Haplotype-resolved 3D genome maps reveal RNAPII-mediated allelic regulation in hybrid rice Authors: Shiping Luan, Runxin Gao, Guoting Chen, Chuanhao Liu, Denggao Li, Zhifei Xue, Ying Zhang, Yue Li, Cheng Ding, Beibei Liu, Yanyan Liu, Jiapei Yan, Mohamed F. Foda, Xingwang Li, Weizhi Ouyang Published: 2026 ABSTRACT To understand how the two parental genomes coordinate transcription in hybrids, chromatin architecture must be resolved at the haplotype level. Here, using phased Bridge-Linker Hi-C, we reconstructed a haplotype-resolved three-dimensional (3D) genome of the elite hybrid rice (Oryza sativa) line Shanyou 63 (SY63). We identified extensive allele-specific chromatin conformations. Furthermore, we generated allele-resolved RNAPII ChIA-PET maps and phased transcriptomes to explore how chromatin interactions contribute to allelic regulation. Although maternal and paternal homologs share broadly similar chromatin features, we detected widespread haplotype-biased RNAPII binding and chromatin looping at high resolution. These allele-specific RNAPII-mediated contacts were significantly associated with biased expression. Stronger RNAPII binding on one haplotype promoted the formation of long-range regulatory loops with distal genes, thereby contributing to allele-biased transcription at a subset of loci, even when promoter-proximal RNAPII occupancy was comparable between alleles. These results demonstrate that subtle differences in RNAPII engagement and 3D regulatory wiring between parental haplotypes can reshape transcriptional output in hybrids, providing new insights into the mechanisms underlying the allelic regulation of gene expression. Introduction The 3D organization of the genome plays a critical role in regulating gene expression and development [[1], [2], [3]]. Chromatin folding dictates the spatial proximity of regulatory DNA elements and genes, thereby shaping transcriptional programs at both the local and global scales [[4], [5], [6], [7]]. With the advent of high-throughput chromosome conformation capture technologies, including Hi-C, ChIA-PET, and related approaches, our understanding of plant nuclear architecture has greatly increased [[8], [9], [10]]. Studies in Arabidopsis thaliana, rice (Oryza sativa), and maize (Zea mays) have revealed conserved hierarchical features of plant 3D genomes, such as chromosome territories, distinct configurations (Rabl-like, bouquet-like, and rosette-like), dynamic chromatin compartmentalization, topologically associated domain (TAD)-like structures, and chromatin-interacting domains [[11], [12], [13], [14], [15]]. However, the extent to which chromatin architecture differs between maternal and paternal alleles in hybrid plants remains largely unexplored. Allele-specific analysis of transcriptomes, epigenomes, and 3D genomes is a powerful strategy for dissecting the contributions of parental genomes to hybrid genome function [[16], [17], [18], [19], [20]]. Extensive allele-specific expression (ASE) has been documented in numerous crops, including rice, maize, Brassica napus, and barley (Hordeum vulgare), underscoring its importance for heterosis, domestication, and environmental adaptation [[21], [22], [23], [24], [25]]. Although genetic variation undoubtedly contributes to ASE, accumulating evidence from animals and plants suggests that allelic differences in 3D genome architecture constitute an additional regulatory layer that influences transcriptional divergence [[26], [27], [28], [29], [30]]. Single-cell and haplotype-resolved Hi-C analyses in mammals have revealed that homologous chromosomes do not fold in an identical manner. Instead, these chromosomes display distinct compartmental states, domain structures, and long-range interactions that correlate with allele-biased transcriptional output [31]. RNAPII- and CTCF-anchored chromatin interaction maps further highlight allele-specific regulatory hubs that couple promoter-enhancer communication with transcriptional activity [29,32]. Similarly, multi-omic analyses of hybrid mice have revealed extensive haplotype-biased chromatin topology linked to regulatory divergence between the parents [33]. Although few allele-resolved 3D genomic studies have been performed in plants, studies of polyploid and duplicated genomes have demonstrated that domain organization, compartment dynamics, epigenomic states, and transcriptional regulation differ between homologous chromosomes or chromosomes in different subgenomes, indicating that structural asymmetry is a conserved feature of plant nuclear architecture [24,[34], [35], [36]]. These findings support a model in which allele- or subgenome-specific chromatin conformations reinforce the functional divergence between homologous chromosomes. We previously mapped haplotype-resolved H3K27me3-associated chromatin interactions in rice and identified widespread allele-specific chromatin loops [37]. However, because H3K27me3 is a repressive histone modification, these interactions account for only a subset of allele-specific gene silencing, leaving unanswered how active chromatin architecture shapes broader allelic transcriptional differences. Hybrid rice provides an ideal system for investigating the interplay between allelic chromatin organization and transcriptional regulation [37]. Hybrid rice, one of the most successful examples of heterosis utilization in agriculture, exhibits extensive phenotypic and transcriptional divergence from its parental lines [21]. However, whether the 3D genome organization also differs between maternal and paternal alleles, and how such differences might influence allelic gene expression have not been systematically addressed. In this study, we established a comprehensive haplotype-resolved 3D genome framework for diploid rice using high-resolution Bridge-Linker Hi-C and RNAPII ChIA-PET datasets. By phasing chromatin interactions using single-nucleotide polymorphisms (SNP) markers, we reconstructed a diploid 3D genome at 20-kb resolution, mapped allele-specific chromatin architecture genome-wide, and integrated these maps with allele-specific RNAPII binding and RNA-seq data. This integrative analysis allowed us to (i) resolve the spatial organization of maternal and paternal chromosomes, (ii) uncover loci exhibiting haplotype-biased chromatin contacts, and (iii) determine how allele-specific loops contribute to differential