Identiﬁcation and Validation of Quantitative Trait Loci for Grain Number in Rice ( Oryza sativa L.)

: Grains number is one of the most important agronomic traits in the determination of rice productivity. To explore the underlying genetic basis of grain number in rice, quantitative trait locus (QTL) analysis was performed using three recombinant inbred line populations derived from indica rice crosses of Teqing / IRBB lines, Zhenshan 97 / Milyang 46, and Xieqingzao / Milyang 46, respectively. A total of 58 QTLs distributed on all 12 rice chromosomes were identiﬁed, including 22 for number of grains per panicle (NGP), 17 for number of spikelets per panicle, and 19 for spikelet fertility. The individual QTL counted for 1.5 to 22.1% of phenotypic variation. Among them, 15 QTLs shared by two or three populations and eight QTLs showed large e ﬀ ects with R 2 larger than 10%. Furthermore, three QTLs with minor e ﬀ ects for NGP, qNGP5.5 , qNGP9.1 , and qNGP12.1 , were detected and validated by eliminating the segregation of major-e ﬀ ect QTL using four residual heterozygote-derived populations. These results not only enrich our understanding of the mechanism of grain number, but also provide a foundation for cloning and selecting candidate for marker-assisted selection breeding in rice.


Introduction
Rice (Oryza sativa L.) is one of the most important cereal crops in the world, playing a key role in meeting the demand of food for a growing global population. During the last 60 years, grain yield has progressively increased in Yangtze River basin, the main indica rice-producing area in China, and such an increase is mainly attributed to the expanded sink size as a result of more number of spikelets per panicle (NSP), especially for the case of super rice [1]. Yang et al. analyzed the data of national rice regional trial in southern China from 1986 to 2002 and found that number of grains per panicle (NGP) played an important role in improving both the yield and rice quality based on suited number of effective panicles and grain weight [2]. Therefore, it is essential to understand the molecular mechanism of grain number in rice.
However, these studies mainly focused on major-effect QTLs, few studies investigated minor-effect QTLs. Recently, studies have shown that QTLs with minor effect also play a role in the regulation of important agronomic traits in rice [30]. Hence, identification of minor-effect QTLs would enrich our knowledge of genetic and molecular network regulating rice grain number. Usually, minor-effect QTLs are more sensitive to genetic background and environment than major-effect QTLs [31,32]. Identification of minor-effect QTLs would be facilitated by eliminating the effect of major-effect QTLs. The use of residual heterozygote (RH) to construct near-isogenic lines (NILs) or secondary segregation populations is an effective strategy for identification and verification of minor-effect QTLs [33,34]. Based on this approach, several QTLs for yield-related trait have been fine-mapped [35,36].
In the present study, QTL analysis for NGP, NSP, and spikelet fertility (SF) was performed using three RIL populations. Three minor-effect QTLs on chromosome 5, 9 and 12 were identified using one RH-derived population fixing the major-effect QTLs and further validated in the new secondary segregation population under more homozygous genetic background, respectively.

Plant Materials
A total of seven populations developed from crossing of indica rice were used. For primary mapping, the previously developed three RIL populations were already used to analyze grain quality and yield-related traits, including Teqing/IRBB lines (TI), Zhenshan 97/Milyang 46 (ZM) and Xieqingzao/Milyang 46 (XM), respectively [10,[37][38][39][40][41]. Teqing, Zhenshan 97 and Xieqingzao are female parents, while IRBB lines and Milyang 46 are male parents. All of the parental lines have been widely used in commercial breeding and production of three-line hybrid rice in China, among which Zhenshan 97 and Xieqingzao are maintainer lines and used as early-season varieties, others are restorer lines and used as middle-season rice varieties in the middle-lower reaches of Yangtze River basin. For TI, phenotypic data of 203 lines in 2009 and 2010 and genetic maps with 127 markers, spanning 1198 cM were used to identify QTLs for yield heterosis [38]. In this study, extra two years' phenotypic data were added, and genetic linkage map had been updated to comprise 135 markers spanning 1345 cM [42]. For ZM, phenotypic data of 243 lines in 1999 and 2000 and genetic map with 158 makers, spanning 1288 cM were used to detect QTLs for yield traits [10]. In this work, phenotypic data for two more years were added. Genetic linkage map had been updated to consist of 256 markers and span 1815 cM [39]. For XM, phenotypic data of 209 lines was not used in any QTL mapping for NGP, NSP, and SF. A genetic map consisting of 240 markers and spanning 2080 cM [39], and phenotypic data of three years were used in this study.
For further mapping and validation, four RH-derived populations were used, and they were derived from one RH of the cross TQ/IRBB52 as described below and illustrated in Figure 1. One F 7 plant carrying 13 heterozygous segments distributed on 10 chromosomes was selected and selfed to produce S 1 population consisting of 251 individuals, which was named Ti52-2. QTLs were determined from data generated from the S 1:2 and S 1:4 families of Ti52-2. QTLs for heading date had been detected using Ti52-2 in previous study [42].

