Developing Climate-Resilient, Direct-Seeded, Adapted Multiple-Stress-Tolerant Rice Applying Genomics-Assisted Breeding

There is an urgent need to breed dry direct-seeded adapted rice varieties in order to address the emerging scenario of water–labor shortage. The aim of this study was to develop high-yielding, direct-seeded adapted varieties utilizing biparental to multiparental crosses involving as many as six different parents in conventional breeding programs and 12 parents in genomics-assisted breeding programs. The rigorous single plant selections were followed from the F2 generation onwards utilizing phenotypic selection and quantitative trait locus (QTL)/gene-based/linked markers for tracking the presence of desirable alleles of targeted QTL/genes. In conventional breeding, multiparent lines had significantly higher yields (2,072–6,569 kg ha−1) than the biparental lines (1,493–6,326 kg ha−1). GAB lines derived from multiparent crosses had significantly higher (3,293–6,719 kg ha−1) yields than the multiparent lines from conventional breeding (2,072–6,569 kg ha−1). Eleven promising lines from genomics-assisted breeding carrying 7–11 QTL/genes and eight lines from conventional breeding with grain-yield improvement from 727 to 1,705 kg ha−1 and 68 to 902 kg ha−1, respectively, over the best check were selected. The developed lines may be released as varieties/parental lines to develop better rice varieties for direct-seeded situations or as novel breeding material to study genetic interactions.


INTRODUCTION
Rice is mainly cultivated under anaerobic conditions and primarily adapted and evolved under these conditions (Kumar and Ladha, 2011). A shortage of water and labor input makes rice production through transplanted rice (TPR) more expensive, less profitable, and unsustainable (Farooq et al., 2011). The estimates specify that the water requirement for irrigation by 2025 could be approximately 561 km 3 for a low-demand scenario and 611 km 3 for a high-demand scenario (Kumar et al., 2005). The demand for water in various non-agricultural sectors in the country reduces its availability for agriculture; as a result, the agricultural sector is likely to face another 10-15% reduction in irrigation water by 2025 (Boretti and Rosa, 2019). With 90% of the world's rice grown and consumed in Asia and the large water requirements of conventional transplanted rice cultivation systems, it is evident that this water scarcity will severely affect rice production in this continent. The water scarcity situation will lead to rice cultivation with less water (Wang et al., 2002).
Cultivation of dry direct-seeded rice (DDSR) helps in avoiding three basic operations: puddling, transplanting, and maintaining standing water. DDSR cultivation in saturated soil has been widely adopted in southern Brazil, Chile, Cuba, Venezuela, some of the Caribbean countries, and in certain areas of Colombia (Fischer and Antigua, 1996). Currently, DDSR is becoming more popular in South Asia, South East Asia, and, to some extent, in West Africa. The cultivation of DDSR is being practiced with several modifications of tillage or land preparation and crop establishment (CE) with a site-specific package, but has not gained the required popularity because of one or two unsolved deficiencies. In addition to higher economic returns, DDSR crops vis-à-vis the transplanted (TPR) system are faster and easier to plant, have a shorter duration, are less labor intensive, and consume less water (Bhushan et al., 2007). The mechanized DDSR cultivation system has been estimated to provide potential irrigation water savings of 40 cm ha −1 , labor savings of 25 persondays ha −1 , energy savings of 1,500 MJ ha −1 , a reduction of GHG (green house gas) emissions of 1,500 kg CO 2 equivalent ha −1 , and a yield increase of 0.5 t ha −1 with an increased net economic return of USD 50 ha −1 in most rice-growing countries (Kumar and Ladha, 2011).
Existing rice varieties are not specifically developed for direct-seeded ecosystems. However, some earlier developed rice varieties for transplanted conditions have been adapted to directseeded conditions exhibiting a yield decline under direct-seeded conditions. The newly developed drought-submergence-tolerant, rainfed-adapted rice varieties have shown good performance in India and Nepal, providing a yield advantage of 1.0 t ha −1 over the currently grown drought-submergence susceptible varieties (Sandhu et al., 2019a). Varieties such as Sahbhagi dhan, DRR 42, DRR 44, CR dhan 201, CR dhan 203, and Swarna Shreya  developed by the International Rice Research Institute (IRRI) and genomics-assisted derived CR dhan 801 varieties have become highly popular in India and are reported to be cultivated on at least an area of 0.8 m ha. Similarly, in Nepal, the rice varieties Sukha dhan 3 and Sukha dhan 5 , and the genomics-assisted derived lines Bahuguni dhan-1 and Bahuguni dhan-2 (Sandhu et al., 2019a) have become highly popular and contributed to the increased rice production in Nepal. Based on these achievements, the development of direct-seeded adapted rice varieties could provide opportunities for a significant increase in rice productivity in rice-growing countries.
Breeding methods such as QTL/gene pyramiding and multiparent application have been reported to be effective in developing resistant/tolerant varieties against biotic and abiotic stresses (Koide et al., 2009;Bandillo et al., 2013;Sandhu et al., 2019b). Conventional breeding is an efficient approach for the development of novel genetic variants (Breseghello and Coelho, 2013). The conventional breeding approach suffers from the problem of linkage drag, leading to the transfer of undesired traits closely linked with traits of interest. The use of molecular markers in breeding programs improves the efficiency of traditional breeding by enabling breeders to select trait-linked molecular markers (Collard and Mackill, 2008). The existence of abundant genetic diversity, genomic variation in the rice gene pool, and availability of modern genomic tools/techniques provides opportunities to choose suitable donors that are free from undesirable linkage and to dissect the quantitative nature of associated traits of interest. Most of the rice landraces perform better under various nutrient-deficient conditions (Wissuwa and Ae, 2001). Screening of diverse genotypes under direct-seeded conditions by assessing various biometrical traits, such as root architecture, plant height, and yield-related traits, will help to identify genotypes with improved yield, better root traits, higher nutrient uptake, higher nutrient utilization efficiency, and better plant type traits.
However, identification of donors, traits, and QTL for traits increasing rice yield under direct-seeded conditions and their introgression in elite genetic backgrounds using advances in genomics-assisted breeding (GAB) present a unique scenario to achieve significant yield advantages under direct-seeded conditions. The present study was undertaken to develop new direct-seeded adapted rice varieties utilizing the identified donors, genes/QTL, and their introgression in elite genetic backgrounds, compare the performance of bi-parental and multi-parental developed lines under dry direct-seeded conditions, and compare the performance of multiparent developed lines through conventional breeding and GAB approaches.