transcriptional outputs. Our findings provide fundamental insights into the allele-dependent structural organization of hybrid plant genomes and the three-dimensional (3D) regulatory logic underlying allelic transcriptional imbalances. Results Three-dimensional genome reconstruction and spatial organization of diploid rice chromosomes To reconstruct the haplotype-resolved 3D genome architecture of rice, we generated 3D chromatin interaction data from seedling tissue of the hybrid rice line SY63 using previously reported Bridge-Linker Hi-C technology (Table S1) [38]. We phased the Hi-C paired-end reads into maternal and paternal origins based on informative SNPs using our previously reported pipeline [29,32,37]. The paired-end tags (PETs) were classified into three categories based on their parental identity at both ends: MM, MP, and PP (M, maternal; P, paternal) (Table S2). Analysis of allelic phasing efficiency revealed that maternal- and paternal-derived PETs were evenly represented, with no significant bias in interaction frequency (Fig. S1A), thereby ensuring balanced parental contributions for the downstream reconstruction of diploid and allele-resolved chromatin organization. This balanced distribution of phased BL-Hi-C PETs indicates that our SNP-based haplotype assignment did not introduce any obvious genome-wide parental bias. To further characterize the large-scale chromosomal interaction patterns, we divided each chromosome into ten equal bins and quantified the interchromosomal PET densities. The interaction frequency was lower near the centromeric regions and gradually increased toward the telomeric regions (Fig. S1B). Consistently, centromere–centromere (CC) contacts exhibited the weakest interactions. By contrast, telomere–telomere (TT) interactions were significantly stronger than both CC and centromere–telomere (CT) contacts, and CT interactions were intermediate but still stronger than CC contacts (Fig. S1C). These features are reminiscent of the organization observed in mouse nuclei, where interactions become progressively stronger toward the chromosome ends. Principal component analysis (PCA) of the phased interchromosomal contact frequency matrices showed that homologous chromosomes clustered closely together. Using these matrices, we inferred the spatial distribution of individual chromosomes within the nucleus (see Materials and Methods). This analysis revealed a significant correlation between the distances between homologous chromosomes and the chromatin centroid (Fig. S1E). The distances between homologous chromosome pairs were significantly smaller than those between randomly selected non-homologous chromosome pairs, representing the nearest-neighbor level of interchromosomal distances (Fig. S1F). These findings indicate that homologous chromosome pairs exhibit highly similar interaction patterns and tend to be positioned in close spatial proximity within the nucleus. Using these phased interaction data, we modeled the three-dimensional chromatin architecture of the diploid rice nucleus, resolving the spatial organization of all 24 chromosomes at 20 kb resolution (Fig. 1A and Video 1). The reconstructed nuclear architecture revealed distinct chromosomal territories, with Chr01 positioned relatively closer to the nuclear center, whereas the transcriptionally less active chromosomes Chr11 and Chr12 were located toward the nuclear periphery (Fig. 1C–D). Virtual nuclear slicing followed by CpG frequency profiling showed that CpG sites were broadly distributed throughout the nucleus (Fig. 1B), in contrast to the punctate CpG island pattern typical of animal cells, indicating that rice nuclei possess a more uniform epigenomic landscape than animal cells.Fig. 1Three-dimensional genome reconstruction and spatial organization of diploid rice chromosomes. A The 3D genome of diploid rice reconstructed at haplotype resolution, revealing the structural organization of all 24 chromosomes (12 maternal and 12 paternal). Each chromosome is color-coded to illustrate its spatial position in the nucleus. B Virtual nuclear sections generated from the reconstructed nuclear architecture; the CpG frequency for each section is shown. CpG sites (green) are dispersed throughout the genome, displaying a broad distribution pattern distinct from the punctate CpG islands typically observed in animal cells. C Distance of each chromosome from the nuclear center. Each point represents the average nuclear position of a genomic segment (M, maternal; P, paternal). Chromosome 1 (Chr01) is located relatively closer to the nuclear center, whereas the transcriptionally less active chromosomes 11 and 12 are positioned more peripherally. D Representative nuclear projections highlighting the spatial positions of Chr01 (central) and Chr11-Chr12 (peripheral) within the reconstructed nuclear structure. Three-dimensional genome reconstruction and spatial organization of diploid rice chromosomes. A The 3D genome of diploid rice reconstructed at haplotype resolution, revealing the structural organization of all 24 chromosomes (12 maternal and 12 paternal). Each chromosome is color-coded to illustrate its spatial position in the nucleus. B Virtual nuclear sections generated from the reconstructed nuclear architecture; the CpG frequency for each section is shown. CpG sites (green) are dispersed throughout the genome, displaying a broad distribution pattern distinct from the punctate CpG islands typically observed in animal cells. C Distance of each chromosome from the nuclear center. Each point represents the average nuclear position of a genomic segment (M, maternal; P, paternal). Chromosome 1 (Chr01) is located relatively closer to the nuclear center, whereas the transcriptionally less active chromosomes 11 and 12 are positioned more peripherally. D Representative nuclear projections highlighting the spatial positions of Chr01 (central) and Chr11-Chr12 (peripheral) within the reconstructed nuclear structure. Therefore, our haplotype-resolved 3D reconstruction of the diploid rice genome reveals balanced parental chromatin interactions, dispersed CpG distribution, and a tightly coordinated spatial organization of homologous chromosomes, reflecting a highly symmetric, plant-specific nuclear architecture. Haplotype-resolved 3D genome maps reveal allele-specific chromatin organization To compare the 3D chromatin organization