Genetic dissection of heterosis of indica–japonica by introgression line, recombinant inbred line and their testcross populations

The successful implementation of heterosis in rice has significantly enhanced rice productivity, but the genetic basis of heterosis in rice remains unclear. To understand the genetic basis of heterosis in rice, main-effect and epistatic quantitative trait loci (QTLs) associated with heterosis for grain yield-related traits in the four related rice mapping populations derived from Xiushui09 (XS09) (japonica) and IR2061 (indica), were dissected using single nucleotide polymorphism bin maps and replicated phenotyping experiments under two locations. Most mid-parent heterosis of testcross F1s (TCF1s) of XS09 background introgression lines (XSILs) with Peiai64S were significantly higher than those of TCF1s of recombinant inbred lines (RILs) with PA64S at two locations, suggesting that the effects of heterosis was influenced by the proportion of introgression of IR2061’s genome into XS09 background. A total of 81 main-effect QTLs (M-QTLs) and 41 epistatic QTLs were identified for the phenotypic variations of four traits of RILs and XSILs, TCF1s and absolute mid-parent heterosis in two locations. Furthermore, overdominance and underdominance were detected to play predominant effects on most traits in this study, suggesting overdominance and underdominance as well as epistasis are the main genetic bases of heterosis in rice. Some M-QTLs exhibiting positive overdominance effects such as qPN1.2, qPN1.5 and qPN4.3 for increased panicle number per plant, qGYP9 and qGYP12.1 for increased grain yield per plant, and qTGW3.4 and qTGW8.2 for enhanced 1000-grain weight would be highly valuable for breeding to enhance grain yield of hybrid rice by marker-assisted selection.


Abbreviations
.2%, and − 0.6% in LS, and 41.0%, 37.7%, and 3.6% in WZ, respectively, showing strong heterosis for PN and FGNP between the two parents and low heterosis for GYP with low fertility of F 1 plants in the two locations, resulting from incompatibility between the two subspecies parents. As compared with XS09, the tester line PA64S had significantly higher PN, FGNP, and GYP, and significantly lower TGW in LS. In the WZ experiment, PA64 (the recurrent parent of PA64S) was used instead of PA64S due to its sterility, which showed significantly  Fig. 1. Wider distributions of all traits were found in the RIL population than XSIL population in both locations. The RILTCF 1 s had significantly lower TGW and higher PN, FGNP and GYP than the RILs at the two locations, while similar results were observed for all traits except for TGW between XSILTCF 1 s and XSILs in WZ (Fig. 1, Table 1). The mean TGW and GYP in RILTCF 1 s and the mean PN, FGNP and GYP in XSILTCF 1 s were all higher than those of the check F 1 (PA64S × XS09) plants in LS. There were 7 (3.3%), 18 (8.6%), 27 (12.9%), and 8 (3.8%) RILTCF 1 s and 16 (7.2%), 16 (7.2%), 0, and 3 (1.4%) XSILTCF 1 s having significantly higher PN, FGNP, TGW, and GYP than the F 1 (PA64S × XS09) plants in LS, respectively. Similarly, in the WZ experiment, the mean PN in both RILTCF 1 s and XSILTCF 1 s and the mean GYP in XSILTCF 1 s were higher than the check F 1 (PA64S × XS09) plants. There were 28 (13.4%), 2 (1.0%), 1 (0.5%), and 1 (0.5%) RILTCF 1 s and 39 (17.6%), 1 (0.5%), 0, and 4 (2.0%) XSILTCF 1 s having significantly higher PN, FGNP, TGW, and GYP than the F 1 (PA64S × XS09) plants, respectively. On average, the RILTCF 1 s had higher TGW and GYP and lower PN and FGNP than XSILTCF 1 s in LS, while a similar trend was observed in RILTCF 1 s for PN, FGNP and TGW, but lower GYP compared with XSILTCF 1 s in WZ.
