Domains Rearranged Methylase 2 maintains DNA methylation at large DNA hypomethylated shores and long‐range chromatin interactions in rice

Summary

DNA methylation plays an important role in gene regulation and genomic stability. However, large DNA hypomethylated regions known as DNA methylation valleys (DMVs) or canyons have also been suggested to serve unique regulatory functions, largely unknown in rice (Oryza sativa). Here, we describe the DMVs in rice seedlings, which were highly enriched with developmental and transcription regulatory genes. Further detailed analysis indicated that grand DMVs (gDMVs) might be derived from nuclear integrants of organelle DNA (NORGs). Furthermore, Domains Rearranged Methylase 2 (OsDRM2) maintained DNA methylation at short DMV (sDMV) shores. Epigenetic maps indicated that sDMVs were marked with H3K4me3 and/or H3K27me3, although the loss of DNA methylation had a negligible effect on histone modification within these regions. In addition, we constructed H3K27me3‐associated interaction maps for homozygous T‐DNA insertion mutant of the gene (osdrm2) and wild type (WT). From a global perspective, most (90%) compartments were stable between osdrm2 and WT plants. At a high resolution, we observed a dramatic loss of long‐range chromatin loops in osdrm2, which suffered an extensive loss of non‐CG (CHG and CHH, H = A, T, or C) methylation. From another viewpoint, the loss of non‐CG methylation at sDMV shores in osdrm2 could disrupt H3K27me3‐mediated chromatin interaction networks. Overall, our results demonstrated that DMVs are a key genomic feature in rice and are precisely regulated by epigenetic modifications, including DNA methylation and histone modifications. OsDRM2 maintained DNA methylation at sDMV shores, while OsDRM2 deficiency strongly affected three‐dimensional (3D) genome architectures.

Introduction

Eukaryotic genomes are packaged in the nucleus to form chromatin and are covalently modified with a diverse set of chromatin marks, such as DNA methylation and histone modifications. Cytosine methylation is the most well‐understood type of DNA methylation and has been evolutionarily conserved from fungi to animals and plants (Schmitz et al., 2019). DNA methylation is crucial in multiple biological processes, including gene expression, TE (transposable element) silencing, and genome stability (He et al., 2022; Li et al., 2021; Zemach et al., 2013). Despite the widespread presence of DNA methylation in the genome, a recent proposal of several epigenotypes demonstrated that regions with low DNA methylation levels were considered functionally important. These regions resembled scUMC (sparsely conserved undermethylated UMC), CGIs (CpG islands), LMRs (low methylated regions), and UMRs (undermethylated regions) found in many animal and plant cell types (Crisp et al., 2020; Irizarry et al., 2009; Lin et al., 2017; Stadler et al., 2011). Single‐base methylome analysis using WGBS (whole‐genome bisulfite sequencing) showed that these regions were enriched for functional regulatory elements, such as promoters, enhancers, and transcription factor binding sites. Further analysis of grass genomes indicated that the UMRs represented all potentially active promoters and cis‐regulatory units (Crisp et al., 2020). In addition, a large portion of hypomethylated regions that extend well beyond proximal promoters have been termed DMVs (DNA methylation valleys), canyons, or broad non‐methylated islands (Jeong et al., 2014; Long et al., 2013; Xie et al., 2013). PRC2 (Polycomb repressive complex 2), which is responsible for H3K27me3, is required for the maintenance of DMV hypomethylation, which is thought to recruit TETs (10–11 translocation proteins) to remove DNA methylation (Li et al., 2018; Margueron and Reinberg, 2011; Wiehle et al., 2016). The DMV shores become eroded in the absence of DNMT3a (DNA methyltransferase 3a; Jeong et al., 2014). The preceding finding raises the fascinating prospect that TET‐dependent DNA demethylation and DNMT3a‐dependent DNA methylation may occur in Polycomb‐targeted DMVs, resulting in a cycle of methylation erasure and restoration at their boundaries. The properties and functions of DMVs in plants, particularly in rice, are still unknown.

It is universally accepted that chromatin occupies specific nuclear spaces and interacts non‐randomly. In recent years, the development of chromosome conformation capture methods such as Hi‐C (high‐through chromosome conformation capture), ChIA‐PET (chromatin interaction analysis by paired‐end tag sequencing), and HiChIP (Hi‐C and chromatin immunoprecipitation) have provided an unprecedented opportunity to explore spatial chromatin organization at the genome‐wide level (Fullwood et al., 2009; Lieberman‐Aiden et al., 2009; Mumbach et al., 2016). Given that both DNA and histones can undergo specific epigenetic modifications, previous studies have proposed that there is an interdependent relationship between histone modifications and chromatin organization (Bi et al., 2017; Rao et al., 2014), but the relationship between higher order chromatin structure and DNA methylation remains elusive. Chromatin compartments may influence DNA methylation modifications by regulating the accessibility of DNMTs (DNA methyltransferases), and DNA methylation deletions do not affect the organization of the A/B compartment, thus suggesting that DNA methylation is dispensable for chromatin architecture (Nothjunge et al., 2017; Zhang et al., 2018b). A joint analysis of chromatin conformation and DNA methylation in a single human cell demonstrated that methylated cytosine signature is associated with cell‐type specific chromatin interactions (Lee et al., 2019).

Rice is one of the most important food crops worldwide and has been established as a model plant for genomic research. Multiple studies have previously confirmed that rice has compartment, domain, and loop structures similar to mammalian cells (Dong et al., 2018; Liu et al., 2017; Zhao et al., 2019; Zhou et al., 2019c). Additionally, one recent study based on Hi‐C in osmet1‐2 (null mutant of DNA methyltransferase 1 gene OsMET1‐2) confirmed that CG methylation played an essential role in chromatin packing (Wang et al., 2022). However, the rice genome encodes nine putative DNMTs containing a highly conserved catalytic domain and are targeted to distinct regulatory pathways (Sharma et al., 2009; Zhang et al., 2018a). OsDRM2 is the major DRM‐type gene in rice responsible for de novo DNA methylation through the RNA‐directed DNA methylation pathway (Law and Jacobsen, 2010; Matzke et al., 2015). The effect of OsDRM2 on chromatin structures remains unclear. Moreover, two recent studies focusing on the genome‐wide hypomethylation domains have demonstrated that very large DNA methylation nadirs (or grand canyons) form at a megabase scale, and inter‐chromosomal 3D genomic interactions did not require the CTCF (CCCTC‐binding factor) and cohesion (Mclaughlin et al., 2019; Zhang et al., 2020). It is essential to explore the effect of non‐CG methylation on genome architecture and the association between DMVs and chromatin organization.

