Identification, pyramid, and candidate gene of QTL for yield-related traits based on rice CSSLs in indica Xihui18 background

Chromosome segment substitution line (CSSL) is important for functional analysis and design breeding of target genes. Here, a novel rice CSSL-Z431 was identified from indica restorer line Xihui18 as recipient and japonica Huhan3 as donor. Z431 contained six segments from Huhan3, with average substitution length of 2.12 Mb. Compared with Xihui18, Z431 increased panicle number per plant (PN) and displayed short-wide grains. The short-wide grain of Z431 was caused by decreased length and increased width of glume cell. Then, thirteen QTLs were identified in secondary F2 population from Xihui18/Z431. Again, eleven QTLs (qPL3, qPN3, qGPP12, qSPP12, qGL3, qGW5, qRLW2, qRLW3, qRLW5, qGWT3, qGWT5-2) were validated by six single-segment substitution lines (SSSLs, S1-S6) developed in F3. In addition, fifteen QTLs (qPN5, qGL1, qGL2, qGL5, qGW1, qGW5-1, qRLW1, qRLW5-2, qGWT1, qGWT2, qYD1, qYD2, qYD3, qYD5, qYD12) were detected by these SSSLs, while not be identified in the F2 population. Multiple panicles of Z431 were controlled by qPN3 and qPN5. OsIAGLU should be the candidate gene for qPN3. The short-wide grain of Z431 was controlled by qGL3, qGW5, etc. By DNA sequencing and qRT-PCR analysis, two best candidate genes for qGL3 and qGW5 were identified, respectively. In addition, pyramid of different QTLs in D1-D3 and T1-T2 showed independent inheritance or various epistatic effects. So, it is necessary to understand all genetic effects of target QTLs for designing breeding. Furthermore, these secondary substitution lines improved the deficiencies of Xihui18 to some extent, especially increasing yield per plant in S1, S3, S5, D1-D3, T1, and T2.


Introduction
Rice (Oryza sativa L.) provides a staple food source for more than half of the world's population (Ranjith et al. 2020). There has been a considerable improvement in rice yield by traditional breeding strategy in the past few decades. However, with the growth of population and the decreasing of available field, improving rice yield is still the main goal of rice breeding (Li and Li 2016). At the beginning of the post-genomic era, a new concept of ''breeding by design" was proposed, which aims to control all allelic variations for all genes of agronomic importance. The core of the design breeding is the use of naturally occurring variation (Zhang. 2021). Thus, the appropriate creation of natural variation and identification of favorable quantitative trait loci (QTLs) for agronomic traits are important.
In the last 3 decades, QTLs for yield traits in rice have been identified on almost all 12 chromosomes using primary segregating populations, such as F 2 lines, recombinant inbred lines (RILs), and doubled haploid lines (DHs) (Xing and Zhang 2010). However, seldom of these QTLs were cloned due to non-accurate QTL mapping caused by excessive genetic background noise in a primary population . To resolve this problem, advanced populations such as chromosome segment substitution lines (CSSLs) have been used in the identification of QTL for a wide range of traits in food and commercial crops (Wu et al. 2020). CSSLs are genetic stocks representing the complete genome of any genotype in the background of a cultivar as overlapping segments. Generally, each CSSL has a single or several specific marker-defined chromosome segments from the donor with a maximum recipient parent genome recovered in the background (Balakrishnan et al. 2019). When each line only carries a single substitution segment, it can be known as single-segment substitution lines (Zhang. 2021). Particularly, CSSLs or NILs (near-isogenic lines) are valuable prebreeding tools for broadening the genetic base of existing cultivars and a powerful platform for breeding by design (Balakrishnan et al. 2019;Zhang. 2021). So far, the efficiency of QTL fine mapping and cloning has been improved greatly using CSSLs or NILs, such as GS3 for grain size , GL3.1 for grain length (Qi et al. 2012), qGL3, qGL6, and qGL3-2 for grain length (Kashif et al. 2020;Zhang et al, 2020;Liang et al. 2021), qKL3 for kernel length , GW7 for grain width (Wang et al. 2015), qGW1-2, qGW3-2, and qGW4-1 for grain width (Li et al. 2019), GW2 for grain weight (Song et al. 2007), GW5 for grain width and grain weight (Liu et al. 2017), GW6a for grain weight (Song et al. 2015), qGWT5 and qTGW11 for 1000-grain weight (Okada et al. 2018;Wang et al. 2021b), Gn1 for grain number (Ashikari et al. 2005), qSP1 for spikelets per panicle (Ma et al. 2019), qGNP7(t) for grain number per panicle (Hu et al. 2018), qSSD5 and qSSD8 for seed setting density (Wang et al. 2021a), and qPPP2 for panicles per plant (Tao et al. 2016). Although these QTLs provided some candidate genes or revealed some molecular regulatory mechanisms of rice yield traits. However, they are insufficient to explain the complexity of the genetic mechanism. Again, due to the separating of favorable alleles in different varieties Dhat et al. 2021), it is necessary to identify more QTLs for yield traits using novel CSSL developed from different elite cultivars.
Restorer line is key in hybrid rice breeding. Xihui18 as an indica rice restorer line was bred by 1 3 Vol.: (0123456789) the Rice Research Institute of Southwest University. It had many advantages as high general combining ability, good flowering habit and long panicles, and multiple grains per panicle. However, the number of panicles per plant in Xihui18 is few and the grains are long and narrow. Huhan3, a japonica cultivar, is characterized by multiple panicles, short and wide grains, and strong resistance to stress. Here, we identified a rice CSSL-Z431, which is mainly characterized by short-wide grain and multiple panicles per plant, derived from Xihui18 as recipient and Huhan3 as donor parent. In this study, we will clarify the following theoretical problems: since Z431 contained 6 substitution segments, how many QTLs will respond to the different traits in Z431, and how are they distributed in these substitution segments? If QTLs controlling the same trait are more than one, will they exist independent inheritance or epistatic interaction? Which of these QTLs were reported or novel? Hence, we analyze Z431 systematically and map QTL using secondary F 2 population from a cross between Xihui18 and Z431, as well as verify the QTL with the developed SSSL and analyze the inheritance model and pyramid effect of target QTLs using the DSSL and TSSL. Finally, we also analyze the candidate genes of major target QTL. It is important to understand the genetic information of target QTL for design breeding.