between maternal and paternal alleles, we generated haplotype-resolved chromatin contact matrices using phased Bridge-Linker Hi-C reads. The genome-wide allele-specific contact map revealed largely similar global chromatin architectures between the maternal (red) and paternal (blue) genomes (Fig. 2A). At higher resolution, allele-specific differences became evident. The 50-kb resolution contact maps of chromosome 7 (Fig. 2B) and chromosome 4 (Fig. S2A) showed broadly conserved domain structures. A finer resolution (5 kb) of the distal region (boxed region in Fig. 2B) revealed extensive allele-biased chromatin interactions (Fig. 2C and Fig. S2B). Analysis at 2.5-kb resolution (boxed region in Fig. 2C) revealed additional maternal allele-specific contact loops (Fig. 2D) and paternal allele-specific contact loops (Fig. S2C).Fig. 2Haplotype-resolved 3D genome maps revealing allele-specific chromatin organization and transcriptional bias in hybrid rice. A Genome-wide allele-specific chromatin contact matrix constructed using phased Bridge-Linker Hi-C reads. Maternal (M) and paternal (P) contacts are shown in red and blue, respectively. B Allele-specific contact map of chromosome 7 at 50-kb resolution, showing the overall conserved domain structures between maternal and paternal alleles. C Zoom-in view of the distal region (black dashed box in panel B) at 5-kb resolution, showing localized allele-biased chromatin interactions. D Zoom-in view (of the black dashed box in panel C) at 2.5-kb resolution, highlighting fine-scale allele-specific contact loops. E Integration of RNAPII ChIP-seq peaks, chromatin loops, and allele-specific RNA-seq tracks revealing a large region with maternal-specific RNAPII binding and loop formation, accompanied by maternally biased gene expression. The lower track shows the annotated genes in this region. Red arcs represent interactions between maternally inherited chromatin regions within the indicated interval, whereas blue arcs represent interactions between paternally inherited chromatin regions. Haplotype-resolved 3D genome maps revealing allele-specific chromatin organization and transcriptional bias in hybrid rice. A Genome-wide allele-specific chromatin contact matrix constructed using phased Bridge-Linker Hi-C reads. Maternal (M) and paternal (P) contacts are shown in red and blue, respectively. B Allele-specific contact map of chromosome 7 at 50-kb resolution, showing the overall conserved domain structures between maternal and paternal alleles. C Zoom-in view of the distal region (black dashed box in panel B) at 5-kb resolution, showing localized allele-biased chromatin interactions. D Zoom-in view (of the black dashed box in panel C) at 2.5-kb resolution, highlighting fine-scale allele-specific contact loops. E Integration of RNAPII ChIP-seq peaks, chromatin loops, and allele-specific RNA-seq tracks revealing a large region with maternal-specific RNAPII binding and loop formation, accompanied by maternally biased gene expression. The lower track shows the annotated genes in this region. Red arcs represent interactions between maternally inherited chromatin regions within the indicated interval, whereas blue arcs represent interactions between paternally inherited chromatin regions. Aggregate peak analysis (APA) revealed pronounced allele-specific differences in the strength of chromatin looping (Fig. S3). In loops classified as paternal-specific, focal interactions were strongly enriched exclusively in paternal-derived PETs, with little or no enrichment detected in maternal contact maps. Conversely, maternal-specific loops exhibited clear APA signals only in maternal PETs. By contrast, loops shared between the two parental genomes displayed comparable interaction intensities in both maternal and paternal PETs, indicating conserved loop formation across the alleles. Importantly, parent-specific loops showed markedly attenuated APA signals when assessed in the opposite haplotype, excluding the existence of confounding technical factors such as sequencing depth or genomic distance and confirming their genuine allele specificity. Together, these results demonstrate that although the global 3D chromatin architecture is largely conserved between parental alleles, a substantial subset of chromatin loops exhibits strong allele specificity, giving rise to fine-scale divergence in local chromatin organization within the hybrid genome. To investigate the potential regulatory significance of these structural differences, we performed integrated RNA polymerase II (RNAPII) ChIP-seq and allele-specific RNA-seq data analyses. Consistent with the chromatin interaction patterns, we identified extensive regions exhibiting maternal-specific RNAPII binding and loop formation (Fig. 2E). These regions coincided with clusters of genes exhibiting maternally biased expression, suggesting that allele-specific chromatin architecture contributes to differential transcriptional regulation in the rice hybrid genome. Together, these results demonstrate that our haplotype-resolved 3D genome maps reveal widespread allele-specific differences in chromatin conformation and transcriptional activity, highlighting the structural basis for allelic imbalances in hybrid rice. Allele-resolved RNAPII ChIA-PET reveals widespread haplotype-biased chromatin interactions in the hybrid genome To comprehensively investigate allele-specific chromatin architecture and its relationship with transcriptional regulation in the hybrid genome, we generated high-depth RNAPII ChIA-PET libraries and assigned each PET to the maternal (M) or paternal (P) haplotype using phased SNPs (Fig. S4A, Table S1). PETs whose two ends mapped to the same haplotype (M-M or P-P) were classified as cis-PETs, whereas those whose two ends mapped to different haplotypes (M − P or P-M) were categorized as trans-PETs. As shown in Fig. 3A, a great majority of PETs were cis-PETs (98.6%), indicating that RNAPII-mediated contacts predominantly occur within the same parental haplotype rather than between maternal and paternal chromosomes. Among the intra-chromosomal cis-PETs, M-M and P-P PETs showed nearly equal abundance (48.5% vs. 51.5%), supporting the previous observation that hybrid genomes generally maintain balanced RNAPII binding and contact formation across parental haplotypes in the absence of strong imprinting or structural asymmetry.Fig. 