The performance of the hybrid plants was determined by their mid-parental value and the heterosis level. Here, the most H MP of the TCF 1 s showed positive heterosis in all traits except for TGW in both locations (Fig. 1, Table 1. Phenotypic performance and mid-parent heterosis of four yield traits in the parent lines, RIL and XSIL populations from a cross of Xiushui09/IR2061, and their testcross F 1 s with common maternal tester line PA64S in Lingshui (LS) and Wenzhou (WZ). a PA64S Peiai64S, PA64 the recurrent parent of an isogenic to PA64S, XS09 Xiushui09, RILs recombinant inbred lines derived from a cross between Xiushui09 and IR2061, XSILs introgression lines under Xiushui09 background with IR2061 as a donor, TCF 1 s testcross F 1 s between RILs or XSILs and one common maternal tester line PA64S, H MP mid-parent heterosis, FGNP filled grain number per panicle; TGW 1000-grain weight, PN effective panicle number per plant, GYP grain yield per plant. Trait values are presented as mean ± sd. Characters behind the sd value indicate significant differences based on Duncan's multiple comparison tests (P < 0.05) '*' and '**' indicate significant differences between mean F 1 hybrid and average performance of corresponding parental lines using Student's t-tests at P < 0.05 and P < 0.01, respectively. www.nature.com/scientificreports/  Table 2. The H MP s of TCF 1 s were determined by the trait values of the TCF 1 s and the corresponding lines (RILs or XSILs). In other words, the performance of individual TCF 1 plants largely depends on the cumulative effect of the trait value of the relative line and the heterosis. There were significant and highly positive correlations between the lines and the relative TCF 1 s for all traits in the two locations except FGNP between XSILs and the XSILTCF 1 , and GYP between RILs and the RILTCF 1 in WZ, and the correlation coefficients in LS were generally higher than those in WZ. The significant and highly positive correlations (r ≥ 0.53, P < 0.01) between TCF 1 s and their H MP s were also observed for PN, FGNP and GYP in both locations, while the correlation of TGW was slightly weak between RILTCF 1 s and their H MP s (r = 0.26) and between XSILTCF 1 s and their H MP s (r = 0.15) in LS. In contrast, most correlations between the H MP of TCF 1 s and the relative lines for FGNP, TGW and GYP were significantly negative and were much stronger than PN in both locations ( Table 2). In addition, the correlations between the lines and their H MP for these yield traits differed according to different locations.
The above results suggested that the role of heterotic loci for yield should be affected by the environment.
Effects of proportion of Japonica genome in male lines on heterosis of testcross. PA64S shared the same genotypes with XS09 at 9646 (38.1%) of all 25,296 SNPs. Within the RIL population, individual RILs varied considerably in their ratios of homozygous XS09 alleles ranging from 0 to 91% with a median of 42%, much smaller than the ratios of homozygous XS09 alleles in the XSILs ranging from 72 to 100% with a median of 92% due to two-time consecutive backcrossing ( Fig. 2A). Furthermore, most XSILTCF 1 s had a higher ratio of homozygous XS09 alleles (a median of 36% with a range from 27 to 40%) and a higher ratio of heterozygous  www.nature.com/scientificreports/ genotype (a median of 60% with a range from 50 to 68%) than RILTCF 1 s. In view of a continuous distribution of japonica genome (XS09 genome) in the two inbred line (XSILs and RILs) populations, the two populations could be also merged into one combined population for further analysis of the proportional effect of japonica genome on heterosis of the combined testcross populations. Thus, at all tested SNPs, the ratio of japonica genome (XS09-genome) could range from 0 to 100% in the combined inbred population consisted of all RILs and XSILs, and from 0 to 40% in a combined TCF 1 population consisted of RILTCF 1 s and XSILTCF 1 s. Correlation analyses between the ratios of XS09 genome of individual inbred line and the trait value or H MP of the testcross for the four yield traits were shown in Fig. 2B and Table S1. The results indicated that there were significant negative correlations for TGW, significant weak positive correlations for PN and FGNP and no significant correlations for GYP between the trait values or H MP of the TCF 1 s or the combined TCF 1 s and the ratios of XS09 genome in the RILs, XSILs and the combined line population (RILs and XSILs). It should be noted that these significant correlations found in the combined line population were much stronger than those only in the RILs or XSILs. It was indicated that there was no inevitable connection between heterosis of the testcross and the proportion of japonica genome in paternal lines in our tested populations.      1, qPN1.4, qPN3, qPN4.2, qPN4.4, qPN5, qPN6.2 and qPN6.3) detected only in the RILs or XSILs, 1 (qPN2) both in the RILs and XSILs, and 6 (qPN4 .1, qPN6.1, qPN7.1, qPN9.1, qPN11 and qPN12) only in the RILTCF 1 s or XSILTCF 1 s. Notably, five genes Os08g0323700 (OsCCC1), Os09g0410500 (OsTb2), Os09g0470500 (Oshox4), Os09g0466400 (OsZHD1) and Os09g0457900 (OsEATB) reportedly contributing to PN were located nearby the UD-type M-QTL qPN9.2 in WZ, which had a largest negative effect − 8.5 on PN from heterozygote in the TCF 1 s. Moreover, qPN6.2, with a negatively additive effect of 1.4 from XS09 allele and the largest logarithm of the odds (LOD) score detected in the lines in WZ, contained two known genes Os06g0157700 (Hd3a) and Os06g0181300 (AID1) for PN.