Development of Z431
The rice CSSL Z431 was developed using Xihui18 as the recipient parent and Huhan3 as the donor parent. Xihui 18 is an indica rice restorer line bred by Southwest University, China. Huhan3 is a japonica rice cultivar of China. Firstly, 241 polymorphic markers between them were screened from 429 simple sequence repeat (SSR) markers that covered the whole rice genome. Then, they were used to develop CSSLs from BC 2 F 1 to BC 3 F 7 , as showed in Fig. 1. Finally, a novel CSSL Z431 with six substitution segments was identified in BC 3 F 7 , which displayed multiple panicle numbers per plant and short-wide grain.
The identification of substitution segments in Z431 was conducted according to the method of Ma et al. (2019). The estimated substitution length was calculated by the method of Paterson et al. (1991). The chromosome map was constructed using the Map chart 2.32 software (https:// www. wur. nl/ en/ show/ Mapch art. htm).

Field planting
In July of 2018, Xihui18 was crossed with Z431 to get a hybrid at Xiema of Chongqing experimental station of Southwest University of China. In September of the same year, the hybrid seeds were planted to get F 1 seeds at the Lingshui experimental base of Hainan province. On March 10 of 2019, Xihui18, Z431, and the F 2 population were planted at the experimental station of Southwest University of Chongqing, China. On April 13, thirty plants of Xihui18 and Z431 and 150 individuals of the F 2 population were transplanted in the same field. The spacing between the hills and rows was 16.67 cm and 26.67 cm. On March 12 of 2020, Xihui18, Z431, and eleven individuals selected from the F 2 population for the development of SSSL, DSSL, and TSSL were planted into the same experimental field and transplanted 30 plants for each material on April 15 of the same year. On March 10 of 2021, Xihui18, Z431, and 6 SSSLs (single-segment substitution lines), 3 DSSLs (dualsegment substitution lines), and 2 TSSLs (triplesegment substitution lines) developed in the F 3 generation were planted into the same experimental field and transplanted 30 plants for each material on April 17 of the same year. Conventional field management practices were applied.

Measurement of yield-related traits
Ten plants each from Xihui18 and Z431 plots, together with 150 F 2 plants were harvested at the maturity stage. Ten yield-related traits were measured, including the number of panicles per plant, panicle length, number of spikelets per panicle, number of grains per panicle, grain length, grain width, ratio of length to width, 1000-grain weight, seed setting rate, and yield per plant. The methods of measurement were the same with Ma et al. (2020a). Panicle length, number of spikelets per panicle, number of grains per panicle, and seed setting rate were measured using all the effective panicles of the plant. The length and width of 30 grains for each plant were measured with a Vernier caliper and then calculated the mean value and ratio of grain length to width for each plant. The 1000-grain weight of Xihui18 and Z431 was measured from random samples of 3000 grains, from which 1000-grain subsets were weighed on an electronic balance, with three repetitions. The 1000-grain weight of each F 2 plant was determined as the weight of 200 grains, multiplied by 5, with three repetitions. Yield per plant was weighed by electronic balance. The mean phenotypic value and t-test of the above traits for Xihui18 and Z431 were calculated in Microsoft Excel 2016.

Scanning electron microscopy
At the completion of the booting stage and before the heading period, the phenotypic characteristics of the inner and outer epidermal cells of the glume in Xihui18 and Z431 were investigated using a Hitachi SU3500  scanning electron microscope (Hitachi, Tokyo, Japan) with a frozen stage (− 40 °C) under a low-vacuum environment ). The total cell number between Xihui18 and Z431 in the outer epidermis of the glume along the longitudinal axis was counted in applicable pixel size (1280 × 960). At the same magnification (50.0 mm × 320 SE), we counted the protuberances (a protuberance represents a cell) with three repetitions, except the ones on the boundary.

QTL mapping
A secondary F 2 population with 150 individuals was used for QTL mapping, which was derived from a cross between Xihui18 and Z431. Total genomic DNA of Xihui18, Huhan3, Z431, and the 150 individuals from the F 2 population was extracted using the cetyl-trimethyl-ammonium-bromide (CTAB) method. PCR amplification, non-denaturing polyacrylamide gel electrophoresis, and rapid silver staining were conducted as the methods described by Zhao et al. (2016a). Bands same with Xihui18 were scored as " − 1," bands same with Z431 were scored as "1," heterozygous bands were scored as "0," and the absence of marker bands was scored as "." The marker assignments of all SSR markers on the substitution segments of Z431, together with the phenotypic values of each individual in the F 2 population, were used for QTL mapping. QTL mapping was performed using the restricted maximum likelihood method by mixed linear models (REML) implemented in the HPMIXED procedure of SAS 9.3 (SAS Institute Inc, Cary, NC, USA), with significance determined at α = 0.05. SSSLs, DSSLs, TSSLs, and verification and pyramid of QTLs using SSSLs, DSSLs, and TSSLs According to the QTL mapping, 11 individuals from the F 2 population were selected to develop SSSLs, DSSLs, and TSSLs by MAS and grown as the lines Z801 ~ Z811 in 2020, with 30 plants each. Then, 11 homozygous secondary segment substitution lines developed from F 3 were planted as lines, with 30 plants each. Finally, 10 individuals from the middle of each plot were sampled randomly from Xihui18 and SSSLs, DSSLs, and TSSLs after maturity, and ten yield-related traits above involved were measured.

