Harnessing Endogenous Homeobox Genes for Synthetic Apomixis in Hybrid Rice

Heterosis significantly boosts crop productivity and adaptability, yet this advantage erodes in progeny due to genetic segregation, necessitating costly annual hybrid seed production (Underwood and Mercier 2022). Synthetic apomixis has attracted considerable attention in breeding for its potential to fix heterosis. We previously engineered a synthetic apomictic hybrid rice, ‘Fixation of Hybrids (Fix) ’, in which heterozygosity was fixed by integrating the OsMTL knockout with a ‘Mitosis instead of Meiosis (MiMe) ’ system through simultaneous disruption of OsOSD1, OsPAIR1 and OsREC8 (Wang et al. 2019). Concurrently, apomixis was achieved by combining MiMe with egg cell-specific expression of OsBBM1 (Khanday et al. 2019). Later, Fix2 and Fix3 were developed through OsBBM4 or OsWUS mediated engineering (Huang et al. 2025; Wei et al. 2023). Here, we explore whether other endogenous genes can be harnessed for synthetic apomixis in hybrid rice. Homeobox genes, as transcription factors, are known to alter the developmental fates of cells, tissues, and organs across a wide range of organisms. In Drosophila, by fusing the gene coding sequence to an inducible promoter and reintroducing it into the germline, the body plan can be altered in a predictable manner (Gehring 1987). BEL homeobox proteins, which represent one of the oldest subclasses in plant homeobox families, play key roles in the initiation and maintenance of reproductive development. In Physcomitrella patens, ectopic overexpression of the BEL homeobox gene induces gametophyte-to-sporophyte transition and fertilisation-independent embryogenesis (Horst et al. 2016). Building on these findings, we sought to determine whether targeted engineering of endogenous BEL-like homeobox genes could lead to apomictic reproduction in rice. Analysis of the PlantTFDB identified 13 BEL homeobox genes in rice (Figure S1), which exhibit evolutionary conservation in key functional domains such as the BEL domain and the homeodomain (Figure S2). To delineate their expression profiles, we leveraged publicly available transcriptome datasets and performed quantitative real-time PCR (qRT-PCR). LOCOs03g47740 and LOCOs12g43950 displayed significantly higher transcript abundance across multiple tissues compared to the other 11 genes (Figure S3a), suggesting critical roles in growth and development. Notably, both genes showed stronger expression in vegetative organs than in reproductive tissues (Figure S3b), a pattern similar to that of OsBBM4, which has been used in haploid induction and synthetic apomixis (Wei et al. 2023). Reverse transcription PCR (RT-PCR) validated their sequences, which contained 5 and 4 coding exons, respectively (Figure S4). Given their conserved BEL homeobox domain, we designated these two rice genes as BEL-LIKE 1 (BEL1) and BEL-LIKE 2 (BEL2), respectively. The expression similarity to OsBBM4 led us to investigate whether they could also be engineered to induce haploid formation and apomixis in rice. To test whether BEL1 or BEL2 can induce haploids in rice, we used the Arabidopsis DD45 promoter to drive egg cell-specific overexpression (Figure 1a). The elite intersubspecific hybrid rice Chunyou84 (CY84) was used for Agrobacterium-mediated transformation, producing 14 EE-BEL1 and 11 EE-BEL2 T0 transgenic plants. These plants exhibited significantly higher expression levels of BEL1 and BEL2, respectively, in pre-flowering pistils, along with normal vegetative growth, compared with the wild-type (WT) (Figures S5 and S6a, c). However, their seed-setting rates showed wide variation, ranging from 14. 3% ± 1. 9% to 83. 6% ± 1. 0% (WT: 82. 1% ± 4. 0%) (Figure S6b; Table S1). To validate haploid production from EE-BEL1/EE-BEL2 T0 plants, we genotyped their self-pollinated progeny using 12 InDel markers (one per chromosome). While no WT progeny (n = 186) showed complete monomorphism, 5 out of 277 plants from three EE-BEL1 T0 lines and 4 out of 505 plants from three EE-BEL2 T0 lines were fully monomorphic across all markers (Figure S7a; Table S1). Further flow cytometry confirmed that all these plants were indeed haploids (Figure S7b). To clarify the genotypes of these identified haploids, we performed whole-genome resequencing and plotted a binmap using 581 737 SNPs (bin size: 1 Mb) across four haploids and two CY84 recombinant inbred diploids (RIDs) (Figure 1b). The RIDs exhibited stochastic heterozygosity, whereas all haploids displayed complete genome-wide homozygosity and showed recombination relative to the parental genome, suggesting that they were each derived from a single gamete. These haploid plants were characterised by reduced plant height, shorter panicles, smaller florets and complete sterility (Figure 1c). The above results demonstrate that enhancing BEL1/BEL2 expression in egg cells induces haploid production in rice. Synergizing haploid induction with clonal gamete formation enables apomixis initiation. To implement this via BEL1/BEL2 engineering, we combined their enhanced expression in egg cells with the MiMe system (Figure 1d) and introduced these two constructs into CY84. Among the 13 and 12 T0 transgenic plants, three plants from each group were identified as having triple osd1/pair1/rec8 mutations (Figures S8 and S9) and were designated. We referred to these plants as Fix6. 