INTRODUCTION
Heterosis, or hybrid vigor, refers to the superior performance of hybrid offspring compared to their inbred parents. This phenomenon has been extensively exploited in major crops such as rice and maize, driving substantial improvements in global agricultural productivity
1. For instance, hybrid rice varieties typically outperform their inbred counterparts by 20–30% in yield
2,3. However, in sexual reproduction, hybrids undergo genetic segregation in subsequent generations, leading to variable phenotypes and a loss of heterotic traits. As a result, hybrid seeds must be regenerated each season through technically demanding and costly crossing procedures
4,5. Engineering clonal reproduction through seeds — known as synthetic apomixis — offers a promising strategy to permanently fix heterosis and preserve elite heterozygous genotypes across generations
6-11. To realize this goal at scale, it is essential to achieve clonal seed production with complete or near-complete penetrance, while maintaining normal seed set and yield potential
12-14.
Currently, two main strategies have been developed for synthetic apomixis in hybrid rice. The first employs multiplex genome editing to simultaneously disrupt three meiosis-related genes, thereby generating
Mitosis instead of Meiosis (
MiMe), which produces clonal diploid gametes
15,16, as well as genes that trigger haploid induction such as
OsMTL17,
OsPLDα218, and
OspPLAIIκ19. While this strategy could theoretically avoid transgene integration, its clonal efficiency remains considerably low
17-20. The second combines the
MiMe system with egg cell-specific expression of parthenogenesis-inducing genes such as
BABY BOOM 1 (
OsBBM1)
21,22,
OsBBM423,
WUSCHEL (
OsWUS)
24, and
OsBEL1/225, as well as the heterologous
PARTHENOGENESIS genes
PpPAR and
ToPAR26-29. Among these,
OsBBM1-based synthetic apomixis has achieved the highest clonal efficiencies reported to date. Optimized
OsBBM1-based systems can achieve clonal efficiency exceeding 95%; however, clonal efficiency varies widely among independent lines
30-32. Critically, a clear trade-off exists between clonal efficiency and seed fertility, with highly clonal lines frequently exhibiting substantially reduced fertility
30,31. Similar challenges persist with other parthenogenesis inducers, where improvements in fertility are often accompanied by low or suboptimal clonal efficiency
23-28. As underscored, for apomixis to be agriculturally applicable, the system must be broadly effective and virtually fully penetrant — yielding close to 100% clonal seeds — while maintaining normal fertility and yield potential
12-14. This highlights the need for parthenogenesis inducers with broader applicability, higher penetrance, and reduced pleiotropic effects.
A long-standing hypothesis posits that egg cells remain mitotically quiescent and require sperm-derived factors to initiate cell division and embryogenesis
33-37. Indeed, the egg cell is transcriptionally poised and transcriptional programs are rapidly activated upon fertilization
38-42. Interestingly, all genes identified to date that can trigger parthenogenesis when ectopically expressed in the egg cell encode transcription factors with broad expression patterns and diverse developmental roles
21,23-25,43-45. Among them,
OsBBM1 exhibits the most restricted expression profile: highly expressed not only in sperms and zygotes but also throughout subsequent embryo development, suggesting a role beyond the initial induction of embryogenesis
21,40. These observations raise the question of whether any factors are exclusively dedicated to initiating embryogenesis. Such factors — potentially restricted to sperms and early zygotes — could provide a refined toolkit for engineering synthetic apomixis with improved applicability, penetrance, and fertility.
Here, we identified a sperm-specific transcription factor in rice and found that its ectopic expression in egg cells efficiently induced parthenogenesis and haploid formation. When combined with MiMe-mediated clonal gametogenesis, the strategy achieved broadly applicable synthetic apomixis with near-100% penetrance — clonal seed production rates exceeding 99% across all tested hybrid cultivars, derived lines, and successive generations. Notably, this system enables efficient generation of apomictic hybrid lines that achieve near-100% clonal efficiency without compromising seed yields relative to conventional F1 hybrids. These findings establish a broadly applicable synthetic apomixis with near-100% penetrance and agronomic viability, offering a scalable route to fix heterosis in self-perpetuating hybrid seed systems.
RESULTS
Identification of sperm-specific and delivered transcription factors
To identify sperm-specific transcription factors capable of inducing parthenogenesis, we analyzed 8,795 publicly available rice RNA-seq datasets (Supplementary Table S1)
43. Batch effects were mitigated by converting the FPKM matrix to TPM values. Tissue-specific enrichment analysis identified 729 sperm-specific genes (Supplementary Fig. S1a and Table S2). Sperm-delivered factors were expected to show high expression in sperms, presence in zygotes (derived from paternal transcripts), and absence or minimal expression in egg cells. Time-series expression profiling across sperms, eggs, zygotes, and early embryos revealed ten clusters (Supplementary Fig. S1b and Table S2); genes in clusters 1 and 2 matched this delivery signature and were prioritized as candidates for embryogenesis initiation (Fig. 1a; Supplementary Fig. S1b). Of the 170 genes in these clusters, five encoded predicted transcription factors:
LOC_Os12g12970 and
LOC_Os09g27190 (cluster 1);
LOC_Os05g41166,
LOC_Os11g07700, and
LOC_Os03g18340 (cluster 2). Co-expression network analysis using the normalized TPM matrix grouped four of these five factors into a single regulatory module (Fig. 1b; Supplementary Table S1). Within this module,
LOC_Os12g12970 exhibited the highest connectivity and average co-expression weight, indicating a central regulatory role (Fig. 1b).
LOC_Os12g12970 encodes the RKD-family transcription factor OsRKD8
46. We renamed it as
HUAXU (
华胥), after a mythological female figure who, according to ancient Chinese texts, conceived Fuxi (伏羲) and Nvwa (女娲) upon stepping into a divine footprint. This name is notable for the factor’s ability to induce parthenogenesis and establish a maternal inheritance system (described below)—this new nomenclature not only simplifies reference but also symbolically reflects the gene’s maternal-side engineering in inducing embryogenesis without paternal DNA contribution.