Fifteen M-QTLs were detected for GYP, including 13 QTLs (10 additive, 2 OD-type and 1 UD-type) together explained 20.9% (20.6%), 24.6% (10.1%) and 19.7% (5.7%) of the total phenotypic variance in RILs (XSILs), RILTCF 1 s (XSILTCF 1 s) and their AH MP s in LS, respectively, and 2 additive QTLs explained 6.5% in XSILs and 3.5% (1.2%) in RILTCF 1 s (XSILTCF 1 s) in WZ. Of these, only UD-type QTL (qGYP10.1) was identified with a negative dominance effect of 5.4 g in LS, which contained a known gene OsPQT3 controlling rice grain yield in the field conditions. In addition, two OD-type M-QTLs qGYP9 and qGYP12.1 were only detected in AH MP s of RILTCF 1 s and both with positive dominance effects of 6.0 g and 7.4 g in WZ.
Not any co-located QTL was detected among 32 QTLs for the traits in the RIL and the AH MP populations at the two locations, and only one co-located QTL (qTGW1.1) was detected at LS among 34 QTLs for the traits in the XSIL and the AH MP populations (Table 3).

Epistatic QTL mapping of yield-related traits and mid-parent heterosis.
In order to test whether these M-QTLs had epistatic effects on the relevant traits, we checked all epistatic QTL (E-QTL) pairs identified in the RILs or XSILs, their TCF 1 s and AH MP s. Eight, 21 and 12 E-QTL pairs were identified in line, TCF 1 and AH MP populations, respectively, which involved one M-QTL and the other random loci (Table 4). For PN, 3 and 4 E-QTLs in the RILs and XSILs, 6 E-QTLs in XSILTCF 1 s, and 4 E-QTLs in AH MP s of XSILTCF 1 s were identified. The total phenotypic variation explained (PVE) of E-QTLs for PN was 6.84% for RILs in LS and 11.10% for XSILs in WZ, 6.32% for XSILTCF 1 s in WZ, and 6.40% and 1.12% for AH MP s of XSILTCF 1 s in LS and WZ, respectively. For GYP, only one E-QTL in the RILs, 15 E-QTLs in XSILTCF 1 s, and 8 E-QTLs in AH MP s of XSILTCF 1 s were identified. The total PVE of E-QTLs for GYP was 13.05% for RILs in LS, 3.52% for XSILTCF 1 s in WZ, and 1.86% for AH MP s in WZ. Among M-QTLs involved epistasis on GYP and PN, qGYP1 and qPN1.3 on chromosome 1 were the most important loci contributing to the TCF 1 performances and heterosis of GYP and PN both in XSIL and the corresponding AH MP populations. No E-QTLs were identified for TGW and FGNP. No E-QTL between two M-QTLs was observed.