Development of
As for each SSSL i (S1-S6), given that the theoretical hypothesis H0: no QTL existed on the substitution segment of SSSL i . Then, when the probability value was less than 0.05 by Student's t test, we denied the hypothesis and considered that a QTL for a certain trait existed in SSSL i . According to the genetic model under certain environment (same year and same experimental field and no replicate plot designed), P 0 = μ + ε for Xihui18, and P i = μ + a i + ε for SSSL i , where P 0 and P i represent the phenotype value of any plant in the plot of Xihui18 and the SSSL i carrying the "i" substitution segment. μ represents the mean value of Xihui18, a i represents the additive effect of the QTL, and ε represents the random error. The additive effect of the QTL was calculated as half the difference between the mean phenotypic values of SSSL and Xihui18 (Wang et al. 2021a;Liang et al. 2021). All calculations were conducted in Microsoft Excel 2016.
As for each DSSL ij and TSSL ijk , given that the theoretical hypothesis H0: QTL1 located in "i" substitution segment and QTL2 located in "j" substitution segment and QTL3 located in "k" substitution segment belonged to independent inheritance, showed as "2 + 0 = 1 + 1" and "3 + 0 + 0 = 1 + 1 + 1." Then, when the probability value was more than 0.05 by Student's t test, we accepted the hypothesis that QTL1, QTL2, and QTL3 belonged to independent inheritance. At this time, the phenotype value of (Xihui18 + DSSL ij ) was the same with (SSSL i + SSSL j ), as well as (Xihui18 + Xihui18 + TSSL ijk ) same with (SSSL i + SSSL j + SSSL k ). However, when the probability value was less than 0.05 by Student's t test, we denied the hypothesis and considered that epistatic interaction occurred between allelic QTL1, QTL2, and QTL3. According to the genetic models, P ij = μ + a i + a j + I ij + ε for DSSL ij and P ijk = μ + a i + a j + a k + I ijk + ε for TSSL ijk , respectively, where P ij and P ijk represented the phenotype value of any plant in the plot of the DSSL ij and TSSL ijk , a i , a j , and a k represent the additive effect of the QTL in "i," "j," and "k" substitution segments, respectively. I ij and I ijk represented the a i a j epistatic effect between QTLs in "i" and "j" substitution segments of DSSL ij and a i a j a k epistatic effect between QTLs in "i," "j," and "k" substitution segments of TSSL ijk , respectively. Thus, the epistatic effect between QTLs in DSSL ij and TSSL ijk could be estimated as half of the mean phenotypic values of [(Xihui18 + DSSL ij ) 1 3 Vol:. (1234567890) -(SSSL i + SSSL j )] and half of the mean phenotypic values of [(Xihui18 + Xihui18 + TSSL ijk ) -(SSSL i + SSSL j + SSSL k )], respectively (Wang et al. 2021a;Liang et al. 2021).
Candidate gene prediction and DNA sequencing of qPN3, qGL3, and qGW5 At the substitution interval of three major QTLs, including qPN3, qGL3, and qGW5, we analyzed the candidate gene information by the Rice Annotation Project (https:/rapdb.dna.affrc.go.jp/) and the China National Rice Database Center (http:/www. riced ata. cn/) and the Gramene (http:/www. grame ne. org/). As for possible genes, the whole sequence was downloaded, and the primers were designed on Vector NTI to amplify the target fragments using the DNA of Xihui18 and corresponding SSSLs as templates, respectively. The PCR products were forwarded for sequencing to Tsingke biological technology Co, Ltd (Chongqing, China).

Total RNA extraction and qRT-PCR analysis
Total RNA was extracted from the root, stem, leaf, sheath, and panicle of Xihui18 and Z431 using the RNAprep pure plant RNA purification kit (Tiangen, Beijing, China). The first-strand complementary cDNA was synthesized from 2 µg of total RNA using oligo (dT)18 primers in 20 µL of reaction volume using the Prime script reagent kit with gDNA Eraser (Takara, Dalian, China). The qRT-PCR analysis was performed with a 7500 real-time PCR system (Applied Biosystems, Carlsbad, CA, USA) and a SYBR premix Ex Taq II kit (TaKaRa). Rice gene Actin (LOC_Os03g50885) was used as the internal control to normalize all data. Each set of experiments was repeated three times. All primer pairs used for sequencing and qRT-PCR are listed in Table S1.

Identification of substitution segments and phenotype analysis Z431
Based on the development of Z431 using 241 polymorphic SSR markers, we further identified the substitution segments and the genetic background of 10 plants of Z431 using 21 SSR markers on the six substitution segments of Z431 and 220 SSR markers outside of the substitution segments. The results showed that the substitution segments of 10 plants of Z431were identical and no other residual segments from Huhan3 were detected. Z431 contained six substitution segments from Huhan3, distributed on chromosomes 1, 2, 3, 5, and 12. The chromosome 5 contained two substitution segments and the other chromosomes each contained one (Fig. 2). The total length of these substitution segments in Z431was estimated as 12.71 Mb. Among them, the length of the longest substitution segment of Z431 was estimated as 3.76 Mb, the shortest one was 0.95 Mb and the average length of these substitution segments was 2.12 Mb (Fig. 2).
Compared with Xihui18, the heading date of Z431 was earlier 8 d than that of Xihui 18 (115 d) in Chongqing, China. The panicle length, number of grains per panicle, number of spikelets per panicle, grain length, and ratio of length to width in Z431 decreased significantly by 20.37%, 38.60%, 37.77%, 1.37%, and 33.24%, respectively. The number of panicles per plant, grain width, 1000-grain weight, and yield per plant of Z431 increased significantly than Xihui18 by 80.95%, 30.00%, 16.55%, and 34.11%, respectively (Fig. 3). There was no significant difference in seed setting rate (86.40%) between Z431 and that (87.99%) of Xihui18. Thus, the increased yield per plant of Z431 was caused by the increase in the number of panicles per plant and grain weight.