Trials and Phenotypic Evaluation
the populations were grown from May to September in the paddy fields at the Rice Research Institute in Hangzhou, Zhejiang, China. A randomized complete with two replications was used for all trials. In each replication, one line was grow of 12 plants. The planting density was 16.6 cm × 26.7 cm in all trials. Field manag local agricultural practice. At maturity, five middle plants of each line were harve hree RIL populations, and ten middle plants of each line for other populations. NGP ere measured in three RIL populations and Ti52-2, whereas only NGP was measu and ZC12 populations.

Marker Analysis
ker data and the linkage map have been available except for ZC5 [39,42]. For ZC5, A lymorphic DNA markers were employed, including three on chromosome 5, o me 6 and four on chromosome 11, respectively (Table S1). DNA was extracted usin samples collected from the middle 10 plants of each line following the method of Zh CR amplification followed the method of Chen et al. [44]. The products were visuali non-denaturing polyacrylamide gels by silver staining for seven simple sequence

Field Trials and Phenotypic Evaluation
All the populations were grown from May to September in the paddy fields at the China National Rice Research Institute in Hangzhou, Zhejiang, China. A randomized complete block designed with two replications was used for all trials. In each replication, one line was grown as a single row of 12 plants. The planting density was 16.6 cm × 26.7 cm in all trials. Field management followed local agricultural practice. At maturity, five middle plants of each line were harvested in bulk for three RIL populations, and ten middle plants of each line for other populations. NGP, NSP, and SF were measured in three RIL populations and Ti52-2, whereas only NGP was measured in ZC5, ZC9 and ZC12 populations.

DNA Marker Analysis
Marker data and the linkage map have been available except for ZC5 [39,42]. For ZC5, A total of eight polymorphic DNA markers were employed, including three on chromosome 5, one on chromosome 6 and four on chromosome 11, respectively (Table S1). DNA was extracted using 2 cm long-leaf samples collected from the middle 10 plants of each line following the method of Zheng et al. [43]. PCR amplification followed the method of Chen et al. [44]. The products were visualized on 6% or 8% non-denaturing polyacrylamide gels by silver staining for seven simple sequence repeat markers, and 2% agarose gels using GelRed (Biotium, Fremont, CA, USA) staining for one sequence-tagged site marker. Linkage map was constructed using Mapmaker/Exp 3.0 (Whitehead Institute for Biomedical Research, Cambridge, MA, USA) [45], Kosambi function was used to calculate the genetic distance.

Data Analysis
Phenotypic values of two replications were averaged for each line and used for data analysis in each trial. Basic descriptive statistics, including mean value, standard deviation, coefficient of variation, the minimum and maximum trait values, skewness, and kurtosis were computed in each trial. The mean value of each trait was used to calculate the correlation coefficients between each trait.
For three RIL and Ti52-2 populations, QTLs were determined using the MET (multi-environment trials) functionality in QTL IciMapping V4.1 software (Chinese Academy of Agricultural Sciences, Beijing, China), taking different years as different environments [46]. LOD thresholds were calculated with 1000 permutation tests (p < 0.05) for each trait and used to declare a putative QTL, which were 4.3-4.8, 4.3-4.8, and 4.4-4.6 for NGP, NSP, and SF in three RIL populations, as well as 2.8, 2.7 and 2.8 in Ti52-2, respectively. QTLs detected were designated following the rule proposed by McCouch and CGSNL [47].
For ZC5, ZC9 and ZC12 populations, QTLs were determined using CIM (composite interval mapping) approach in Windows QTL Cartographer 2.5 (North Carolina State University, Raleigh, NC, USA) [48]. A threshold of LOD > 2.0 was used for claiming a putative QTL.

