The materials used in this study included nine BC₂F₂ populations derived from crosses between two elite Xian (indica) recipients, (IR64 from Philippines and Teqing from China), and seven diverse donors from six countries ( Supplementary Table S1 ). The donors included Shwe-Thwe-Yin-Hyv (STYH) from Myanmar, BR24 from Bangladesh, Type3 (a Geng or japonica variety) from India, Cisanggarung (Cis, a lowland Xian variety) from Indonesia, Binam (a Basmati landrace) from Iran, and OM1723 (a lowland Xian variety) from Vietnam and FR13A (the best ST Xian landrace) from India. The procedure of the BC breeding was described previously (Ali et al., 2006). Briefly, the two recipients were crossed with all donors to produce the F₁s. The F₁s were backcrossed with the recipients to produce the BC₁F₁s. Approximately 25 plants from each BC₁F₁ line were backcrossed with each recipient to produce ~25 BC₂F₁ lines. From each of the crosses, 25 BC₂F₁ lines were planted in the following season and seeds all 25 BC₂F₁ lines of each cross were bulk-harvested to form a single bulk BC₂F₂ population.

In the dry season (DS) of 2004, seeds from each BC₂F₂ population were sown on the seedling nursery with the parental lines and 208 25-day old seedlings from each population and the recipients were transplanted into a paddy field at the International Rice Research Institute (IRRI) experimental farm in the Philippines. Normal irrigation was applied until the peak tillering stage (30 days after transplanting) and then water was withheld until maturity, subjecting all plants to severe terminal drought at the reproductive stage. In addition, the bulk harvested seeds of each 9 BC₂F₂ population were directly sown into a five-row plot (with ~250 plants/plot) flanked by two rows of the recipients in the upland (aerobic) conditions in the IRRI upland facility, and then the fields was irrigated with flash water to allow seeds to germinate until the five-leaf age and then irrigated was withheld, subjecting all plants to chronic water stress from the seedling stage to maturity. Under both the lowland and upland drought stresses, the recipients, IR64 and Teqing were unable to produce any seeds, while some plants in each of the BC₂F₂ populations survived and were able to produce seeds, which were individually selected to produce respective BC₂F₃ lines.

All BC₂F₃ introgression lines (ILs) selected from the DT screening were progeny tested in the replicated experiments in the upland (stress) and normal irrigated lowland conditions in the coming wet season (WS) and dry season (DS) of 2005-2006 at IRRI. In the progeny testing, seeds of all selected BC₂F₃ ILs were sown in the seedling nursery in late May and 25-day old seedlings of each BC₂F₃ IL or their parents were transplanted into a three-row plot with 33 plants per plot at a spacing of 20 × 25 cm in the field. The plots were arranged with a complete randomized block design with three replications for each IL plus insertions of the recipient plots (IR64 and Teqing) in every 20 plots as checks in the field experiment. The same experimental design was used in the normal irrigated control. All selected ILs were shown to have improved DT based on the progeny testing (Lafitte et al., 2006).

In the following WS of 2005, all drought selected BC₂F₃ ILs were screened for their ST. Briefly, the evaluation of ST was performed in the 3m-deep-pond facility of IRRI. Seeds of each of the drought selected BC₂F₃ ILs, the recipients, and a susceptible check (IR42), were sown in a single row (~15 plants per row) on seedling nursery of the facility with two replications for each BC₂F₃ IL, recipients and IR42. The 30-day old seedlings were submerged under 1.5 m-deep water and maintained under the complete submergence for two weeks until the susceptible check (IR42) and recipients were killed. Water was then drained and the surviving/recovering plants of each BC₂F₃ IL were counted and converted into surviving or recovering percentage. In the following DS, seeds collected form all survived BC₂F₄ ILs from the submergence evaluation in the WS of 2006 were tested in the same way for their ST with two-week complete submergence. In the confirmation experiment, 5g of seeds for each selected BC₂F₄ IL were sown in a 75-cm long row with 15 cm between rows. At 30 days after sowing, the water level was increased up to 1.3 m and was maintained up to 14 days until the susceptible check IR42 and recipients were all dead. Then, the water was totally drained out of the pond. A IL was considered to be ST when its average surviving/recovering rate was greater than 50% in the BC₂F₃ screening and BC₂F₄ progeny testing under complete submergence in the DS and WS experiments.