MATERIALS AND METHODS
The study on the development of direct-seeded adapted rice lines through conventional and GAB approaches began at the International Rice Research Institute (IRRI), Los Baños, Laguna (Philippines) in 2009, with the identification of suitable traits and genomic regions associated with the traits improving grain yield and adaptability of rice under dry direct-seeded cultivation conditions.

Selection of Parents for the Conventional Breeding Program
The parents used in the conventional breeding program included widely grown high-yielding rice varieties under the transplanted system, drought-tolerant lines adapted to upland and shallow lowland conditions, and direct-seeded adapted breeding lines possessing traits required to increase rice adaptability to directseeded conditions. The conventional breeding program involved two, three, four, five, and six parent crosses. Detailed information on breeding lines/varieties used and the traits they possessed is presented in Table 1.

Conventional Breeding Program
The conventional breeding hybridization program to develop suitable adapted breeding lines involving the parents presented in Table 1 was initiated in the dry season (DS) of 2009. New crosses were included in each successive season. The F 1 s were selfpollinated and phenotypic selection of F 2 s involving selection for bacterial late blight, blast, plant height, and visual grain yield was conducted under direct-seeded conditions. The best plants with enhanced plant type, disease resistance, grain type, and grain yield were carried forward in successive generations until the F 4 generation. The plant-to-progeny testing of selected F 4 plants was attempted in the F 5 and F 6 generation followed by an observational yield trial (OYT) and advanced yield trial (AYT) (Figure 1). The selection of disease resistance (blast and bacterial late blight) and grain quality was conducted in AYTs. The disease screening protocol followed was as described by Sandhu et al. (2018). The advanced breeding lines were analyzed for grain quality parameters at the IRRI GQNC (Grain Quality and Nutrition Center) laboratory facility. The detailed information on the grain quality testing is presented in Yadav et al. (2020). In the F 5 and F 6 generations, the progenies were tested in replicated yield trials with a plot size of 1.6 m 2 , whereas the plot sizes in OYT and AYT were 3.2 and 4.8 m 2 , respectively. In each generation, the plants were grown maintaining 20 cm (hill to hill) × 20 cm (row to row) distance. Detailed information on soil characteristics of upland fields, land preparation, field management, and phenotypic screening was provided by Sandhu et al. (2019b).