Here, we characterized DMVs in the rice genome and observed a class of gDMVs that were likely derived from organelle‐to‐nucleus DNA transfer events. Further, we mapped the epigenetic landscape of sDMVs and confirmed that OsDRM2 maintained the shores. High‐resolution chromatin interactome maps in wild type (WT) and osdrm2 revealed the role of DNA methylation (or non‐CG methylation) in mediating the 3D genome organization.

Results

Identification of DMVs with <5% bulk methylation level in rice

To study the DMVs in the rice genome, we searched the large hypomethylated regions with a previously described approach (Chen et al., 2018) using WGBS data sets (Table S1; Tan et al., 2016). We then characterized 1543 DMV regions with <5% bulk methylation in all cytosine contexts (CG, CHG, and CHH) across the rice genome, which occupied 2.9% of the rice genome (10 Mb; Figure 1a,b). The average size of these regions was approximately 6 kb, with the largest fragment reaching 134 kb (Figure 1c). To explore the features of these DMVs in the rice genome, we analysed the enrichment of TEs and PcGs (protein‐coding genes) in the DMV regions. We randomly simulated 1543 genome intervals of equal length with DMVs as controls. Compared with the DMV shores, which enriched approximately nine TEs per fragment, the average number of TEs in DMV regions was reduced to four, significantly lower than that observed in the random control (Figure 1d). In contrast, the density of PcGs in DMVs was higher than in adjacent shore regions (Figure 1e). Considering that PcGs were enriched in DMVs and whether these genes have specific functions was considered. In total, 1684 coding genes were found within the 1543 DMVs, and approximately 42.8% of these genes were expressed (FPKM values >1) in seedlings (Table S2). In addition, DMV genes with a significantly higher expression level than the non‐DMV genes, which processed a comparable TE overlapping ratio with DMV genes, served as random controls (Figure 1f). To characterize the expression patterns of these DMV genes, we analysed RNA‐seq data from 20 various rice tissues (Table S2). Subsequently, we found that DMV genes tended to be expressed in the majority of examined tissues, showing a completely different distribution from that observed in the random control (Figure 1g). GO (Gene ontology) analysis of all DMV genes also revealed that DMV genes were involved significantly in morphological development, DNA binding, transcription regulation, and RNA biosynthetic and metabolic processes (Figure S1). Overall, these data revealed a unique class of genomic regions in rice that showed wide depletion of DNA methylation, while TEs and DMV regions were highly enriched with developmental and transcription regulation genes.

Figure 1.

Identification of DMVs with <5% bulk methylation level in rice seedlings. (a) Genome Browser track depicting the methylation profile in Chr1 in rice seedlings. Methylation ratios from 0 to 1 are shown in orange, DMVs (methylation ratio ≤5%) are indicated by green bars. Gene structures and transcription directions (arrows) are shown below. (b) Statistics of DMVs on each chromosome; Y‐axis represents a number (green bar) and portion (black line) of DMVs on each chromosome. (c) Size distribution of DMVs. (d) The average number of TEs across DMVs and adjacent regions. NS is not statistically significant, P < 2.2e−16 from the Wilcoxon test. (e) Density of PcGs (protein‐coding genes) across DMV regions. DMVs are normalized to position 0, representing the DMV centre, and positions −1 and 1 illustrate the DMV edges. (f) Violin plot for DMV and non‐DMV genes' transcription levels, P < 2.2e−16 from Wilcoxon test. DMV genes have significantly higher expression levels than non‐DMV genes (Random control). (g) Expression breadth of DMV genes. Random genes served as control. P < 2.2e−16 from Kruskal–Wallis test.

Genome‐wide distribution of DMVs and grand DMVs derived from NORGs

We then examined the genome‐wide distribution of DMVs and found that they were unevenly distributed across the rice genome. We also observed that some specific chromosomal regions were significantly enriched with DMVs (Figure 2a; Figure S2). We also noticed sharp enrichment signals on chromosomes with more than 100 kb DNA hypomethylated regions (Figure 2b). Hereafter, we define these large DMVs as gDMVs (>20 kb). A total of 14 gDMVs covering 206 genes (the gene bodies entirely located in gDMVs) were then used for further analysis. Intriguingly, GO analysis of the gDMV genes indicated they were enriched regarding photosystem, electron transfer, and photosynthesis (Figure 2c).

Figure 2.

Uneven distribution of DMVs and gDMVs derived from NORG. (a) Distribution of DMVs across Chr10. Heatmaps at the bottom show the density of PcGs and TEs respectively. (b) Genome browser view of a zoom‐in region in Chr10 showed a high density of DMVs in panel (a). (c) Most enriched GO terms of gDMV genes at three aspects. The bar length represents −log10 (P‐value). CC, cellular component; MF, molecular function; BP, biological process. (d) The percentage of different types of DMVs (sDMVs and gDMVs) overlapped with NORGs. (e) Distribution of NUPTs and NUMTs across the chromosomes with gDMVs. The yellow and blue vertical lines denote NUPTs and NUMTs, respectively, with the heights of the lines indicating their densities. The green bars at the chromosome model indicate gDMVs. Black triangles represent giant NUPTs (>20 kb), and red arrows indicate giant NUMTs (>20 kb).

Considering that rice chloroplast DNA from young leaves is hypomethylated (Muniandy et al., 2019), similarly to gDMVs, we speculated that these large DNA hypomethylation fragments might be derived from NORGs. To test this hypothesis, we first investigated genome‐wide organelle‐to‐nucleus DNA transfer events in the rice genome using a previously reported approach (Ma et al., 2020), including NUPTs (nuclear integrants of plastid DNA) and NUMTs (nuclear integrants of mitochondrial DNA). As a result, we detected 1086 NUPTs and 1399 NUMTs, alongside 24 giant NORGs (defined as >20 kb, 11 giant NUPTs, and 13 giant NUMTs). The aggregated lengths of NUPTs and NUMTs were 0.95 and 1.02 Mb, with average individual lengths of 881 and 730 bp respectively (Figure S3a,b). In addition, we plotted the DNA methylation levels across all NUPTs and NUMTs. The NORG regions were found to have clearly and consistently lower DNA methylation levels than the flanking regions (Figure S3c). We overlapped these two fragments to investigate the relationship between DMVs and NORGs and found that few DMVs co‐located with NORGs (Figure S3d). We then further analysed the ratios of different lengths of DMVs that overlapped with NORGs and found that all gDMVs overlapped with NORGs, while the sDMVs displayed an exceptionally low degree of overlapping (Figure 2d; Figure S3e). These results suggested that the gDMVs demonstrated in this study were likely nuclear integrants of organelle DNA. Finally, we detected the distribution of NUPTs and NUMTs across chromosomes; 10 of the 11 giant NUPTs were co‐located with gDMVs and the other four gDMVs overlapped with giant NUMTs (Figure 2e; Figure S3f). We mapped the DNA fragment transfer from the organelle to the nucleus in the Nipponbare genome, with these inserts being hypomethylated. Moreover, we found that gDMVs defined in this study were all co‐located with NORGs, especially plastid DNA transfer events.