Egg cell-specific expression of HUAXU triggers parthenogenetic embryogenesis in hybrid rice
To test whether
HUAXU could induce parthenogenesis when expressed in the egg cell, we generated a binary construct (HUAXU-ee; Fig. 1c) that placed
HUAXU under the egg cell-specific
OsECA promoter
47. This construct was transformed into the hybrid rice cultivar Chunyou 84 (CY84; wild-type (WT)) via
Agrobacterium-mediated callus transformation, yielding four independent transgenic lines (Supplementary Fig. S1c, d). To determine whether egg cell-expressed
HUAXU initiates fertilization-independent embryogenesis, we examined unpollinated pistils from emasculated T
0 plants at 3 days after emasculation (3 DAE) using Eosin B staining. WT pistils consistently showed a single egg cell and two undivided central nuclei with no embryogenesis (
n = 30; Fig. 1d, left). In contrast, 18 of 38
HUAXU-ee pistils contained multicellular embryo-like structures accompanied by unfertilized central cells, indicative of autonomous embryo development (Fig. 1d, right). These findings demonstrate that egg cell-specific
HUAXU expression efficiently triggers parthenogenetic embryogenesis in the absence of fertilization. Notably, emasculated
HUAXU-ee pistils failed to initiate endosperm development (Fig. 1d; Supplementary Fig. S1e). Consistently, mature seeds from
HUAXU-ee plants exhibited triploid endosperm comparable to those of WT (Supplementary Figs. S1f, S3b), indicating that endosperm formation remains fertilization-dependent.
Egg cell-specific expression of HUAXU efficiently induces haploid progeny
Engineered parthenogenesis enables haploid embryo and progeny production, a key approach for haploid induction in plant breeding
48,49. To assess whether egg cell-expressed
HUAXU triggers haploid formation, we first examined field performance of
HUAXU-ee T
0 plants. These lines showed vegetative and reproductive growth comparable to WT controls (Supplementary Fig. S1g, h). Following flowering, ~20% (6/30) of
HUAXU-ee seeds displayed obviously precocious embryo development compared with WT (Supplementary Fig. S1j). Seed set ranged from 50.3 ± 4.2% to 74.6 ± 4.3% across the four lines, vs 74.3 ± 2.5% in WT (Supplementary Fig. S1h, i). Mature seeds and germinated seedlings appeared morphologically normal compared to WT (Supplementary Fig. S1k, n), while a small fraction (6/38) of mature seeds showed embryos with impaired viability (Supplementary Fig. S1l, m). Flow cytometry revealed haploid progeny in 20.0–57.4% of individuals (54–70 per line), in contrast to exclusively diploid WT progeny (
n = 96; Fig. 1e, f; Supplementary Fig. S1o). Genotyping with 12 insertion-deletion (InDel) markers confirmed partial heterozygosity in a recombinant inbred diploid (RID) control, but uniform single-band patterns in all haploids (Supplementary Fig. S1p and Table S3). Embryos with impaired viability were likewise haploid as determined by marker analysis (Supplementary Fig. S1l, m). Whole-genome resequencing of haploids revealed no heterozygous regions genome-wide, in contrast to heterozygous and homozygous blocks in RID controls (Fig. 1g).
HUAXU-ee-derived haploid plants showed reduced height, smaller glumes, and complete sterility (Fig. 1h, i). Collectively, these data demonstrate that egg cell-specific
HUAXU expression efficiently induces parthenogenesis and haploid progeny in hybrid rice.
Fully penetrant clonal reproduction across lines
To engineer synthetic apomixis, we combined egg cell-specific HUAXU expression with clonal gametogenesis by integrating the HUAXU-ee cassette into the sgMiMe vector, which simultaneously targets OSD1, PAIR1, and REC8 (Fig. 2a). The resulting HUAXU-ee_sgMiMe construct was transformed into the hybrid rice CY84 via Agrobacterium-mediated callus transformation. Hi-TOM genotyping identified five independent T0 lines carrying homozygous or biallelic frameshift mutations in all three MiMe loci, and PCR confirmed the integration of the HUAXU-ee cassette (Supplementary Fig. S2a–c). These lines were designated as Fixation of Hybrids 8 (Fix8). Fix8 T0 plants exhibited morphology and panicle architecture indistinguishable from WT CY84 (Supplementary Fig. S2d, e). Seed-setting rates ranged from 50.4 ± 3.9% to 74.3 ± 2.4% across the five lines, with Fix8 #3 statistically comparable to WT (74.9 ± 2.4%) (Supplementary Fig. S2e, f).
MiMe mutants alone produce exclusively tetraploid progeny (Fig. 2b, d). In contrast, flow cytometry of 464 progeny (80–112 per line) from the five Fix8 T0 plants revealed uniform diploidy, with no tetraploid individuals detected (Fig. 2c, d; Supplementary Fig. S2g). To distinguish clonal reproduction from recombination-based diploidy, we genotyped progeny using 12 genome-wide InDel markers. All Fix8 offspring retained the parental heterozygous pattern identical to CY84, whereas F2 control plants showed extensive homozygosity (Supplementary Fig. S2h). Whole-genome sequencing (20× coverage) of selected Fix8 progeny confirmed complete conservation of heterozygosity across the genome, with no detectable recombination or loss of hybrid parental haplotypes (Fig. 2e). Endosperm in Fix8 seeds was hexaploid (Supplementary Fig. S3a, b), indicating a fertilization-derived origin, and embryo viability in mature Fix8 seeds was not impaired (Supplementary Fig. S3f). Collectively, the invariant diploid ploidy and faithful transmission of the full F1 hybrid heterozygous genotype in all analyzed progeny demonstrate 100% penetrant clonal reproduction in Fix8 lines.