Genomics-Assisted Introgression Program
To combine anaerobic germination, blast resistance, brown plant hopper resistance, bacterial blight resistance, gall midge resistance, grain yield under direct-seeded cultivation conditions, early vigor, nodal roots, and early and uniform emergence in the background of high-yielding irrigated rice cultivar IR09N538 (IRRI 132/PR 30138-35-2//IR04N114) with preferable grain type, a complex crossing program began in the 2014 DS with 12 donors, including donors for biotic and abiotic stress tolerance and DSR-adapted traits. Based on the synchronization of flowering and availability of a true F 1 generation with desired trait/QTL combinations, different cross combinations were attempted to achieve success in the breeding program. A detailed schematic representation of the pre-breeding steps and the genomics-assisted breeding strategy followed for the introgression of QTL/genes to develop a multi-stress-tolerant, direct-seeded, adapted superior rice genotype suitable for DDSR cultivation is presented in Figure 1.

Genotyping
In the genomics-assisted introgression program, the F 1 s from biparental and multiparental crosses were tested in each season to check for true F 1 s using rice microsatellite (SSR; simple sequence repeats) markers. The SSR markers were identified in the previously reported QTL associated with traits, such as early uniform germination, early vegetative vigor, nodal roots, root hair length, root hair density, grain yield under directseeded conditions, and drought stress tolerance. A total of 1420 SSR markers in the QTL regions were used for parental polymorphism. Eighty-one SSR markers, one indel, and eight gene-specific polymorphic markers were used to screen the F 1 s  Improved breeding lines B 1050 D-KN-1-1-1-1-3, HHZ 8-SAL 6-SAL 3-Y2, IR04N114, IR08L152,  IR08L181, IR08N194, IR09L179, IR09L204, IR09L303, IR09L336,  IR09L337, IR11L269, IR11L101, IR11L184, IR 111249, IR 3, IR05N412, IR06N119, IR06N155, IR08N121,  IR08L216, IR09L120, IR09L179, IR09L204, IR09A228, IR09L303,  IR09L336, IR09L337, IR10N108, IR10N237, IR 67966-44 Table 3. Fresh, young leaves from the 15-day-old rice seedlings were collected from each F 1 plant and the respective parents. Genomic DNA was extracted from F 1 s using a modified CTAB protocol (Murray and Thompson, 1980). The DNA from the phenotypically selected plants from generations F 2 onwards was used for further studies. The amplification of genomic region using PCR (polymerase chain reaction), separation of bands on agarose, and 6-8% (v/v) PAGE (polyacrylamide gel electrophoresis (CBS scientific model MGV-202-33); depending on the SSR, indel, and gene-specific marker product size was performed. After staining with SYBR Safe TM , the separated genomic fragments were visualized under a UV trans-illuminator (AlphaImager TM System). The associations of the polymorphic SSRs, indel, and gene-specific markers with the introgressed FIGURE 1 | Detailed schematic representation of the pre-breeding steps, use of genomics-assisted breeding following funnel mating design to assemble the targated QTL/genes to develop multi-stress-tolerant homozygous breeding lines suitable for direct-seeded cultivation conditions. genomic regions were exploited for the accurate detection of the F 1 s and the progenies. Because there are many introgressed genomic regions/genes and many are not fine mapped, a sequential two-stage genotyping approach was used to make the genotyping labor and cost effective. The first stage selection was performed on phenotypically selected plants using the peak markers associated with the introgressed genomic region. The second-stage selection was performed on the selected plants harboring the selected donor allele using flanking and all other markers underlying the QTL regions.