3Allele-resolved RNAPII ChIA-PET profiling revealing haplotype-biased chromatin interactions in the hybrid genome. A Classification of RNAPII ChIA-PET paired-end tags (PETs). Left: Proportions of intra-chromosomal cis-PETs and trans-PET. Right: Distribution of cis-PETs separated into maternal-maternal (M-M) and paternal-paternal (P-P) PETs. The majority of PETs were cis-PETs (98.6%), with comparable proportions of M-M and P-P PETs (48.5% and 51.5%, respectively). B Summary of haplotype-biased RNAPII-binding peaks (top) and haplotype-biased chromatin interactions (bottom). A total of 3406 phased RNAPII peaks were identified, 430 of which showed significant allele bias (q < 0.01). In total, 195 haplotype-biased interactions were detected, which were associated with 521 genes. C Genome browser view (IGV) showing representative maternal-biased RNAPII interactions. Green arcs represent all interactions, red arcs represent maternal-biased loops; the corresponding RNAPII peaks and SNP positions are shown below the arcs. D Allele-separated interaction heatmaps (at 5 kb resolution) corresponding to panel C, showing stronger loop signals in the maternal haplotype. E Genome browser view showing representative paternal-biased RNAPII interactions. Green arcs represent all chromatin interactions, whereas red and blue arcs indicate maternal-biased and paternal-biased loops, respectively. The corresponding RNAPII peaks and SNP tracks are also shown. F Allele-separated interaction heatmaps corresponding to panel E, showing preferential loop formation in the paternal haplotype. Allele-resolved RNAPII ChIA-PET profiling revealing haplotype-biased chromatin interactions in the hybrid genome. A Classification of RNAPII ChIA-PET paired-end tags (PETs). Left: Proportions of intra-chromosomal cis-PETs and trans-PET. Right: Distribution of cis-PETs separated into maternal-maternal (M-M) and paternal-paternal (P-P) PETs. The majority of PETs were cis-PETs (98.6%), with comparable proportions of M-M and P-P PETs (48.5% and 51.5%, respectively). B Summary of haplotype-biased RNAPII-binding peaks (top) and haplotype-biased chromatin interactions (bottom). A total of 3406 phased RNAPII peaks were identified, 430 of which showed significant allele bias (q < 0.01). In total, 195 haplotype-biased interactions were detected, which were associated with 521 genes. C Genome browser view (IGV) showing representative maternal-biased RNAPII interactions. Green arcs represent all interactions, red arcs represent maternal-biased loops; the corresponding RNAPII peaks and SNP positions are shown below the arcs. D Allele-separated interaction heatmaps (at 5 kb resolution) corresponding to panel C, showing stronger loop signals in the maternal haplotype. E Genome browser view showing representative paternal-biased RNAPII interactions. Green arcs represent all chromatin interactions, whereas red and blue arcs indicate maternal-biased and paternal-biased loops, respectively. The corresponding RNAPII peaks and SNP tracks are also shown. F Allele-separated interaction heatmaps corresponding to panel E, showing preferential loop formation in the paternal haplotype. To assess whether allele-specific chromatin interactions might be influenced by sequence divergence between the two parental haplotypes, we performed an alternative reference validation using the paternal genome (ZS97). Specifically, we converted the annotated MH63 SNP coordinates to the corresponding ZS97 genomic coordinates using the UCSC liftOver tool (https://genome.ucsc.edu/cgi-bin/hgLiftOver), re-phased PETs mapped to the ZS97 reference genome, and compared the phasing assignments of PETs recovered in both reference-based analyses according to read IDs. We detected a high degree of consistency between the two reference frameworks. Among the 166,358,483 BL-Hi-C PETs recovered in both analyses, 161,756,712 (97.23%) showed concordant phasing assignments. Likewise, among the 21,824,210 RNAPII ChIA-PET PETs identified by both analyses, 21,390,056 (98.01%) showed consistent assignments (Fig. S4B). These results indicate that the allele-specific chromatin interaction patterns identified in this study are highly robust to reference choice and are unlikely to be primarily driven by parental sequence divergence. Next, we identified allele-biased RNAPII-binding peaks and chromatin interactions to uncover potential regulatory asymmetries embedded within the hybrid genome (Fig. 3B). Among all RNAPII ChIA-PET peaks, 3406 could be confidently phased, 430 of which exhibited significant haplotype-specific enrichment (q < 0.01), including 211 maternal-biased and 219 paternal-biased RNAPII-binding peaks. Using these phased peaks as interaction anchors, we identified 195 haplotype-biased chromatin loops, including 97 maternal-biased and 98 paternal-biased loops. Each arc shown in the browser view represents one loop. These allele-biased contacts were collectively associated with 521 genes, suggesting that haplotype-specific 3D chromatin architecture has broad functional effects on transcriptional output. Together with the near-equal proportion of M-M and P-P cis-PETs, the highly symmetric numbers of maternal- and paternal-biased RNAPII peaks and chromatin loops argue against a systematic genotyping or phasing bias favoring either parental haplotype. Representative Integrated Genome Viewer snapshots and allele-separated interaction heatmaps illustrate clear cases of haplotype-biased looping (Fig. 3C–F). Regions harboring maternal-biased interactions exhibited stronger interaction arcs and enriched RNAPII signals on the maternal vs. paternal haplotype (Fig. 3C–D), whereas paternal-biased interactions showed the opposite pattern (Fig. 3E–F). These examples highlight how allele-specific differences in RNAPII-binding intensities translate into pronounced differences in 3D chromatin organization, resulting in asymmetric enhancer-promoter contacts or promoter-promoter interactions between haplotypes. Together, these results demonstrate that RNAPII ChIA-PET is an effective and highly resolved approach for dissecting allele-specific chromatin topologies in complex hybrid genomes. The widespread haplotype-biased loops and RNAPII-binding asymmetries uncovered in this study reveal substantial divergence in maternal and paternal chromatin conformations, offering insights into how parent-of-origin effects and