Consistently clonal reproduction and agronomic performance of Fix8 hybrid rice across successive generations
Stable transmission of high seed set and clonal fidelity is critical for synthetic apomixis in agriculture. To assess retention of F1 heterosis in diploid clonal Fix8 progeny (T1 generation), we evaluated key agronomic traits under field conditions. Four of five independent T1 lineages showed no significant differences from the WT F1 hybrid CY84 in plant height, tiller number, or panicle length (Fig. 2f, g; Supplementary Fig. S3c–e); the exceptional dwarf line likely resulted from T-DNA position effects. Seed-setting rates in T1 clonal Fix8 plants ranged from 50.9 ± 3.5% to 72.2 ± 3.0% across lineages, representing 68.0–96.5% of the WT F1 control (74.8 ± 2.7%) (Fig. 2g, h). Line Fix8 #3 exhibited seed set statistically indistinguishable from WT.
Given the consistent near-WT performance of Fix8 #3 across T0 and T1 generations, we selected this line for multi-generation stability testing. In three large T2 populations (n = 1,104, 1,180, and 806 plants) derived from independent T1 individuals, diploid clonal progeny accounted for 1,101 (99.7%), 1,176 (99.7%), and 804 (99.8%) individuals, respectively (Fig. 3a–c). These data establish Fix8 as a near-fully penetrant synthetic apomictic hybrid rice with consistently clonal inheritance. To evaluate environmental robustness, we grew T2 progeny of Fix8 #3 alongside WT F1 plants across multiple field sites differing in soil type, water regime, topography, and management. Clonal Fix8 T2 plants displayed uniform plant and panicle architecture indistinguishable from that of WT F1 across all sites, with no evidence of the phenotypic segregation characteristic of F2 populations (Fig. 3a–d). Seed-setting rates in clonal Fix8 #3 T2 plants ranged from 76.8 ± 6.1% to 79.4 ± 3.8%, closely matching parallel WT controls (77.3 ± 3.2% to 80.7 ± 2.9%) (Fig. 3e). In addition, 1,000-grain weight, apparent amylose content, and total protein levels in clonal Fix8 #3 seeds were comparable to those of WT (Supplementary Fig. S3g–i). Collectively, these results demonstrate that Fix8 #3 confers stable, near-complete clonal reproduction, sustained heterosis-level agronomic performance, and consistent seed production across multiple generations and diverse field environments.
100% clonal efficiency in different hybrid rice backgrounds
We next tested whether genetic background modulates the penetrance of the HUAXU-based system. To this end, we introduced the HUAXU-ee_sgMiMe construct (Fig. 2a) into five additional elite hybrid rice cultivars (Supplementary Tables S4 and S5). Across 23 independent lines from these five cultivars, all progeny (32–64 individuals per line) were exclusively diploid, with no tetraploids detected (Supplementary Fig. S4a–e and Table S6). These diploid plants were morphologically indistinguishable from their respective hybrid controls. However, seed-setting rates varied among lines and cultivars (Supplementary Fig. S4a–j). Nevertheless, normal seed production was also achieved in two additional cultivars and three lines (Supplementary Fig. S4b, c, g, h). Collectively, these results demonstrate that HUAXU-based synthetic apomixis confers fully or near-fully penetrant clonal reproduction that remains stably maintained across diverse elite hybrid rice genetic backgrounds.
Attenuated promoter enables efficient generation of apomictic hybrid lines with near-100% clonal efficiency and normal seed yields
Although all Fix8 lines achieved near-100% clonal efficiency in different backgrounds, seed production was frequently impaired in most lines (Fig. 2h; Supplementary Figs. S2f, S4f–j). Notably, in certain varieties such as CY83, no lines with normal fertility were obtained (Supplementary Fig. S4a, f). Given the consistently near-100% clonal efficiency of Fix8 across all lines, we hypothesized that the expression of HUAXU driven by the OsECA promoter might be excessively strong, thereby accounting for the observed fertility penalty.
To address this, we reasoned that fine-tuning transgene expression levels could alleviate this fertility penalty while maintaining the high efficiency of clonal seed induction. We therefore screened for weaker egg cell-specific promoters using transcriptome data (Supplementary Table S2) and identified 14 genes expressed specifically in egg cells and zygotes (Supplementary Fig. S4k). We selected two candidates — LOC_Os10g29380 and LOC_Os12g07610 (designated EGG1 and EGG2) — to replace the OsECA promoter and generated HUAXU-ee_sgMiMe_Plus1 (HUAXU-ee_sgMiMe_P1) and _P2 constructs (Fig. 4a–c). For direct comparison with the original HUAXU-ee_sgMiMe construct, these constructs were transformed into hybrid rice CY83, in which all previously obtained Fix8 transgenic lines displayed markedly reduced fertility (Fig. 2a; Supplementary Fig. S4f). We subsequently obtained five homozygous MiMe lines for each construct, designated Fix8_Plus1 (Fix8_P1) and Fix8_Plus2 (Fix8P_2), respectively (Supplementary Table S7).