Phenotyping
To combine traits of true F 1 s to increase rice adaptation under direct-seeded cultivation conditions, -early uniform emergence, early vigor, high-nutrient uptake through improved root traits, tolerance to drought, and grain yield under conditions together with resistance to biotic stresses were selected using trait-linked markers. The number of F 1 s developed per cross per season is listed in Supplementary Table 1. At each generation, phenotypic plant selection was conducted initially based on plant type, duration, plant height, number of tillers, grain type, and visual yield, and then selected plants were evaluated genotypically with the trait associated markers. A single plant selection strategy was followed from the F 2 to F 6 generation for the presence of QTL/gene combinations together with phenotypic selection for yield to reject plants showing negative interactions. A plot size of 8 m 2 was used for the screening of each breeding line at the F 4 , F 5 , and F 6 generation. At each advancing generation step, the plants were grown maintaining 20 cm (hill to hill) × 20 cm (row to row) distance. Field management was conducted following the procedure described by Sandhu et al. (2019b). The data on plant height (cm) and grain yield of single plant selection (g) and plot yield (kg ha −1 ) were collected. Plant height was measured from the root-shoot junction to the tip of the uppermost panicle on the main tiller. After harvesting at maturity, the grains were first dried to 14% moisture, and then weighed to record GY (g, kg ha −1 ).

Statistical Analysis
The agronomic data collected from all the conventional and GAB experiments were analyzed using statistical tools and software. The experimental means and standard error of difference of the progeny testing experiments were calculated using IRRI PBTools v1.4. The least significant difference (LSD) at P = 1 and 5% levels of significance were used to compare the means of test entries and to estimate the significant variations existing between parents and the breeding lines for the particular trait of interest. An ANOVA (analysis of variance) was estimated using following mixed linear model: where µ, Gi, Rj, BK(Rj), and eijk are the overall mean, effect of ith breeding line, effect of jth replicate, block effect of the jth replicate, and error, respectively. While estimating the entry mean, the genotypes and replication block effects were kept as fixed and random, respectively.

Mean Comparison of Different Parental Classes
The hypothesis regarding mean differences among the breeding lines developed involving different numbers of parental lines where µ, rk, b(r)kl, qi, g(q)ij, and eijkl symbolize the population mean, effect of the kth replicate, effect of the lth block within the kth replicate, effect of the ith parental class, effect of the jth breeding line nested within the ith parental class, and the error, respectively (Knapp, 2001). The effects of parental class and the breeding lines within the parental class were considered as fixed effects, and the replicates and blocks effects within replicates were set as random.

Conventional Breeding Program
A total of 184 crosses involving 137 biparent crosses, 21 triparent crosses, 16 quadraparent, 5 pentaparent, and 5 hexaparent crosses were attempted and evaluated from the F 2 generation for advanced yield trials. A highly stringent phenotypic selection based on plant type, duration, plant height, number of tillers, grain type, and visual yield across generations was made. The stringency of selection can be estimated from the number of plants/breeding lines selected across generation advancement, as represented in Supplementary Table 2. The crosses involving three or more parents showed grain yield improvement over the biparental lines and the upland adapted check varieties ( Table 4). The high mean grain performance of the parental class involving more than two parents in the F 4 generation was consistent in the F 5 and F 6 generations and also in observational and advanced yield trials across seasons. The highest average grain yield was observed in the hexaparent class across generations and seasons ( Table 4). The average grain yield of the biparent class ranged from 3,517 to 4,589 kg ha −1 in the WS (wet season) and from 5,010 to 5,796 kg ha −1 in the DS across generations. The parental class involving  three, four, and five parents showed average grain yields ranging from 4,434 to 5,904 kg ha −1 , 4,532 to 6,002 kg ha −1 , and 4,566 to 6,124 kg ha −1 , respectively, across generations. The grain yield of the hexaparent class ranged from 4,680 to 5,603 kg ha −1 in the WS and from 5,505 to 6,292 kg ha −1 in the DS across generations. Across season and generation advancement, the grain yield advantage of the biparent class ranged from 100 to 540 kg ha −1 , triparent class from 35 to 1,008 kg ha −1 , and quadraparent class from 352 to 1,095 kg ha −1 over the check varieties. The parental classes involving five and six parents consistently outperformed the check varieties in terms of average grain yield advantage ( Table 4). The multi-parent lines had significantly higher yield (2,072-6,569 kg ha −1 ) than the bi-parental lines (1,493-6,326 kg ha −1 ) at the F 6 generation in the DS.   The breeding lines developed involving more than two parents could ensure 0.3-1.5 kg ha −1 improvement in grain yield over the check varieties and 0.1-1.3 kg ha −1 grain yield improvement over the breeding lines involving two parents. DTF of the different classes were comparable to the check varieties, except in some seasons where 1 or 2 d earliness was observed ( Table 5). Most of the classes showed comparable or slightly higher PHT than did the check varieties (Table 5).