structural variations shape transcriptional regulation in hybrids. Allele-specific RNAPII-mediated chromatin interactions are associated with allelic transcriptional bias To further explore how haplotype-biased chromatin architecture contributes to allele-specific gene expression, we integrated phased RNAPII ChIA-PET interaction data with allele-resolved transcriptomes (Fig. 4A). We examined the genes associated with haplotype-biased RNAPII-mediated loops and determined whether these genes exhibited significant ASE. In total, 195 haplotype-biased chromatin interactions were detected and were collectively associated with 521 genes. Among these 521 loop-associated genes, 301 had phased transcript information and could therefore be evaluated for ASE. Of these 301 genes, 64 (21.3%) were identified as significant ASE genes, whereas the remaining 237 showed no significant allelic bias (Fig. 4B). These results indicate that although haplotype-biased chromatin interactions do not universally lead to ASE, a substantial subset of interaction-associated genes exhibits biased transcription.Fig. 4Allele-specific chromatin interactions are associated with allele-specific gene expression. A Schematic representation of the relationship between allele-specific RNAPII binding, chromatin looping, and allele-specific gene expression. A strong RNAPII-mediated interaction with the maternal haplotype promotes maternal-biased transcription, whereas the paternal haplotype lacks the corresponding distal interaction. B Proportion of genes with phased transcripts exhibiting allele-specific expression (ASE). Among the 301 genes associated with allele-specific RNAPII-mediated interactions, 64 (21.3%) showed significant ASE, whereas 237 (78.7%) showed no allelic bias. C MA plot of allele-specific gene expression. The x-axis indicates the total expression abundance (log2[M+P]), and the y-axis represents the allelic fold change (log2[M/P]). Red: maternal-biased genes (M); blue: paternal-biased genes (P); black: genes whose ASE direction contradicts the corresponding haplotype-biased interaction (C); gray: genes without allelic bias (N). D Example of a maternal-biased chromatin interaction. IGV tracks show that the maternal haplotype exhibits a strong distal RNAPII binding peak, forming enhancer-promoter loops (red arcs), whereas the paternal haplotype has weak distal RNAPII binding and no detectable long-range interaction (no blue arcs). Although both alleles display promoter RNAPII occupancy, only the maternal allele forms functional loops, resulting in maternal-biased expression of the downstream gene. Allele-specific chromatin interactions are associated with allele-specific gene expression. A Schematic representation of the relationship between allele-specific RNAPII binding, chromatin looping, and allele-specific gene expression. A strong RNAPII-mediated interaction with the maternal haplotype promotes maternal-biased transcription, whereas the paternal haplotype lacks the corresponding distal interaction. B Proportion of genes with phased transcripts exhibiting allele-specific expression (ASE). Among the 301 genes associated with allele-specific RNAPII-mediated interactions, 64 (21.3%) showed significant ASE, whereas 237 (78.7%) showed no allelic bias. C MA plot of allele-specific gene expression. The x-axis indicates the total expression abundance (log2[M+P]), and the y-axis represents the allelic fold change (log2[M/P]). Red: maternal-biased genes (M); blue: paternal-biased genes (P); black: genes whose ASE direction contradicts the corresponding haplotype-biased interaction (C); gray: genes without allelic bias (N). D Example of a maternal-biased chromatin interaction. IGV tracks show that the maternal haplotype exhibits a strong distal RNAPII binding peak, forming enhancer-promoter loops (red arcs), whereas the paternal haplotype has weak distal RNAPII binding and no detectable long-range interaction (no blue arcs). Although both alleles display promoter RNAPII occupancy, only the maternal allele forms functional loops, resulting in maternal-biased expression of the downstream gene. To quantify the direction and magnitude of ASE, we generated an MA plot illustrating the differential gene expression between two haplotypes (Fig. 4C). Genes with significantly higher maternal expression were classified as maternal-biased (M, red), and those with higher paternal expression were classified as paternal-biased (P, blue). Genes whose ASE direction contradicted the corresponding haplotype-biased RNAPII interaction were marked as contradictory (C, black), whereas genes without significant allelic differences were labeled as unbiased (N, gray). Overall, the classifications of the majority of allele-biased genes were consistent with the parental origin of the stronger RNAPII-mediated interaction, supporting the notion that allele-specific chromatin looping has a direct regulatory impact on transcription. A representative example is shown in Fig. 4D. At this locus, the maternal haplotype exhibited strong RNAPII binding at a distal regulatory region, forming robust long-range chromatin loops with a downstream gene promoter. By contrast, the paternal haplotype showed markedly weaker RNAPII binding at the corresponding distal site, failing to establish comparable interactions. Notably, although both haplotypes showed RNAPII occupancy at the gene promoter, only the maternal haplotype established enhancer-promoter communication through looping, resulting in maternal-biased gene expression. This case demonstrates a mechanistic scenario in which RNAPII-associated allele-specific chromatin interactions, rather than promoter binding intensity alone, dictate the transcriptional output of the two haplotypes. Together, these results provide functional evidence that haplotype-biased chromatin looping contributes to differential gene regulation between parental alleles, highlighting the role of 3D genome architecture in shaping allelic transcriptional activity in hybrid rice. Discussion Our haplotype-resolved 3D genome maps of hybrid rice provide a comprehensive view of how maternal and paternal alleles are folded, positioned, and transcriptionally regulated within the same nucleus. Several key principles have emerged from this study. First, we demonstrated that hybrid rice nuclei exhibit parent-balanced large-scale chromatin organization. Maternal