Ploidy analysis of progeny showed that Fix8_P1 lines exhibited highly variable clonal efficiencies, ranging from 13.3% to 100% (Supplementary Fig. S4l). In contrast, four of the five Fix8_P2 lines reached 100% clonal efficiency, with one line at 98.2%. Diploid progeny from both Fix8_P1 and Fix8_P2 displayed no obvious phenotypic differences from the CY83 control (Fig. 4e; Supplementary Fig. S4m). Seed-setting rates in all five clonal Fix8_P1 (from 70.9 ± 2.2% to 75.0 ± 2.0%) or Fix8_P2 (from 70.6 ± 4.2% to 75.5 ± 3.9%) lines were comparable to the control CY83 (73.4 ± 2.7% and 73.9 ± 2.1%) (Fig. 4e, f; Supplementary Fig. S4m, n), representing a marked improvement in fertility over the original OsECA-driven Fix8 in CY83 (Supplementary Fig. S4a, f). Notably, Fix8_P2 consistently achieved normal seed set in all lines without obviously compromising clonality (Fig. 4d–f). We further selected two Fix8_P2 lines with normal seed set and 100% clonal efficiency in the T1 generation for further evaluation in field-scale trials. Among 679 and 766 progeny, we identified 673 and 760 diploid individuals, corresponding to clonal efficiencies of 99.1% and 99.2%, respectively. Their overall agronomic performance and seed-setting rates (80.3 ± 2.9% and 75.8 ± 3.1%, respectively) remained indistinguishable from those of the control CY83 F1 hybrid (78.3 ± 2.8%) (Fig. 4g, h). Thus, fine-tuning egg cell-specific promoter activity enabled the efficient generation of apomictic hybrid lines with near-100% clonal efficiency and normal seed yields.
HUAXU likely drives parthenogenesis independently of OsBBM1
To dissect the mechanisms underlying the high parthenogenetic efficiency conferred by HUAXU, we first generated huaxu knockout mutants by genome editing and systematically profiled their phenotypes (Supplementary Fig. S5a). Alexander staining showed that pollen fertility in huaxu mutants is indistinguishable from that of WT plants (Supplementary Fig. S5b). Likewise, embryo and endosperm development in huaxu seeds at 4 days after flowering was comparable to WT (Supplementary Fig. S5c). At maturity, seed-set rates in the mutants (78.3 ± 2.6% and 79.9 ± 3.8%) remained statistically equivalent to WT (79.7 ± 2.4%), with no detectable defects in mature seed morphology, embryo or endosperm structure, or embryo viability (Supplementary Fig. S5d–h). Seedling shoot and root growth, as well as overall adult plant architecture, were likewise unaffected (Supplementary Fig. S5i, j). The absence of overt phenotypic defects in huaxu mutants is most likely explained by genetic redundancy with other parthenogenesis-inducing loci or multiple sperm cell-specific paralogs in the rice genome (Supplementary Fig. S5k).
As transcription factors typically exert their function through downstream gene regulation, we next sought to identify genes regulated by
HUAXU. RNA-seq analysis of ovaries from
HUAXU-ee lines, in which
HUAXU is ectopically expressed in egg cells, yielded low and variable
HUAXU transcript levels (Supplementary Table S8), likely due to the limited abundance and transient nature of egg cells in
HUAXU-ee lines. To circumvent this limitation, we employed a transient rice protoplast system (Supplementary Fig. S6a, b). RNA-seq performed 16 h post-transfection revealed 9,053 significantly upregulated and 411 downregulated genes in HUAXU-GFP samples, compared to GFP control (Supplementary Fig. S6c, d). Notably, several key parthenogenesis-associated regulators, including
OsBBM121,
OsWUS24,
OsWOX9a50, and
OsBBM221,40, were activated (Supplementary Fig. S6e, f). We further examined whether genes induced during the egg-to-zygote transition were affected. Of 181 genes previously identified as
de novo expressed upon fertilization (i.e., absent in egg cells but expressed in zygotes)
39, 164 were detected in the protoplast dataset, and 93 of these were upregulated in HUAXU-GFP samples (Supplementary Fig. S6g and Table S9). These findings indicate that ectopic
HUAXU expression possibly recapitulates the transcriptional program of fertilization-driven egg-to-zygote transition, thereby promoting parthenogenetic development.
Given that OsBBM1 is a sperm cell-expressed factor known to induce parthenogenesis when ectopically expressed in egg cells, we next asked whether HUAXU functions through OsBBM1. Dual-luciferase assays in tobacco showed that HUAXU does not transactivate the OsBBM1 promoter (Supplementary Fig. S6h, i), and yeast two-hybrid assays detected no direct protein–protein interaction (Supplementary Fig. S6j). To further test this relationship, we generated OsBBM1 knockout mutants in the Fix8 background. Two independent mutant lines were obtained, and in each, all 46 progeny examined were diploid, indicating that clonal efficiency remained near 100% and was unaffected by loss of OsBBM1 (Supplementary Fig. S6k, l). Collectively, these results suggest that HUAXU induces parthenogenesis independently of, or at least not primarily through, OsBBM1.
DISCUSSION
Crop uniformity is a cornerstone of modern agriculture, essential for predictable cultivation management and yield stability. This necessity is reflected in stringent global seed purity standards; for example, in China, inbred rice seeds are required to achieve ≥ 99% purity, and hybrid seeds must surpass 97%. A major hurdle for the application of synthetic apomixis has been simultaneously achieving broad applicability and high penetrance to meet commercial benchmarks without compromising seed set or yield potential. In this study, we present a HUAXU-based synthetic apomixis system that overcomes this challenge. Our system achieves broadly applicable and near-complete penetrance with clonal seed production rates of 99–100% across cultivars, independent lines, and successive generations, even in large populations. This > 99% efficiency represents a sustained minimum rather than a transient maximum, and not only meets but reliably exceeds current seed purity standards, demonstrating its potential for field deployment.
