Background Genomic prediction plays a pivotal role in assessing crop germplasm resources, accelerating crop breeding, and enhancing genetic improvement. The mixture of experts network (MoE), has been successfully applied in the fields of natural language processing and image recognition. Results In this study, we proposed a novel, flexible, and robust approach for genomic prediction based on MoE, named Mixture of Experts for Genomic Prediction (MoEGP). This method incorporates a set of experts that process data features and a gating network that distributes samples and assigns weight score to each expert. MoEGP combines the predictions of the selected top-k experts, weighted by their corresponding weight score as final outputs. We evaluated the performance of MoEGP using 36 trait cases from five public datasets across three major cereal crops in comparison to five well-known genomic prediction methods. The results indicated that MoEGP exhibited superior performance in terms of both pearson correlation coefficient (PCC) and mean absolute error (MAE) in all 36 trait experiments, encompassing morphological, developmental, and yield traits. MoEGP achieved an average PCC improvement of 33.75%, and an average reduction of 19.55% in MAE compared with the other five reported methods. With its automated hyper-parameter search system, MoEGP offers flexibility in supporting traits governed by distinct quantitative trait locus (QTL) architectures, facilitating its broad application in crop genomic prediction.The corresponding code is available as open-source on GitHub (https://github.com/CNRRI-RGRT/MoEGP). Conclusions We proposed a novel deep-learning network designed for genomic prediction, named MoEGP, which outperformed five previously reported methods and demonstrated strong potential for application in crop genomic selection.

Rice germplasms serve as repositories of genetic variation. Population structure analysis is crucial for the effective utilization of rice germplasm and provides an evolutionary framework for future biological research. In this study, we constructed a genetic variation map using a large population of 20 056 rice accessions. A total of 34 426 306 variants were identified, including 29 251 099 biallelic and 5 175 207 multiallelic variants. Population structure analysis of a subset of 5 668 accessions classified Asian cultivated rice into eight distinct groups: temperate japonica 1 (TEJ1), temperate japonica 2 (TEJ2), tropical japonica (TRJ), aromatic (ARO), aus (AUS), indica 1 (IND1), indica 2 (IND2), indica 3 (IND3). Comparative analysis revealed that japonica accessions from China were predominantly composed of TEJ1 and TEJ2, while indica varieties were mainly from IND1. In contrast, introduced japonica accessions were primarily TEJ1 and TRJ, with IND2 representing the major indica group. Foreign accessions represented all eight groups, whereas Chinese accessions lacked the ARO, AUS, and IND3 groups. Genetic diversity analysis revealed that IND3 and TEJ1 exhibited the highest and lowest nucleotide diversity, respectively. Geographically, TRJ was largely distributed in the Americas, ARO was concentrated in the Middle East, and notably, TEJ2 was identified as a group specific to China. Furthermore, we identified 52 regions with significant genetic differentiation. Two of these regions, harboring the cold-tolerant genes bZIP73 and COLD1, were selected for further analysis. Nucleotide diversity and haplotype network analyses indicated that both genes may have undergone positive selection and contribute to indica-japonica differentiation. Haplotype analysis also uncovered novel alleles, including a new mutation in COLD1 that may represent a favorable variant for enhancing cold tolerance in japonica accessions. Our study demonstrates that large-scale population analysis of genetic variation is a powerful method for dissecting population structure, elucidating the mechanisms of genetic differentiation, and identifying novel functional alleles.