Discussion
QTLs play an important role to enrich our understanding of the genetic basis of heterosis in rice. This study permitted the direct measurement of heterosis for all the measured traits using two sets of lines (RILs and XSILs) together with their TCF 1 s and H MP s that exaggerated the capacity to more precisely resolve different types of gene actions for identified QTLs that were responsible for trait performance and heterosis. In this study, the phenotypic performance of four yield traits was evaluated in two locations, LS and WZ. The performance of hybrid plants was mainly determined by their mid-parental value and the heterosis level. In both the locations, the XSILTCF 1 s had higher H MPS of PN, FGNP, and GYP but lower TGW than the RILTCF 1 s, which suggested that the portion of introgression of IR2061 (indica)'s genomic fragment into XS09 (japonica) background may influence the effect of heterosis. Recently, Lin et al. reported that the introduction of japonica germplasm played an important role in indica hybrid breeding 36 . In their study, only 3.31% of the genome in the parents of the indica hybrids were contributed by japonica germplasm, which affected about half of the grain yield heterotic loci 36 . To evaluate the effects of the proportion of indica and japonica genome on heterosis, we explored the correlations between the ratios of heterozygous or homozygous XS09 alleles and the heterosis for four yield-related traits (Fig. 2). We observed a weak trend for higher ratios of homozygous XS09 genotype in RILs/XSILs or heterozygous genotype in their TCF 1 s with higher heterosis for most yield-related traits except for TGW. This trend is even more pronounced in the combined line populations (RILs and XSILs). Indeed, moderate genomic differences between parents of indica-japonica cross do improve, at least to some extent, grain yield and its degree of heterosis in rice 30,37,38 . However, the genomic differences between parents of indica-japonica cross for attaining strongest heterosis varies according to the cross, and actually there is no a fixed proportion of introgression of indica genome into japonica background or japonica genome into indica background for the parents, as indicated by indica-japonica hybrid breeding practices in China 37,39,40 .
Our study provides a possibility to identify significant genetic factors for heterosis by comparing QTLs detected from different datasets. A total of 81 M-QTLs were identified for phenotypic variation of four traits of lines (RILs and XSILs) or TCF 1 s and AH MP values in both locations. The most striking finding was the presence of two predominant types of M-QTLs for yield-related traits except for TGW, the additive M-QTLs and OD/ UD-type M-QTLs, without M-QTLs exhibiting complete and partial dominance (Fig. 3, Table 3). Previously, it has been reported that the genetic basis of heterosis is mainly determined by dominance and overdominance effects 14,17,41 . Furthermore, Wen et al. used F 1 hybrids in the NCII design to dissect the genetic basis of heterosis by investigating the factors that mainly affect the heterosis were dominance, dominance-by-dominance, overdominance, and complete dominance QTL 42 . Here, 83.3%, 80.0%, 71.4% and 16.7%, 20.0%, 23.8% of the detected M-QTLs were attributed by additive QTL actions and OD/UD-type QTL actions in FGNP, GYP and PN, respectively. However, 9 of 10 D-type M-QTLs in this study were detected in TGW, which may result in a relatively low proportion (51.5%) of the detected 33 M-QTLs for TGW with additive QTL actions. On the other hand, it is worth noting that 13 of 81 M-QTLs were also detected as E-QTLs with another random locus without significant M-QTL. Our results confirmed the previous reports of epistasis and overdominance as the major genetic basis of heterosis in rice 8,43 , and these findings are also consistent with Li et al. 44 and Melchinger et al. 45 . However, almost no genetic overlap was found between the QTLs affecting the traits in inbred line populations (RILs and XS09) and the ones underlying heterosis in their testcross populations (XSILTCF 1 s and RILTCF 1 s) under the two locations, suggesting that different genetic mechanisms involved in trait itself and its heterosis.