The role of OsDRM2 in DNA methylation at sDMV shores in rice

Previous reports have shown that large genomic regions with low DNA methylation were maintained by DNMT3a in mice (Dixon et al., 2021; Gu et al., 2018; Heyn et al., 2019; Jeong et al., 2014). Here, we aimed to explore whether OsDRM2 (orthologue of mammalian DNMT3a) was involved in this process in rice. First, we obtained the WGBS data in osdrm2 from the same research study alongside WT (Tan et al., 2016). Extensive loss of non‐CG methylation in osdrm2 was observed here (Figure S4a–c). Additionally, DMR (differential methylation region) analysis revealed that OsDRM2 mainly functioned at intergenic and promoter regions and was responsible for the methylation of short TEs (<500 bp) or MITEs (miniature inverted repeat transposable elements), especially PIF/Harbinger and Mariner superfamilies (Figure S4d–g). Considering the particularity of gDMV, subsequent studies have all been focused on sDMVs. Next, we investigated whether differential methylation was associated with sDMVs. We found that the shores of the sDMVs were hotspots for DNA methylation changes (Figure 3a). A detailed analysis of two DMR classes (CG and non‐CG) showed comparable results (Figure S5a,b). Approximately, 75.8% of sDMV shores (1159) were accompanied by DMRs, of which 80.5% were non‐CG DMRs, while 18.6% had all three context DMRs (Figure 3b).

Figure 3.

OsDRM2 maintains DNA methylation at the sDMV shores in rice. (a) Positions of DMRs were identified when comparing WT and osdrm2 at sDMVs. The DMR position in an sDMV is the relative distance between the DMR centre and the sDMV centre. sDMVs are normalized to position 0, representing the sDMV centre, and positions‐1 and 1 illustrate the sDMV edges. Random fragments served as control. (b) Proportion of sDMVs that contain DMRs at shores and relative abundance of three context DMRs. (c) Pie chart showing the proportion of sDMVs defined in WT changed in fragment lengths in osdrm2. (d) Patterns and genome‐wide average cytosine methylation levels (CG, CHG, and CHH) of three types of sDMVs in WT and osdrm2. Average cytosine methylation levels within each 100 bp interval are plotted. *** for P < 0.001 from the t‐test. LB, left border; RB, right border. Arrows highlight higher CG and CHG methylation in osdrm2 in contracted sDMVs. (e) Genome browser screenshot showing a representative region in Chr9 where the sDMV shores changed with the deficiency of OsDRM2. DNA methylation levels of CG, CHG, and CHH in WT and osdrm2 were shown. The arrows indicate DMRs and different colour boxes represent sDMVs as indicated. (f) Predominant motifs identified by de novo motif analysis for DMRs at sDMV shores. The motifs discovery study was performed on the top 500 methdiff of DMR sequences.

To examine the impact of OsDRM2 deficiency on sDMV size, we scanned sDMVs in osdrm2 using the method described above. Loss of methylation in osdrm2 led to the addition of 1613 new sDMVs, namely osdrm2 gained sDMVs, which occupied 5.8% of the rice genome (20 Mb). GO analysis indicated that osdrm2 gained sDMVs differently from sDMVs with DMR. osdrm2 gained sDMV genes enriched interpreting response to hypoxia and red or far‐red light. However, sDMVs with DMR were enriched with morphological development and transcription‐related genes (Figure S5c,d). The Wilcoxon test showed that the length distribution of sDMVs, particularly those longer than 8 kb, increased significantly in osdrm2 (P < 6.719e−9; Figure S5e). Furthermore, we found that sDMVs defined in WT changed in fragment lengths in osdrm2. A total of 38.3% of these sDMVs lost DNA methylation at shores in osdrm2, such that these regions increased in size (expanded); 24.6% showed hypermethylation in sDMV regions, such that these sDMVs decreased in size (contracted); and 37.1% experienced no significant changes in size (aligned; Figure 3c). Subsequently, we analysed the metaplot of DNA methylation for the three types of sDMVs. Fascinatingly, different patterns of non‐CG methylation were found at all three types of sDMVs (P < 0.001 from t‐test; Figure 3d,e). Meanwhile, CG and CHG methylation seemed higher in osdrm2 than in WT in the contracted sDMV regions. More detailed analysis indicated that the CG and CHG methylation within the contracted sDMV regions was significantly increased in osdrm2 (Figure 3d; Figure S5f). Finally, by searching for DNA sequence motifs present in sDMV shores, we found that these regions were enriched for transcription factor binding motifs such as VRN1, AP2EREBP, and other unknown but significantly enriched GGAGGGAG motifs (Figure 3f). Furthermore, GO analysis revealed that DMV shore genes were significantly enriched concerning 1,3‐beta‐D‐glucan metabolic processes (Figure S5g). In summary, deficiency in OsDRM2 affected DNA methylation at sDMV shores and further disturbed sDMV boundaries.

Epigenetic landscape of sDMVs and effect of OsDRM2 ablation on sDMVs

To study the epigenetic landscape of sDMVs and the effect of OsDRM2 ablation on sDMVs, we analysed the occupancy of four histone modification markers (H3K27me3, H3K4me3, H3K4me1, and H3K9me2) in rice seedlings with the same materials above, both WT and osdrm2 (Table S3). The mutant displayed severe pleiotropic growth phenotypes at both the seedling and reproductive stages (Figure 4a,b). Additionally, a genome‐wide Spearman correlation heatmap (with 1 kb bin size; Figure S6a) revealed high reproducibility between replicates. The high‐quality replicates were afterwards merged, and the respective peaks for subsequent analysis were identified. We also checked the epigenomic features of sDMVs in WT rice seedlings. Notably, most sDMVs marked with H3K4me3 and/or H3K27me3 (91.5%; Figure 4c–e), but H3K4me1 and H3K9me2 were depleted in contrast (Figure S6b,c).

