Method for improving whole genome selection predictability and breeding application

Abstract

The invention discloses a method for improving whole genome selection predictability and breeding application, and belongs to the field of rice molecular breeding. The method comprises the following specific steps: 1) performing gene sequencing and variation detection on a rice parent to obtain SNP variation information; 2) performing heterosis group division based on SNP variation of the whole genome; 3) constructing a training group and planting; 4) constructing a genome selection model and predicting the yield of the hybrid combination; according to the method, a hybrid rice yield whole genome selection model is finally constructed based on group division, training group matching and high-quality reference genome assembling of core indica rice varieties; according to the method, excellent hybrid rice is matched according to the hybrid rice yield whole genome selection model; the genome prediction model provided by the invention can be used for improving the selection efficiency of excellent hybrid combinations and accelerating the breeding process, and has important application value.

Description

Technical Field

[0001] The present invention belongs to the field of rice molecular breeding, and specifically relates to a rice whole genome selection breeding method for improving breeding efficiency, which is suitable for data analysis such as breeding value estimation of whole genome selection in model organisms represented by rice. Background Art

[0002] Most of the target traits of rice breeding are complex quantitative traits, and the efficient selection of complex agronomic traits is the technical bottleneck of breeding (Asano K et al. Artificial selection for a green revolution gene during japonica rice domestication. PNAS, 2011, 108, 11034-11039; Zhao K, et al. Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa. Nat. Commun. 2011, 2, 467.). Although molecular marker assisted selection (MAS) breeding is more targeted than traditional conventional breeding, MAS breeding is usually limited to the selection of a few major effect sites, and the selection effect on complex agronomic traits controlled by multiple genes is not obvious (Maher B. Personal genomes: The case of the missing heritability. Nature News, 2008, 6: 18-23.). Meuwissen et al. (2001) first proposed a whole genome selection breeding strategy, which is to use molecular markers at the whole genome level and the phenotypic data of each individual in the training population to establish a prediction model, calculate the genetic effect of each marker, and then use the genetic effect of the whole genome marker in the subsequent derived breeding population to predict the breeding value of the individual, and finally select the individual plant according to the breeding value (Meuwissen THE, Hayes BJ, Goddard MEPrediction of total genetic value using genome-wide dense marker maps. Genetics, 2001, 157: 1819-1829.). Since whole genome selection is to establish a prediction model using markers covering the whole genome, it can effectively avoid biased estimation of marker effects and consider the comprehensive effects of different site interactions at the whole genome level on complex traits, thereby ensuring the accuracy of breeding individual selection and breeding efficiency (Bernardo R., Yu J. Prospects for genomewide selection for quantitative traits in maize. Crop Science, 2007, 47: 1082-1090.).

[0003] Compared with molecular marker-assisted selection, whole genome selection has the advantages of accelerating breeding process and reducing R&D costs (Xu D, et al. Yield characteristics of japonica / indica hybrids rice in themiddle and lower reaches of the Yangtze River in China. Journal of Integrative Agriculture, 2020, 19(10): 2394-2406; Wei X, et al. A quantitative genomics map of rice provides genetic insights and guides breeding. Nat Gent. 2021, 53(2): 243–251.). After the whole genome selection breeding technology was applied to dairy cattle breeding, the R&D expenses of dairy cattle breeding companies were reduced by nearly 80% (VanRaden PM et al. Invited review: Reliability of genomic predictions for North American Holstein bulls. Journal of Daily Science, 2009, 92(1): 16-24.). In terms of plant application, the progress of whole genome selection breeding is 4-25% higher than that of traditional phenotypic selection, and the cost per unit genetic progress is 26-65% lower than that of traditional breeding (Crossa J, et al. Genomic selection in plant breeding: methods, models, and perspectives. Trends in Plant Science, 2017, 22 (11): 961-975.). In terms of predictive ability, compared with the traditional prediction of offspring breeding values ​​based on pedigree information, the predictive ability of wheat yield has increased from 7.7% to 35.7%, and the predictive ability of corn has increased from 15% to 79% (Desta ZA and Ortiz R, Genomic selection: genome-wide prediction in plant improvement. Trends in Plant Science, 2014, 19 (9): 592-601.). Therefore, the prediction model established based on whole genome molecular marker information is very effective for the selection breeding of complex agronomic traits.