Cytological analysis of Z431
To examine the factors responsible for the decrease in grain length and increase in width of Z431, scanning electron microscopy was used to observe the cell morphology of glumes in Xihui18 and Z431 before the heading stage. The cell length of the glume in Z431 was 88.06 μm, shorter than that (178.60 μm) of Xihui18. While the cell width in Z431was 40.00 μm, wider than that (29.00 μm) of Xihui18 . There was no significant difference in the cell number of a certain glume area along the longitudinal axis between Xihui18 and Z431 ( Fig. 4e-f, i). These results indicated that the short-wide grain of Z431 was caused by the increase in cell width and decrease in cell length and not by cell number of glume.  Identification of QTL for yield-related traits and correlation analysis A total of thirteen QTLs were detected in the secondary F 2 population from a cross between Xihui18 and Z431. These QTLs were distributed on five substitution segments of Z431 and explained the phenotypic variation from 2.10 to 30.75% (Table 1). Among them, there were one QTL for panicle length, number of panicles per plant, number of spikelets per panicle, number of grains per panicle, grain length, and grain width, respectively, and 3 QTLs for ratio of length to width and 1000-grain weight, respectively. The short panicle of Z431 was controlled by qPL3, which decreased the panicle length by 0.89 cm. qPN3 increased the number of panicles per plant of Z431 by 0.51, explained 6.75% of the phenotypic variation. The less number of grains and spikelets per panicle of Z431 were responded by qGPP12 and qSPP12, whose additive effects were − 7.01 and − 7.25, respectively. qSSR5 decreased seed setting rate by 1.19 percent point, explained 2.35% of the phenotypic variation. qGL3 for grain length from Huhan3 reduced the grain length of Z431 by 0.27 mm, explained 30.75% of the variation. qGW5 increased the grain width of Z431 by 0.09 mm, explained 2.10% of the phenotypic variation. The grain shape of Z431 was controlled by qRLW2, qRLW3, and qRLW5, while qRLW2 increased the ratio of length to width by 0.06 and qRLW3 and qRLW5 decreased the trait by 0.08 and 0.06. The 1000-grain weight of Z431 was controlled by two major QTL (qGWT3, qGWT-5-2) and one minor QTL (qGWT-5-1) and the additive effects from alleles of Huhan3 decreased the 1000grain weight of Z431 for qGWT3 by 1.09 g, and increased it by 0.84 and 0.76 g for qGWT-5-1 and qGWT-5-2, respectively. In addition, we also found that some QTLs were always detected in cluster, for example, qPN3, qPL3, qGL3, qRLW3, and qGWT3 linked to the same marker, so was qGPP12 and qSPP12, as well as qSSR5, qRLW5, and qGWT-5-1. Whether these traits correlated? We analyzed the Pearson correlation coefficient for 10 yield-related traits in the F 2 population by IBM SPSS Statistics 26 ( Table 2). The number of panicles per plant (PN) displayed significantly positive correlations with the number of grains (GPP) and spikelets per panicle (SPP), while no correlation with panicle length (PL), grain length (GL), ratio of length to width (RLW), and 1000-grain weight (GWT) ( Table 2). Panicle length (PL) showed a significantly positive correlation with the number of grains (GPP) and spikelets per panicle (SPP). The number of grains (GPP) exhibited a significantly positive correlation with the number of spikelets per panicle (SPP) and seed setting rate (SSR). Grain length (GL) displayed a significant negative correlation with grain width (r = − 0.414**) and significantly positive correlation with ratio of length to width (r = 0.711**). Grain width also showed a significant negative correlation with the ratio of length to width   (Table 2). Thus, qPN3 should not be controlled by the same gene with qPL3, qGL3, qRLW3, and qGWT3. While qGL3, qRLW3, and qGWT3 belonged to pleiotropy. Also, qGPP12 and qSPP12 were pleiotropy; similarly, qSSR5, qRLW5, and qGWT-5-1 should belong to pleiotropy.
Pyramiding of qPN3 (a = 1.63) and qPN5 (a = 1.43) in D3 yielded an epistatic effect of − 1.86; thus, the genetic effect of PN in D3 was 1.20. Due to 1.20 < 1.43 < 1.63, D3 had less number of panicles (6.75) than those (7.60 and 7.20) of S3 (qPN3) and S5 (qPN5), however, with no significant differences among them (Fig. 5a, c). Pyramiding of qPN3 (a = 1.63) and a substitution locus on chromosome 2 without PN in D1 yielded an epistatic effect of − 1.08, which increased PN genetically by 0.55. Because of 0 < 0.55 < 1.63, the number of panicles (5.60) of D1 was more than that (4.50) of S2 (without QTL for panicles) and significantly less than that (7.60) of S3 (qPN3) (Fig. 5a, c). Pyramiding of qPN5 (a = 1.43) and a substitution locus without PN TSSLs. a Schematic graph of substitution segment lines and QTLs located on SSSLs, DSSLs, and TSSLs. Xihui18 represented the recipient parent of the SSSLs, DSSLs, and TSSLs. b Grains of Xihui18, S1-S6, D1-D3, and T1-T2. Scale bar, 2 mm. c-k Analysis of additive and epistatic effects of QTLs for the number of panicles per plant (c), grain length (d), grain width (e), ratio of length to width (f), 1000-grain weight (g), yield of per plant (h), panicle length (i), the number of grains per panicle (j), the number of spikelets per panicle (k). Different lowercase letters indicate a significant difference (P < 0.05) as determined by Duncan's multiple comparisons. µ represents the mean value of the corresponding trait, a i represents the additive effect of QTL, I represents the epistatic effect of QTLs, P-value in S1-S6 indicates the probability of t-test between Xihui18 and SSSL, when P < 0.05 indicated existing a QTL, P-value in D1-D3 indicates the probability of t-test between (DSSL ij + Xihui18) and (SSSL i + SSSL j ), when P > 0.05 showed existing epistatic interaction and P > 0.05 showed independent inheritance between two loci. P-value in T1-T2 indicates the probability of t-test between (Xihui18 + Xihui18 + TSSL ijk ) and (SSSL i + SSSL j + SSSL k ), when P < 0.05 showed existing epistatic interaction and P > 0.05 showed independent inheritance between three loci. S1: Chr. on chromosome 2 in D2 yielded an epistatic effect of 0.12, which increased PN genetically by 1.55. Owing to 0 < 1.43 < 1.55, the number of panicles (7.60) of D2 was more than that (4.50) of S2 (without PN QTL) and that (7.20) of S5 (qPN3), displaying a significant difference with S2 (Fig. 5a, c). Pyramiding of qPN3 (a = 1.63), qPN5 (a = 1.43), and a substitution locus on chromosome 1 without PN in T1 produced an epistatic effect of − 2.09, which increased the number of panicles per plant genetically by 0.97. On account of 0 < 0.97 < 1.43 < 1.63, the number of panicles (7.20) in T1 was significantly more than that (5.25) of S1 and less than that (7.60) of S3 (qPN3) and equal to that (7.20) of S5 (qPN5), however, without significant difference with S3 and S5 (Fig. 5a, c). Pyramiding of qPN3 (a = 1.63), qPN5 (a = 1.43), and a substitution locus without PN on chromosome 2 in T2 produced an epistatic effect of − 2.02, resulting in the increase of 1.04 number of panicles genetically. Due to 0 < 1.04 < 1.43 < 1.63, T2 exhibited significantly more panicles (6.60) than that (4.50) of S2, and less than those (7.60 and 7.20) of S3 (qPN3) and S5 (qPN5), with no significant difference with S3 and S5 (Fig. 5a, c).
As for panicle length, the number of grains per panicle, and the number of spikelets per panicle, all the loci in D1, D2, D3, and T1 and T2 displayed independent inheritance (Fig. 5a, i, j, k).