Figure 4.

OsDRM2 ablation has a weak impact on histone modifications within sDMV regions. (a, b) Aberrant developmental phenotypes of osdrm2 compared with WT. Severely reduced plant height in osdrm2 at seedling (a) and mature stage (b) is shown. (c) Heatmaps of H3K4me3 and H3K27me3 in WT enriched in sDMV regions. All sDMVs are normalized to the same length. The heatmaps showing the histone enrichment signals for sDMVs and extended region (1 kb upstream and downstream). (d) The percentage analysis of four group sDMVs enriched with distinct histone modifications. (e) Profiles of epigenetic landscapes of sDMVs in Chr1 in WT. The grey boxes indicate four types of sDMVs. (f) Number of differential peaks between WT and osdrm2. Increased representing a higher level of peak intensity in osdrm2 and vice versa. (g) Representative examples of the epigenetic landscape of Chr8 in WT and osdrm2. The black boxes under the peak indicate differential peaks.

DNA methylation and histone modifications are two major epigenetic mediators. To explore how altered DNA methylation in osdrm2 affected histone modifications in rice, we first characterized differential (or ectopic) histone modifications between WT and osdrm2 by quantitative comparison (with an adjusted P value of 0.05 and log2 fold‐change of >1). We detected a small number of significant differential peaks, with H3K27me3 showing the most changes (262 increased and 96 reduced peaks; Figure 4f,g). Notably, the heterochromatin mark H3K9me2 between osdrm2 and WT exhibited no changes in peak intensities under the above conditions. This might be accounted for by the absence of DNA methylation in osdrm2, which was seen mainly in the euchromatin regions in WT (Tan et al., 2016).

To demonstrate the landscape of OsDRM2 target sites, DMRs were clustered into three groups based on histone modifications. Group 1 was covered with strong H3K27me3 and was mainly located in intergenic regions, suggesting that these regions were modified with mutual DNA methylation and H3K27me3. Meanwhile, group 2 was located upstream of the promoter region with a high level of H3K4me3 modification, as presented in Figure S7a,b. These results indicated that OsDRM2 mainly targeted H3K27me3 marked intergenic and H3K4me3 marked promoter regions, while the loss of DNA methylation in osdrm2 had no obvious effect on the distribution of H3K9me2.

Furthermore, we also considered whether H3K4me3 and H3K27me3 within sDMV regions were affected in osdrm2, which had disturbed boundaries. Comparing these two histone modifications in WT and osdrm2, we found that peaks in sDMVs were stable, with only 0.6% and 1.7% being differential peaks in H3K4me3 and H3K27me3 respectively. Representative screenshots for histone modifications in the sDMV regions show the differential between WT and osdrm2 (Figure S7c). Collectively, our data suggested that sDMVs were marked with H3K27me3 and/or H3K4me3, and the ablation of OsDRM2 had a weak impact on the deposition and distribution of histone modifications within the sDMVs.

The effect of OsDRM2 deficiency on 3D genome architecture

To study the effect of OsDRM2 deficiency (extensive loss of non‐CG methylation) on 3D genome architecture, we applied the long‐read ChIA‐PET method (Zhao et al., 2019) to generate two independent high‐resolution interaction maps of WT and osdrm2 using H3K27me3 antibodies in rice seedlings. We collected ~293 million and ~279 million uniquely mapped paired‐end tags (2 × 150 bp) from ChIA‐PET libraries in WT and osdrm2 respectively (Table S4). Considering the high reproducibility of the biological replicates for each ChIA‐PET data set category (Figure S8a), we combined biological replicate data for downstream analysis.

To assess the effects of genome‐wide loss of non‐CG methylation on H3K27me3‐associated chromatin structures, we first examined the overall contact frequency map at resolutions of 100, 50, and 10 kb with our ChIA‐PET data. We found that the 3D chromatin interaction patterns of osdrm2 were generally similar to those of WT at low resolution, except for the several obvious losses of interactions, as illustrated in Chr9 (chromosome 9; Figure 5a; Figure S8b). Subsequently, through compartmentalization pattern analysis, we found that most of the A and B compartments remained conserved, with only 10% of the annotated genome changing compartments A‐B (8.18%) or B‐A (1.87%; from WT to osdrm2; Figure 5b; Figure S8c). Interestingly, more changes occurred from the A compartment to the B compartment in WT. Detailed analysis of the switched regions confirmed that compartment changes were accompanied by differences in DNA methylation (Figure S8d).

Figure 5.

OsDRM2 depletion affected H3K27me3‐associated chromosome conformation. (a) Heatmaps from H3K27me3 with LR ChIA‐PET data in WT and osdrm2. In the top, normalized contact frequencies at 50 kb resolution of Chr9. In the bottom, a locally enlarged contact heatmap of Chr9 at 10 kb resolution. The orange boxes showed a divergence between WT and osdrm2. (b) A/B compartment switching (from WT to osdrm2) between WT and osdrm2 at 100 kb resolution. (c) Networks showing overall H3K27me3‐associated interactions in WT and osdrm2 with the same parameters and layout model. Chromosomes are coloured individually as indexed. We identified high‐confidence clusters with intra‐chromosome pet count ≥6 and inter‐chromosome pet count ≥20. (d) Loop span distribution of H3K27me3‐associated chromatin interactions in WT and osdrm2. P value <2.2e−16 from Wilcoxon rank‐sum test. (e) Ratio of inter‐chromosomal and intra‐chromosomal interactions in WT and osdrm2. (f) Number and percentage of H3K27me3 peaks involved (anchor) or not involved (basal) in chromatin interactions in WT and osdrm2.

Next, we constructed a genome‐wide network with clustered interactions using Cytoscape software in WT and osdrm2 plants with the same parameters and layout model (Figure 5c). Each chromosome was assigned to relatively independent specific nuclear spaces in the WT depending on the frequency of chromatin interactions. In contrast, the network of osdrm2 seemed messy to some extent. These observations implied that osdrm2 chromatin was decondensed. The loop span of the H3K27me3‐marked regions involved in loops was generally between 10 kb and 1 Mb (summit at 100 kb), and the loop span in WT was significantly higher than that in osdrm2, as determined through a Wilcoxon rank‐sum test (Figure 5d). In addition, more than 80% of the interactions from our data sets were intra‐chromosomal interactions, with the inter‐chromosomal interaction frequency increased in osdrm2 (Figure 5e).