[0004] The effect of whole genome selection prediction depends largely on: (1) the density of whole genome markers; (2) the quality of training population construction; (3) the choice of prediction model. Although constructing a large-scale training population can increase the accuracy of prediction, a large number of matchings between parents in actual breeding require a lot of manpower and identification costs, which is time-consuming and labor-intensive, and significantly increases breeding costs. Secondly, parents with no breeding application value will reduce the quality of the training population and reduce the accuracy of prediction. Therefore, the accuracy of parent selection and phenotypic identification in constructing the training population is crucial (Xu Y, et al. Genomic selection: A breakthrough technology in rice breeding. Crop Journal, 2021, 9 (3): 669-677.).

[0005] Utilizing hybrid vigor is the key to increasing crop yields. Hybrid parents are divided into different groups through molecular markers. The genetic structure of germplasm within the same group is similar, and there is no obvious yield advantage in mutual hybridization. However, the genetic structure of germplasm between groups is obviously different, and mutual hybridization can produce significant yield advantages. Strong hybrid vigor is more likely to occur between parents with a large genetic distance (Li, Y. Hybrid rice breeding in China. Chin. Ann. Hist. Sci. Technol. 2020, 4: 135-159; Wang et al. De novo genome assembly and analyses of 12founder inbred lines provide insights into maize heterosis. Nat Gent. 2023, 55(2): 312-323). The use of hybrid vigor is the main means to increase rice yield. Hybrid rice has a yield advantage of more than 20% over inbred lines (Jumin T, et al. Field performance of transgenic elite commercial hybrid rice expressing bacillus thuringiensis dendotoxin, Nat. Biotechnol. 18 (2000) 1101–1104.). However, the biggest challenge in rice hybrid breeding is to select the desired hybrid combination from a large number of potential hybrid combinations. Due to limited resources, it is unrealistic to evaluate the field performance of all potential hybrids. Therefore, how to model and predict rice agronomic traits and improve breeding efficiency has become a technical problem that needs to be solved urgently in this field. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention constructs a core parent group and a training group suitable for full-genome selection breeding of hybrid indica rice, provides a full-genome selection breeding method for hybrid rice yield, and improves the selection efficiency of excellent hybrid rice. Another purpose of the present invention is to apply the full-genome selection technology to the breeding process of hybrid rice, and establish a full-genome selection breeding process for hybrid rice yield. Reduce the number of F1 populations in the field, reduce production costs, and establish an economical and effective full-genome selection technology.

[0007] Specifically, the present application is implemented through the following technical solutions:

[0008] A method for improving the prediction power of whole genome selection, the specific steps of which are as follows:

[0009] 1) Perform second-generation sequencing and mutation detection on rice parents to obtain SNP mutation information

[0010] The rice varieties of the sample population were resequenced using Illumina's second-generation sequencing technology, and then the clean reads obtained from each sample were aligned to the R498 reference genome using the bwa tool. The BAM files generated by the alignment were sorted using the Samtools software, and the genotype files were generated using GATK. The variation information was extracted and low-quality SNPs were filtered to obtain the SNP variation data of rice in the sample population. The PLINK software was then used to filter the obtained rice SNP variation data of the sample population, with quality control parameters such as minor allele frequency greater than 0.05 and missing rate less than 80%, and sites that did not conform to the Hardy-Weinberg equilibrium were deleted, and finally the SNP variation information of the sample population was obtained.

[0011] 2) Heterotic vigor grouping based on genome-wide SNP variation

[0012] Based on the SNP variation information obtained in step 1), the Admixture tool was used to perform structural analysis on the sample population, K=5, and the sample population was divided into groups; then the population evolution tree was constructed using TASSEL software, and PCA analysis was performed using PLINK to obtain group division results, including sterile line groups and restorer line groups.

[0013] In this step, hybrid rice is matched according to the group division results for training population construction, yield traits are further examined, and a whole-genome selection model for hybrid rice yield is established.