DNA was isolated from bulked fresh leaf tissues from all (>30) plants of each BC₂F₃ line selected for DT using the CTAB method (Murray and Thompson, 1980) to reconstruct the genotype for each of the original selected BC₂F₂ plants. Then, a total of 625 single sequence repeat (SSR) markers across the rice genome (Ware et al., 2002; https://www.gramene.org/) were used to screen the polymorphisms between the recipients and donors, from which 309 SSR markers representing 91 well-distributed bins (each covering a genomic region of 10-20 cM) across the rice genome ( Figure 1 ) were used to genotype all drought-selected BC₂F₃ ILs. On average, ILs from each BC population was genotyped with 86-158 polymorphic SSR markers that covered >95% of the introgressed donor segments in the ILs.

According to the population genetics theory (Falconer and Mackay, 1996; Hartl and Clark, 1997), identification of QTLs affecting DT (target trait) was performed using the strategy of selective introgression (Zhang et al., 2021a). Specifically, the donor introgression in the 260 BC₂F₂ plants initially selected under severe drought from the nine populations were expected to show significant (P ≤ 0.001) over-introgression at QTL associated with DT. Thus, χ² tests were performed to scan the whole genome to identify marker that showed significant (P ≤ 0.001) deviation in genotypic frequencies in the drought BC₂F₂ plants of a specific population from the whole genome donor introgression of the population.

Identification of ST was performed in three steps. First, χ² tests were performed to scan the whole genome to identify marker that showed significant (P ≤ 0.001) deviation in genotypic frequencies in the DT-ST selected BC₂F₄ lines of a specific population from the whole genome donor introgression of the population. Secondly, because those ST BC₂F₄ lines were selected from the drought selected BC₂F₂ plants, those QTLs showing over-introgression detected by χ² tests in the DT-ST selected lines from a BC population were not resulting from selection for ST, but from their high frequencies in the original selected DT plants. To overcome this, we performed a one-tailed conditional Z test, Z = O F ( S T ) − E F ( S T ) S D , where SD = ( O F ( S T ) ) ( 1 − O F ( S T ) ) n , OF _{( ST )} was the observed donor introgression frequency at an identified ST QTL and EF _{( ST )}=OF _{( ST )}·si _{( ST )} was the expected donor introgression frequency at the ST QTL, SI_(ST) and n were the selection intensity and number of the drought selected BC₂F₂ plants of the population. Once Z ≥ 1.645, i.e. OF _{( ST )} was significantly greater than EF _{( ST )} of the ST QTL at P ≤ 0.05, suggesting that this QTL was significantly associated with ST conditional to its frequency in the drought selected BC₂F₂ plants. Thirdly, a multi-locus independence probability (MIP) test, P _{( AG )}=(p_i ) ^rm •(1–p_i ) ^{r ( n – m )}, was performed to detect individual association groups (AGs) for ST in the DT-ST selected lines of a population. Here, an AG is a group of r (r≥2) unlinked but perfectly associated loci of equal introgression in the DT-ST selected lines of a population, where p_i is the expected frequency of the donor introgression in the DT-ST selected lines of the BC population, n is the number of DT-ST selected lines, m is the number of lines that have co-introgression of the donor alleles, and (n–m) is the number of lines having no introgression at the r unlinked QTLs in the AG. Here, (P_i ) ^m is the probability of m lines having co-introgression of the donor alleles and (1-P_i ) ^{n - m} is the probability of (n-m) lines having no introgression at r unlinked loci. To minimize the false positive rate in detecting AG for ST QTLs, double criteria were used to claim a significant AG, including the χ² test for each of the QTLs in an AG should reach P< 0.05 and P_AG ≤ 0.001.

For important major ST/DT QTLs that were detected in >3 populations, we searched the rice databases (http://rice.uga.edu/; https://www.ricedata.cn/) and identified genes that have reported functions on DT and/or ST and are co-localized with the major QTLs as candidate genes for each of the major ST/DT QTLs. To obtain additional evidence in candidate gene analyses, we performed the gene CDS haplotype analyses (Zhang et al., 2021b) to compare the parental differences at each of the candidate genes to see if the recipient (IR64) allele differed from the donor alleles at each of the candidate genes.