Interestingly, a further detailed inter‐chromosomal interaction analysis revealed that there was a high inter‐chromosomal interaction frequency between Chr1, Chr2, and Chr3 in both WT and osdrm2, which may suggest that these three chromosomes were relatively close in nuclear space (Figure S8e). We then calculated and compared the ratio of H3K27me3 peaks involved in interactions (anchor peaks) or not (basal peaks) and observed that reduced peaks were functional as anchors in osdrm2 (Figure 5f). Collectively, fewer H3K27me3 peaks involved in chromatin interactions and shorter genome span of the loops led to a decrease in intra‐chromosomal interactions and decondensed chromatin conformation in osdrm2 with a global loss of non‐CG methylation.

Long‐range chromatin interactions in osdrm2 and WT

The extremely high‐resolution H3K27me3 interactome enabled us to make detailed comparisons of chromatin loops between osdrm2 and WT. We characterized high‐resolution pairwise chromatin loops and various genomic data sets, including histone modifications (H3K4me1, H3K4me3, H3K27me3, and H3K9me2), non‐CG methylation, and gene expression levels (Figure 6a). Despite the chromatin interaction patterns between WT and osdrm2 being globally similar at a low resolution (Figure S8b), chromatin loops showed drastic changes, especially in long‐range loops (loop span >200 kb; Figure 6a,b; Figure S9). Long‐range interactions were dramatically depleted in osdrm2 (Figure 6b). Additional analysis provided more insight showing that histone modification did not change but caused extensive loss of DNA methylation in this region (Figure 6c). We further calculated the number of long‐range genome‐wide loops with high confidence (pet count ≥20), with each chromosome displaying a significant decrease in osdrm2 (Figure 6d). DNA methylation analysis for pairwise chromatin loop anchor regions revealed that the differences were significantly increased in osdrm2 (Figure 6e). In contrast to the general loss of long‐range chromatin interactions in hypomethylation regions, we also observed remarkable examples of increased chromatin interactions in hypermethylation regions where it was accompanied by an increase in H3K27me3 peaks, such as on Chr8 and Chr11 (Figure S10). These results provide new insight showing that DNA methylation (or non‐CG methylation) maintains long‐range interactions independent of histone modification changes.

Figure 6.

Long‐range chromatin interactions decreased in osdrm2, including sDMV‐associated regions. (a) Global view of H3K27me3‐associated chromatin interactions and epigenetic and transcriptional features in Chr9 in WT (upper) and osdrm2 (bottom). Long‐range loops (loop span >200 kb) and local loops (loop span <200 kb) were shown with indicating colours. Tracks from the inner to the outer circles represent non‐CG methylation, histone modification, and gene expression respectively. (b) A representative example of a dramatic loss of long‐range interaction loops in osdrm2. (c) Zoomed‐in view of the indicated regions' detailed histone modification and DNA methylation distributions. Histone modifications are comparable between WT and osdrm2, but DNA methylation levels were affected. The grey boxes denote hypo‐DMRs. (d) Statistics of the number of long‐range interaction loops in WT and osdrm2. Each chromosome showed evident loss of long‐range interaction loops in osdrm2. (e) DNA methylation variation of pairwise chromatin loops. Average DNA methylation levels within anchor sites were calculated. mr: meth ratio, P < 2.2e−16 from Kruskal–Wallis test. (f) Ratio of H3K27me3 peaks within sDMVs involved in chromatin interactions in WT and osdrm2. (g) Comparing three types of sDMV‐associated interactions in WT and osdrm2. (h) Genome browser screenshot showing disruption of H3K27me3 mediated chromatin interactions due to loss of DNA methylation at sDMV shores. The grey boxes indicated sDMVs or hypomethylated regions. (i) Model for the impact of OsDRM2‐dependent DNA methylation on sDMV state and 3D organization based on H3K27me3 ChIA‐PET data in WT and osdrm2. OsDRM2 deficiency gave rise to decreased DNA methylation at sDMV shores, but histone modifications were almost stable within these regions. Long‐range interactions were dramatic loss in osdrm2, and the collapse of sDMV shores led to the disruption of chromatin interactions, which may further affect the chromosome conformation.

Given that sDMVs are enriched with H3K27me3 modifications, we investigated whether the loss of DNA methylation at sDMV shores affected the interactions within sDMV regions. Firstly, the ratio of H3K27me3 peaks within sDMVs served as anchors to participate in chromatin interactions was decreased in osdrm2 (Figure 6f). Correspondingly, chromatin interactions were either reduced or lost in osdrm2, while all three types of sDMVs showed coincidental changes (Figure 6g). Visual inspection of ChIA‐PET data on the genome browser revealed the loss of sDMV‐associated chromatin interactions due to hypomethylation at sDMV shores (Figure 6h). Overall, our results demonstrated a new feature of the rice genome: OsDRM2 maintained DNA methylation at sDMV shores, and OsDRM2 deficiency resulted in the collapse of sDMV boundaries but had a weak impact on histone modification within these regions. H3K27me3‐associated long‐range chromatin interactions decreased due to the loss of non‐CG methylation in osdrm2. OsDRM2 ablation disrupted sDMV boundaries and affected H3K27me3‐associated chromatin interactions within these regions. These results suggested that DNA methylation (or non‐CG methylation) is important in genome architecture organization (Figure 6i).

Discussion

DNA methylation is a major epigenetic marker, and keeping it at the proper level is essential for gene regulation and genomic stability. Several studies on genome hypomethylated regions, such as CGIs, LMRs, and UMRs, have demonstrated that these regions are also functionally important and have been suggested to be potentially active promoters and cis‐regulatory units (Blackledge and Klose, 2011; Crisp et al., 2020; Stadler et al., 2011). Compared with CGIs, LMRs, and UMRs, which are generally short, DMVs show a much wider depletion of DNA methylation that covers the entire gene body. DMVs are highly conserved across vertebrates and play critical roles in various biological processes, such as cell identification and cancer development (Jeong et al., 2014; Li et al., 2018; Long et al., 2013; Zhang et al., 2018b). Furthermore, DMVs have been identified in plants recently (Chen et al., 2018; Han et al., 2022). Using rice as a model plant, it is unknown if the rice genome contains a similar characteristic. Using rice seedling WGBS data, we discovered 1543 DMVs in this study, taking up 2.9% (10 Mb) of the rice genome. This ratio is less than that of A. thaliana, which was observed to be 41%, 49 Mb (Chen et al., 2018). This might have been due to the rice genome's high cytosine methylation level and roughly 46% repetitive regions (Feng et al., 2010; Song et al., 2021; Tan et al., 2016). These DMVs ranged in size from 5 to 134 kb, with an average of 6 kb across all chromosomes, and TEs were depleted while being enriched with PcGs (1684 genes; Figure 1b–e; Figure S1). Our results identified a key feature of the rice genome: these large hypomethylated regions were enriched with developmental genes.