1 and Fix6. 2, respectively, for agronomic evaluation. The statistical results for plant height, panicle length, and productive tiller number showed no significant differences compared with the WT (Figure 1e; Figure S10). Notably, both the Fix6. 1 and Fix6. 2 exhibited high seed-setting rate, ranging from 81. 5% ± 2. 2% to 85. 9% ± 1. 2%, which was comparable to that of the WT (83. 4% ± 1. 7%) (Table S2). These findings suggest that integrating MiMe with enhanced BEL1/BEL2 expression in egg cells does not compromise plant developmental integrity. Next, to ascertain whether Fix6. 1 and Fix6. 2 undergo apomixis and produce clonal seeds, we performed ploidy identification on their T1 plants using flow cytometry. For Fix6. 1, the diploid progeny counts were 2/96, 6/192 and 1/105 from three T0 lines, with cloning efficiencies ranging from 1. 0% to 3. 1%; the rest were tetraploids (Figure S11a; Table S2). Concurrently, for Fix6. 2, the diploid progeny counts were 6/96, 1/140 and 1/96 from three T0 lines, with cloning efficiencies ranging from 0. 7% to 6. 3% (Figure S11a; Table S2). To access genome-wide heterozygosity in these clonal T1 plants, we conducted whole-genome resequencing with an average coverage of 30-fold. Analysis of 1. 62 million SNPs revealed heterozygosity patterns identical to those of their T0 generation and the hybrid rice CY84 (Figure 1b), validating complete fixation of the heterozygous genotype. These results indicate that BEL1 or BEL2 engineering enables the implementation of apomixis while maintaining normal growth and seed production. Lastly, to evaluate whether the features of Fix6. 1 and Fix6. 2 could be stably inherited by clonal progenies, we investigated clonal T1 plants, two T1 tetraploids and WT under field conditions. Agronomic trait surveys revealed that these clonal T1 plants exhibited no significant differences in plant, panicle and grain morphology compared to WT (Figure 1f, g; Figure S11b, c), indicating stable inheritance of vegetative growth traits. The tetraploids exhibited typical features such as thicker stems and elongated seed awns Figure S12). We randomly selected one line each from the Fix6. 1 and Fix6. 2 clonal T1 plants as representatives to determine the seed-setting rate and cloning efficiency. Their seed-setting rates were 84. 4% ± 2. 4% and 83. 1% ± 3. 1%, respectively, showing no significant difference from that of the WT (82. 3% ± 1. 3%) (Table S3). Flow cytometric ploidy analysis revealed cloning efficiencies of 2. 6% and 2. 0%, consistent with those of the T0 generation (Tables S2 and S3). In addition, whole-genome resequencing verified that these clonal T2 plants retained heterozygosity levels equivalent to those of CY84 and their T0 and T1 generations (Figure 1b). These findings reveal the stable inheritance of normal vegetative growth, high fertility and cloning efficiency in Fix6. 1 and Fix6. 2. Our studies in rice demonstrate that fusing an endogenous BEL homeobox gene (BEL1 or BEL2) with an egg cell-specific promoter enables haploid induction. By integrating this haploid induction with the MiMe system, we established two innovative synthetic apomixis systems, Fix6. 1 and Fix6. 2, in hybrid rice. These systems maintain normal development and high fertility while achieving fixation and stable inheritance of superior traits through seed-derived clonal offspring. Overall, our study advances both functional insights into BEL homeobox genes and new strategies for synthetic apomixis. Given that these proteins are widely present across species, this approach has broad applicability in crop breeding. K. W. and Q. Q. managed the project. X. W. , C. L. , Y. H. and D. R. performed the experiments. X. W. , T. S. and H. L. analysed the data. X. W. wrote the manuscript. K. W. revised the manuscript. This work was supported by the National Key Research and Development Program of China (2022YFF1003304 and 2023YFD1200800), the National Natural Science Foundation of China (32188102, 32025028, 32301861 and U20A2030), the Postdoctoral Fellowship Program and China Postdoctoral Science Foundation (BX20230422) and the Central Publicinterest Scientific Institution Basal Research Fund (Y2022QC20). The data supporting the findings are available in the Supporting Information. Whole-genome resequencing data generated in this study have been deposited in the NCBI SRA database under accession number PRJNA1271228. Gene sequence annotations correspond to those from the Rice Genome Project (http: //rice. plantbiology. msu. edu/) for LOCOs03g47740 (BEL1), LOCOs12g43950 (BEL2), LOCOs02g37850 (OSD1), LOCOs03g01590 (PAIR1) and LOCOs05g50410 (REC8). Figures S1–S12. Tables S1–S4. Appendix S1 Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.