In a backcross breeding program initiated in 1998, IR64, the most widely grown rice variety in South/Southeast Asia, was used as recurrent parent to cross and backcross twice with six donors, Madhukar (India), Binam (Iran), OM1723 (Vietnam), FR13A (India), Hao-An-Nong (China), and BR24 (Bangladesh) (Table 1) to develop BC₂F₂ bulk populations as described previously (Lafitte et al., 2006). In 2001 dry season, the BC₂F₂ bulk populations were screened under severe terminal drought that killed most individuals of each BC population. A small portion of the BC progeny survived drought stress and was selected. The selected BC₂F₃ progeny were planted under the normal irrigated conditions for seed increase and genotyped with polymorphic SSR markers. The BC₂F₄ progeny were confirmed for improved drought tolerance (DT) in a replicated experiment (Lafitte et al., 2007) and were screened under EC (electric conductivity) 12 dSm^-1 at the seedling stage in the 2002 wet-season. From these progeny, eight BC₂F₅ introgression lines (ILs) that showed significantly improved ST or DT and good yield performances under the normal conditions, were used as the parents to make four intercrosses (Table 1) to develop ST varieties by pyramiding ST QTLs from different donors. Two types of progeny were developed from each of the crosses, including a random set of F₂ progeny obtained from F₁ plants grown under the normal conditions and a set of ST F₃ progeny selected under severe salt stress from 600 F₂ individuals of each of the pyramiding F₂ populations.

The ST F₃ progeny selected from the four populations were then progeny tested in the replicated experiments under salt stress during the dry season of 2006–2007. For each F₃ line, 20 plants were evaluated with its parental ILs and IR64. The salt treatment was the same as the F₂ screen described above, except after the salt concentration in the tanks was increased to EC 18 dSm^-1 and maintained for 5 days, and then the salt concentration was further raised to EC 24 dSm^-1 and maintained until IR64 and the parental ILs were all dead. Then, the survival plants of each F₃ lines were counted. Leaf tissues from 20 plants of each of the F₃ lines from each cross was bulk-harvested for DNA extraction and genotyping to reconstruct the genotypes of their original F₂ plants. A total of 637 well-distributed anchor SSR markers¹ from the Cornell University were used to survey the polymorphic markers differentiating the parental ILs of each cross, from which 35, 29, 6, and 25 differentiating SSR markers were used to genotype the selected ST F₃ progeny and the random F₂ plants from each cross.

Table 2 and Figure 2 show the segregation patterns of all segregating SSR markers in the 422 ST F₃ lines and 318 random F₂ progeny from the four populations, which revealed three types of extremely distorted or ES characterized with loss of heterozygosity (LOH) and loss of allele (LOA, fixation of one of the parental alleles) in the random and/or salt-selected progeny. Of the 20 segregating markers (regions) examined in the progeny from cross A, 13 markers on 9 chromosomes (1–3, 5–8, 10, and 11) were of type I, which had a mean heterozygosity (H) of 1.9 and 0.4% in the random and salt-selected progenies, respectively (Table 2A). The paternal alleles almost reached fixation with a mean frequency of 0.931 (0.958) at 11 of the 12 loci in the random (salt-selected) progeny, except for RM171 on chromosome 10 at which the maternal allele was almost fixed with an allele frequency of 0.913 (0.993), indicating those alleles of ES were inherited from their parents. The remaining seven loci on seven chromosomes (2, 4–7, 10, and 11) were of type II that showed the 1:2:1 Mendelian segregation ratio in the random F₂ progeny but strong ES in the salt-selected progeny with a mean H of 12.8%. The paternal alleles at five of these type II ES loci were favored with a mean frequency of 0.717, while the maternal alleles at the remaining two loci were favored with a mean frequency of 0.649.

Of the 20 segregating markers examined in the progenies of cross B, three loci in chromosomes 1, 4, and 6 were of type I ES, which showed extreme LOH in both the random (H = 1.5%) and salt-selected (H = 1.8%) progenies significant at P < 1.2^-17. However, both parental alleles had significant frequencies ≥ 25% in the random F₂ progeny, while one of the parental alleles became virtually fixed at all three loci in the salt-selected progeny (Table 2B). We detected 12 loci of type II ES on eight chromosomes (1–4, 6, 7, 9, 10, and 12) in cross B that showed the expected 1:2:1 segregation in the random F₂ progeny but extreme ES in the salt-selected progeny. The paternal alleles at 9 of the 12 loci were fixed or nearly so with frequencies > 0.943, and so were for the maternal alleles at the remaining three loci in the salt-selected progeny. There were three type III loci that showed ES in both the random and salt-selected progenies, but their favorable alleles were switched in the random and selected progenies. We noted two loci near RM174 and RM519 on chromosomes 4 and 12, at which the random progeny did not show LOH, but had significantly higher frequencies of the maternal alleles. In the salt-selected progeny, the two loci showed extreme ES. Thus, these two loci should be considered type II ES loci.