[0014] 3) Constructing a training group

[0015] Based on the group division results obtained in step 2), the sterile line group and the restorer line group parent materials are selected to obtain several hybrid combinations, which are training groups; these training groups are planted and the yield is measured. 4) Construction of genomic selection model and prediction of hybrid combination yield

[0016] Construct a genome selection model for the training population obtained in step 3):

[0017] y=Xβ+G a α a +ε (1)

[0018] In formula (1), y is the n×1 phenotypic observation vector of hybrids, X is the n×p design matrix for fixed effects, β is the fixed effect, α a is the genomic additive effect, as a random effect, α a It follows the following normal distribution: G a is the n×g genotype additive coding matrix of the hybrid, and ε is the n×1 residual vector, which follows a normal distribution:

[0019] 5) Optimization of selection model

[0020] PacBio's third-generation HiFi sequencing technology was used to assemble the T2T golden version genome maps of two excellent sterile lines, Yuan 88S and Yangxian 9A. This was used as a reference genome to optimize the whole-genome selection model and algorithm, thereby realizing whole-genome selection prediction of rice and improving prediction accuracy.

[0021] Secondly, the present application also provides the application of the above method in rice breeding, that is, predicting the Top 100 hybrid combinations of target seeds through a whole genome selection model, and then comparing them with hybrid rice randomly matched in the field to create hybrid rice with excellent yield, improve the selection efficiency of excellent hybrid combinations, reduce costs and accelerate the breeding process.

[0022] This application uses a whole genome selection method to identify a small part of all potential hybrid combinations in the field in the hybrid rice phenotype, and predicts the performance of other combinations through genetic simulation, which can greatly reduce field costs and accelerate the breeding process. This application is the first to use molecular markers (SNPs) covering the entire genome to divide the hybrid advantage groups for the indica rice parents in the middle and lower reaches of the Yangtze River ecological zone and the excellent germplasm introduced from abroad (Table 1), and further based on the group division results, the hybrid rice is matched for training population construction, the yield traits are examined and a whole genome selection model is established. On this basis, the T2T genome assembly of two excellent sterile lines, Yuan 88S and Yangxian 9A, further improves the predictive power of the whole genome selection model, and creates a series of excellent hybrid rice. This method can achieve efficient variety improvement, scientific parent matching and precise breeding mode, and realize the creation of excellent germplasm. Compared with the existing hybridization methods, the beneficial effects of the present invention are:

[0023] 1. Created a training population for hybrid rice genomic selection model based on group division

[0024] In the embodiment of the present application, based on the grouping of core parents of indica hybrid rice, a high-quality training population of 2232 hybrid rice F1 was constructed, and its yield traits were investigated in the field. This population laid a solid foundation for the subsequent construction of a hybrid rice yield genomic selection model.

[0025] 2. Provides a genomic selection model for rapid matching of excellent hybrid rice

[0026] The examples of this application have mapped high-quality genomes of two excellent sterile lines, Yuan 88S and Yangxian 9A, and based on this, optimized the genome selection model and improved the prediction power. By comparing model prediction with field matching, it was found that the prediction model significantly improved the efficiency of matching excellent hybrid rice, reduced field workload, and bred a series of excellent hybrid combinations, which has great application value in production. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 .Division of parental groups of 222 core hybrid indica rice accessions and construction of GS model

[0028] A: Structure genetic structure analysis of 222 core hybrid indica rice parents based on whole genome SNPs, when K=5, the cross-validation error is the smallest; B: Evolutionary tree of 222 sequenced parents; C: PCA analysis of 222 sequenced parents, I: two-line restorer group; II: two-line sterile group; III: three-line restorer group; IV: South China late indica group; V: three-line sterile group; D: Screening of core parents with yield advantage based on genomic selection model, red dots indicate hybrid combinations with Yangdao No. 6 as the female parent in the TOP 100 hybrid combinations.

[0029] Figure 2 .High-precision genome assembly and genome selection model optimization of Yuan88S and Yangxian 9AA: High-precision genome assembly and annotation of Yuan88S; B: High-precision genome assembly and annotation of Yangxian 9A; C: Distribution of hybrid rice yield in 2,232 combinations; D: Analysis results of the predictive power of the genome model using Yuan88S, Yangxian 9A and R498 as reference genomes.