We systematically evaluated three key determinants affecting prediction accuracy and the algorithm performance differences based on fifteen state-of-the-art GP methods, and found LSTM suitable for capturing additive and epistatic effects. Genomic prediction (GP) has been developed as an important method supporting crop breeding. By utilizing the phenotype values result from GP, breeders could make decisions in the seedling stage that consequently benefit for cost saving. In recent years, machine learning emerged as an efficient technology to solve modeling problems in many fields, including crop breeding. However, numerous modeling approaches have hindered the application of GP since breeders struggle to choose. Therefore, a comprehensively methodological research with guiding significance is extremely necessary. In the present study, we systematically evaluated three key determinants affecting prediction accuracy and the algorithm performance differences based on fifteen state-of-the-art GP methods. As for genomic feature processing, we found feature selection (SNP filtering approach) performed better than feature extraction (PCA method). Specifically, the feature relationship dependent methods (GBLUP, RNN, and LSTM) as well as DNN architecture showed superior performance with feature selection. Marker density analysis showed positive correlation with prediction accuracy in a limited threshold. Comparison on effect of population size demonstrated a positive correlation between trait genetic complexity and the optimal population size required. By testing fifteen modeling methods, we found LSTM network displayed superior performance, achieving the highest average STScore (0.967) across six datasets. Further research using all cell states or the latest cell states of LSTM inputs demonstrated its architecture particularly adept with capturing additive and epistatic QTL effects among SNPs. In conclusion, our findings provide basic principles for implementing GP in breeding project to maximize prediction accuracy while maintaining cost-effectiveness.

Modern cultivated rice plays a pivotal role in global food security. China accounts for nearly 30% of the world's rice production and has developed numerous cultivated varieties over the past decades that are well adapted to diverse growing regions. However, the genomic bases underlying the phenotypes of these modern cultivars remain poorly characterized, limiting the exploitation of this vast resource for breeding specialized, regionally adapted cultivars. In this study, we constructed a comprehensive genetic variation map of modern rice using resequencing datasets from 6044 representative cultivars from five major rice-growing regions in China. Our genomic and phenotypic analyses of this diversity panel revealed regional preferences for specific genomic backgrounds and traits, such as heading date, biotic/abiotic stress resistance, and grain shape, which are crucial for adaptation to local conditions and consumer preferences. We identified 3131 quantitative trait loci associated with 53 phenotypes across 212 datasets under various environmental conditions through genome-wide association studies. Notably, we cloned and functionally verified a novel gene related to grain length, OsGL3.6. By integrating multiple datasets, we developed RiceAtlas, a versatile multi-scale toolkit for rice breeding design. We successfully utilized the RiceAtlas breeding design function to rapidly improve the grain shape of the Suigeng4 cultivar. These valuable resources enhance our understanding of the adaptability and breeding requirements of modern rice and can facilitate advances in future rice-breeding initiatives.

>Tiller number is a crucial agronomic trait for achieving high yield in rice. NAC proteins play critical roles in regulating plant growth and development. However, the role of NAC transcription factors in regulating tiller number remains poorly understood. In this study, we isolated a rice NAC gene, OsNAC022, which is conserved in cereal crops and functions as a transcriptional activator.

Rice blast, caused by the fungus Magnaporthe oryzae, reduces rice yields by 10% to 35%. Incorporating blast resistance genes into breeding programs is an effective strategy to combat this disease. Understanding the genetic variants that confer resistance is crucial to this strategy. The gene Bsr-d1 encodes a C2H2-like transcription factor, and its recessive allele confers broad-spectrum resistance against infections by various strains of M. oryzae. In this study, we investigated the molecular evolution of the rice blast resistance gene bsr-d1 in a representative population consisting of 827 cultivated and wild rice accessions. Our results revealed that wild rice exhibited significantly higher nucleotide diversity, with polymorphic regions primarily concentrated in the promoter region, in contrast to indica and japonica rice varieties. The Bsr-d1 gene displayed significant differentiation between indica and japonica rice varieties, with the bsr-d1 resistance allele being unique to indica rice. Haplotype network and phylogenetic analyses suggested that the bsr-d1 resistance allele most likely originated from Oryza nivara in the region adjacent to the Indian Peninsula and the Indochina Peninsula. Moreover, we explored the utilization of bsr-d1 resistance alleles in China and designed a pair of DNA primers based on the polymorphic sites for the detection of the bsr-d1 resistance gene. In summary, our study uncovering the origin and evolution of bsr-d1 will enhance our understanding of resistance gene variation and expedite the resistance breeding process.