Genomics-Assisted Breeding Program
The GAB program involving 12 donors and IR09N538 as the recipient was initiated in the 2014 DS. The final multiparent F 1 s were generated in the 2016DS Figure 2. A single plant selection strategy was followed across generations. The multiparental 4,200 F 1 s were grown in an F 1 nursery at IRRI under puddled transplantation conditions in the 2016WS. The number of QTL/genes varied from two to 15 and grain yield ranged from 1.0 to 58.8 g/plant. Thirty-five F 2 s derived F 1 s were selected based on different QTL/gene combinations. The number of QTL/genes in the selected plants ranged from 8 to 15 and the grain yield varied from 8.0 to 58.8 g/plant. An F 2 population of approximately 40,000 plants, F 3 population of 13,780 plants, F 4 population of 144 breeding lines, F 5 population of 243 breeding lines, and F 6 population of 70 breeding lines were raised and screened under DDSR cultivation conditions. A total of 13,468 F 2 plants of required plant type, duration, plant height, number of tillers, grain type, and visual yield were screened genotypically with the trait-associated markers, and a total of 153 F 2 derived F 3 plants were selected. The number of QTL/genes in the selected plants varied from 3 to 12, plant height from 80 to 130 cm, and grain yield from 5.9 to 68 g/plant. The 1,177 F 3 derived F 4 plants and 1,411 F 4 derived F 5 plants were selected and advanced by combining phenotypic genotypic selection in the 2017WS and the 2018DS, respectively. More than 30 genomicsassisted derived F 5 breeding lines with different combinations of QTL/genes yielded more than check varieties MTU1010, IRRI 155, UPLRi7, and the recipient parent, IR09N538. F 4 derived F 5 breeding lines with 223 different combinations of QTL/genes were evaluated in the 2018WS. Significant variability in grain yield, ranging from 1,026 to 5,956 kg ha −1 , was observed (data not shown). After genotypic selection at the F 5 level, a total of 427 single plants were selected phenotypically, and then 70 plants to progeny were carried forward to the F 6 generation. The grain yield variability in the selected F 6 breeding lines varied from 3,293 to 6,719 kg ha −1 , plant height from 65 to 115 cm, and days to 50% flowering from 77 to 104 d (data not shown). The QTL/genes in selected F 6 breeding lines varied from 6 to 12, and a total of 55 different combinations of QTL/genes with acceptable phenotype and grain characteristics were selected.

Comparison of Conventional and Genomics-Assisted Derived Breeding Lines
The analysis of the differences in mean grain yield of genomicsassisted derived breeding lines was significantly higher than that of breeding lines derived from conventional breeding programs FIGURE 2 | Detailed schematic representation of crossing strategy followed for the introgression of QTL/genes from 12 donors in the background of IR09N538 and the number of plants selected across successive generations. #C: represents the cross. Table 6. The grain yield of GAB lines derived from multiparent crosses ranged from 3,293 to 6,719 kg ha −1 ; however, the grain yield of multi-parent lines from conventional breeding ranged from 2,072 to 6,569 kg ha −1 at the F 6 generation in the DS. The GAB lines IR 129477-2815-41-4-7-1, IR 129477-3343-500-36-5-1, IR 129477-4139-439-1-1-2, and IR 129477-1629-14-1-2-2 with eight, nine, nine, and 11 QTL/genes, respectively, showed grain yield of more than 5,500 kg ha −1 in the WS at the F 5 generation advancement stage and more than 6,000 kg ha −1 in the DS at the F 6 generation Table 7. The breeding lines IR 115844-B-B-475-1-2, IR 107891-B-B-1253-1-1, IR 115845-B-B-296-2-1, and IR 115846-B-B-197-1-1 from four different crosses of the conventional breeding program involving five/six parents yielded more than 4,500 kg ha −1 in the WS at the F 5 generation advancement stage and more than 5,500 kg ha −1 in the DS at the F 6 generation ( Table 7). The genomics-assisted derived breeding lines showed better yield performance (5,100-7,004 kg ha −1 ) when tested at Varanasi, India, under observational yield trials ( Table 7). The breeding lines from the conventional breeding program showed delayed flowering compared to the genomics-assisted breeding lines, except the lines derived from the cross Dular (acc. 32561)/IRRI 148///IRRI 154/UPLRi 7//IR87707-446-B-B-B/Kali Aus ( Table 7). The genomics-assisted derived breeding lines were more dwarf than the breeding lines developed from the conventional breeding program (Table 7) Table 7).