and paternal chromosomes displayed comparable interaction frequencies, similar global contact patterns, and nearly identical radial positions within the nucleus. Homologous chromosomes preferentially colocalize, which is consistent with the coordinated spatial constraints governing their nuclear arrangement. These properties highlight a fundamental difference between plant and animal nuclei: whereas mammalian genomes often display strong allele-specific compartmental divergence (e.g., imprinting domains or X-inactivation), rice appears to maintain a more balanced parental chromatin environment, likely reflecting the absence of global parental imprinting in plants. Second, despite genome-wide similarity, our allele-resolved chromatin maps reveal numerous localized differences in 3D organization. Haplotype-specific domains and loops became evident when we increased the structural resolution to 5 kb and below, indicating that allelic divergence manifests primarily at fine-scale regulatory elements rather than at large chromosomal domains. These allele-biased chromatin interactions were enriched for RNAPII binding, supporting the notion that local regulatory element activity is a major driver of allele-specific chromatin topology in the hybrid genome. Third, by integrating RNAPII ChIA-PET and allele-specific RNA-seq data, we directly linked haplotype-biased chromatin interactions with transcriptional output. Approximately 21% of the genes associated with allele-biased RNAPII loops exhibited significant ASE, and the majority showed directional consistency between loop strength and expression bias. Analysis of a representative example demonstrated how a maternal-specific distal loop enables enhancer-promoter communication and drives maternal-biased transcription, despite both alleles showing RNAPII promoter occupancy. This finding provides mechanistic insight into how allele-differential 3D interactions, together with RNAPII promoter binding, determine the transcriptional outcomes of different alleles. Fourth, the widespread haplotype-biased RNAPII-mediated loops uncovered in this study suggest that structural asymmetry may contribute to heterosis by enabling lineage-specific regulatory element activity. Although most allelic loops do not necessarily produce ASE under standard growth conditions, they may represent a latent regulatory layer that becomes relevant under environmental stress, developmental transitions, or hybrid-specific physiological contexts. Such “poised” allele-specific interactions could provide hybrids with expanded regulatory capabilities. In summary, we established an integrative framework for allele-specific 3D genome analysis in plants and demonstrated that parental genomes within a hybrid are not structurally identical at the fine regulatory scale. The structural and transcriptional asymmetries identified in this study highlight the importance of chromatin topology in shaping allelic gene regulation. These results provide a foundation for future investigations into how 3D genome divergence contributes to phenotypic variation, hybrid vigor, and adaptive evolution in crops. Materials and methods Plant materials Seeds of the hybrid rice line SY63 were soaked in water for 3 days and transferred to a constant-temperature incubator at 37 °C for germination. After 2 days, uniformly germinated seeds with healthy roots were transferred to a growth chamber for hydroponic cultivation. The plants were grown under a 12 h light/12 h dark photoperiod, with temperatures of 26 °C during the light period and 24 °C during the dark period and a relative humidity of 70%. The nutrient solution was replaced every 4 days. After 15 days of growth, aerial tissues were harvested at 08:00 a.m. for subsequent experiments. Experimental procedures eChIP-seq library preparation ChIP-seq was performed using a modified eChIP-seq protocol [37,39]. Approximately 0.5 g of formaldehyde-crosslinked seedling tissue was ground in liquid nitrogen and sequentially lysed with Buffer S and Buffer F. The chromatin was fragmented to 200–600 bp using a Bioruptor sonication system (Diagenode), followed by centrifugation at 20,000 × g for 10 min at 4 °C to remove debris. The fragmented chromatin was incubated with Protein G magnetic beads pre-bound to 5–10 μg of the appropriate antibody for 8–16 h at 4 °C with rotation. After immunoprecipitation, the bead-bound complexes were sequentially washed with low-salt ChIP buffer, high-salt ChIP buffer, ChIP wash buffer, and TE buffer. DNA-protein complexes were eluted using freshly prepared SDS-containing elution buffer, and the crosslinking was reversed by heat treatment. Following protein digestion, the ChIP DNA was purified and used for library construction using a NEBNext Ultra II DNA Library Preparation Kit. Size selection (approximately 250–650 bp) was performed using AMPure XP beads. The libraries were sequenced on an Illumina HiSeq X Ten platform to generate paired-end 150-bp reads. RNA-seq library preparation RNA-seq libraries were constructed as previously described [40]. Total RNA was extracted from seedlings using a RNeasy Plant Mini Kit (QIAGEN, 74904) according to the manufacturer's protocol. To deplete ribosomal RNA, 0.5–1.5 μg of purified total RNA was processed using a TruSeq Stranded Total RNA Library Prep Kit with Ribo-Zero Plant according to the manufacturer's instructions to enrich non-ribosomal RNA species. Strand-specific RNA-seq libraries were constructed and sequenced on an Illumina HiSeq X Ten platform, generating 150 bp paired-end reads. RNAPII ChIA-PET library preparation ChIA-PET libraries were prepared as previously described [7,41]. Crosslinked seedling tissue (∼5 g per assay) was flash-frozen in liquid nitrogen, ground into a fine powder, and sequentially lysed with Buffer S and Buffer F to release the crosslinked chromatin. The crosslinked chromatin was fragmented by sonication, clarified by centrifugation, and incubated with 60–80 μg of RNAPII antibody to enrich RNAPII-associated DNA complexes. After immunoprecipitation, the bead-bound chromatin was sequentially washed under stringent conditions with 0.1% SDS FA cell lysis buffer, high-salt ChIP buffer, ChIP wash buffer, and 1× TE buffer to reduce the nonspecific background. The recovered ChIP DNA was subjected to end repair using T4 DNA polymerase and Klenow enzyme, followed immediately by