To understand the genome-wide responses of the donor introgressions in the BC progeny to selection for DT, we genotyped the 260 drought selected BC₂F₃ lines with 309 polymorphic SSR markers that covered >95% of the introgressed donor segments in the originally selected BC₂F₂ plants. The donor introgression patterns in the 260 BC₂F₂ plants, characterized by the polymorphic SSR markers across the genome, were reflected in three major ways ( Supplementary Table S1 ). First, strong selection under drought resulted in significantly increased donor introgression in the ST ILs. On average, the donor introgression in the 260 BC₂F₂ plants, estimated from all polymorphic markers in the resultant BC₂F₃ lines, was 0.199 ± 0.052, ranging from 0.139 in the 13 lines of population Teqing/OM1723 to 0.298 in the 38 lines of population IR64/Type3, significantly higher than the Mendelian expectation of 0.125. This excess donor introgression resulted primarily from the excess homozygous donor alleles (0.167 ± 0.043) by 167.2% across the genome at the expenses of greatly reduced heterozygosity by 49.6% in the drought selected plants ( Supplementary Table S1 ).

The χ² tests using the genotypic data of SSR markers across the rice genome identified 59 genomic regions or QTLs (in 144 cases) that showed significant excess donor introgression in the 260 drought selected lines from the nine BC₂F₂ populations ( Figure 1 ; Supplementary Table S2 ). Of the identified DT QTLs, 13 QTLs were more important and each detected in ≥4 of the populations, while 17 of the DT QTLs were each detected in a single population ( Figure 1 ). Specifically, the identified DT QTLs included 11 QTLs on eight rice chromosomes with an average donor frequency of 0.571 in the 46 drought selected BC₂F₂ plants of population IR64/STYH ( Supplementary Table S2A ), 14 QTLs on ten chromosomes with an average donor frequency of 0.360 in the 41 drought selected BC₂F₂ plants of population IR64/BR24 ( Supplementary Table S2B ), 26 QTLs on ten chromosomes with an average donor frequency of 0.624 in the 38 drought selected BC₂F₂ plants of population IR64/Type3 ( Supplementary Table S2C ), nine QTLs on seven chromosomes with an average donor frequency of 0.513 in the 13 drought selected BC₂F₂ plants of population IR64/Cis ( Supplementary Table S2D ), 13 QTLs on eight chromosomes with an average donor frequency of 0.484 in the 19 drought selected BC₂F₂ plants of population Teqing/BR24 ( Supplementary Table S2E ), 25 QTLs on all chromosomes with an average donor frequency of 0.510 in the 27 drought selected BC₂F₂ plants of population Teqing/Binam ( Supplementary Table S2F ), 12 QTLs on eight chromosomes with an average donor introgression of 0.529 in the 13 drought selected BC₂F₂ plants of population Teqing/OM1723 ( Supplementary Table S2G ), 21 QTLs on 11 chromosomes with an average donor introgression of 0.422 in the 32 drought selected BC₂F₂ plants of population IR64/FR13A ( Supplementary Table S2H ), and 13 QTLs on seven chromosomes with an average donor frequency of 0.408 in the 31 drought selected BC₂F₂ plants of population Teqing/FR13A ( Supplementary Table S2I ), respectively. Notably, non-Mendelian segregation distortions with extremely high donor introgression (>0.65) and/or zero heterozygosity were observed in 74 (51.4%) of the 144 cases of DT QTLs identified in specific populations. In particular, 17 of the 26 DT QTLs detected in population IR64/Type3 were of this type.