Selection of Promising Lines/Donors
The breeding lines carrying multiple QTL/genes from GAB programs and multiparent lines from conventional breeding programs performed well in terms of grain yield and adaptability under DDSR and also exhibited phenotypically high levels of tolerance/resistance to abiotic and biotic stresses with desired grain quality characteristics. A total of 11 promising lines carrying 7-11 QTL/genes for various DDSR adaptable traits, as well as biotic stresses tolerance/resistance (Table 7,  Figures 3a-c), and eight promising lines from the multiparental conventional breeding program were selected (Table 8, Figures 3d,e) for further evaluation in multiple environments for varietal release. The recipient variety IR 09N538 is presented in Figure 3f and the upland adapted check varieties, UPLRi7 and Katihan 1 are presented in Figures 3g,h, respectively. It is interesting to note that the genomics-assisted breeding lines with a combination of DDSR adaptable traits and abiotic stress tolerance/resistance QTL/genes showed better yield advantages, viz. IR 129477-2815-41-4-7-1 (Xa4 + BPH3 + Pita + Pita2 + qAG 9.1 + qDTY 3.1 + qGY 6.1 + qGY 10.1 ) showed ∼33% and IR 129477-1629-14-1-2-2 (Xa4 + xa5 + Xa21 + BPH3 + Pita + qAG 9.1 + Pi9 + qDTY 3.1 + qNR 5.1 + TABLE 6 | Analysis of the differences for grain yield (kg ha −1 ) at the F 6 generation in the dry season between the categories of breeding lines (with a confidence interval of 95%) developed following genomic assisted breeding and conventional breeding approaches.   Table 8). Most of the selected breeding lines possessed less chalkiness, medium amylose content, and intermediate GT content (Table 8). Lines with chalkiness >10% and amylose content <10% were not carried forward in the breeding program. The promising breeding lines from both the conventional and GAB programs can be further tested for adaptability and yield stability by conducting multi-location trials in their targeted environments. These pyramided lines may serve as novel donors for the development of lines with multiple abiotic and biotic stress tolerance/resistance traits.