Previous
OsBBM1-based systems typically display a pronounced trade-off between clonal efficiency and fertility, with highly clonal lines often exhibiting substantially reduced seed set
30,31. Moreover, high clonal efficiency is generally limited to select individual lines rather than being universally high across independent transformants
27,28,30-32. In contrast, egg cell-specific ectopic expression of
HUAXU efficiently induces parthenogenesis and haploid formation at rates of up to 57.4% in T
0 plants, without markedly compromising fertility in multiple lines. When combined with clonal gametogenesis, the resulting synthetic apomixis system enables universal clonal efficiency exceeding 99% across all tested rice cultivars, independent lines, and generations, with only moderate variation in fertility, indicating the absence of a trade-off between induction efficiency and seed set. Importantly, fine-tuning
HUAXU expression with the weaker egg cell-specific
EGG2 promoter restores fertility to WT levels in every line in the CY83 background without obviously compromising clonality. This advance enables the efficient production of apomictic hybrids that integrate near-complete clonal propagation with seed yields comparable to those of conventional F
1 hybrids. However, whether this near-complete penetrance of clonal reproduction with normal fertility can be extended to additional varieties, independent lines, generations, and larger populations, and whether it is robust under abiotic stresses such as high temperature during flowering and grain filling, remains to be determined. Future work combining controlled-environment assays with multi-location field trials will be essential to evaluate the stability of clonal propagation and seed set under diverse environmental conditions, and to assess the feasibility of deploying this system at agricultural scale under increasingly variable climates.
Although
HUAXU was identified under the sperm-delivered paradigm, its precise role in fertilization and egg-to-zygote transition remains unexplored.
HUAXU encodes an RKD transcription factor, a family known to regulate somatic embryogenesis. For example, overexpression of
AtRKD4 induces somatic embryogenesis in
Arabidopsis51, while its rice ortholog
OsRKD3 elicits similar phenotypes
52. In citrus,
CiRWP expression is associated with polyembryony, and its homolog
FhRWP promotes embryogenic callus formation
53-55. Together, these findings suggest a major role for
RKD genes in regulating somatic embryogenesis. However, in contrast to the sperm-specific
HUAXU, both
AtRKD4 and
OsRKD3 are expressed in egg cells, precluding their capacity to induce parthenogenesis
51,52,56. Moreover,
AtRKD5 has been shown to suppress
BBM-mediated parthenogenesis
57, suggesting that
RKD family members can also exert antagonistic effects on this process. These findings highlight substantial functional diversification within the
RKD family. Determining the extent to which
HUAXU homologs retain conserved functions and parthenogenetic potential beyond rice will be critical for extending synthetic apomixis to diverse crops.
Although the HUAXU-based system meets essential agronomic requirements for clonal hybrid seed production, it currently relies on T-DNA-mediated ectopic expression in the egg cell. The sgMiMe constructs and selection markers are stably maintained and clonally inherited through apomixis; the persistence of these exogenous sequences poses regulatory challenges for field deployment. Given that HUAXU is native to rice, targeted modifications of its endogenous locus using transient editing offer a pathway toward truly marker- and transgene-free apomixis. Ultimate field application will also require multi-location, multi-year field trials, large-scale seed purity assessments, and comprehensive regulatory evaluation. In conclusion, the HUAXU-based synthetic apomixis system described here provides a broadly applicable, near-100% penetrant, and agronomically viable strategy to permanently fix heterosis in hybrid crops through seeds, paving the way for transformative changes in breeding paradigms and hybrid seed production systems.
MATERIALS AND METHODS
Plant materials
The hybrid rice (Oryza sativa L.) cultivar ‘Chunyou 84’ (CY84), derived from a three-line cross between the late japonica male-sterile line ‘Chunjiang 16A’ (16A) and the indica-japonica intermediate restorer line C84, served as the primary host variety and WT control in this study unless otherwise indicated. The conventional japonica cultivar Nipponbare (Nip) was used as the genetic background to generate huaxu mutants. Five additional hybrid rice cultivars were used to evaluate varietal applicability and to perform promoter optimization fertility experiments (Supplementary Table S4).
Transgenic plants were grown in approved transgenic paddy fields at the National South China Biological Breeding Base in Sanya, China, during the winter season and at the China National Rice Research Institute in Hangzhou, China, during the summer season.
Transcriptome analysis
We retrieved 8,795 publicly available rice RNA-seq datasets from a prior study
43. Datasets were filtered based on stringent criteria: a uniquely mapped read rate greater than 0.2 and a minimum of 1 million total reads. These datasets were used to construct a unified transcripts per million (TPM) expression matrix. Sperm-specific genes were identified using TissueEnrich software
58. Time-series and co-expression network analyses were performed with Mfuzz software using default parameters
59.
Plasmid construction
To generate the egg cell-specific expression vector for
HUAXU, a 2,081-bp promoter region of
OsECA47 and a 5,671-bp genomic fragment of
HUAXU were amplified from rice (
Oryza sativa L. cv. 16A) genomic DNA. Additionally, a 253-bp
NOS terminator was amplified from pHBT-sGFP(S65T)-NOS
60. These fragments were assembled into the
PmeI site of pCambia1300 using one-step cloning
61, yielding pC1300-OsECA:HUAXU (HUAXU-ee).
The binary vector for
Mitosis instead of Meiosis (
MiMe), pC1300-ACT:Cas9-sgRNA
OSD1-sgRNA
PAIR1-sgRNA
REC8, was described previously
17. To enhance homozygous
MiMe efficiency, a second
OSD1 target site (5’-TTGACCGCCACGGCTCCCGGCGG-3’) was designed. The corresponding sgRNA cassette, digested with
KpnI and
NheI, was inserted into the original MiMe vector (digested with
KpnI and
XbaI), producing pC1300-ACT:Cas9-sgRNA2
OSD1-sgRNA1
OSD1-sgRNA
PAIR1-sgRNA
REC8 (sgMiMe construct).
For synthetic apomixis and
Fixation of Hybrid 8 (
Fix8) generation, the egg cell-specific
HUAXU expression cassette was amplified from HUAXU-ee and assembled into
PmeI-digested sgMiMe via Gibson assembly
61, yielding pC1300-ACT:Cas9-sgRNA2
OSD1-sgRNA1
OSD1-sgRNA
PAIR1-sgRNA
REC8-OsECA:HUAXU (HUAXU-ee_sgMiMe).