Candidate gene analysis of major QTLs-qPN3, qGL3, and qGW5 Considering the larger contribution rates of qPN3, qGL3, and qGW5 for major phenotypic variation of Z431, and also detecting both in F 2 and SSSLs, they deserved our priority attention (Fig. 5a). qPN3 and qGL3 were all located on S3 whose estimated substitution length was 1.48 Mb (Fig. 6c, 7d). qGW5 was located on S5 whose estimated substitution length was 1.68 Mb (Fig. 8d).
About qPN3, we found a reported gene (OsIAGLU) in the substitution interval of S3, which influences the tiller formation by negatively regulating IAA. Thus, we firstly wanted to know whether the cloned OsI-AGLU was the candidate gene for qPN3. By DNA sequencing, we found actually existing four differences of single-nucleotide polymorphisms (SNPs) in CDS of OsIAGLU between Xihui18 and S3. Moreover, three of four SNP differences caused changes in amino acids (Fig. 6c). Furthermore, by qRT-PCR analysis, the expression levels of OsIAGLU in root, stem, leaf, sheath, and panicle in S3 were higher than that in Xihui18, especially in root, leaf, and sheath (Fig. 6d). At the same time, the panicle number of S3 was significantly more than that of Xihui18 (Fig. 6a, b) that was consistent with the function of OsIAGLU. Thus, OsIAGLU should be the candidate gene for qPN3.
Concerning qGL3, considering that the cloned genes related to rice grain size development are mainly involved in phytohormones such as BR, IAA, and CK signal pathway, MAPK signal pathway, protein kinase G protein signaling, the ubiquitin-proteasome pathway, and transcriptional and S3. Data were mean ± SD. ** showed 1% significant lev-els according to the Student t-test. c Position of qPN3 and sequence blast of OsIAGLU between Xihui18 and S3. d Relative expression levels of OsIAGLU between Xihui18 and S3 by qRT-PCR analysis regulatory factors (Li and Li. 2016), we checked all the genes related to the above pathways in this region of qGL3 and found 6 possible genes, including auxin-responsive protein, ABA signal transduction, and serine/threonine-protein kinase-like protein ACR4. Therefore, we preliminarily selected these 6 genes as candidate genes. Then, by the whole DNA sequencing of these genes between S3 and Xihui18, the candidate gene 1 (serine/threonine-protein kinase) and the candidate gene 5 (auxin-responsive protein) showed no differences between the two lines. The result showed that the candidate genes 1 and 5 should not be the candidate genes for qGL3. However, for the candidate gene 2 (an auxin-responsive protein),  there were 2 SNPs differences and 2 bp deletion in CDS of S3 compared with Xihui18, where the 175 th base of CDS changed from C in Xihui18 to T in S3, resulting in Leu of Xihui18 to Phe of S3, and another SNP (from C to T) in the 357 th of CDS without causing amino acid change. The 2 bp deletion in the 401 st base of CDS in S3 caused delaying the termination of the translation (Fig. 7d). However, the qRT-PCR analysis showed that the expression level in the root, stem leaf, sheath, and panicle displayed no significant differences between S3 and Xihui18 (Fig. 7e). These results showed that the candidate gene 2 could be temporary as the candidate gene of qGL3. With regards to the candidate gene 3 (serine/threonineprotein kinase in ABA signal transduction), there was 1 SNP difference with no resulting in amino acid change (Fig. 7d). At the same time, expression levels of the gene displayed no significant differences in the root, stem, leaf, sheath, and panicle between S3 and Xihui18 (Fig. 7f). These results indicated that candidate gene 3 should not act as the candidate gene of qGL3. For the candidate 4 (an auxin-responsive protein), we found differences of 4 SNPs between Xihui18 and S3, where the base C, T, and C in 125 th , 215 th , and 274 th of the CDS in Xihui18 changed to A, G, and A in S3, which caused amino acid mutation from Ala, Val, and Ala in Xihui18 to Glu, Gly, and Glu in S3, respectively. And the other SNP did not cause amino acid changes (Fig. 7d). Intriguingly, the qRT-PCR analysis showed that expression levels of the gene in S3 were significantly higher than in Xihui18 in the leaf, sheath, and panicle (Fig. 7g). About the candidate gene 6 (a serine/threonine-protein kinase-like protein ACR4), there was a deletion of base T in 1900 th of CDS in S3, which caused shift mutation. In addition, there were still 3 SNP differences (Fig. 7d). Simultaneously, the qRT-PCR analysis showed that expression levels of the gene in S3 were significantly higher than in Xihui 18 in the root, stem, leaf, sheath, and panicle (Fig. 7h). In conclusion, candidate genes 4 and 6 should be the best candidate genes of qGL3, and candidate gene 2 also acted as its candidate gene. qGW5 was located on the substitution segment of S5, whose grain width and brown grain were significantly wider than that of Xihui18 (Fig. 8a, b, c). Since grain width also being within the grain size range, the candidate genes for qGW5 could be found by the above signal pathway of grain size. Firstly, we looked for all the genes within the substitution interval of 1.68 Mb for qGW5 by the Gramene (http:/www. grame ne. org/), then referred to the functions of every gene by Rice annotation project (https:/rapdb.dna. affrc.go.jp/) and the China national rice database center (http:/www. riced ata. cn/). Finally, we found 6 potential candidate genes according to the probable genes involved in the signal pathway of grain size (Li and Li. 2016). By DNA sequencing, we found that the candidate gene 3 (DNA-methyltransferase CMT2), candidate gene 4 (auxin-responsive protein), and candidate gene 5 (serine/threonine-protein kinase) did not exist any sequence differences between Xihui18 and S5, while the other three exhibited differences between Xihui18 and S5. The candidate gene 6 (E3 ubiquitin-protein ligase) displayed 3 SNP differences in the CDS between Xihui18 and S5. However, they all belonged to nonsense mutation (Fig. 8d). Again, the qRT-PCR analysis showed that there was also no significant difference in the expression levels for the gene in the root, stem, leaf, sheath, and panicle between S5 and Xihui18 (Fig. 8g). Thus, the candidate genes 3, 4, 5, and 6 should not be the candidate genes of qGW5. For the candidate 1 (RING-type E3 ubiquitin transferase), there were 2 SNP differences in the CDS between Xihui18 and S5, where the base G and T in the 62 nd and 175 th of CDS in Xihui18 changed to C and C in S5, which caused amino acid mutation from Gly and Met in Xihui18 to Ala and Thr in S5 (Fig. 8d). The qRT-PCR analysis showed that the expression levels of the gene were significantly higher in panicle, while lower in the root, stem, leaf, and sheath of S5 than those in Xihui18 (Fig. 8e). With regard to the candidate 2 (eukaryotic translation initiation factor 3 subunit), there were 3 SNP differences in the CDS between Xihui18 and S5. The 1500 th base changed from G in Xihui18 to C in S5, resulting in amino acid mutation from Leu to Phe, whereas the other SNP did not cause amino acid changes (Fig. 8d). Furthermore, the qRT-PCR analysis showed that its expression levels were significantly higher in the stem, leaf, sheath, and panicle of S5 than that of Xihui18 (Fig. 8f). In conclusion, the candidate gene 1 (RING-type E3 ubiquitin transferase) and candidate gene 2 (eukaryotic translation initiation factor 3 subunit) should be the best candidate genes for qGW5.

