no. of polymorphic SSR markers used for background estimation.
Graphical representation of background genome was constructed using software Graphical Genotype (GGT v 2.0) 58 (https://www.plantbreeding.wur.nl/). Epistatic (digenic) interactions between markers loci of all the NILs including high and low yielding NILs background wise were analyzed considering all the conditions (SS, MS and NS) in model using QTL IciMapping 59 ver. 4.0.1. In epistasis mapping method, a two-stage stepwise regression strategy was adopted to identify the most significant markers and marker-pair followed by two-dimensional scanning to identify significant digenic epistasis using adjusted phenotypic values based on best fitted multiple regression model 60 . In the first stage, the significant markers and marker pairs explaining epistatic and additive variations were selected in model and then in next stage the stepwise regression with stricter probability levels were applied to the residuals from the first stage to select significant marker pairs and estimate their effects in model 50. The stricter probability level was applied in the second stage to avoid over-fitting due to the large number of regression variables. Extensive simulations have shown epistatic QTLs and QTL epistatic networks can be identified efficiently by ICIM mapping even in case two interacting QTL have some additive effects 60 . The threshold LOD value was determined by a permutation test involving 1000 runs at a significance level of P = 0.01 to detect significant digenic interactions between marker loci. In mapping parameters, window size and walk speed used for the genome scan was 10 cM and 1 cM respectively.