Phenotypic Variance
Descriptive statistics of the NGP, NSP, and SF in each trial of three RIL populations were presented in Table 1. The three traits were continuously distributed with low skewness and kurtosis, showing a typical pattern of quantitative variation. Generally, the mean values for all three traits were higher in TI than in ZM and XM. In terms of phenotypic differences between the female and male parents, significant differences (p < 0.05) were observed for NGP and NSP only in TI. The female parent TQ had higher values of NGP and NSP compared with the male parent. NGP was positively correlated with NSP and SF in all three RIL populations but the coefficients were obviously higher between NGP and NSP than between NGP and SF in each population ( Figure 2). Non-significant correlation between NSP and SF was detected in three RIL populations except in TI.
Agronomy 2020, 10, x FOR PEER REVIEW 5 of 15 NGP was positively correlated with NSP and SF in all three RIL populations but the coefficients were obviously higher between NGP and NSP than between NGP and SF in each population ( Figure  2). Non-significant correlation between NSP and SF was detected in three RIL populations except in TI.  (Table S2). Furthermore, the standard deviation values of ZC5, ZC9 and ZC12 derived from three RHs of Ti52-2 were decreased to 8.5, 9.6 and 9.9 in 2017, respectively (Table S2). Likewise, the coefficients of variation decreased from 0.150~0.194 to 0.118 and 0.079, then to 0.056~0.063 (Table 1 and Table S2). A similar trend was also observed between TI and Ti52-2 for NSP. These results were in line with the expectation that the background variations get more homozygous from TI to Ti52-2 and then to ZC5, ZC9, and ZC12.

QTLs Detected in Three RIL Populations
A total of 58 QTLs were detected in three RIL populations and distributed on all 12 rice chromosomes, including 22 for NGP, 17 for NSP, and 19 for SF (Table 2).  (Table S2). Furthermore, the standard deviation values of ZC5, ZC9 and ZC12 derived from three RHs of Ti52-2 were decreased to 8.5, 9.6 and 9.9 in 2017, respectively (Table S2). Likewise, the coefficients of variation decreased from 0.150~0.194 to 0.118 and 0.079, then to 0.056~0.063 (Table 1 and Table S2). A similar trend was also observed between TI and Ti52-2 for NSP. These results were in line with the expectation that the background variations get more homozygous from TI to Ti52-2 and then to ZC5, ZC9, and ZC12.