[0030] Figure 3 .Breeding of excellent hybrid rice based on genomic selection model

[0031] A: Comparison of yield between the top 100 hybrid rice combinations predicted by the whole genome selection model and randomly matched hybrid rice; B: The yield of the excellent hybrid rice Yuanliangyou 289, Yuanliangyou 829 and Yuanliangyou 868 bred with Yuan88S as the female parent was significantly increased; CF: Comparative analysis results of effective panicle number (C), fruit setting rate (D), 1000-grain weight (E) and number of grains per panicle (F) between Yuanliangyou 289, Yuanliangyou 829 and Yuanliangyou 868 and the control Quanyou 822. DETAILED DESCRIPTION

[0032] The rice varieties involved in the embodiments are all preserved in the applicant's laboratory.

[0033] Example 1 Division of core hybrid indica rice parent groups and construction of GS model

[0034] 1. Analysis of SNP variation information of 222 hybrid indica rice parents

[0035] Table 1 222 core hybrid indica rice parents

[0036]

[0037]

[0038]

[0039] This example uses 222 hybrid indica rice parents collected from home and abroad (Table 1), and each variety is resequenced at 30× depth. After data quality control, the clean reads obtained from each sample are aligned to the R498 reference genome using the bwa tool, and the BAM files generated by the alignment are sorted by Samtools software, and the genotype files are generated using GATK. The variation information is further extracted and low-quality SNPs are filtered, and finally the population SNP variation data of 222 hybrid indica rice parents are obtained, and a total of 3,296,995 SNP sites are detected. In order to improve the reliability of the analysis, the population SNP data is further filtered using PLINK software, and the quality control parameters are minor allele frequency greater than 0.05; missing rate less than 80%; sites that do not conform to Hardy-Weinberg equilibrium are eliminated, and finally 545,566 high-quality SNP sites are obtained.

[0040] 2. Division of core hybrid indica rice parent groups

[0041] Based on 545,566 high-quality SNP loci obtained from population resequencing, the Admixture tool was used to analyze the population structure of 222 hybrid indica rice parents. Cross-validation analysis showed that the error value was the smallest when K = 5, and they could be divided into 5 groups ( Figure 1 (shown in A).

[0042] The population evolution tree was further constructed using TASSEL software, and PCA analysis was performed using PLINK. Both showed high consistency with the above population structure analysis. Finally, the 222 hybrid indica rice parents were divided into five groups: I: two-line restorer group; II: two-line sterile group; III: three-line restorer group; IV: South China late indica group; V: three-line sterile group (such as Figure 1 (shown in B and C).

[0043] 3. Build a training group

[0044] Based on the division of core parent groups, groups II and V were mainly sterile parent materials, and groups I, III, and VI were restorer parent materials. Representative sterile parents were selected from the II and V group materials, and representative restorer parents were selected from I, III, and VI, and a total of 175 hybrid combinations were obtained.

[0045] These 175 hybrid combinations were planted in Yangzhou, Jiangsu Province, with 200 plants planted in each plot. The planting specifications were (row spacing of 29.9 cm, plant spacing of 13.32 cm). Three replicates were randomly arranged for each hybrid combination. The field management was 11 kg / mu of nitrogen fertilizer. After maturity, the actual yield of the plot was measured.

[0046] 4. Construction of genomic selection model and prediction of hybrid combination yield

[0047] Based on the genomic best linear unbiased prediction method (GBLUP), a genomic selection model was constructed for the training population of the above 175 hybrid combinations:

[0048] y=Xβ+G a α a +ε (1)

[0049] In formula (1), y is the n×1 phenotypic observation vector of hybrids, X is the n×p design matrix for fixed effects, β is the fixed effect, α a is the genomic additive effect, as a random effect, α a It follows the following normal distribution: G a is the n×g genotype additive coding matrix of the hybrid, and ε is the n×1 residual vector, which follows a normal distribution:

[0050] In this example, the whole genome selection model (1) was used to further predict the yield of 3741 hybrid combinations. Among the hybrid combinations with the top 100 yields in the prediction results, it was found that Yangdao No. 6 as the female parent and the restorer parent of other groups showed excellent heterosis, and the hybrid rice yield was in the forefront ( Figure 1 The above prediction results show that Yangdao 6 can be used as the female parent of hybrid rice to combine with other restorer lines to make fuller use of its heterosis and has great potential for combining with excellent hybrid rice.