As shown in Supplementary Table S1 , 124 of the 260 drought selected BC₂F₄ lines from the nine populations had confirmed ST under two rounds of progeny testing under complete submergence for 14 days in the DS and WS. χ² tests at individual loci, conditional Z tests plus the multi-locus independence probability (MIP) tests using the genotypic data of SSR markers of the 124 DT-ST selected BC₂F₄ ILs from the nine populations identified 68 QTLs (in 145 cases) associated with ST ( Figure 1 and Supplementary Table S2 ). Specifically, the identified ST loci included 10 QTLs with an average donor frequency (F_(donor)) of 0.545 in the 26 ST BC₂F₄ lines of population IR64/STYH ( Supplementary Table S2A ), 10 QTLs with an average donor frequency of 0.431 in the 25 ST BC₂F₄ lines of population IR64/BR24 ( Supplementary Table S2B ), 16 QTLs on eight chromosomes with an average F_(donor) of 0.642 in the 20 BC₂F₄ lines of population IR64/Type3 ( Supplementary Table S2C ), nine QTLs on seven chromosomes with an average F_(donor) of 0.533 in the 11 ST BC₂F₄ lines of population IR64/Cis ( Supplementary Table S2D ), 10 QTLs plus two AGs on seven chromosomes with an average F_(donor) of 0.478 in the eight ST BC₂F₄ lines of population Teqing/BR24 ( Supplementary Table S2E ), 11 QTLs and seven AGs on all 12 chromosomes with an average F_(donor) of 0.455 in the 11 ST BC₂F₄ lines of population Teqing/Binam ( Supplementary Table S2F ), 12 QTLs on eight chromosomes with an average F_(donor) of 0.574 in the nine BC₂F₄ lines of population Teqing/OM1723 ( Supplementary Table S2G ), 14 QTLs and five AGs on 11 chromosomes with an average F_(donor) of 0.355 in the 10 ST BC₂F₄ lines of population IR64/FR13A ( Supplementary Table S2H ), and 12 QTLs and three AGs with an average F_(donor) of 0.397 in the 13 ST BC₂F₄ lines of population Teqing/FR13A ( Supplementary Table S2I ), respectively. Notably, non-Mendelian segregation distortions with extremely high donor introgression (>0.65) and/or zero heterozygosity were observed in 97 (67.4%) of the 144 cases of ST loci identified in specific populations.

When compared the DT and ST QTLs identified in the same populations, we observed an average 55.4% of the genetic overlap (the shared proportion of the same QTLs for DT and ST identified in the same populations), close to the mean 50.9% of the phenotypic overlap between drought selected and ST BC₂F₄ lines, ranging from 29.2% in population Teqing/FR13A to 80% in population IR64/Cis ( Figure 1 and Supplementary Table S2 ). In fact, the genetic overlap was positively correlated with the phenotypic overlap (r = 0.716) across the BC populations ( Supplementary Table S1 ). To better illustrate the genetic overlap between rice DT and ST, we compared the genetic mapping results from two separate experiments using the same two BC₂F₂ populations between recipients (IR64 and Teqing) and donor FR13A. Figure 2A shows the ST genetic network (multi-locus structure) containing 12 loci and two AGs identified in 14 surviving plants selected under submergence stress from the IR64/FR13A BC₂F₂ population (Wang et al., 2015), of which nine QTLs were also detected in the ten DT-ST selected BC₂F₄ lines of the same population in this study ( Supplementary Table S2H ). Moreover, eight of the nine ST QTLs detected in the ST BC₂F₂ plants were also associated with DT in the 32 drought selected BC₂F₂ plants from the same population. Only five ST QTLs identified in the previously submergence selected BC₂F₂ plants were not detected in the ten DT-ST BC₂F₄ lines. Similarly, the genetic network detected in the eight submergence-surviving plants of the Teqing/FR13A BC₂F₂ population containing 6 QTLs and eight AGs ( Figure 2B , Wang et al., 2015), ten of which were also detected as ST QTLs in the 13 DT-ST BC₂F₄ lines of the same population ( Supplementary Table S2I ). Moreover, four of the ST QTLs detected in the ST BC₂F₂ plants were associated with DT in the 31 drought selected BC₂F₂ plants from the same population. Ten ST QTLs identified in the eight submergence selected BC₂F₂ plants were not detected in the ten DT-ST BC₂F₄ lines in the Teqing/FR13A BC₂F₂ population ( Figure 2B ). Strikingly, in both populations, the ST loci detected in submergence-selected BC₂F₂ plants but not in the DT-ST selected BC₂F₄ lines of the same populations shared a common branch of associated QTLs (the blue colored loci of Figure 2 ), in which one QTL near bin 9.3 in the upper regulatory position of the branch harbors the well-known Sub1 gene for ST (Xu et al., 2006). Apparently, the Sub1A (an ethylene response transcriptional factor) and its regulated loci were eliminated in the drought-selected plants of both IR64/FR13A and Teqing/FR13A BC₂F₂ populations, but in over-introgression in the ST selected BC progenies of the same two populations (Wang et al., 2015).