proximity ligation with biotin-labeled bridge linkers (forward strand: 5ʹ-[Phos]CGCGATATC/iBIOdT/TATCTGACT-3ʹ; reverse strand: 5ʹ-[Phos]GTCAGATAAGATATCGCGT-3ʹ). After the crosslinking was reversed, fragments containing spatial interaction information were purified and processed for library construction using Tn5 transposase-based tagmentation, and the resulting libraries were sequenced on an Illumina HiSeq X Ten platform in paired-end 150-bp mode. BL-Hi-C library preparation BL-Hi-C was performed as previously described with minor modifications [38]. Briefly, freshly chopped seedling tissue was crosslinked with 1% formaldehyde at room temperature to fix the chromatin conformations within nuclei, and after 15 min, crosslinking was quenched by adding 13.2 ml of 2.5 M glycine to a final concentration of 0.2 M. To enhance protein–protein and protein–DNA crosslinking, the samples were treated with 1.5 mM ethylene glycol bis (succinimidyl succinate) for 40 min at 25 °C. The isolated chromatin was digested with the restriction enzyme MboI, and the resulting DNA ends were sequentially subjected to end repair and adenylation using T4 DNA polymerase and Klenow enzyme. Biotin-labeled bridge linkers (forward strand: 5ʹ-[5Phos]CGCGATATC/iBIOdT/TATCTGACT-3ʹ; reverse strand: 5ʹ-[5Phos]GTCAGATAAGATATCGCGT-3ʹ) were then used to ligate spatially proximal DNA fragments in situ. After reversing protein-DNA crosslinking and purification, the ligation products were processed for library construction using Tn5 transposase-based tagmentation, and the resulting libraries were sequenced on an Illumina HiSeq X Ten platform in paired-end 150 bp mode. Data processing and statistical analysis eChIP-seq data processing eChIP-seq data were processed using a standardized analysis pipeline. Raw sequencing reads were subjected to quality filtering and adapter trimming using Trimmomatic (v0.39) [42], during which reads with low base quality or a final length < 50 bp were discarded. The remaining high-quality reads were aligned to the MH63 reference genome using the BWA-MEM algorithm implemented in BWA (v0.7.17) [43]. Alignments with high mapping confidence (MAPQ > 30) were sorted by genomic coordinates, and PCR duplicates were removed to retain uniquely mapped fragments using Picard (v2.23.9, https://broadinstitute.github.io/picard/). Genome-wide signal tracks (bw files) for visualization in IGV were generated using BEDTools (v2.29.0) [44]. RNAPII peak calling was performed using MACS2 (v2.1.1) [45] with the following parameters: -B -f BAMPE -q 0.05 --trackline --verbose 3 --SPMR --keep-dup all. Total RNA-seq data processing and SNP calling To obtain SNP information for the parental genotypes, the total RNA-seq data were processed. Paired-end reads were quality-controlled using fastp (v0.23.4) [46], and the filtered reads were aligned to the MH63RS1 reference genome using HISAT2 (v2.1.0) [47]. Low-quality alignments (MAPQ < 30) were removed using samtools (v1.16.1) [48], and uniquely mapped reads were retained and converted to BAM format (samtools view -q 30 ∗. bam | grep 'NH:i:1$'). Subsequently, the BAM headers were modified using the AddOrReplaceReadGroups function of Picard (v2.23.9, https://broadinstitute.github.io/picard/). The updated BAM files were used as input for GATK (v4.5.0.0) [49] to call genomic variants using HaplotypeCaller (using --java-options "-Xmx100g" and -ERC GVCF). Joint genotyping was performed using GenotypeGVCFs to generate VCF files with variant site information. Finally, SNPs were extracted from the VCF files using gatk SelectVariants (--select-type-to-include SNP), followed by quality filtering with gatk VariantFiltration (--filter-name "QD2" --filter-expression "QD < 2.0" --filter-name "FS60" --filter-expression "FS > 60.0" --filter-name "MQ40" --filter-expression "MQ < 40.0" --filter-name "SOR3" --filter-expression "SOR > 3.0" --filter-name "ReadPosRankSum-8" --filter-expression "ReadPosRankSum < −8.0" --filter-name "MQRankSum-12.5" --filter-expression "MQRankSum < −12.5″) to obtain high-confidence SNP datasets for downstream genotyping analysis. Only high-confidence parental SNPs supported by uniquely mapped reads and passing GATK-based quality filtering were used for downstream haplotype assignment. BL-Hi-C data analysis and SNP phasing For the total RNA-seq data, parental read information was extracted from uniquely mapped BAM files using a custom script (SNPphase.py) based on identified SNPs. Because the Ghd7 allele is present only in the paternal genome, all reads that mapped to this allele were classified as paternal. The extracted parental read data were converted to bedGraph format using the genomecov function of BEDTools (v2.29.0) [44], followed by conversion to BigWig format using bedGraphToBigWig (v4) [50]. The resulting paternal and maternal BigWig files were used for visualization in IGV. For the BL-Hi-C data, raw paired-end sequencing reads were trimmed using Cutadapt (version 1.18) with the following parameters: -a CTGTCTCTTATACACATCT -A CTGTCTCTTATACACATCT -g AGATGTGTATAAGAGACAG -G AGATGTGTATAAGAGACAG -O 8 -m 17 --no-indels. Overlapping read pairs were merged into single contiguous sequences using FLASH (v1.2.11) [51]. Next, linker sequences (>A GATCACGCGATATCTTATCTGACTGATC; >A_ GATCAGTCAGATAAGATATCGCGTGATC) were identified using Cutadapt, and PETs containing exactly one linker with ≥ 16 bp of interaction sequence on both sides were retained. After deduplication, the PETs were aligned separately to the reference genome using Bowtie2 (v2.4.4) [52] with the following parameters: --very-sensitive -L 20 --end-to-end --reorder --rg-id BMG. Alignment pairs in which both ends were uniquely mapped with a mapping quality > 20 were merged using the Merge_TRUEsam.py script in HiC-Pro (v3.1.0) [53]. Finally, parental genotype information was assigned to both ends of each PET using a custom script (SNPphase.py) based on the SNP data, and PETs in which both ends overlapped with informative SNP loci were retained as the final dataset. To assess the reliability of SNP-based haplotype assignment and potential genotyping bias, the global balance of maternal- and paternal-phased PETs was evaluated in both the BL-Hi-C and RNAPII ChIA-PET datasets, the symmetry of maternal- and paternal-biased features was examined, and representative loci were manually inspected in IGV. Representative shared loops