[0051] Example 2: The high-precision genomes of Yangxian 9A and Yuan 88S improve the predictive power of GS

[0052] 1. Gap-free genome assembly of Yangxian 9A and Yuan 88S

[0053] In order to obtain high-quality genomes of Yangxian 9A and Yuan 88S, 30× PacBio third-generation HiFi sequencing was performed on Yangxian 9A and Yuan 88S, and 50× HiC sequencing was performed to assist in genome assembly. After filtering low-quality raw sequencing reads, Hifiasm (v0.16.0) was used for de novo genome assembly (see the literature “Wenger, AM, et al. (2019) Accurate Circular Consensus Long-read Sequencing Improves Variant Detection and Assembly of a Human Genome. Nature Biotechnology, 37, 1155–1162.”). In order to fill the missing gaps on each chromosome, Nanopore's ONT sequencing was used to perform 50× deep sequencing on Yangxian 9A and Yuan 88S. The reads data after sequencing were assembled into long fragments for gap filling after quality control. At the same time, in order to correct the base errors caused by HiFi and ONT sequencing, the results of Illumina's second-generation sequencing were further used to perform base correction on the assembled genome, and finally two gap-free T2T-level rice genomes were obtained.

[0054] The LAI values ​​of the Yangxian 9A and Yuan 88S genomes obtained in this example were evaluated to be 23.41 and 24.33, respectively, reaching the gold version level (Ou, S., Chen, J., and Jiang, N. (2018). Assessing genome assembly quality using the LTR Assembly Index (LAI). Nucleic Acids Res. 46, e126.).

[0055] 2. Genome annotation of Yangxian 9A and Yuan 88S

[0056] The assembled genome sizes of Yangxian 9A and Yuan 88S were 411.89Mb and 402.91Mb, respectively, with contig N50 reaching 22.64Mb and 21.67Mb, respectively, and GC contents of 43.61% and 43.80%, respectively. The assembly results showed that the assembly integrity was 98.6%, the alignment rates of second-generation sequencing reads were 99.6% and 99.03%, the coverage rates were 99.47% and 99.44%, respectively, and the QV values ​​were 38.7615 and 38.8941, respectively, indicating that the two rice T2T genome versions have high assembly integrity, consistency and accuracy. Extensive de novo TE Annotator (EDTA, version 1.9.6) was used to annotate the whole-genome repeat sequences, and the proportions of repeat sequences in the two rice genomes were 46.45% and 45.58%, respectively. Gene annotation was performed by combining de novo annotation, homology annotation, and transcriptome annotation. Finally, the results of the three annotations were integrated using EvidenceModeler software to obtain the final gene annotation. 39,168 and 37,281 genes were identified in the two genomes, respectively (e.g. Figure 2 (as shown in A and B).

[0057] 3. Construction of a genome-wide selection model using Yangxian 9A and Yuan 88S genomes as reference genomes

[0058] The 2232 hybrid rice pairs were used as the training population ( Figure 2 The assembled Yangxian 9A and Yuan88ST2T genomes were used as reference genomes, and a whole-genome selection model based on the selection index was applied (Wang et al. (2019) Efficiency of linear selection index in predicting rice hybrid performance. Molecular Breeding, 39(6):77). The selection index was reconstructed using the auxiliary trait thousand-grain weight to achieve comprehensive selection of rice hybrids and auxiliary prediction of target traits.

[0059] By comparing the prediction power of hybrid rice yield, it was found that compared with R498 as the reference genome, the Yangxian 9A and Yuan 88S genomes as reference genomes combined with 1000-grain weight auxiliary prediction significantly improved the prediction power of whole genome selection ( Figure 2 (shown in D).

[0060] Example 3 Breeding of excellent hybrid combinations based on hybrid rice whole genome selection model

[0061] 1. Genomic selection improves the efficiency of excellent hybrid rice pairing

[0062] In this example, Yangxian 9A and Yuan 88S were used as female parents, and the genomic selection model constructed in step (3) of Example 2 was used to predict the top 100 hybrid rice combinations in yield, and hybrid rice matching was performed. At the same time, 100 hybrid combinations were randomly matched in the field using Yangxian 9A and Yuan 88S as female parents.

[0063] The matched hybrid rice was planted in Yangzhou, Jiangsu Province. Each plot was planted with 200 plants and the planting specifications were (row spacing of 29.9 cm and plant spacing of 13.32 cm). Three replicates were randomly arranged for each hybrid combination, and the actual yield of the plot was measured after maturity.