In cross C, only five loci in three regions on chromosomes 1, 6, and 7 were segregating. All five loci were of type II ES that showed the Mendelian segregation in the random F₂ progeny but extreme ES in the salt-selected progeny (Table 2C). The paternal alleles at four loci were predominant, while the maternal allele at the remaining locus was favored in the selected progeny. There were 14 loci segregating in cross D. Of these, only one locus on chromosome 3 was of type I ES that exhibited ES segregation in both the random and salt-selected progeny, though ES in the selected progeny was much more extreme with the maternal allele reaching the fixation (Table 2C). We detected four loci of type II ES in cross D that showed the Mendelian segregation in the random progeny but extreme ES in the salt-selected progeny. Surprisingly, there were nine loci of type III ES detected in cross D, at which extreme ES with complete LOH and LOA (fixation of the paternal alleles) at all loci except for one on chromosome 6 in the random progeny, but complete LOH and LOA (fixation of the maternal alleles) at all loci in the salt-selected progeny. In fact, all 138 salt-selected progeny in cross D were fixed at the maternal alleles from a Geng (japonica) donor, Hao-An-Nong at all 14 segregating loci.

Taking the above results together, we detected a total of 32 genomic regions that showed one or more types of salt-induced ES in the random and salt-selected progenies from the four pyramiding crosses (Figure 2). When results from different crosses were compared, we noted several interesting characteristics of the salt-induced ES. First, the size of genomic regions covered by multiple closely linked markers showing salt-induced ES detected in specific crosses varied considerably, ranging from a single marker to a group of closely linked markers covering a region > 30 cM. Second, the same type of ES or different types of ES were detected in progenies from different crosses with two large genomic regions on chromosomes 1 and 6 where more and the strongest ES were detected across different populations. Thirdly, in our previous study, we reported 27 genomic regions that harbor 53 QTLs affecting six ST related traits identified in the random backcross progenies of two crosses, IR64/Binam and Teqing/Binam with Binam as the donor (Sun et al., 2007; Zang et al., 2008), which are highly related to the populations used in this study. These ST related traits included scores of salt toxicity of leaves (SST), survival days of seedlings (SDS), shoot N⁺ concentration (SNC), shoot K⁺ concentration (SKC), plant height difference of the stress to control (DPH), tiller number difference of the stress to control (DTN), and plant fresh weight difference of the stress to control (DFW). When taking a close look at their genomic locations, we found 14 genomic regions where the salt-induced ES regions were overlapping with QTL affecting ST related traits. This is much higher than the probability (P ≤ 0.03) caused by chance, suggesting that ST QTL regions tended to show ES in the salt-selected progeny.

In this study, we have shown the results from our breeding effort to develop ST rice varieties by the approach of “designed QTL pyramiding (DQP)” (Li and Xu, 2007; Li and Zhang, 2013). The development of 18 completely homozygous ST lines that showed > 80% survival rates under the natural whole-growth duration salt stress of EC 12–18 dSm^-1 in 3 years has demonstrated the power and efficiency of this breeding strategy. During this breeding process, we also sought to understand the genetic basis of ST in rice. The discovery of the three types of salt-induced ES was surprising, each of which showed several interesting characteristics and had important implications in evolution and future breeding for developing stress-resilient rice and crops.

In this study, although a small number of markers were segregating in each population, the fact that all of them showed strong LOH and/or LOA in the salt-selected progenies of all four populations, indicating that salt induced ES occurred across the whole genome. Of the genomic regions showing ES, 14 (25%) of them were type I ES which showed LOH and/or LOA in both the random and salt-selected progenies of three crosses (Figure 2), 12 of which were detected in cross A whose parental ILs were all salt-selected (Ali et al., 2006). The remaining four regions detected in crosses B and D whose parents were all drought selected (Lafitte et al., 2006). Clearly, these type I ES loci were inherited from their parents. In other words, this type of ES loci showed transgenerational inheritance. In rice, reduced heterozygosity was reported in the drought-, salt- and submergence-selected BC progenies (Li and Zhang, 2013; Wang et al., 2015; Ali et al., 2017), but this type of stress-induced LOH were not fully characterized and appropriately interpreted. Thus, the inclusion of the random populations in this experiment provided compelling evidence that type I ES loci were inherited from their parental ILs. We noted that in almost all cases, the levels of ES (LOH and LOA) in the salt-selected progenies were stronger than in the random progenies from the same crosses, further implicating the important role of salt stress for the observed LOH and LOA in the populations. We noted that the behaviors of type III ES loci were very similar to type I ES loci except that the observed LOA in the random progeny was in the opposite direction as that in the salt-selected progeny from the same cross. When taking a close look at the 12 (9 in cross D and 3 in cross B) loci of type III ES identified, we found that 2 (RM501 on bin 7.3 and RM334 on bin 5.9) of them in cross B were not typical (Table 2). In the former case, LOH was observed and the frequencies of the parental alleles did not differ significantly in the random progeny, but showed LOH and LOA in the salt-selected one. Thus, these loci should be considered as a type I ES locus. In the latter case, there was equally reduced heterozygosity and no LOH or LOA observed in the random and salt-selected progenies. The remaining 10 loci (9 from cross D and 1 from cross B) were of typical type III ES or transgenerational ES, at which LOH and LOA were present in both the random and salt-selected progenies, but different parental alleles were fixed in the random and salt-selected progenies, respectively. As mentioned earlier, the parental ILs of crosses B and D were all drought-selected. ST and DT are known to be partially overlapped genetically and physiologically (Zhou et al., 2009; Hong et al., 2016). As expected, no type III ES loci were identified in cross A whose parental lines were all salt-selected. Moreover, there were apparent correspondences between the salt-induced ES loci and QTL affecting ST traits identified in the random ILs of the same IR64 backgrounds (Figure 2). All these results suggested that type I ES regions contain the parental alleles contributing to both ST and DT, while different parental alleles in type III ES regions were associated separately with ST and DT, respectively. If so, our results would suggest that the ES regions indeed contain genes for ST and/or DT as the targets of the strong phenotypic selection for ST and/or DT in our breeding processes.