Among the identified QTLs, we found 16 QTLs for either DT and/or ST were co-localized near the same markers or in the same bin regions in three or more of the populations simultaneously ( Table S3 ). These hotspots or clustered QTLs were definitively designated as the major QTLs. These included seven QTLs (qDT1.7, qDT2.2, qDT3.8, qDT7.5, qDT7.6, qDT10.5, and qDT11.6) for DT each detected in three populations, six QTLs (qDT2.9, qDT3.2, qDT4.7, qDT5.3, qDT9.6, and qDT12.4) each identified in four populations, two QTLs (qDT1.3 and qDT7.4) detected in five populations, and one QTL (qDT2.11) identified in six populations. Eleven major QTLs were found to affect ST, including three (qST2.9, qST5.3, and qST6.5) and six QTLs (qST1.3, qST3.2, qST4.7, qST7.4, qST7.6, and qST9.6) identified in three and four populations, respectively. Two QTLs (qST2.11 and qST12.4) were each identified in five populations simultaneously ( Table S3 ).

Table S4 shows the eight most likely candidate genes for the identified major QTLs and their population genetics parameters in different rice populations. The first one is LOC_Os01g09620 co-localizing with qST1.3/qDT1.3 and encoding a CCCH-type zinc finger protein to have the function for DT and is responsive to ethylene (Kong et al., 2006.). At this locus, IR64 has Hap2 for susceptibility to both drought and submergence, while the three donors (BR24, Type3, and OM1723) have Hap1 contributing DT/ST. In the qDT2.11/qST2.11 region, the most likely candidate gene was LOC_Os02g57530 encoding an ethylene receptor which was potentially to be associated with ST and DT (Yau et al., 2004). At this locus, IR64 has Hap2 for susceptibility to both drought and submergence, while six donors contributing tolerance have Hap1, Hap3, Hap4, and Hap5. We identified LOC_Os02g48010 as the most likely candidate for qDT2.9/qST2.9. It encodes a nuclear matrix constituent protein 1-like, and IR64 has Hap3 while three donors (BR24, Type3, and OM1723) have Hap1, Hap2, and Hap4 contributing DT/ST. The most likely candidate gene for qDT2.9/qST2.9 is LOC_Os04g54474 encoding a bZIP transcription factor, and its homologous gene (OsbZIP86) is associated with DT/ST (Nijhawan et al., 2008; Gao et al., 2022). The gcHaps of the DT/ST contributing donors (Type3, Cisanggarung, and Binam) have either Hap1 or Hap3, while IR64 has Hap2. For qDT5.3/qST5.3, the most likely candidate gene is LOC_Os05g28350 encoding an AP2 domain containing protein. The evidence was from the allelic differences between IR64 (Hap3) and the donors (Hap2, Hap4, and Hap5) contributing to DT/ST. For qDT7.6/qST7.6, the most likely candidate gene is LOC_Os07g48630 encoding a protein responsive to ethylene stimulus (Mao et al., 2006) with obvious allelic differences between IR64 (Hap4) and the donors (Hap1, Hap6, and Hap7) contributing to DT/ST. The most likely candidate gene for qDT9.6/qST9.6 is LOC_Os09g35010, encoding AP2/EREBP transcription factor which is reportedly associated with DT (Figureueiredo et al., 2012) with obvious allelic differences between IR64 (Hap1) and the donors (Hap2 and Hap3) contributing to DT/ST. As the most likely candidate gene for qDT12.4/qST12.4, LOC_Os12g35610 encodes an ethylene-induced protein functioning in root aerenchyma formation under oxygen-deficient condition (Liu et al., 2012) with obvious allelic differences between IR64 (Hap3) and the donors (Hap2) contributing to DT/ST ( Table S4 ). Except for LOC_Os02g48010 which is a conserved gene with a low diversity (E_H = 0.09) and only two gcHaps in rice populations, the remaining seven candidate genes were high diverse (E_H is between 0.42 and 1.18, and gcHapN is between 4 and 11) within and among different rice populations ( Table S4 ).