For HUAXU-ee_sgMiMe_P1 and HUAXU-ee_sgMiMe_P2 constructs, the 691-bp promoter region upstream of the
EGG1 start codon and the 1,104-bp promoter region upstream of the
EGG2 start codon were separately amplified from genomic DNA, and the HUAXU-NOS fragment was amplified from HUAXU-ee. These fragments were assembled into
PmeI-digested sgMiMe via Gibson assembly
61 to produce the respective vectors.
For transient protoplast assays, the UBI:HUAXU-GFP and UBI:GFP constructs were generated as follows: the UBI promoter from pC1300-UBI:Cas9 and GFP-NOS from pPLV04 were amplified for both constructs; additionally, the HUAXU CDS from mature pollen cDNA was amplified for the UBI:HUAXU-GFP construct. These fragments were assembled into the PmeI site of pCambia1300 using one-step cloning.
For targeted knockout of HUAXU and OsBBM1, sgRNA target sequences were synthesized and individually ligated into pC1300-UBI:Cas9 and pC2300-UBI:Cas9 using compatible-end cloning, yielding pC1300-UBI:Cas9-sgRNAHUAXU and pC2300-UBI:Cas9-sgRNAOsBBM1, respectively.
All primers are listed in Supplementary Table S10. All constructs were confirmed by Sanger sequencing, and sequence-verified plasmids were used for Agrobacterium-mediated transformation or transient protoplast expression assays.
Rice transformation and genotyping
Agrobacterium-mediated transformation was performed by EDGENE Biotechnology Co., Ltd. (Wuhan, China) to generate transgenic rice. Genome modifications in
MiMe-associated genes (
OSD1,
PAIR1, and
REC8),
HUAXU, and
OsBBM1 were detected using the high-throughput tracking of mutations (Hi-TOM) method
62. Briefly, genomic DNA was extracted from approximately 100 mg of rice leaf tissue using the cetyltrimethylammonium bromide (CTAB) method. PCR amplification of genomic regions flanking the target sites was conducted using primers listed in Supplementary Table S10. The sequencing was performed following the Hi-TOM protocol as previously described
62. The egg cell-specific ectopic expression fragment
OsECA:HUAXU in T
0 plants of
HUAXU-ee and
Fix8 was confirmed by PCR using primers listed in Supplementary Table S10.
Ploidy analysis
Ploidy levels were determined by flow cytometry as previously described
17. Briefly, ~2 cm
2 of fresh leaf tissue was finely chopped in 1 mL of ice-cold LB01 lysis buffer (15 mM Tris, 2 mM disodium EDTA, 0.5 mM spermine tetrahydrochloride, 80 mM KCl, 20 mM NaCl, 0.1% (v/v) Triton X-100, 15 mM β-mercaptoethanol, pH 7.5, pre-filtered through a 0.22-μm membrane). The homogenate was filtered through a 40-µm nylon mesh and centrifuged at 135×
g for 5 min at 4 °C. Nuclei were resuspended in 450 µL of fresh LB01 buffer, stained with 25 µL of 1 mg/mL propidium iodide (PI; Sigma-Aldrich, P4170) and 25 µL of 1 mg/mL DNase-free RNase A (Sigma-Aldrich, V900498), and incubated for ≥ 15 min at room temperature in the dark. Samples were analyzed on a BD Accuri C6 flow cytometer (552-nm excitation, 610/20-nm emission). Gating and peak assignment strategies are presented in Supplementary Figs. S1o, S2g. Nuclei with a peak matching the WT diploid leaf reference (~6 × 10
5 relative fluorescence units, RFU) were classified as 2n (diploid). Peaks at approximately half (~3 × 10
5 RFU) and double (~12 × 10
5 RFU) this value were classified as n (haploid) and 4n (tetraploid), respectively.
For large-scale ploidy screening in the field, putative diploids and tetraploids were initially identified based on diagnostic morphological traits
17. Seeds were germinated, and seedlings were transplanted into the field using single-seedling transplantation; plants were scored at the grain-filling stage. Individuals displaying elongated awns, markedly enlarged grains, and strongly reduced fertility were classified as tetraploid candidates, whereas those phenotypically indistinguishable from diploid WT controls were classified as diploids. Representative plants from each class were subsequently confirmed by flow cytometry.
Endosperm ploidy was determined by flow cytometry similarly. Mature seeds were dissected to remove the embryo; the remaining tissue (starchy endosperm, aleurone layer, and seed coat)
63,64 was finely chopped in LB01 buffer and processed as described for leaf tissue. Gating and peak assignment strategies are presented in Supplementary Fig. S3b. Nuclei with a peak matching the WT leaf diploid reference (~6 × 10
5 RFU) were classified as 2n (diploid). Peaks matching the typical WT endosperm reference (~9 × 10
5 RFU, corresponding to the triploid product of central cell fertilization) were classified as 3n (triploid). Peaks at approximately double the triploid value (~18 × 10
5 RFU) were classified as 6n (hexaploid).
Eosin B staining and imaging
To visualize early embryo and endosperm development, whole-mount Eosin B staining with methyl salicylate clearing was performed with modifications to a previously described protocol
65. Pistils or young caryopses were fixed in modified FAA solution (water:ethanol:acetic acid:formaldehyde:glycerol, 48:45:6:2:5, v/v) at 4 °C for 24 h or longer. Fixed samples were manually dissected under a stereomicroscope and rehydrated sequentially through 70% ethanol (twice), 50% ethanol, 30% ethanol, and distilled water (1 h each). Samples were pretreated with 2% (w/v) aluminum potassium sulfate for 30 min, stained with 10 mg/L Eosin B in 4% (w/v) sucrose for 16 h at room temperature, post-treated with 2% (w/v) aluminum potassium sulfate for 30 min, and rinsed three times in distilled water. Samples were dehydrated in a graded ethanol series (30%, 50%, 70%, 80%, 90%, 100% (v/v); 1 h each). Samples were then cleared stepwise in absolute ethanol:methyl salicylate mixtures (2:1, 1:1, 1:2 (v/v); 1 h each) and incubated overnight in pure methyl salicylate. Cleared specimens were imaged using a Zeiss LSM 980 confocal laser scanning microscope (590 nm excitation; 574–720 nm emission detection).