In addition, we identified a subset of gDMVs (>20 kb, total of 14) that showed a striking enrichment pattern for genes involved in processes related to chloroplast functioning (Figure 2c). Previous reports have also shown that organelle‐to‐nucleus DNA transfer is a common and continuous process that shapes nuclear genome architecture and profoundly affects genome evolution (Liang et al., 2018). We defined NORGs in the rice genome, and surprisingly, all gDMVs co‐occurred with NORGs (Figure 2d; Figure S3d). NUPTs were frequently detected in rice, and these transfer events resulted in numerous structural variations (Ma et al., 2020). Our results suggest that these large organelle‐to‐nucleus DNA fragments were still hypomethylated, which may help fully understand NORGs.

Considering the particularity of gDMVs, subsequent studies have focused on sDMVs. The mechanistic basis for the epigenetic landscape of sDMVs and how this methylation‐free state is maintained in plants remains to be studied. Our results indicated that the loss of DNA methylation in osdrm2 was enriched at sDMV shores, and ablation of OsDRM2 altered sDMV sizes (Figure 3a–e; Figure S5a,b). Furthermore, non‐CG methylation significantly differed between WT and osdrm2 (Figure 3d,e), consistent with the function of OsDRM2, which had little effect on CG methylation (Figure S4a). In addition, sDMVs were enriched for H3K4me3 and/or H3K27me3 in rice while depleted for H3K4me1 and H3K9me2 (Figure 4c–e; Figure S6b). Supporting the hypothesis that DNA methylation inhibits the deposition of H3K27me3 and that H3K27me3 and DNA methylation are mutually exclusive processes that operate throughout the genome (Brinkman et al., 2012; Jermann et al., 2014). We also observed more or higher H3K27me3 peaks in osdrm2, although the depletion of DNA methylation did not lead to the spreading’ of H3K27me3 into non‐target regions in sDMVs (Figure 4f,g). Previous studies have indicated that Polycomb group proteins promoted the hypomethylation of DMVs by recruiting TET proteins (Jeong et al., 2014; Li et al., 2018; Neri et al., 2013; Wiehle et al., 2016). Therefore, our results demonstrate that sDMVs are a special class of genomic regions subject to exquisite epigenetic control, including DNA methylation and histone modifications; however, the mechanisms need further exploration.

Since DNA methylation and histone modifications covalently modify eukaryotic chromatin and additional fold to form a hierarchical 3D organization, it is worthwhile to investigate the relationship between DNA methylation and higher order chromatin architecture. Earlier studies on DNA methylation and chromatin architecture have been mainly focused on the dynamics during differentiation and early development; however, these were limited to low resolution (compartment level; Nothjunge et al., 2017; Zhang et al., 2018b). Although DNA methylation modification is conserved, it occurs in both CG and non‐CG contexts in plants, whereas this occurs mainly in the CG context in mammals. The availability of some DNA methylation‐related mutants is ideal for understanding this mechanism in plants. Hi‐C interaction maps for Arabidopsis thaliana revealed that DNA methylation affected genomic architectures such as NOR‐related chromosomal interaction patterns in met1 and ddm1 (Feng et al., 2014). The rice genome displays a globally different DNA methylation profile with A. thaliana, and each context is established and maintained by distinct DNA methyltransferase families (Hu et al., 2021). A recent study revealed the potential function of DNA methylation in TAD formation in rice (Wang et al., 2022). Deficiency in OsDRM2 also provides an excellent model to investigate the effects of globally reduced DNA methylation, especially non‐CG methylation, on 3D genome organization. At the larger scale, we found that the chromosomal territories occupied independent spaces in WT but were altered in osdrm2 (Figure 5c). Increased inter‐chromosome interactions or decreased anchor peak ratio and significantly reduced loop genome span suggested H3K27me3‐associated chromatin decompaction after the loss of non‐CG methylation in osdrm2 (Figure 5d–f). Chromatin decomposition is not a result of global alteration in the 3D chromatin organization but is a consequence of perturbing one part of the epigenome (Mclaughlin et al., 2019). Analysis of the A/B compartments suggested that most compartments were stable, and the transition was mainly from A to B in WT (Figure 5b), which is consistent with dynamic changes in non‐CG methylation in DNMT3 knockdown cell lines that were restricted to the A compartment (Nothjunge et al., 2017). However, there were more B‐A compartment transitions in osmet1‐2 (Wang et al., 2022), suggesting that the effects of distinct DNMTs on 3D organization were different. Long‐range chromatin connections have dramatically decreased, according to an additional study of high‐resolution chromatin loops (Figures 5d and 6a,b,d; Figure S9). These findings provided evidence for chromatin decondensation. In addition, we observed an increase in H3K27me3‐mediated interactions with hypermethylation regions (Figure S10). This alteration may be linked to the weakened binding of methyl‐sensitive DNA binding proteins, such as REF6 (relative to early flowering 6). Several previous studies have reported that DNA methylation inhibits the targeting of REF6 (H3K27 demethylase; Cui et al., 2016; Lu et al., 2011; Qiu et al., 2019), although experiments are still required to confirm this hypothesis. DNA methylation is important for chromatic accessibility (Zhong et al., 2021), and long‐range interactions are redistributed due to the loss of chromatin accessibility (Mas et al., 2018). Long‐range interactions may have been impacted by the loss of DNA methylation (or non‐CG methylation), and chromatin accessibility might be connected to chromatin organization.