Discussion
Z431 and its secondary substitution lines are potential to be used as novel rice restorer line in breeding novel hybrid rice varieties Successful breeding of hybrid rice is an outstanding breakthrough in rice breeding history, which has dramatically improved the theory and techniques for crop genetics and breeding (Cui et al. 2020). To date, most of the hybrid rice varieties were interspecific hybrids . Due to the close interspecies relationships and relatively small genetic differences, the heterosis was not strong enough. Thus, the use of heterosis between subspecies and wild species will be imperative. However, direct subspecies crosses often result in sterility and a reducing seed setting rate due to reproductive isolation (Nadir et al. 2018). The development of indica-japonica rice chromosome segment substitution lines can overcome the limitation (Singh et al. 2020). Excellent restorer line is an essential part of heterosis utilization, which often contains the main fertility recovery genes Rf-1 (Akagi et al. 2004), Rf2 (Etsuko et al. 2011), Rf3 (Cai et al. 2013), and Rf4 (Tang et al. 2014). Xihui18 was an elite rice indica restorer line with strong combining ability, long panicle with multiple grains, and long-narrow grain. However, its number of panicle per plant was fewer (4.26 for average in 2019 and 2021). In this study, a novel CSSL-Z431 containing six substitution segments was identified from Xihui18 as the recipient parent and Huhan3 as the donor parent. Compared with Xihui18, Z431 had panicle number per plant of 7.56 (average value in 2019 and 2021) and short-wide grains. Again, six SSSLs (S1-S6), three DSSLs (D1-D3), and two TSSLs (T1 and T2) were developed by MAS in the F 3 generation. One of the notable findings of this study was that all the fertility-restoring genes Rf1 ~ Rf4 were all not substituted in Z431, SSSLs, DSSLs, and TSSLs, especially to some extent these substitution lines make up for the deficiencies of Xihui18. For example, the number of panicles per plant of S3, S5, D1, D2, D3, T1, and T2 was significantly more than that of Xihui18. Accordingly, the yield per plant (33.28, 34.82, 41.25, 42.70, 31.24, 41.74, 40.45, 38.12, and 37.18 g) of Z431, S1, S3, S5, D1, D2, D3, T1, and T2 were all significantly increased than that (27.66 g) of Xihui18. Seed setting rate of these secondary substitution lines were 87.53% for S1, 85.96% for S2, 85.14% for S3, 88.79% for S4, 91.78% for S5, 91.12% for S6, 87.44% for D1, 94.73% for D2, 93.13% for D3, 89.38% for T1, and 85.52% for T2 (calculated by the data in Fig. 5 j and k). Furthermore, the grain size of these lines displayed diversity. Therefore, Z431 and these secondary substitution lines have the potential to be used as novel restorer lines for breeding novel rice hybrid varieties.
Identification of QTL for yield-related traits of Z431 and comparison with the reported genes Since Z431 had still six chromosomal substitution segments from donor Huhan3 and significant differences in nine yield-related traits with recipient parent Xihui18. Then, which QTLs are responsible for these different traits in these substitution segments? We further identified 13 QTLs for these traits using the secondary F 2 population from crosses between Xihui18 and Z431. Among them, eleven QTLs could be validated by 6 developed SSSLs (S1-S6). In addition, fifteen QTLs could be detected by these SSSLs, while were not identified in the F 2 population. Thus, there were in total 28 QTLs identified. Among them, qGL1, qGW1, qRWL1, and qGWT1 were all linked with RM283 (5.27 Mb) in S1. Again, these traits all displayed significant correlations. Thus, these QTLs should belong to pleiotropy. At this substitution interval, we found OsOFP1 (7.01 Mb) distancing 1.78 Mb from RM283. OsOFP1 negatively regulates cell reproduction and elongation by the BR signal transduction pathway. The seeds of overexpressing OsOFP1 transgenic plants became shorter, wider, and thicker . Similarly, the grains of S1 also exhibited significantly shorter, wider, and heavier than that of Xihui18. Thus, OsOFP1 could act as the candidate gene for qGL1 etc. Similar with qGL1, these genes as qGL3, qRLW3, qGWT3, qPL3, and qPN3 were all linked to RM6266 (27.64 Mb) on chromosome 3 in S3. However, PN displayed no significant correlation with PL, GL, RLW, and GWT, and the others had significant correlations with each other. Thus qPN3 should not be controlled by the same gene with qPL3, qPN3, qGL3, qRLW3, and qGWT3, and the latter should be pleiotropy. At this substitution interval, qPN3 should be an allele of OsIAGLU (27.78 Mb). OsIAGLU encodes IAA-glucose synthase, which is known to generate IAA-glucose conjugate from free IAA. The number of tillers and panicles significantly increased in the transgenic lines of overexpression of OsI-AGLU compared to the wild-type plants (Choi et al. 2012). By DNA sequencing of OsIAGLU, there were actual differences of DNA sequences between Xihui18 and S3. Especially, the expression level of OsIAGLU in S3 was significantly higher than that in Xihui18, and S3 displayed significantly increasing panicles number than Xihui18. Thus, OsIAGLU should be the candidate gene for qPN3. Although the gene has been cloned, compared with its overexpression transgenic plants, S3 are more beneficial to be used in rice design breeding. In addition, just as mentioned above, the substitution interval of S3 still contained qGL3, qRLW3, qGWT3, and qPL3. By the cloned genes for grain size, rice grain size development is mainly involved in phytohormone signal pathways such as BR, IAA, and CK, as