Alexander staining
Mature anthers were collected from unopened flowers and fixed in Carnoy’s solution (60% (v/v) ethanol, 30% (v/v) chloroform, 10% (v/v) acetic acid) overnight or longer at 4°C. Fixed anthers were dissected, stained in Alexander’s solution (SL7660, Coolaber) for 30 min at 25 °C, rinsed with ddH2O, and mounted for imaging. Samples were examined and photographed under a bright-field microscope.
2,3,5-triph-enyltetrazolium chloride (TTC) staining
Mature dehulled grains were imbibed in ddH2O for 2 h at room temperature. Caryopses were bisected longitudinally with a razor blade to expose the embryos, transferred to 10-mL tubes, and immersed in 5 mL of 0.2% (w/v) TTC solution (prepared by diluting a 1% stock (Cat. No. RS4131, G-GLONE) 1:5 in PBS (pH 7.4)). Tubes were wrapped in foil and incubated at 28°C for 2 h in the dark. The staining solution was discarded, and samples were rinsed five times with ddH2O. Stained caryopses were examined and imaged under a stereomicroscope.
Field trial design and statistical analysis
Field experiments were conducted in Hangzhou, China (120.1 °E, 30.3 °N) during the summer growing season and in Sanya, China (109.5 °E, 18.2 °N) during the winter growing season. Rice seedlings were transplanted with one plant per hill at a spacing of 20 cm between rows and 20 cm between plants within rows, corresponding to a planting density of approximately 25 plants/m2. For routine phenotypic evaluation of small populations, each line was planted in a plot consisting of six rows with eight plants per row (48 plants in total). For large-scale field evaluation of the Fix8 T2 population, each replicate consisted of 32 rows with eight plants per row, and five biological replicates were established for each line. For the Fix8_P2 T2 population, each replicate consisted of 16 rows with eight plants per row, with five biological replicates for each line.
All field management followed local standard practices for hybrid rice production. Basal fertilizer (N:P:K = 15:15:15) was applied at 500 kg/ha before transplanting. Nitrogen was supplied as urea (46% N) at a total rate of 207 kg N/ha, top-dressed in three equal splits at the tillering, panicle initiation, and booting stages. Potassium fertilizer (150 kg/ha) was applied simultaneously with the nitrogen top-dress at the booting stage. Irrigation was managed by maintaining a 3–5 cm standing water layer with alternating wetting-drying cycles until the end of grain filling, after which the field was drained in preparation for harvest. Pests, diseases, and weeds were controlled according to local recommendations to minimize yield loss.
For T0 plants, seed-setting rate was determined from independently harvested panicles and calculated as the ratio of filled grains to total grains. For field-grown populations, agronomic traits were evaluated using representative plants randomly selected from the central area of each plot, excluding border rows and border plants. Seed-setting rate was calculated as the proportion of filled grains among total grains from three primary panicles per plant. Plant height was measured as the distance from the soil surface to the tip of the primary panicle. Tiller number was recorded as the number of productive tillers bearing mature primary panicles. Panicle length was measured as the distance from the panicle base to the panicle tip, and 1,000-grain weight was determined using fully matured grains. All measurements were obtained from randomly selected plants within each line. All data were analyzed using a two-tailed Student’s t-test, with significance thresholds set at *P < 0.05 and **P < 0.01.
Definitions of reproductive indices
Parthenogenesis induction efficiency is defined as the proportion of pistils (or ovules) that developed autonomous embryo-like structures accompanied by unfertilized central cells. Haploid induction frequency is defined as the proportion of individuals (or progeny) that carried haploid genomes, quantified by flow cytometry. Clonal frequency is defined as the proportion of offspring that are genetically identical to the maternal parent and maintain the identical diploid ploidy level.
Apparent amylose content determination
One biological replicate consisted of approximately 25 g of mature rice grains pooled from two randomly selected plants. Mature rice grains were dehulled and polished to remove the pericarp, embryo, and aleurone layers, yielding polished rice that was ground to flour using a laboratory mill. After sieving through a mesh screen, flour aliquots (100 mg) were analyzed for apparent amylose content using a fully automated flow-injection amylose analyzer (BAODE INSTRUMENTS, BDFIA-7000) according to the manufacturer’s iodine-based colorimetric protocol. Each biological replicate was analyzed in technical duplicate. The results are expressed as the apparent amylose content (%) on a dry-weight basis from three independent biological replicates.
Total protein determination
Milled rice flour was prepared as described above and analyzed for total nitrogen using an automated Kjeldahl system (FOSS, Kjeltec KT8400) according to the manufacturer’s digestion, distillation, and titration protocol. Protein content was calculated as nitrogen content (%) × 5.95 (the standard conversion factor for rice). Each biological replicate was analyzed in technical triplicate. The results are reported as total protein content (%) on a dry-weight basis from three independent biological replicates.
Subcellular localization in protoplasts
Rice protoplasts were isolated from etiolated seedling leaf sheaths using enzymatic digestion as described
66. Plasmids expressing HUAXU-GFP and free GFP (control) were introduced into protoplasts by PEG-mediated transfection. Transfected protoplasts were incubated in the dark at 28 °C for 16 h. Nuclei were counterstained with 4',6-diamidino-2-phenylindole (DAPI). GFP (488 nm excitation; 490–546 nm emission) and DAPI (353 nm excitation; 410–496 nm emission) fluorescence signals were imaged using a Zeiss LSM 980 confocal laser scanning microscope.