In the present study, GYP showed a strongest heterosis among the four yield-related traits studied, and TGW had a weakest heterosis in XSILTCF 1 s and RILTCF 1 s (Table 1). Heterosis for GYP in TCF 1 s was mainly attributed to yield-component traits FGNP and PN both in the two sets of testcrosses, consistent with the findings of previous studies performed on rice 8,30,37,43 . In TCF 1 s, the heterosis of GYP and yield component traits (PN and FGNP) was mainly produced by the overdominance of heterotic loci, indicating that non-additive gene actions are pivotal to grain yield. This finding is consistent with He et al. as they used RILs based NCII design for E-QTLs to estimate genomic position, digenic interactions of QTL, additive and dominance effects, and they noted that non-additive gene actions mainly contribute to heterosis 46 . Further, we explored the correlations among the lines, TCF 1 s and H MP s, and found highly positive correlations between TCF 1 s and H MP s and lower positive correlations between the lines and TCF 1 s for FGNP and GYP in both locations (Table 2). These results suggested that non-additive QTL was a contributor to TCF 1 s for FGNP and GYP traits. Contrarily, a high positive correlation was found between the lines and TCF 1 s for TGW compared to the low positive correlation between TCF 1 s and H MP s in LS, indicating that additive QTL mainly contribute to TCF 1 s for TGW only in LS. However, the correlations between the lines and TCF 1 s for TGW in WZ exhibited contradictory results as indicted in LS. The negative correlation between the lines and their H MP s evidently indicated that additive and dominant QTLs acted independently in the testcross populations as previously reported 17 . In our study, the correlations between the lines and H MP s Table 4. Digenic epistatic QTLs affecting four yield-related traits detected in the Xiushui09/IR2061 RILs and Xiushui09 background introgression lines (XSILs), and their testcross F1s (PA64S × the RILs and XSILs) in Lingshui (LS) and Wenzhou (WZ). 1 PN effective panicle number per plant, GYP grain yield per plant. 2 Interval is based on the Nipponbare reference genome IRGSP 1.0 3 A i and A j are the main effects of locus i and locus j. AA ij is the epistatic effect between loci i and j, as defined by Mei et al. 69 . For RILs or XSILs, the main effects of the loci i and j, arising from the substitution of the IR2061 allele by the XS09 allele. For TCF 1 s and H MP , the main effects of the loci i and j, estimated by the difference between heterozygote (PA64S/XS09) and homozygote using the mean F 1 and H MP values. 4 Percentage of the total variation explained by AA ij . Bold markers are those flanking M-QTLs identified in Table 3 www.nature.com/scientificreports/ for yield traits were different according to genetic background and location, which suggested that the role of heterotic loci for yield was affected by genetic background and environment 8,41,47,48 . Actually, a lot of epistasis between two random loci were detected for the four traits under the two locations (data not shown). So, complexity of heterosis in rice, reflected by a large number of loci involved, complex epistatic relationships, and genetic background-and environment-dependent gene actions on heterosis, suggested that marker-assisted selection for significantly improving heterosis of yield traits in hybrid rice breeding programs may be very challenging.
A few already reported genes in rice were co-located with the heterotic QTLs identified in this study such as qGYP10.1 (OsPQT3, increases grain yield in the field) 49 , qPN9.2 and qTGW9 (OsCCC1 for cell elongation and panicle number 50 , OsTb2 for tillering and grain yield per panicle 51 , and Oshox4 affecting bushy tillers 52 , OsZHD1 for tiller number 53 , and OsEATB, ERF protein associated with tillering and panicle branching 54 , qTGW1.1 (GW5L, negatively regulates grain width and weight) 55 , qTGW2.2 (OsGS1, growth rate and grain filling) 56 , qTGW3.3 (GS3, grain length and weight) 57 , qTGW3.6 (OsMADS34, grain size and yield 58 , Pho1, the size of mature seeds and the starch content) 59 , and qTGW6.1 (OsKASI, 1000-grain weight and tiller number) 60 . All the above mentioned genes provided heterotic effects with heterozygous alleles in UD-type or D-type M-QTLs. However, some M-QTLs detected in this study exhibited a positive overdominance heterosis such as qPN1.2, qPN1.5 and qPN4.3 for increased AH MP s of PN in TCF 1 s; qGYP9 and qGYP12.1 for positive overdominance effect of GYP in both TCF 1 s and AH MP s. They enhanced the performance of TCF 1 s hybrid by increasing PN and GYP. Similarly, qTGW3.4 and qTGW8.2 enhanced TGW and showed a positive overdominance effect in AH MP s. These overdominance heterotic QTLs identified in this study across different populations would be highly valuable for breeding to enhance grain yield of hybrid rice by marker-assisted selection.