Previous studies have reported that DNA methylation is required for Polycomb‐associated interactions (Mclaughlin et al., 2019) and that large DNA methylation nadirs anchor chromatin loops to maintain cell identity (Zhang et al., 2020). Our results also indicated that the collapse of sDMV shores in osdrm2 led to the disruption of H3K27me3‐mediated chromatin interaction networks (Figure 6f–i). Loop extrusion and phase separation are known to be involved in forming a 3D genome organization (Stam et al., 2019). Given that CTCF is absent in plant species (Heger et al., 2012) and that grand canyon 3D interactions do not require CTCF and cohesion, high H3K27me3 levels are required (Zhang et al., 2020). Polycomb group proteins have been shown to exhibit liquid–liquid phase separation features (Dong et al., 2020; Tatavosian et al., 2019). These results have suggested that high‐level H3K27me3 modifications in sDMVs might be involved in chromatin architectures through phase separation. It should also be noted that the VRN1 (orthologue of rice OsMADS14) binding motif was enriched at sDMV shores, which has been reported to undergo liquid–liquid phase separation with DNA (Zhou et al., 2019a). TEs were also reported to contribute to the formation of higher order chromosomal structures (Choudhary et al., 2020; Song et al., 2020; Wang et al., 2021), which were repressed by DNA methylation and enriched at DMV shores (Figure 1d). This warrants a study of the mechanism of DNA methylation in maintaining high‐order chromatin structures at DMV shores. Overall, we revealed that DMVs, as the key genomic feature of the rice genome, are precisely regulated by epigenetic modifications and are involved in long‐range chromatin interactions mediated by H3K27me3. Our study sheds light on the role of DMVs in the inter‐relationships between DNA methylation and chromatin structures in plants.

Experimental procedures

Plant material and tissue collection

T‐DNA insertion line of osdrm2 (PFG_3A‐04110) and its corresponding WT (Oryza sativa L. ssp. japonica ‘DonJin’) used in this study were obtained from the Postech rice mutant database (http://www.postech.ac.kr/life/pfg/risd/). Germinating seeds were obtained by soaking dry seeds in water for 72 h at 37 °C. The germinated seeds were grown in a phytotron with the day/night cycle set at 14 h/10 h and a temperature of 32 °C/28 °C. Approximately, 10‐day‐old young leaves were harvested for genomic DNA extraction. The insertion of osdrm2 was confirmed by PCR using the primers P1: 5′‐TATGCATCACAGGAAGCAGG‐3′ and P2: 5′‐GTAATCGTAGAAGAGCGCGG‐3′ and the T‐DNA‐specific primer P3: 5′‐CTAGAGTCGAGAATTCAGTACA‐3′. Two‐week‐old seedling leaves from each genotype were harvested and immediately cross‐linked.

eChIP‐Seq and ChIA‐PET libraries preparation

eChIP‐Seq libraries were performed using previous methods (Zhao et al., 2020). Approximately 0.1 g samples were used for each eChIP‐Seq assay. Samples were ground in liquid nitrogen into fine powder and resuspended with 300 μL of Buffer S for 10 min at 4 °C. The homogenate was mixed with 1.2 mL of Buffer F, and the chromatin was fragmented into 200–600 bp by sonication using a Bioruptor (Diagenode). Then fragments were immunoprecipitated using antibodies as the following: H3K4me1 (ABclonal, A2355), H3K4me3 (ABclonal, A2357), H3K9me2 (Abcam, ab1220), H3K27me3 (ABclonal, A2363), and incubated at 4 °C overnight. ChIP DNA was extracted, and NEBNext® Ultra™ II DNA library prep kit for Illumina® sequencing (New England BioLabs, E7645) was used for libraries preparation. Finally, the DNA fragments (250–650 bp) were sequenced using an Illumina HiSeq X Ten system (pairedend 150‐bp reads, Annoroad Gene Technology) after size selection.

Rice long‐read ChIA‐PET libraries were constructed with a modified ChIA‐PET method (Zhao et al., 2019). About 5 g of samples were ground in liquid nitrogen into fine powder and resuspended in 100 mL of EB1 buffer. About 60–100 μg antibodies (H3K27me3, ABclonal, A2363) were used and incubated at 4 °C for 8 h with rotation. Then, mix sheared chromatin solution with antibody‐loaded beads and incubate it at 4 °C overnight with rotation. ChIP DNA was used for end‐repair and A‐tailing and proximity ligation with biotinylated bridge‐linker (forward strand: 5′‐[5Phos] CGCGATATC/iBIOdT/TATCTGACT‐3′, reverse strand: 5′‐[5Phos] GTCAGATAAGATATCGCGT‐3′). Protein–DNA complex was reverse cross‐linked and fragmented using Tn5 transposase. The libraries were then subjected to paired‐end sequencing (2 × 150 bp) using Illumina Hiseq X‐Ten.

DNA methylation analysis

Quality control, mapping, and processing of BS‐seq read were performed as follows. Briefly, low‐quality read and artificial sequence trimming were performed using Fastp with default parameters. Clean reads were mapped to the MSU7.0 rice reference genome (http://rice.uga.edu/pub/data/Eukaryotic_Projects/o_sativa/annotation_dbs/pseudomolecules/version_7.0/all.dir/) using BatMeth2‐align with default parameters (Zhou et al., 2019b). DNA methylation calling was performed with BatMeth2‐calmeth with parameter ‘‐c 5 ‐nC 2 ‐R 100’. Then we used BatMeth2‐Meth2BigWig to generate BigWig files for IGV visualization.

DMV and DMV genes identification

DMVs were identified with modified strategies, as previously reported (Chen et al., 2018; Xie et al., 2013). The methylome was scanned using a sliding window of 5 kb with smaller 1 kb incremental steps, and regions without cytosine bases was discarded. The bulk methylation levels in CG, CHG, and CHH contexts were calculated for all remaining windows. DMVs were genomic regions with <5% bulk methylation levels in all three cytosine contexts. Overlapping DMVs were merged and reported as contiguous DMV regions. Only genes that more than 90% of their entire bodies located in DMVs were designated as DMV genes.

GO analysis

The clusterProfiler R Bioconductor package (Wu et al., 2021), rice GO annotations (https://ftp.psb.ugent.be/pub/plaza/plaza_public_monocots_05/GO/go.osa.csv.gz), and clusterProfiler‐enricher were used to identify GO terms enriched in rice DMV genes. Emapplot was used to visualize the reports.

DMR analysis

We used BatMeth2‐batDMR to identify the DMRs. Only cytosine regions with adjusted P values of 0.05 or less and DNA methylation differences of more than 0.6, 0.5, or 0.2 (CG, CHG, CHH) were designated for DMRs.

RNA‐seq analysis

RNA‐seq data were first quality evaluated using FastQC and cleaned by removing adapter sequences, and low‐quality reads by Trimmomatic (Bolger et al., 2014). The clean data were mapped against the MSU7.0 rice genome with corresponding annotation by Hisat2 using default parameters (Kim et al., 2019). Then, the gene counts were calculated with Subread‐ featureCounts (Liao et al., 2014). Expressed genes were defined as those with FPKM >1, and the differentially expressed genes (log [fold change] ≥1, P < 0.05) were calculated by DESeq2.