RNA sequencing
Rice protoplasts expressing HUAXU-GFP or GFP were obtained as described above. 16 h post-transfection (HAT), protoplasts were collected by centrifugation, and total RNA was extracted using TRIzol reagent. RNA integrity was assessed using an Agilent 5300 Fragment Analyzer (Agilent Technologies), and only samples with an RNA Quality Number (RQN) > 6.0 were used for downstream processing. High-quality total RNA samples (0.5 μg per sample) were shipped to Shanghai Majorbio Bio-pharm Technology Co., Ltd. (Shanghai, China) for strand-specific library preparation and sequencing. Briefly, poly(A)-enriched mRNA libraries were constructed using the Illumina TruSeq RNA Sample Prep Kit, followed by 150-bp paired-end sequencing on the Illumina NovaSeq platform, typically yielding > 30 million clean reads per sample.
Raw sequencing reads were processed to remove adapters and low-quality bases using fastp (v0.23.4)
67 with a quality score cutoff of Q20. Clean reads were aligned to the rice reference genome (
Oryza sativa Nipponbare MSU7) using HISAT2 (v2.2.1)
68. Gene-level read counts were quantified with HTSeq-count (v2.0.5)
69 using the union mode and reverse strand specificity. Transcript assembly and abundance estimation were performed using StringTie (v2.2.3)
70. Principal component analysis (PCA) was performed on variance-stabilizing transformed (VST) counts. Differential expression analysis was performed using DESeq2 (v1.34.0)
71 with Benjamini-Hochberg adjustment for multiple testing. Differentially expressed genes (DEGs) were called with |log
2fold change| ≥ 2 and adjusted
P value < 0.05. All experiments included four independent biological replicates per construct.
Dual-luciferase reporter assay
The effector construct was generated by cloning the HUAXU coding sequence into the pGreenII 62-SK backbone. The reporter construct was generated by inserting the OsBBM1 promoter region into the pGreenII 0800-LUC vector, which contains a 35S:Renilla luciferase (REN) cassette as the internal normalization control. Agrobacterium tumefaciens GV3101 strains carrying the respective plasmids were mixed as indicated and infiltrated into the abaxial surface of young leaves from 5-week-old Nicotiana benthamiana plants grown under long-day conditions (16 h light/8 h dark, 22–24 °C). At 48 h post-infiltration, leaves were sprayed with 150 μg/mL D-luciferin and imaged using a Tanon 5200 bioluminescence system. For quantification, leaf discs were excised, lysed in passive lysis buffer, and LUC and REN activities were measured sequentially with the Dual-Luciferase Reporter Assay System (Promega) on a Tecan Infinite 200 PRO plate reader according to the manufacturer’s instructions. Promoter activity was expressed as the LUC/REN ratio. Experiments were performed in three independent biological replicates, with at least three leaves analyzed per effector-reporter combination. Data were analyzed using a two-tailed Student’s t-test (* P < 0.05; ** P < 0.01).
Yeast two-hybrid assay
Yeast two-hybrid assays were performed using the Matchmaker Gold system (Clontech) according to the manufacturer’s instructions. The HUAXU coding sequence was cloned in-frame with the GAL4 DNA-binding domain (BD) in pGBKT7 to generate the bait construct (HUAXU-BD). The OsBBM1 coding sequence was fused to the GAL4 activation domain (AD) in pGADT7 to generate the prey construct (OsBBM1-AD). Empty pGBKT7 and pGADT7 vectors served as negative controls, and the well-characterized interacting pair D53-AD/IPA1-BD served as a positive control. Bait and prey constructs were co-transformed into Saccharomyces cerevisiae strain Y2HGold using the lithium acetate method. Transformants were selected on synthetic dropout medium lacking Trp and Leu (SD/–Trp/–Leu). Positive interactions were scored by growth on quadruple dropout medium (SD/–Trp/–Leu/–His/–Ade) after 2 d at 30 °C. The assays were performed in three independent biological replicates, with representative plates shown.
Genotyping with InDel markers
Insertion-deletion (InDel) markers, previously described
17, were used to distinguish between heterozygous and homozygous genotypes. Genomic DNA was extracted from rice leaf tissue using the cetyltrimethylammonium bromide (CTAB) method. PCR was performed using InDel-specific primers (Supplementary Table S2), and amplicons were resolved by electrophoresis on 5% agarose gels.
Whole-genome resequencing and genotype calling
Paired-end 150-bp reads were generated using the Illumina HiSeq 2500 platform, achieving an average sequencing depth of ~20× per sample. Raw reads were filtered to obtain high-quality data using Fastp v0.23.2
67 with default parameters. Clean reads were aligned to the rice reference genome (
Oryza sativa Nipponbare MSU7) using BWA v0.7.17
72 with parameters ‘-M’ and ‘-t 4’ for multi-threaded processing. To ensure high-quality single nucleotide polymorphism (SNP) detection, ambiguously mapped reads were excluded using SAMtools v1.5
73. Only uniquely mapped single-end and paired-end reads were used for SNP calling. Genotype calling was performed across the genome using SNPs that were heterozygous in the parental line. A 1-Mb sliding window approach was applied to construct recombination maps for each chromosome.
Accession numbers
Sequence information of the genes used in this article can be found in HUAXU (LOC_Os12g12970), EGG1 (LOC_Os10g29380), EGG2 (LOC_Os12g07610), OSD1 (LOC_Os02g37850), PAIR1 (LOC_Os03g01590), REC8 (LOC_Os05g50410), and OsECA (LOC_Os03g18530).
DATA AVAILABILITY
The whole genome sequencing and RNA-seq data generated in this study are publicly accessible through the NCBI Sequence Read Archive under the accession code PRJNA1305152. Additional data supporting the findings of this study can be obtained from the corresponding author upon reasonable request.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