[0064] The results showed that the yield of hybrid rice with Yangxian 9A and Yuan 88S as female parents based on the genomic selection model was significantly higher than that of random hybrid combinations ( Figure 3 (As shown in A). Hybrid rice pairing based on the genomic selection model has eliminated poor hybrid combinations to a great extent, greatly reduced the workload and cost of field pairing, and improved breeding efficiency.

[0065] 2. Breeding of excellent hybrid rice combinations

[0066] Based on the constructed whole genome selection model of hybrid rice, this embodiment bred a series of excellent hybrid rices from the top 100 hybrid rices with yield using Yangxian 9A and Yuan88S as the maternal combination, among which Yuanliangyou Jingxiang Si Miao, Yuanliangyou 819, Yuanliangyou 968, Yuanliangyou 289, Yuanliangyou 829 and Yuanliangyou 868 using Yuan88S as the maternal combination showed excellent yield advantages in field trials.

[0067] The results of field plot yield measurement showed that the average per-acre yield of Yuanliangyou Jingxiangsi Miao, Yuanliangyou 819, Yuanliangyou 968, Yuanliangyou 289, Yuanliangyou 829 and Yuanliangyou 868 increased by 7.96%, 6.47%, 7.1%, 10.72%, 10.24% and 11.41% respectively compared with the control Quanyou 822. Figure 3 (shown in B).

[0068] The four important yield components, namely, number of effective ears, thousand-grain weight, fruit setting rate and number of grains per ear, were further analyzed. Figure 3 The analysis showed that Yuanliangyou 868 and Yuanliangyou 289 had a more compact plant shape and a slightly reduced number of effective panicles compared to the control (Quanyou 822). Figure 3 C), there was no significant difference in fruit setting rate and thousand-grain weight compared with the control ( Figure 3 The yield increase of Yuanliangyou 289, Yuanliangyou 829 and Yuanliangyou 868 was mainly due to the increase in the number of grains per panicle. Compared with Quanyou 822, the number of grains per panicle of the three hybrid rice varieties increased by 38.73%, 48.19% and 40% respectively. Figure 3At present, the above hybrid combinations are participating in the intermediate test of variety approval, which also shows that the hybrid rice yield prediction model constructed by this patent has improved the efficiency of excellent hybrid combination breeding and has great application value in breeding practice.

Claims

1. A method to improve the prediction power of whole-genome selection, It is characterized in that The specific steps are as follows: 1) Perform gene sequencing and mutation detection on rice parents to obtain SNP mutation information The rice varieties in the sample population were resequenced, and then the files generated by comparing the data obtained from each sample with the R498 reference genome were sorted to generate genotype files, extract variation information, and filter low-quality SNPs to obtain SNP variation data of rice in the sample population; the obtained rice SNP variation data of the sample population were then filtered, and the quality control parameters were that the minor allele frequency was greater than 0.05 and the missing rate was less than 80%, and the sites that did not conform to the Hardy-Weinberg equilibrium were deleted, and finally the SNP variation information of the sample population was obtained; 2) Heterotic vigor grouping based on genome-wide SNP variation Based on the SNP variation information obtained in step 1), the sample population is subjected to structural analysis, K=5, and the sample population is divided into groups; then, a population evolution tree is constructed, and PCA analysis is performed to obtain group division results including sterile line groups and restorer line groups; 3) Constructing a training group Based on the group division results obtained in step 2), the parent materials of the sterile line group and the restorer line group are respectively selected to obtain several hybrid combinations, namely, training groups; these training groups are planted and yields are measured; 4) Construction of genomic selection model and prediction of hybrid combination yield Construct a genome selection model for the training population obtained in step 3): y=Xβ+G a a a +e (1) In formula (1), y is the n×1 phenotypic observation vector of hybrids, X is the n×p design matrix for fixed effects, β is the fixed effect, α a is the genomic additive effect, as a random effect, α a It follows the following normal distribution: G a is the n×g genotype additive coding matrix of the hybrid, and ε is the n×1 residual vector, which follows a normal distribution: That is, to achieve whole-genome selection prediction of rice.

2. Application of the method as claimed in claim 1 in rice breeding.

Images