Identification and analyses of NUPTs and NUMTs

We identify genome‐wide NUPTs and NUMTs with the method reported by Ma et al. (2020). Briefly, we obtained the complete chloroplast and mitochondrion genomes of japonica Nipponbare from NCBI GenBank (https://www.ncbi.nlm.nih.gov/genome/10) and were mapped to the MSU7.0 rice genome using BLAST+ with the BLASTN task and 4 bp word size (E value <1e–10). Given the fragmentation of NORGs resulting from recombination and TE transposition, adjacent NUPTs or NUMTs, defined as those <400 bp apart, were merged into a single transfer event.

eChIP‐Seq analysis

eChIP‐Seq data were aligned to the MSU7.0 rice genome using the bwa mem algorithm (Li and Durbin, 2009). Mapped reads with a MAPQ quality score below 30 and PCR duplicates were filtered using SAMTools to ensure high‐quality, aligned data (Li et al., 2009). For the analysis of H3K4me3 libraries, narrow‐peak calling settings were used in MACS2 (2.1.1; Zhang et al., 2008). For analysis of H3K9me2, H3K4me1, and H3K27me3 libraries, the broad‐peak mode was used in MACS2. RSC was set to >0.8, and NSC was set to >1.05 to assess the quality of each eChIP‐Seq data set; FRiP was used to assess the peak quality in the dataset (Landt et al., 2012). eChIP‐Seq data were visualized using the IGV.

Differential analysis of each histone mark

Histone modification regions were defined based on the merged peak for each duplicate. The read counts were calculated with Subread‐featureCounts, and the differential histone signal was also calculated using the DESeq2 package in R; an adjusted P value of 0.05 and log2 fold‐change of >1 were used to find significant differential histones (Figure 4f,g). The heatmap and peak profile were generated by deepTools (Ramirez et al., 2016). First, ComputeMatrix was used to calculate the peak density distribution matrix, and then the corresponding visualization results were generated by PlotHeatmap and PlotProfile.

ChIA‐PET data processing

ChIA‐PET data were processed with updated ChIAPET Tool software (Li et al., 2010). After filtering the linkers, BWA mapped the sequences to the MSU7.0 rice genome. In the ChIA‐PET Tool pipeline, we chose ChIP‐seq peaks extended by ±10% as the given anchors to call clusters. The precise statistics of ChIA‐PET replicate data sets and combined data sets are summarized in Table S4.

ChIA‐PET contact map construction and normalization

We transformed ChIA‐PET unique mapping reads to contact frequency matrix using the bedpe2Matrix program of the ChIA‐PET2 software at 100 and 50 kb resolutions with ‘‐‐all ‐‐matrix‐format complete’ parameters (Li et al., 2017). For matrix normalization, we processed the raw contact maps by iterative correction and eigenvector decomposition methods from HiC‐Pro software, with default settings to adjust the count of the contact matrix (Servant et al., 2015).

A/B compartment delineating

We used an eigenvector program from juicer software to delineate A/B compartments in ChIA‐PET data at 100 kb resolution. The first principal component of the correlation matrix indicated the compartments.

Construction of interactome networks

The interactome networks were constructed with intra‐chromosome pet count ≥6 and inter‐chromosome pet count ≥20 and integrated into a global network using Cytoscape (version 3.9.0). The network was analysed and visualized by AllegroLayout.

Accession numbers

Sequence data in this study were obtained from the Rice Genome Annotation Project website (MSU) with the following accession number: OsDRM2, LOC_Os03g02010.

Funding

This work was supported by the AgroST Project (NK2022050105 to X.L.); the National Natural Science Foundation of China (32070612 to X.L., 32200424 to W.O.); the funds from the Interdisciplinary Sciences Research Institute (2662021JC005 to X.L.); the National Key Research and Development Program of Hubei Province (2022BBA54 to X.L.); the Natural Science Foundation of Hubei Province (2022CFA024 to X.L.); the Hubei Hongshan Laboratory Research Fund (2021HSZD010 to X.L.); HZAU‐AGIS Cooperation Fund (SZYJY2021001 to X.L.); and the National Key Laboratory of Crop Genetic Improvement Research Program (ZW22B0101 to X.L.), and the China Postdoctoral Science Foundation (2022T150245 and 2022M711264 to W.O.).

Conflict of interest

The authors declare no competing interests.

Author contributions

X.L. and G.L. conceived the project. W.Z. generated the data with assistance from Y.M., W.O., and M.G. W.Z. performed the data analysis with help from H.W., B.G., P.G., X.H., and G.C. X.L. and W.Z. interpreted the data and wrote the article. All authors read and approved the final article.

Supporting information

Figure S1 Gene ontology (GO) analysis of DMV genes.

Figure S2 Distribution of DMVs, PcGs and TEs across chromosomes.

Figure S3 Genome‐wide identification of nuclear integrants of plastid DNA (NUPTs) and mitochondrial DNA (NUMTs) in rice genomes.

Figure S4 Differences of DNA methylation between WT and osdrm2.

Figure S5 Comparison of sDMVs in WT and osdrm2.

Figure S6 eChIP‐Seq data reproducibility and features in sDMVs.

Figure S7 Histone modification landscapes of OsDRM2 target sites.

Figure S8 ChIA‐PET libraries reproducibility and comparison of H3K27me3‐associated chromatin interactions in WT and osdrm2.

Figure S9 Global view of H3K27me3‐associated interactions and epigenetic and transcriptional features on each chromosome in WT (left) and osdrm2 (right).

Figure S10 Gained chromatin interactions in osdrm2 accompanied by hypermethylation.

Table S1 WGBS data of WT and osdrm2 were used in this study (Tan et al., 2016).

Table S2 RNA‐Seq data for different tissues used in this study.

Table S3 Summary of ChIP‐seq libraries.

Table S4 Summary of ChIA‐PET libraries.

Data availability statement

The raw sequence data reported in this study have been deposited in the Genome Sequence Archive in BIG Data Center, Beijing Institute of Genomics (BIG), Chinese Academy of Sciences, under accession numbers CRA008595 (https://ngdc.cncb.ac.cn/gsub/). All data supporting the findings of this study are available within the article and its supporting information or are available from the corresponding author upon request.