QTLs Detected in Three RIL Populations
A total of 58 QTLs were detected in three RIL populations and distributed on all 12 rice chromosomes, including 22 for NGP, 17 for NSP, and 19 for SF (Table 2).  For NGP, 12, 11, and four QTLs were identified in TI, ZM, and XM, respectively. Among them, five QTLs were shared between two populations and 17 were identified in single population (Table 2 and Figure 3a). In TI, qNGP2.2 and qNGP3.1 explained the highest two proportions of phenotypic variance with the value of 20.6% and 11.4%, respectively. The other 10 QTLs had the phenotypic variance explanation ranging from 1.5 to 3.9%, of which three QTLs: qNGP1.2, qNGP4, and qNGP5.2 were shared with ZM. In ZM, qNGP1.1 and qNGP6.2 were revealed to have the highest two R 2 of 15.2% and 12.4%, which were also detected in XM. The R 2 values of other QTLs were much smaller ranging as 2.5-5.4%. In XM, all four QTLs showed small R 2 values ranging from 2.6 to 6.6%. No QTL was found in both of TI and XM. The overall R 2 for NGP were much higher in TI and ZM than in XM, which were 59.0%, 56.3%, and 17.4%, respectively. Significant genotype-by-environment (GE) interaction was observed for qNGP1.1 and qNGP6.2 only in ZM (Table S3). For NGP, 12, 11, and four QTLs were identified in TI, ZM, and XM, respectively. Among them, five QTLs were shared between two populations and 17 were identified in single population (Table 2 and Figure 3a). In TI, qNGP2.2 and qNGP3.1 explained the highest two proportions of phenotypic variance with the value of 20.6% and 11.4%, respectively. The other 10 QTLs had the phenotypic variance explanation ranging from 1.5 to 3.9%, of which three QTLs: qNGP1.2, qNGP4, and qNGP5.2 were shared with ZM. In ZM, qNGP1.1 and qNGP6.2 were revealed to have the highest two R 2 of 15.2% and 12.4%, which were also detected in XM. The R 2 values of other QTLs were much smaller ranging as 2.5-5.4%. In XM, all four QTLs showed small R 2 values ranging from 2.6 to 6.6%. No QTL was found in both of TI and XM. The overall R 2 for NGP were much higher in TI and ZM than in XM, which were 59.0%, 56.3%, and 17.4%, respectively. Significant genotype-by-environment (GE) interaction was observed for qNGP1.1 and qNGP6.2 only in ZM (Table S3). For NSP, ten, eight and eight QTLs were detected in TI, ZM and XM, respectively. Among them, one QTL was shared within all three populations, seven QTLs were contributed by two populations and nine QTLs were identified in single population (Table 2 and Figure 3b). qNSP5.1, the only one stably detected in three populations, lay in the interval of RM164-RM18927 on chromosome 5 in TI and showed minor effect in all populations. qNSP2.2 displayed the largest R 2 of 22.1% in TI but nearly the smallest R 2 of 2.1% in ZM, indicating divergent effects produced in different genetic background. In ZM, qNSP1 and qNSP6.2 had the highest two R 2 of 15.1% and 13.6%, the other six QTLs had R 2 ranging from 2.0 to 4.6%. In XM, qNSP1 also had the highest contribution of 11.1%, and the other seven QTLs had R 2 ranging from 2.9 to 6.5%. The overall R 2 of the QTLs were 55.9%, 45.8% and 45.1% in TI, ZM and XM, respectively. Except qNSP2.3, qNSP10, and qNSP11.2, all QTLs for NSP were simultaneously shared the same intervals as QTLs for NGP, in accordance with high positive correlation between NGP and NSP in three populations. In addition, one significant GE interaction was found for qNSP1 in ZM (Table S3).
For SF, ten, six and five QTLs were identified in TI, ZM and XM, respectively. Only two QTLs were shared between two populations (Table 2 and Figure 3c). qSF3.3 was simultaneously identified in ZM and XM, of which the R 2 reached almost the highest with value of 7.4% and 7.6%, respectively. qSF6.2 was detected in TI and ZM, with minor effect to SF in both populations. In TI, qSF5.2 appeared the highest R 2 of 13.9% and the remaining 16 QTLs had the R 2 ranging from 2.9 to 8.2%. The overall R 2 of the QTLs were 51.1%, 27.3% and 30.1% in TI, ZM and XM, respectively. Only one QTL, qSF2.1, showed significant GE interaction in ZM (Table S3).