Materials and methods
Experimental materials. The rice populations used in this study included a set of 209 F 2 : 10 RILs derived from single-seed descent from a cross between a photosensitive late japonica variety XS09 developed in China and an indica inbred line IR2061 developed at IRRI, and a set of 222 BC 2 F 8 XSILs under XS09 background with IR2061 as a donor. Then two testcross F 1 populations were developed by crossing the RILs and XSILs to a common maternal tester line PA64S, which is a stable indica thermo-sensitive genic male sterile line with excellent wide compatibility to both indica and japonica cultivars 61 and has been extensively used as a sterile line in twoline hybrid rice breeding programs in China. The first testcross population consisting of 209 RILTCF 1 s from crosses between the RILs (used as male) and PA64S. The second one consisted of 222 XSILTCF 1 s from crosses between the XSILs (used as male) and PA64S. In addition, the parents XS09 and IR2061, PA64 (the recurrent parent of PA64S, an isogenic line of PA64S), two F 1 s (XS09 × IR2061 and PA64S × XS09) were used as checks in the phenotyping experiments. In the LS experiment, the RILs, XSILs, the two testcross populations (RILTCF 1 s and XSILTCF 1 s), the parents (XS09, IR2061 and PA64S) and their F 1 s (XS09 × IR2061 and PA64S × XS09) were sowed in the seedling nursery on November 25, 2014. The 25-day-old seedlings were transplanted into four-row plots each consisting of a single row of the male RIL, XSIL and the two testcross hybrids (RILTCF 1 and XSILTCF 1 ). The plots were arranged in a randomized complete block design with three replications. Each row within a plot consisted of 12 plants with a spacing of 17 cm between the plants and 25 cm between rows. Five check plots consisting of XS09, IR2061, PA64S, and XS09 × IR2061 F 1 and PA64S × XS09 F 1 were randomly arranged in each replication. In the WZ experiment, materials were sowed in the seedling nursery on June 15, 2015, and the 25-day-old seedlings were transplanted into four-row plots each consisting of a single row of a RIL, XSIL and the two testcross hybrids. The field arrangement in WZ was the same as the LS experiment. In addition, five check plots consisting of XS09, IR2061, PA64, XS09 × IR2061 F 1 , and PA64S × XS09 F 1 were included in each replication. Crop management followed local field production practices in the two sites. At maturity, PN was investigated from the middle 10 plants. Total grain number per panicle was measured from 10 main panicles from the middle 10 plants (one main panicle each plant) each plot. The grain yield each plot was obtained after weighing all grains collected from the rest panicles and the 10 main panicles of the middle 10 plants. Then GYP (g) was calculated by the ratio of grain yield each plot to 10. FGNP was calculated by the ratio filled grain number each plot to 10. An estimate of the TGW (g) was made by weighing three lots of 100 grains per entry.
For each testcross F 1 , AH MP and the relative H MP was calculated as AH MP = F 1 − MP and H MP (%) = (F 1 − MP)/ MP × 100, respectively, where F 1 is the trait value of a testcross F 1 and MP is the mean value of the corresponding paternal RIL or XSIL and the common maternal tester line PA64S in LS and PA64 (an isogenic line of PA64S) in WZ. PA64S was replaced by PA64 for trait measurement because the former shows sterility in later season in WZ, where the temperature is over 23.5 °C at the panicle differentiation stage (a crucial stage of fertility transformation of two-line sterile lines. SNP genotyping. Genomic DNA for SNP genotyping was isolated from approximately 100 mg fresh leaf samples of 5-week-old seedlings for the 209 RILs, 222 XSILs, the parents XS09 and IR2061, and a tester line PA64S using a modified cetyltrimethylammonium bromide (CTAB) method 62  www.nature.com/scientificreports/ screened based on polymorphism between XS09 and IR2061. Among them, 25,296 high-quality non-redundant SNPs were finally selected for genotype analysis (Fig. S1). Genotypes of TCF 1 s at 25,296 SNPs were determined based on the SNP genotypes of the corresponding RILs or XSILs and PA64S. Specifically, if the two parents (RIL or XSIL and PA64S) have the same homozygous genotype, their TCF 1 s shared the same genotype, and if the two parents have different homozygous genotypes, the genotypes of their TCF 1 s were deduced as heterozygotes.