Different from types I and III ES loci, the 21 type II ES loci from the four crosses showed the Mendelian 1:2:1 segregation in the random progenies but LOH/LOA in the salt-selected progenies of the same crosses, provided compelling evidence that the observed ES at all these loci were salt-induced during the selection process. Interestingly, the overall levels of LOH, particularly LOA, in the type II ES regions were less pronounced as compared to those of types I and III (Table 2), suggesting that the effect of the applied abiotic stress (salt and/or drought) on the level of LOH and LOA was accumulative. It should be pointed out that the genotypes of all selected survival F₂ plants were reconstructed from the bulk DNA of their F₃ progeny. Thus, we were actually measuring the gametic genotypes of the selected F₂ plants. Thus, the observed genome-wide ES resulted from strong salt (drought)-induced gametic selection favoring one of the parental alleles in the salt-stressed F₂ plants of all crosses.

Environmentally induced genome-wide LOH and/or LOA have been reported in both natural and experimental populations of many diploid organisms, such as in natural and experimental populations of diploid isolates of yeast as a result of selection on nutrient stress (Granek and Magwene, 2010; Smukowski Heil et al., 2017). In humans, a strong correlation was found between LOH events and deficient protein expression of specific genes and the incidence of secondary tumors of breast cancer (Kamat et al., 2012). LOH/LOA on specific chromosomal regions was reported to contribute to the molecular pathway of glioma progression in Malay population (Zainuddin et al., 2004). The second one is that LOH and/or LOA in the genome of an organism can be a localized event involving specific genes in specific genomic region, or genome-wide events. Then, questions arise regarding what molecular mechanism(s) were responsible for the observed stress-induced ES in rice and other organisms.

We noticed that ‘stress-induced genomic imprinting (SIGI)’ or ‘stress imprinting’ (Zucchi et al., 2012), an epigenetic phenomenon, was the most likely molecular mechanism responsible for the salt-induced LOH and LOA in rice observed in this study based on two indirect pieces of evidence. First, the SIGI is a well-characterized epigenetic molecular mechanism that leads to differential expression of maternal and paternal alleles, depending on their parent-of-origin (Feil and Berger, 2007). In fact, the SIGI behaves remarkably similar to the salt (drought)-induced epiloci of this study in the following ways. First, like the type II ES loci, environmentally induced epigenetic changes are transient in most cases (Iwasaki, 2015), and in some cases they are stably maintained through mitotic cell divisions and show transgenerational inheritance, as demonstrated in vernalization of Arabidopsis by prolonged cold that involves epigenetic silencing of specific genes (Kim and Sung, 2012; Song et al., 2012). In Arabidopsis, SIGI results from stress regulated expression and function of three key enzymes, MET1, DDM1, and MOM1, which regulate, by DNA methylation, the expression of some Arabidopsis loci under heat stress and set or reset stress-induced chromatin changes and epigenetic marks transmitted to the next generation (Tittel-Elmer et al., 2010). Secondly, in a sister line of the parental ILs (crosses B and D) selected from drought, we observed that drought was able to induced site-specific DNA methylation/demethylation across the genome (Wang et al., 2010), and ∼25% of those drought-induced DNA methylation/demethylation sites were irreversible, while the 75% were reversible when the drought stress was removed. This lent a strong support to the above speculation that drought- or salt-induced DNA methylation was the most likely molecular mechanism for the observed salt- or drought induced ES, because the proportions of those irreversible and reversible methylated sites matched almost perfectly with that of type I (III) and II ES, even though it remains a mystery how this happened at the molecular level.