QTLs Detected in Ti52-2 Population
A total of five, three and four QTLs for NGP, NSP, and SF were detected in Ti52-2 population, respectively and distributed on rice chromosome 3, 5, 6, 9, 11 and 12 (Table 3). No significant GE interaction was observed. For NSP, ten, eight and eight QTLs were detected in TI, ZM and XM, respectively. Among them, one QTL was shared within all three populations, seven QTLs were contributed by two populations and nine QTLs were identified in single population (Table 2 and Figure 3b). qNSP5.1, the only one stably detected in three populations, lay in the interval of RM164-RM18927 on chromosome 5 in TI and showed minor effect in all populations. qNSP2.2 displayed the largest R 2 of 22.1% in TI but nearly the smallest R 2 of 2.1% in ZM, indicating divergent effects produced in different genetic background. In ZM, qNSP1 and qNSP6.2 had the highest two R 2 of 15.1% and 13.6%, the other six QTLs had R 2 ranging from 2.0 to 4.6%. In XM, qNSP1 also had the highest contribution of 11.1%, and the other seven QTLs had R 2 ranging from 2.9 to 6.5%. The overall R 2 of the QTLs were 55.9%, 45.8% and 45.1% in TI, ZM and XM, respectively. Except qNSP2.3, qNSP10, and qNSP11.2, all QTLs for NSP were simultaneously shared the same intervals as QTLs for NGP, in accordance with high positive correlation between NGP and NSP in three populations. In addition, one significant GE interaction was found for qNSP1 in ZM (Table S3).
For SF, ten, six and five QTLs were identified in TI, ZM and XM, respectively. Only two QTLs were shared between two populations (Table 2 and Figure 3c). qSF3.3 was simultaneously identified in ZM and XM, of which the R 2 reached almost the highest with value of 7.4% and 7.6%, respectively. qSF6.2 was detected in TI and ZM, with minor effect to SF in both populations. In TI, qSF5.2 appeared the highest R 2 of 13.9% and the remaining 16 QTLs had the R 2 ranging from 2.9 to 8.2%. The overall R 2 of the QTLs were 51.1%, 27.3% and 30.1% in TI, ZM and XM, respectively. Only one QTL, qSF2.1, showed significant GE interaction in ZM (Table S3).
Amid of the three QTLs for NSP detected in Ti52-2, qNSP12 was validated with the largest R 2 of 7.8% and the enhancing allele was from IRBB52 in both TI and Ti52-2. qNSP5.4, a new QTL with an allelic effect in opposite direction to qNSP5.3 in the interval of RM274−RM334 detected in TI, was detected in the RM18927−RM3321 region. In addition, qNSP3.2 having the marginal LOD threshold and the smallest R 2 was not previously detected in TI, ZM and XM.
Out of the four QTLs for SF detected in Ti52-2, qSF6.1 with the highest R 2 of 16.3% was mapped in RM586−RM584 region, which was undetected in TI and lay in a similar interval of RM190−RZ516 detected in XM. The others, qSF9.2, qSF3.4, and qSF11, were not identified previously in three RIL populations. Either position or additive effect direction of qSF9.2 was different from qSF9 identified in ZM. Thus qSF9.2 was a new QTL and qSF9 was renamed as qSF9.1.