Genetic linkage map construction and QTL mapping. Filtered and high-quality SNPs with less than 10% missing were used for the construction of bin maps for each population using the BIN function in QTL IciMapping Version 4.2 65 . For non-redundancy, only one SNP was retained to represent each bin, either one with a minimum missing rate, or a random one when the missing rate was equal. The SNPs which displayed a unique pattern of segregation and did not fall into a bin were removed. We then constructed the linkage map of each population using these bins by the MAP function in QTL IciMapping Version 4.2 65 . The values obtained for the recombination frequencies were converted into map distance by the Kosambi mapping function 66 . A total of 1756 bins were used to construct two high-density linkage maps with 2017.1 cM and 1082.9 cM for RIL and XSIL populations, respectively. The genotypes for each cross in the RILTCF 1 s and XSILTCF 1 s were deduced from the RILs/XSILs and the original parents that were used as the parents for the crosses.
QTL mapping for four yield-related traits was performed separately for the RIL, XSIL, RILTCF 1 , and XSILTCF 1 populations. For the RIL and XSIL datasets, the mean trait values from three replications at each location were used as input data. For RILTCF 1 and XSILTCF 1 populations, the mean trait values and AH MP s of the TCF 1 s were used as input data. All datasets were analyzed using the biparental populations (BIP) function in QTL IciMapping Version 4.2 65 . The inclusive composite interval mapping of additive (ICIM-ADD) QTL method was performed to identify M-QTLs by using default settings. The analyses of M-QTLs were performed with pre-adjusted IciMapping parameters, in which the P values for entering a variable (PIN) were set at 0.001 and the scanning step was set at 1.0 cM. The inclusive composite interval mapping of the digenic epistatic (ICIM-EPI) QTL method was used to find possible digenic E-QTLs by using default settings. The corresponding scan step and PIN for E-QTLs mapping were set at 5 cM and 0.0001, respectively. The LOD threshold values 3.0 and 5.0 were used to declare significant M-QTLs and E-QTLs, respectively. The physical position of a QTL was retrieved based on the left and right markers of the detected interval. The known genes underlying the related traits within an identified QTL interval were considered as candidate genes based on the Nipponbare reference genome (IRGSP 1.0) 67 . The QTLs were named as "q + trait abbreviation + chromosome number + QTL number" following the rules recommended by McCouch and CGSN 68 . The type of digenic epistasis without M-QTL was ignored.  18,69 : additive (detected only in lines or TCF 1 s) QTLs, heterotic (D-type, OD-type, and UD-type) M-QTLs. D-type and OD-type QTLs were determined using the values of d/a and a + d. QTLs with 0 <|d/a|< 1 or |2d/(a + d)|≤ 1 or |2a/(a + d)|≥ 1 were designated as D-type QTLs. QTLs with 0 <|d/a|< 1 or |2d/(a + d)|> 1 or |2a/ (a + d)|< 1, or those detected only in AH MP datasets, were designated as OD-type QTLs. QTLs detected in AH MP datasets but showing negative 'd' values were defined as UD-type QTLs. Of the detected QTLs, only D-type, OD-type, and UD-type M-QTLs were used for subsequent comparative studies and direct effect analysis. In this study, all these heterotic QTLs are genetic loci underlying heterosis of yield-related traits in rice.
Statistical analysis. Analyses of variance were performed to determine significant variation between locations and genotypes for all measured traits by the Agricolae Package in R. Significant phenotypic differences among the check parents and the relative hybrids using Duncan's multiple comparison test, and among RILs, XSILs and the TCF 1 s were statistically assessed using Student's t-test by the agricolae package in R. Here, RILs and XSILs when combined were termed "lines". H MP was tested with a Student's t-test based on the contrast between F 1 hybrid mean and average performance of corresponding parental lines 70 . Pearson's correlation analyses among the phenotypic traits measured were performed by the Hmisc Package in R.