Nucleotide-binding site and leucine-rich repeat (NLR) proteins are activated by detecting pathogen effectors, which in turn trigger host defenses and cell death. Although many NLRs have been identified, the mechanism responsible for NLR-triggered defense responses are still poorly understood. In this study, through GWAS approach, we identified a novel NLR gene, Blast Resistance Gene 8 (BRG8), conferring resistance to rice blast and bacterial blight diseases. Consistently, the BRG8 overexpression and complementation lines exhibited enhanced resistance to both pathogens. Subcellular localization assays showed that BRG8 localized in both cytoplasm and nucleus. More evidence revealed that nuclear-localized BRG8 enhanced rice immunity without hypersensitive response (HR)-like phenotype. Furthermore, we also demonstrated the CC domain of BRG8 not only physically interacted with itself, but also interacted with the KNOX Ⅱ protein HOMEOBOX ORYZA SATIVA59 (HOS59). Knockout of HOS59 in BRG8 background showed enhanced resistance to M. oryzae strain CH171 and Xoo strain CR4, similar to BRG8 background. In contrast, overexpression of HOS59 in BRG8 background, compromised the HR-like phenotype and resistance response. Further analysis revealed that HOS59 promotes the degradation of BRG8 via the 26S proteasome pathway. Collectively, our study highlights HOS59 as NLR immune regulators, fine-tune BRG8-mediated immune responses against pathogens, and provides new insights into NLR association and function in plant immunity.

>Plant architecture is a collection of major agronomic traits that determines rice grain production, and it is mainly influenced by tillering, tiller angle, plant height and panicle morphology(Wang and Li 2006). Tiller angle is one of the critical components that determines rice plant architecture, which in turn influences grain yield mainly due to its large impact on plant density(Wang et al. 2022). Tiller angle plays a key role in the growth and development of various organs within plants and among plants within the population, which influences the canopy structure of rice, thereby affecting plant morphology and the photosynthetic efficiency of rice leaves, and ultimately impacts the yield of rice(Wei et al. 2022).

Tiller angle is a key agricultural trait that establishes plant architecture, which in turn strongly affects grain yield by influencing planting density in rice. The shoot gravity response plays a crucial role in the regulation of tiller angle in rice, but the underlying molecular mechanism is largely unknown. Here, we report the identification of the BIG TILLER ANGLE2 (BTA2), which regulates tiller angle by controlling the shoot gravity response in rice. Loss-of-function mutation of BTA2 dramatically reduced auxin content and affected auxin distribution in rice shoot base, leading to impaired gravitropism and therefore a big tiller angle. BTA2 interacted with AUXIN RESPONSE FACTOR7 (ARF7) to modulate rice tiller angle through the gravity signaling pathway. The BTA2 protein was highly conserved during evolution. Sequence variation in the BTA2 promoter of indica cultivars harboring a less expressed BTA2 allele caused lower BTA2 expression in shoot base and thus wide tiller angle during rice domestication. Overexpression of BTA2 significantly increased grain yield in the elite rice cultivar Huanghuazhan under appropriate dense planting conditions. Our findings thus uncovered the BTA2-ARF7 module that regulates tiller angle by mediating the shoot gravity response. Our work offers a target for genetic manipulation of plant architecture and valuable information for crop improvement by producing the ideal plant type.

Bacterial blight, a major disease in rice, poses a serious impact on rice production. In this study, a doubled haploid (DH) population derived from a cross between the introduced japonica cultivar 'Maybelle' and the indica landrace 'Baiyeqiu' was used to investigate the pathogenicity of four pathogen races causing bacterial blight. The results showed that the pathogenicity of all the pathogen races exhibited continuous, transgressive distribution in the DH population. Moreover, strong correlations existed between every two pathogen races, with the correlation coefficients ranging from 0.3 to 0.6. A total of 12 quantitative trait loci (QTLs) distributed on chromosomes 1, 2, 3, 5, 6, 7, 9, and 12 were detected for rice bacterial blight, explaining 4.95% to 16.05% of the phenotype. Among these QTLs, a major QTL located in the interval RM6024-RM163 on chromosome 5 was detected in three pathogen races. In addition, the pyramiding of the positive alleles can apparently improve the rice resistance to bacterial blight. This study is of great significance for broadening the genetic resources with resistance to bacterial blight in China.