DISCUSSION
To meet the gaps between rice crop yields and global rice consumption, rice breeders need to continuously release new rice varieties with better yield potential, adaptability under DDSR, acceptable grain quality traits, high nutrient-use efficiency, and resistance to various biotic/abiotic stresses (Leng et al., 2017). The longstanding idea behind using an insecticide mixture to broaden the insect resistance spectrum laid the foundation of gene pyramiding (Ye and Smith, 2008). The nature of donors/traits selected for introgression, number of genes transferred, distance between the introgressed genes and flanking markers, and the number of plants selected with different gene combinations across each breeding generation, are critical for an effective QTL/genes pyramiding program.
To the best of our knowledge, the present study is the first to report the pyramiding of 7-11 QTL/genes associated with both traits, providing better adaptability and biotic stress tolerance/resistance under DSSR, exploiting genomics-assisted foreground selection. Previous studies reported the markerassisted pyramiding of QTL for either tolerance to abiotic stress traits, such as drought and submergence Septiningsih et al., 2015;Shamsudin et al., 2016;Dixit et al., 2017;Sandhu et al., 2019a), or resistance to blast Fukuoka et al., 2015), bacterial blight (Suh et al., 2013;Pradhan et al., 2015;Das et al., 2018), brown plant hoppers (Wang et al., 2015;Jena et al., 2017), and gall midges (Divya et al., 2015) in rice.
Various high-yielding rice varieties from the conventional breeding program involving phenotypic selection  and genomics-assisted selection (Sandhu et al., 2019b) have been released. However, labor and time use, linkage drag, and low efficiency are some of the major barriers impeding conventional breeding (Prohens, 2011). Advances in molecular mapping, precise phenotyping, novel marker development, and multi-parent population development have paved the way for genomics-assisted breeding. These advances have gradually shifted the focus of traditional plant breeding from phenotypebased selection to genotype-based selection (Xu and Crouch, 2008). The GAB strategies involve backcrossing or introgression of QTL/genes, enrichment of multiple favorable alleles, and selection for introgressed traits/QTL/genes using trait-linked markers (Hospital et al., 1992;Eathington et al., 2007;Gupta et al., 2010).
A GAB program combining 25 QTL/genes and a conventional breeding program involving multi-parent cross-governing better adaptation of rice to direct-seeded cultivation conditions was undertaken at IRRI. The final step of the introgression program was to fix the targeted QTL/genes into a homozygous state across successive generations. Generally, recombinant inbred lines (RILs) and double haploid (DH) production techniques are used for the development of homozygous lines. We did not succeed in combining all 25 QTL/genes in one breeding line because: (i) some with large introgressed regions were lost during successive recombination events, (ii) some QTL/genes combination altered the flowering cycle and led to the failure of planned crossing strategy in successive seasons because of  non-synchronization of flowering, (iii) some of the complex F 1 s crosses with possible negative interactions among the introgressed QTL/genes were rejected and not utilized for further crossing, and (iv) some of the QTL that were in a heterozygous state were lost in succeeding generations. We succeeded in pyramiding 15 QTL/genes in the F 2 /F 3 generation in the background of IR09N538 but some QTL/genes, which were in a heterozygous state, were lost in succeeding generations. In this regard, DH production has been reported to provide satisfactory results in becoming fully homozygous in a very short time (Dunwell, 2010;Mishra and Rao, 2016), which was not used in the present study. Additionally, genetic engineering (GE) also aids the precise transfer of genes of interest to crop plants (Lemaux, 2008) to generate crops with desired trait/s. Although DH and GE are precise, universal, and fast methods to transfer the desired gene/s into different crop plants (Nicholl, 2008), it will not replace GAB, but it will definitely add to the efficiency of rice crop improvement. Some QTL/genes combination that showed negative interactions leading to poor grain yield and adaptability under DDSR with unacceptable plant and grain type were rejected in successive generations. Finally, the pyramided lines possessing 7-11 QTL/genes under homozygous conditions were selected, indicating the existence of positive interactions among different biotic and abiotic stress-related QTL/genes. The DDSR lines developed through conventional and GAB strategies were evaluated at an advanced stage under dry direct-seeding conditions. The increased yield of breeding lines developed under conventional breeding programs involving multi-parents performed over the bi-parental lines across generations could result from the accumulation of alleles for traits that increase rice adaptability to direct-seeded situations (Huang et al., 2012). It can be predicted that, with the use of a greater number of parents, there will be greater identification of the complex traits (Huang et al., 2012), a greater number of recombinations, and greater chances of accumulation of favorable alleles. Nevertheless, multi-parent crosses require more time and higher costs for the development of the population, but increased genetic variations may provide opportunities to improve grain yield and adaptability under DDSR. In the present study, we had succeeded in breaking the unfavorable linkages, FIGURE 3 | Selected promising breeding lines from (a-c) genomics-assisted breeding program (d,e), conventional breeding program, and (f) recipient variety used in genomics-assisted breeding program (g,h) for upland adapted check varieties. elimination of the inferior plants with poor plant and grain type during the selection process. The better grain yield performance of GAB-derived multi-parental breeding lines compared to the breeding lines developed from conventional breeding programs could be the result of precise marker-based selection that helps reduce undesirable linkage drags. GAB aids in obtaining the desirable QTL/genes combination without any unwanted genes, minimizing the linkage drag around the target QTL/genes. GAB has been proven effective for transferring QTL/genes from pyramided lines into improved varieties and new plants (Magar et al., 2014).