detected by both BL-Hi-C and RNAPII ChIA-PET showed concordant haplotype assignments, supporting the robustness of the SNP-based phasing strategy. PCA of an interchromosomal interaction matrix To perform PCA on interchromosomal interaction frequencies, PETs whose both ends overlapped with SNP loci were categorized and quantified, and these counts were recorded in a matrix. The resulting matrix was then normalized [33], after which PCA was conducted using the prcomp function in R (center = TRUE, scale = TRUE). Inference of the spatial positions of chromosomes The relative 3D positions of rice chromosomes were inferred using the iteratively weighted adjustment algorithm proposed by Han et al. (2020) [33]. In this procedure, each chromosome was first assigned a random initial set of XYZ coordinates, after which the coordinates were iteratively updated based on the interaction frequencies derived from the phased interchromosomal PETs. To avoid potential bias introduced by high interaction frequencies between homologous chromosomes, the interaction values between homologs were replaced with the mean interaction frequency between each chromosome and all other non-homologous chromosomes. 3D nucleus reconstruction The 3D chromatin architecture of rice nuclei was reconstructed using Dip-C and Hickit [31,54]. Briefly, parental haplotype information in the required format was extracted from SNP-containing VCF files using hickit.js. Based on these SNPs, the BL-Hi-C PET alignment files were processed using hickit.js sam2seg and hickit.js chronly -y, which identified the phases of contact of overlapping SNPs with known haplotypes. The resulting files were then converted to the required format using Hickit (hickit -o) and subjected to missing phase imputation (hickit -u -o). The imputed output was subsequently used as input for 3D genome modeling with Hickit (hickit -M -i -S r1m -c1 -r10 m -c2 -b4m -b1m -b200k -D5 -b50k -D5 -b20k -O). Next, repetitive regions were removed from the reconstructed 3D structures using dip-c clean3, yielding cleaned, bin-level coordinate files. These cleaned 3D coordinates were converted into PyMOL-compatible.cif files using dip-c color and dip-c vis software. Finally, the resulting 3D chromatin structures were visualized using PyMOL. Calculation of relative chromosome positioning and distance to the nuclear centroid To characterize the spatial positioning of chromosomes within the nucleus, the coordinates obtained from three-dimensional genome reconstruction were used to calculate the distance between each chromosomal region and the nuclear centroid. Briefly, the 3D coordinates of each genomic bin were extracted, including chromosome identity, genomic position, and the corresponding x, y, and z values. Adjacent genomic bins were then merged into 200-kb windows, and the mean coordinates of all bins within each window were used to represent its spatial position. The nuclear centroid was defined as the average position of all genomic bins in the 3D model. The Euclidean distance from each 200-kb window to the nuclear centroid was then calculated as follows:d=(x−xc)2+(y−yc)2+(z−zc)2 where X, Y, and Z represent the spatial coordinates of a given 200-kb window, and Xc, Yc, and Zc represent the coordinates of the nuclear centroid. A smaller distance indicates that the region is located closer to the nuclear center, whereas a larger distance indicates that it is positioned closer to the nuclear periphery. Based on these measurements, the relative spatial positions of different chromosomes within the nucleus were compared. Allele-specific expression analysis To identify ASE, quality-controlled clean RNA-seq reads from the hybrid were aligned to the SNP N-masked MH63RS1 reference genome. Informative SNP sites distinguishing the maternal and paternal alleles were used for phased allelic read counting. Specifically, allele-supporting reads at each informative SNP were quantified using GATK ASEReadCounter. The resulting SNP-level allelic counts were then converted into GeneiASE input format and aggregated to the gene level using GeneiASE (v1.0.1) in static mode. Only genes with informative SNP support and phased transcript information were retained for ASE evaluation. Genes lacking phased transcript information or with insufficient informative allelic reads were excluded from further analysis. Gene-level ASE significance was determined using the GeneiASE static model; genes with an FDR < 0.05 were considered to be significant ASE genes. Maternal- and paternal-biased expression was assigned according to the direction of the allelic imbalance. Analysis of allele bias and allele-specific chromatin loops Allele-specific RNAPII peak analysis was performed as previously described [29,32,37]. Briefly, a union set of RNAPII peaks (q < 0.01) was generated from the combined phased uniquely mapped RNAPII ChIA-PET reads. Maternal and paternal-phased read counts within each peak were then quantified for each biological replicate. Allelic differences in peak signals were tested using DESeq2 with a paired design. Peaks with an adjusted P value < 0.01 were considered to be allele-biased, and those with log2(maternal/paternal) ≥ 2 or ≤ −2 were classified as maternal-biased or paternal-biased peaks, respectively. To analyze haplotype-resolved chromatin loops, maternal- and paternal-phased PETs were analyzed separately using ChIA-PET Tools V3. Interaction clusters with PET count ≥ 3 and q < 0.05 were defined as chromatin loops. Chromatin loops whose anchors overlapped with only maternal-biased or only paternal-biased RNAPII peaks were defined as maternal-specific or paternal-specific chromatin loops, respectively. CRediT authorship contribution statement Shiping Luan: Writing – original draft, Visualization, Formal analysis. Runxin Gao: Investigation. Guoting Chen: Formal analysis. Chuanhao Liu: Investigation. Denggao Li: Investigation. Zhifei Xue: Investigation. Ying Zhang: Investigation. Yue Li: Investigation. Cheng Ding: Investigation. Beibei Liu: Investigation. Yanyan Liu: Investigation, Funding acquisition. Jiapei Yan: Writing – review & editing, Funding acquisition. Mohamed F. Foda: Writing – review & editing. Xingwang Li: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Weizhi Ouyang: Writing – review & editing, Writing – original draft, Supervision, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.