Accurately identifying varieties with targeted agronomic traits was thought to contribute to genetic selection and accelerate rice breeding progress. Genomic selection (GS) is a promising technique that uses markers covering the whole genome to predict the genomic-estimated breeding values (GEBV), with the ability to select before phenotypes are measured. To choose the appropriate GS models for breeding work, we analyzed the predictability of nine agronomic traits measured from a population of 459 diverse rice varieties. By the comparison of eight representative GS models, we found that the prediction accuracies ranged from 0.407 to 0.896, with reproducing kernel Hilbert space (RKHS) having the highest predictive ability in most traits. Further results demonstrated the predictivity of GS is altered by several factors. Moreover, we assessed the method of integrating genome-wide association study (GWAS) into various GS models. The predictabilities of GS combined peak-associated markers generated from six different GWAS models were significantly different; a recommendation of Mixed Linear Model (MLM)-RKHS was given for the GWAS-GS-integrated prediction. Finally, based on the above result, we experimented with applying the P -values obtained from optimal GWAS models into ridge regression best linear unbiased prediction (rrBLUP), which benefited the low predictive traits in rice.

Grain size and flag leaf angle are two important traits that determining grain yield in rice. However, the mechanisms regulating these two traits remain largely unknown. In this study, a rice long grain 5 (lg5) mutant with a large flag leaf angle was identified, and map-based cloning revealed that a single base substitution followed by a 2 bp insertion in the LOC_Os05g40384 gene resulted in larger grains, a larger flag leaf angle, and higher plant height than the wild type. Sequence analysis revealed that lg5 is a novel allele of elongated uppermost internode-1 (EUI1), which encodes a cytochrome P450 protein. Functional complementation and overexpression tests showed that LG5 can rescue the bigger grain size and larger flag leaf angle in the Xiushui11 (XS) background. Knockdown of the LG5 transcription level by RNA interference resulted in elevated grain size and flag leaf angle in the Nipponbare (NIP) background. Morphological and cellular analyses suggested that LG5 regulated grain size and flag leaf angle by promoting cell expansion and cell proliferation. Our results provided new insight into the functions of EUI1 in rice, especially in regulating grain size and flag leaf angle, indicating a potential target for the improvement of rice breeding.

Because plant mechanical strength influences plant growth and development, the regulatory mechanisms underlying cell-wall synthesis deserve investigation. Rice mutants are useful for such research. We have identified a novel brittle culm 25 (bc25) mutant with reduced growth and partial sterility. BC25 encodes an UDP-glucuronic acid decarboxylase involved in cellulose synthesis and belongs to the UXS family. A single-nucleotide mutation in BC25 accounts for its altered cell morphology and cell-wall composition. Transmission electron microscopy analysis showed that the thickness of the secondary cell wall was reduced in bc25. Monosaccharide analysis revealed significant increases in content of rhamnose and arabinose but not of other monosaccharides, indicating that BC25 was involved in xylose synthesis with some level of functional redundancy. Enzymatic assays suggested that BC25 functions with high activity to interconvert UDP-glucuronic acid (UDP-GlcA) and UDP-xylose. GUS staining showed that BC25 was ubiquitously expressed with higher expression in culm, root and sheath, in agreement with that shown by quantitative real-time (qRT)-PCR. RNA-seq further suggested that BC25 is involved in sugar metabolism. We conclude that BC25 strongly influences rice cell wall formation.