Discussion
Grain number is one of the most important agronomic traits in the determination of rice productivity, which is governed by multiple QTLs with major or minor effect. In the present study, a total of 58 QTLs were detected in three RIL populations, including 22, 17, and 19 for NGP, NSP, and SF, respectively. The individual QTL counted for 1.5% to 22.1% of phenotypic variation. Among 58 QTLs, 15 with various effects shared by two or three populations and eight showed large effects with R 2 larger than 10%. When major-effect QTLs were fixed, three minor-effect QTLs for NGP were further detected and validated using secondary segregation populations derived from one RH of the cross TQ/IRBB52 under increment homogeneous backgrounds.
Among the three minor-effect QTLs for NGP, qNGP9.1 and qNGP12.1 were stably identified in TI, Ti52-2, ZC9, and ZC12 populations, having the same direction of additive effects and similar value. Moreover, the regions of RM8206−RM219 containing qNGP9.1 and RM511−RM28313 containing qNGP12.1 were previously reported to affect NGP. qFG9 and Gpp12.2 were detected in the region of RM219−RM342 on chromosome 9 and RG869−RM277 on chromosome 12, respectively [49,50]. It proves that qNGP9.1 and qNGP12.1 are good targets for underlying the genetic basis of grain number.
qNGP5.5 was a new QTL identified in Ti52-2. This QTL was well verified in ZC5, the direction of allelic effect remaining consistent and R 2 increased from 4.9 to 6.3%. Actually, qNGP5.4 was detected in TI with an allelic effect in the opposite direction to qNGP5.5. Comparing the two loci, we found that qNGP5.5 was located in RM18927−RM3321, while the qNGP5.4 was detected in RM274−RM334. The segregation region of Ti52-2 population covered RM274 but uncovered RM334. In previous studies, two QTLs with opposite allelic direction for grain weight were detected in RM18927−RM334 [37,51]. Hence, there might be two tightly linked QTLs with opposite direction of additive effect in the region of RM18927−RM334.
Verification of three minor-effect QTLs was benefited from the fixation of the major-effect QTLs and increment homozygous of genetic background. Along with the background became more homogenous from TI to Ti52-2 then to ZC5, ZC9 and ZC12, the phenotypic variations were getting smaller, and the power of detection of QTLs was further improved. For instance, the standard deviation values and coefficient of variation of NGP and NSP has been decreasing with background became more homogenous (Table 1 and Table S2). Out of the 12 QTLs for NGP detected in TI, only qNGP4, qNGP5.4, qNGP9.1 and qNGP12.1 were located in the segregation region of Ti52-2. Except qNGP4 with the smallest R 2 , other three QTLs were detected in Ti52-2. In addition, two more QTLs qNGP3.3 and qNGP5.5 were detected in Ti52-2. These results supported that RHs are efficient materials in the establishment of NILs and secondary mapping populations for QTL validation and fine-mapping especially for minor-effect QTLs.
Interestingly, among the five QTLs for NGP identified in Ti52-2, qNGP3.3, qNGP5.5, and qNGP12.1 were located in the same regions with qNSP3.2, qNSP5.4, and qNSP12, respectively. The regions of qNGP6.1 and qNGP9.1 showed the same intervals with qSF6.1 and qSF9.2. The results suggest that the five QTLs for NGP detected in Ti52-2 is a result of the integration of NSP and SF. Usually, NGP is determined by NSP multiply SF. Thus, these QTL regions sharing for both NGP and NSP or SF might be due to a pleiotropic effect of one QTL rather than the close linkage of different QTLs.
The eight QTLs with large effects detected in three RIL populations were located in six regions on chromosome 1, 2, 3, 5 and 6. Except for qSF5.2, all QTLs controlled NGP and/or NSP.
The region of RG532−RM151 on chromosome 1 containing qNGP1.1 and qNSP1 showed the largest R 2 for NGP and NSP in both of ZM and XM, while the interval of RM6−RM240 on chromosome 2 covering qNGP2.2 and qNSP2.2 showed the largest R 2 for NGP and NSP in TI. The special significance demonstrated in the given population might reveal the different mechanism of variant ecological type of rice varieties. Gn1a, the first cloned QTL for grain number in rice, is located in the former region [18], whereas no QTL for grain number was found to be fine-mapped or cloned in the latter interval. By comparing the physical position, we found a heading date QTL, DTH2, situated within this interval [52]. Anyway, no QTL controlling heading date was detected around this region in previous studies [42], which indicated that DTH2 could not the candidate gene of qNGP2.2 and qNSP2.2. In addition, we found that GS2, a major QTL for grain size was lain 0.7 Mb upstream of RM6. Hu et al. reported that no significant differences of NGP were found between the recurrent parents and NILs [53], suggesting it would not probably be the pleiotropic effect of GS2. Hence, this locus might be a new QTL controlling the grain number. Fine-mapping of qNGP2.2 is to be undertaken.
The region of RM15303−RM16 on chromosome 3 was another one for both of NGP and NSP identified in TI. However, it presented the second largest R 2 for NGP, but much smaller effect on NSP. GL3.1, a cloned QTL for grain length, was positioned 1.9 cM downstream of RM16 and was reported to have a significant effect on NGP [54]. More work is needed to confirm whether this locus is a new QTL for grain number or just a pleiotropic effect of GL3.1.
The region of RZ398−RM217 on chromosome 6 exhibited the second largest R 2 for NGP but no significant effect on NSP in ZM. It also displayed minor effect on NGP and NSP in XM. No QTL controlling grain number had been cloned or fine-mapped in this interval. In the other hand, the region of RM253−RM276 downstream of RZ398−RM217 exhibited the second largest R 2 for NSP but no significant effect on NGP in ZM. It was noteworthy that it also displayed the second largest R 2 for NSP and minor effect on NGP in TI. It was previously reported that a QTL controlled NGP and yield-related traits in the same region [55]. Hence, these QTLs provided new candidates for gene cloning and marker-assisted breeding.

Conclusions
A total of 58 QTLs for NGP, NSP, and SF were detected in three RIL populations of indica rice. Among them, eight QTLs showed large effects with R 2 larger than 10%. Three QTLs with minor effects for NGP, qNGP5.5, qNGP9.1, and qNGP12.1, were further detected and validated using segregation populations derived from one RH of the cross TQ/IRBB52. These results proved that the use of residual heterozygotes to construct secondary mapping populations is an efficient strategy to detect minor-effect QTLs for complex traits. The results also enrich our understanding of the mechanism of grain number and provide foundation for cloning and selecting candidate for marker-assisted selection breeding in rice.
Author Contributions: J.Z. and Y.F. conceived and designed the experiments; X.N., Y.Z. and Z.S. performed the experiments; S.Y. and J.Z. analyzed the data; X.N. and Y.F. wrote the paper.

Conflicts of Interest:
The authors declare no conflict of interest.