A stepwise hybridization and selection strategy in the GAB program led to the selection of the most suitable plants with desired QTL/gene combinations. Arbelaez et al. (2019) reported that an average of 800 F 2 individuals plants were required to obtain 50 lines with the desired genotype fixed at two loci. Compared to the F 2 generation, the frequency of desired homozygotes for two linked genomic loci will decrease nearly three times in the F 6 generation (Arbelaez et al., 2019). Keeping the above points in view, a large population size of 40,000 (F 2 ) and 13,780 (F 3 ) plants capturing the hidden genetic variations were maintained in the early generations to select appropriate plants possessing positive interactions among targeted QTL/gene combinations, with desired plant type, grain type, better grain yield, and adaptability under DDSR. The increased yield of the GAB-derived lines from multiparent crosses has been achieved, keeping the plant height and days to 50% flowering, similar to check varieties, also indicates yield increase as a result of accumulation of favorable alleles, increasing rice adaptability to direct-seeded situations caused by introgression of genes/QTL for such traits. Notably, stepwise phenotypic and genotypic selection in the early generations has reduced the genotypic efforts, finally reducing the genotyping cost. The effectiveness of the above-mentioned strategy targeting the pyramiding of various drought QTL has been successfully demonstrated in various genetic backgrounds by Kumar et al. (2018). Most of the GAB-improved lines developed in the present study carry the xa5 gene in combination with xa13 and Xa21 genes. Based on previous reports, xa5 (Jiang et al., 2006;Huang et al., 2016), xa13 (Ogawa et al., 1987;Chu et al., 2006), and Xa21 (Huang et al., 1997;Cao et al., 2005) genes are considered as the most effective R genes, providing broader levels of resistance, and have been widely used for introgression programs against BLB in most rice growing countries in Asia. The selected promising lines carrying either the Pita, Pita2, or Pi9 gene alone showed stable resistance to blast disease. Similarly, Qu et al. (2006) reported that the Pi9 gene can provide broader and durable resistance against rice blast disease. Additionally, the selected breeding lines possessed a genomic region providing resistance to gall midge, tolerance to reproductive stage drought stress, adaptability, and improved yield under DDSR. The selected improved breeding lines carrying the genomic region contributing to biotic stress resistance will further be evaluated phenotypically. The better performance of breeding lines under transplanted control nonstress and reproductive stage drought stress conditions (Table 8), in addition to DDSR, indicated their phenotypic plasticity. Enhanced seedling establishment, improved and stable yield, and phenotypic plasticity of the root system have been reported to result in improved adaptability of selected promising lines across variable growing conditions . Grain quality parameters play a crucial role in varietal release and wider acceptance of the released varieties (Rani et al., 2006). Thus, our field selection criteria also involved selection for grain type across each successive generation. The breeding lines from the conventional and GAB programs were selected considering the preferences of various countries in South and South East Asia. The developed breeding lines possessing medium amylose content are suitable for India and Nepal, those with higher amylose content for Bangladesh, and those with lower amylose content for South East Asia.

CONCLUSIONS
The study showed GAB was a better strategy for improving yield and adaptation of rice under DDSR utilizing multiple QTL/genes compared to conventional breeding. A higher number of parents and more recombination events resulted in higher grain yield advantage compared to the classical bi-parental population. These findings are vital for researchers working on multi-parental populations, because this will remarkably reduce the efforts of developing large numbers of bi-parental populations. The rigorous selection from the F 2 generation onwards resulted in the development of promising breeding lines carrying 7-11 QTL/genes for different traits. Promising GAB lines carrying multiple QTL/genes under homozygous conditions with preferable grain type and quality traits can address the constraints faced by farmers caused by both biotic and abiotic stresses simultaneously under DDSR in addition to the labor-water shortage. The GAB-derived breeding lines and multi-parent conventional breeding lines could further be recommended for varietal release after multi-location evaluation in national and provincial coordinated trials in different countries. These breeding lines may serve as a novel genetic resource that could aid breeders and molecular biologists to further conduct genetic and interaction studies for a wide range of breederrelevant parameters.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.