非洲栽培稻作为重要的水稻种质资源,其基因渗入系可以为普通栽培稻的遗传背景提供新的有利基因,如果能将这些优良基因引入普通栽培稻中,可为水稻分子设计育种提供新的基因资源.本研究以非洲栽培稻基因渗入系YIL60与轮回亲本中9B(Z9B)杂交衍生的包含188个株系的F2和F2∶3群体为材料,对粒形性状包括粒长、粒宽、籽粒长宽比、千粒重,剑叶形态性状包括剑叶长、剑叶宽进行数量性状点位(QTL)检测.结果共检测到16个QTL,包括2个粒长QTL、3个粒宽QTL、2个籽粒长宽比QTL、2个千粒重QTL、1个剑叶长QTL、6个剑叶宽QTL,分布于第1、第6、第7、第10和第11号染色体上,贡献率为2.25％～25.64％;有4个多效性 QTL 区间,有4个 QTL qGW7-1、qFLL10、qFLW10、qTGW7在 F2和F2∶3群体中被重复检测到,其中在第7号染色体RM3859-RM3394区间检测到同时控制粒长、粒宽、籽粒长宽比和千粒重的QTL,贡献率最高达17.10％,是一个来源于非洲栽培稻的新粒形QTL.本研究为进一步开展粒形、剑叶形态性状基因的精细定位、克隆和分子标记辅助育种工作奠定了 一定的理论基础.

Grain size, grain number per panicle, and grain weight are key agronomic traits that determine grain yield in rice. However, the molecular mechanisms coordinately controlling these traits remain largely unknown. In this study, we identified a major QTL, SMG3, that is responsible for grain size, grain number per panicle, and grain weight in rice, which encodes a MYB-like protein. The SMG3 allele from M494 causes an increase in the number of grains per panicle but produces smaller grain size and thousand grain weight. The SMG3 is constitutively expressed in various organs in rice, and the SMG3 protein is located in the nucleus. Microscopy analysis shows that SMG3 mainly produces long grains by increasing in both cell length and cell number in the length direction, which thus enhances grain weight by promoting cell expansion and cell proliferation. Overexpression of SMG3 in rice produces a phenotype with more grains but reduces grain length and weight. Our results reveal that SMG3 plays an important role in the coordinated regulation of grain size, grain number per panicle, and grain weight, providing a new insight into synergistical modification on the grain appearance quality, grain number per panicle, and grain weight in rice.

The annual planting area of major inbred rice (Oryza sativa) cultivars reach more than half of the total annual planting area of inbred rice cultivars in China. However, how the major inbred rice cultivars changed during decades of genetic improvement and why they can be prevalently cultivated in China remains unclear. Here, we investigated the underlying genetic changes of major inbred cultivars and the contributions of landraces and introduced cultivars during the improvement by resequencing a collection of 439 rice accessions including major inbred cultivars, landraces, and introduced cultivars. The results showed that landraces were the main genetic contribution sources of major inbred Xian (Indica) cultivars, while introduced cultivars were that of major inbred Geng (Japonica) cultivars. Selection scans and haplotype frequency analysis shed light on the reflections of some well-known genes in rice improvement, and breeders had different preferences for the Xian's and Geng's breeding. Six candidate regions associated with agronomic traits were identified by genome-wide association mapping, five of which were under positive selection in rice improvement. Our study provides a comprehensive insight into the development of major inbred rice cultivars and lays the foundation for genomics-based breeding in rice.

Milling and appearance quality are important contributors to rice grain quality. Abundant genetic diversity and a suitable environment are crucial for rice improvement. In this study, we investigated the milling and appearance quality-related traits in a panel of 200 japonica rice cultivars selected from Liaoning, Jilin and Heilongjiang provinces in Northeast China. Pedigree assessment and genetic diversity analysis indicated that cultivars from Jilin harbored the highest genetic diversity among the three geographic regions. An evaluation of grain quality indicated that cultivars from Liaoning showed superior milling quality, whereas cultivars from Heilongjiang tended to exhibit superior appearance quality. Single- and multi-locus genome-wide association studies (GWAS) were conducted to identify loci associated with milling and appearance quality-related traits. Ninety-nine significant single-nucleotide polymorphisms (SNPs) were detected. Three common SNPs were detected using the mixed linear model (MLM), mrMLM, and FASTmrMLM methods. Linkage disequilibrium decay was estimated and indicated three candidate regions (qBRR-1, qBRR-9 and qDEC-3) for further candidate gene analysis. More than 300 genes were located in these candidate regions. Gene Ontology (GO) analysis was performed to discover the potential candidate genes. Genetic diversity analysis of the candidate regions revealed that qBRR-9 may have been subject to strong selection during breeding. These results provide information that will be valuable for the improvement of grain quality in rice breeding.