well as MAPK signal pathway, protein kinase G protein signaling, the ubiquitin-proteasome pathway, and transcriptional regulatory factors (Li and Li. 2016). By checking all the genes with function annotations in this substitution region of S3, we found six possible genes involved in the signal pathway of grain size, including candidate gene 1 for serine/threonine-protein kinase, candidate genes 2, 4, and 5 for auxin-responsive protein, candidate 3 for abscisic acid-activated protein kinase, and candidate 6 for serine/threonine-protein kinase-like protein ACR4. Some genes for grain size have been reported involving these signaling pathways. For example, BG1 as a primary response gene for auxin regulates auxin transport, positively regulating cell division and cell elongation . OsP-PKL1/GL3.1 encodes a protein phosphatase kelch (PPKL) family-Ser/Thr phosphatase (Zhang et al. 2012), which are involved in BR signaling (Mora-García et al. 2004). GL3.1 controls rice seed size and yield by direct dephosphorylation of the substrate cyclin-T1;3, and down-regulation of GL3.1 in rice results in shorter grains, suggesting that GL3.1 might participate in BR signaling (Qi et al. 2012;Li and Li. 2016b). By sequencing, the candidate 1 (serine/threonine-protein kinase) and the candidate 5 (auxin-responsive protein) showed no difference between the two lines. The candidate gene 3 (abscisic acid-activated protein kinase) should not be the candidate genes of qGL3 because there were no amino acid changes, although existing 1 SNP difference in CDS between Xihui18 and S3. As for candidate gene 2 (auxin-responsive protein), there were 2 SNPs differences and 2 bp deletion in CDS of S3 compared with Xihui18, while expression levels of the gene displayed no significant difference, suggesting that candidate gene 2 could temporarily act as the candidate gene of qGL3. Only candidate gene 4 and gene 6 existed both differences in DNA sequence and expression levels between Xihui18 and S3. Therefore, candidate gene 4 (auxin-responsive protein) or gene 6 (serine/threonine-protein kinase-like protein ACR4) should be the best candidate genes of qGL3. It is worth noting that these three genes were still not cloned. qGL5, qRLW5, and qGWT-5-1 were all linked to RM3322 (4.39 Mb). In the substitution interval, OsDER1 (5.39 Mb) existed. According to the reports, suppression of expression of OsDER1 in transgenic rice showed a decrease in grain length and grain width (Qian et al. 2018). Thus, OsDER1 could act as a candidate gene for qGL5, qRLW5, and qGWT-5-1. Similarly, qGW5, qRLW-5-2, and qGWT-5-2 were all linked to RM169 (7.85 Mb) and they could also be detected by Zhang et al. (2020), which indicated that these two QTLs could be stably inherited. In this substitution interval of S5, we found 6 genes involved in the above signal pathway of grain size, including candidate gene 1 for RING-type E3 ubiquitin transferase, candidate gene 2 for eukaryotic translation initiation factor 3 subunit, candidate gene 3 for DNA-methyltransferase CMT2, candidate gene 4 for auxin-responsive protein, and candidate gene 5 for serine/threonine-protein kinase and candidate gene 6 for E3 ubiquitin-protein ligase. There were already some genes for grain size reported involving in these signal pathways. GW8 codes a SBP-domain transcription factor, regulates grain width as a positive regulator, and can bound directly to the GW7 promoter and repressed its expression (Wang et al. 2015). GRAIN WEIGHT 2 (GW2) encodes a RING-type protein with E3 ubiquitin ligase activity. Loss of GW2 increases cell number, which results in broader glumes, thus increasing grain width, weight, and yield (Song et al. 2007). However, the candidate genes 3, 4, 5, and 6 should not be the candidate genes of qGW5 due to no substantial sequence differences between Xihui18 and S5. While candidate gene 1 (RINGtype E3 ubiquitin transferase) and gene 2 (eukaryotic translation initiation factor 3 subunit) should be acted as the best candidate genes for qGW5 owing to both differences of substantial DNA sequence and expression levels between Xihui18 and S5. Interestingly, these two genes are still not cloned. Although qRT-PCR analysis displayed difference or not, the expression pattern of candidate genes did not well explain the difference of target traits, that is no dominant expression. They need further studying by genetic complementary experiment etc. In conclusion, 12 QTLs (qPN5,qGPP12,qSPP12,qGL2,qGL3,qGW5,qRLW2,qRLW3,qGWT2,qGWT3, were still unreported. These results will be important for both functional studies of target QTLs in theory and breeding novel hybrid varieties in application. SSSLs, DSSLs, and TSSLs can improve efficiency of QTL identification and epistatic effects detection between QTLs, thus are ideal materials for favorable gene pyramiding Breeding by design refers to the breeding of varieties by crop design utilizing favorable alleles dispersed in different genetic resources in a genome (Peleman and van der Voort 2003). Zhang (2021) has applied a three-step strategy for research on rice breeding by design based on single-segment substitution lines (SSSLs) library of HJX 74 and argued that target chromosome segment substitution is a way to breed by design in rice. Thus, SSSLs are ideal materials to realize the strategy owing to each SSSL carrying only one single segment from the donor parent in the genome of the recipient. At first, SSSLs can improve the efficiency of QTL identification (Zhao et al, 2016b;Balakrishnan et al, 2019;Wang et al. 2020). In this study, we developed six SSSLs (S1 ~ S6) from progenies of the Xihui18/Z431 F 2 population. And using these SSSLs, we validated eleven QTLs (qPL3,qPN3,qGPP12,qSPP12,qGL3,qGW5,qRLW2,qRLW3,qRLW5,qGWT3, and qGWT5-2) among 13 QTLs which were detected by the Xihui18/Z431 F 2 population. However, qSSR5 and qGWT5-1 could not be validated by SSSLs, these results indicated that some minor QTLs were affected easily by environmental conditions. Paterson et al. (1991) has demonstrated that QTLs for traits with low heritability appear to show a range of sensitivities to the environment, such as yield per plant and seed setting rate. Liu et al. (2008) argued that the total effect of a QTL in any specific environment includes the main effect and the Q × E interaction effect for that environment. Thus, qSSR5 and qGWT5-1 should have a Q × E interaction effect. Zhao et al. (2016b) found that different agronomic traits displayed different Q × E interactions and the Q × E interaction effect was specific to a particular environmental condition. In addition, we identified 15 QTLs (qPN5,qGL1,qGL2,qGL5,qGW1,qRLW1,qGWT1,qGWT2,qYD1,qYD2,qYD3,qYD5,qYD12) by these SSSLs, which were not detected in the F 2 population. These results indicated that SSSLs showed higher QTL detection efficiency. Many pieces of research also supported the conclusion (Zhao et al, 2016a, b;Balakrishnan et al, 2019;Wang et al. 2020;Wang et al. 2021a, b;Liang et al. 2021).
In addition to efficiently detecting single QTL using SSSLs, the epistatic interaction between QTLs could be studied by dual-segment substitution lines (DSSLs) combined with according SSSLs. Due to many important agronomic traits controlled by more than one QTL, thus, it is necessary to check whether these QTLs belong to independent inheritance or epistatic interaction. The epistatic effect between target QTLs is very important for design breeding. Wang et al. (2018) argued that epistasis is an important genetic component affecting the pyramid effect, especially in QTLs for complex quantitative traits. Zou et al (2020) showed that there were significant differences in epistatic effects among seven 2-QTL pyramiding lines, presenting in both directions and magnitudes. The epistatic effects between QTLs for stigma exertion rate (qSERb3-1/qSERb6-1, qSERb3-1/qSERb8-1, and qSERb3-1/qSERb12-1) in 3 pyramiding lines were all positive, which induced very larger pyramiding effects as 51.49%, 43.54%, and 55.51% (Zou et al. 2020). In this study, we analyzed epistatic effects and pyramid performance of many pair QTLs by 6 SSSLs, 3 DSSLs, and 2 TSSLs. The results showed that some QTLs belonged to independent inheritance, such as pyramiding of qGW1 and qGW5, qGWT2 and qGWT3, qGWT2 and qGWT5, qYD2 and qYD5, etc., while most QTLs or loci exhibited epistatic interactions. For example, pyramiding of qGL2 (a = − 0.47) and qGL3 (a = − 0.31) yielded an epistatic effect of 0.62, resulting in grain length (10.03) in D1 significantly longer than that (9.39) of S2 (qGL2). Pyramiding of qYD2 (a = − 1.69), qYD3 (a = 6.72), and qYD5 (a = 7.45) produced an epistatic effect of − 7.79, leading to the yield per plant (37.18 g) of T2 that was significantly higher than that (24.41) of S2 and that (27.66 g) of Xihui18. Pyramiding of qPN3 (a = 1.63) and qPN5 (a = 1.43) in D3 yielded an epistatic effect of − 1.86, resulting in less number of panicles (6.75) than those (7.60 and 7.20) of S3 (qPN3) and S5 (qPN5). Zhang et al. (2020) showed that pyramiding qGL5 (a = 0.30) and qGL6 (a = 0.13) for long grain yielded positive epistatic effects of 0.31 and produced longer grain (8.67 mm) than that (7.80 and 7.46 mm) of the responding SSSLs. Wang et al. (2021a) pyramided qSSD5 (a = 14.10), qSSD8 (a = 11.38), and qSSD10 (a = 5.11) for seed setting density produced an epistatic effect of − 9.36 and resulted in increasing seed setting density. In conclusion, pyramiding different QTLs yielded different genetic models. Hence, in breeding of design, we need to understand the genetic effects of all target QTLs, including the additive effect of single QTL and epistatic effects between QTLs. With the explicit genetic backgrounds of the excellent recipient and known favorable alleles, SSSLs, DSSLs, and TSSLs can be directly used in our design breeding. According to the additive effects and epistatic effects of target genes, we can predict the ideal phenotype of a needful genotype and design personalized novel varieties.

3
Vol.: (0123456789) Author contribution FM Zhao proposed the structure and content. SF Sun completed QTL mapping, cytological analysis, candidate gene analysis, and DNA sequencing and drafted the manuscript. ZB Wang completed the trait correlation analysis, verification, and pyramid of QTLs, as well as qRT-PCR analysis. FM Zhao, SF Sun, SQ Xiang, M Lv, K Zhou, J Li, and PX Liang developed and identified the chromosome segment substitution line Z431 and the development of SSSLs, DSSLs, and TSSLs. RX Li, MM Li, GH He, and YH Ling assessed the agronomic traits. All authors read and approved the final version.
Funding The study was supported by the National Key Research Plan Project (2016YFD0101107). The study was supported by the Chongqing Science and Technology Commission Special Project (CSTC2019jscx-msxmX0392) and the Chongqing Innovation Project of returned from studying abroad.
Availability of data and material All data generated or analyzed during this study are included in this published article.

Declarations
Consent for publication Informed consent for publication was obtained from all participants.

Conflict of interest
The authors declare no competing interests.