Asian cultivated rice (Oryza sativa L.) is divided into two subspecies of xian and geng. With the development of hybrid rice and utilization of interspecific heterosis, the boundaries between xian and geng are becoming more and more vague. In this study, based on the SNP-index value of 20 million single nucleotide polymorphism (SNP) loci from 3000 rice germplasm resources, we captured 4084 xian-geng specific SNP loci named as 4k-SNP and used the xian-geng index as an indicator for xian-geng identification. Furthermore, the 4k-SNP was reduced to 40-SNP (40 SNP loci) for indica/japonica identification by using the statistical analysis methods such as large-scale simple random sampling based on the dimensionality reduction algorithms. To verify the effectiveness of 40-SNP on xian-geng identification, 82 bred varieties were used in this study to compare the results of 40-SNP xian-geng identification and 4k-SNP identification. The result showed that the geng index obtained from 40-SNP and 4k-SNP were very close, and the correlation coefficient was 0.99. Moreover, a total of 49 varieties, belonging to six subgroups (indica, aus, rayada, aromatic, tropical japonica, and temperate japonica), were used to compare the xian-geng identification results of 40-SNP with those of 4k-SNP and Cheng’s index. And the correlation coefficients of xian-geng identifications between 40-SNP and 4k-SNP as well as between 40-SNP and Cheng’s index were above 0.98 and 0.86, respectively. These results verified the validity and accuracy of 40-SNP on xian-geng identification in Oryza sativa L. In addition, 40-SNP also had a good distinguishability for the six subgroups in rice, and the xian-geng index of indica, aus, rayada, aromatic, tropical japonica, and, temperate japonica was less than 0.20, 0.20-0.40, 0.60-0.85, more than 0.90, and 1.00, respectively. This study provides the data and theoretical basis for the differentiation of xian-geng and the utilization of heterosis and, formulation of seed management regulations.

Grain weight and size, mostly determined by grain length, width and thickness, are crucial traits affecting grain quality and yield in rice. A quantitative trait locus controlling grain length and width in rice, qGS1-35.2, was previously fine-mapped in a 57.7-kb region on the long arm of chromosome 1. In this study, OsPUB3, a gene encoding a U-box E3 ubiquitin ligase, was validated as the causal gene for qGS1-35.2. The effects were confirmed firstly by using CRISPR/Cas9-based mutagenesis and then through transgenic complementation of a Cas9-free knock-out (KO) mutant. Two homozygous KO lines were produced, each having a 1-bp insertion in OsPUB3 which caused frameshift mutation and premature termination. Compared with the recipient and a transgenic-negative control, both mutants showed significant decreases in grain weight and size. In transgenic complementation populations derived from four independent T0 plants, grain weight of transgenic-positive plants was significantly higher than transgenic-negative plants, coming with increased grain length and a less significant decrease in grain width. Based on data documented in RiceVarMap V2.0, eight haplotypes were classified according to six single-nucleotide polymorphisms (SNPs) found in the OsPUB3 coding region of 4695 rice accessions. Significant differences on grain size traits were detected between the three major haplotypes, Hap1, Hap2 and Hap3 that jointly occupy 98.6% of the accessions. Hap3 having the largest grain weight and grain length but intermediate grain width exhibits a potential for simultaneously improving grain yield and quality. In another set of 257 indica rice cultivars tested in our study, Hap1 and Hap2 remained to be the two largest groups. Their differences on grain weight and size were significant in the background of non-functional gse5, but non-significant in the background of functional GSE5, indicating a genetic interaction between OsPUB3 and GSE5. Cloning of OsPUB3 provides a new gene resource for investigating the regulation of grain weight and size.