Deciphering histone mark-specific fine-scale chromatin organization at high resolution with Micro-C-ChIP

Abstract

The regulation of cell-type-specific transcription relies on complex 3D interactions between promoters and distal regulatory elements. Although Hi-C has advanced our understanding of genome architecture, its high sequencing demand limits use in large-scale or time course experiments. We introduce Micro-C-ChIP, a strategy combining Micro-C with chromatin immunoprecipitation to map 3D genome organization at nucleosome resolution for defined histone modifications. We profile H3K4me3 and H3K27me3-specific 3D genome architecture in mouse embryonic stem cells (mESC), hTERT-immortalized human retinal pigment epithelial (hTERT-RPE1) cells, and HCT-116 RAD21-mAID-mClover (HCT-116 RAD21-mAC) cells. We validate that Micro-C-ChIP reveals genuine 3D genome features that are not driven by ChIP-enrichment bias. We identify extensive promoter–promoter contact networks in mESCs and hTERT-RPE1, and resolve the distinct 3D architecture of bivalent promoters in mESCs. Together, our results demonstrate that Micro-C-ChIP is a high-resolution, cost-efficient approach to study histone-modification-specific chromatin folding.

Here, the authors introduce Micro-C-ChIP, a strategy combining Micro-C with chromatin immunoprecipitation to map 3D genome organization at nucleosome resolution for defined histone modifications.

Introduction

The topological organization of genomes within the nucleus is consequential for many DNA-templated processes such as gene transcription^1–5, DNA damage repair^6,7 and DNA replication⁸. The development of sequencing-based approaches to map 3D genome architecture, such as Hi-C, has revolutionized how we view mammalian genome organization^9–11. Hi-C is based on Chromosome Conformation Capture (3 C), in which chromatin is crosslinked, digested with restriction enzymes, and ligated to generate chimeric DNA molecules from topologically proximal regions. At the scale of chromosomes, Hi-C has revealed that active and inactive eu- and heterochromatin are spatially separated into A and B compartments⁹. At the level of genes, topologically associating domains (TADs) are formed by loop-extrusion machinery to regulate the interaction probability of genomic elements, such as promoters and enhancers, in 3D space^11–16. Enhancers are distal regulatory elements that modulate the transcriptional activity of promoters, presumably via direct spatial interaction¹⁷.

However, with the resolution of Hi-C, detecting focal interactions remains challenging¹⁸. The resolution of Hi-C is limited by the choice of restriction enzymes and an uneven distribution of their recognition sites across the genome. Achieving nucleosome-scale resolution requires over a billion sequencing reads, making it costly and inefficient for large-scale studies. Micro-C, an MNase-based version of Hi-C, improves the detection of short-range features such as enhancer-promoter loops by utilizing a dual crosslinking strategy and nucleosome resolution fragmentation^1,19–21. While genome-wide approaches like Hi-C and Micro-C provide a comprehensive view of chromatin organization, they are highly sequencing-intensive. This poses a challenge in studies that require data from multiple samples, such as time-course experiments or large patient cohorts, where a substantial fraction of sequencing reads is spent on regions that are not the primary focus of the study. The recent development of locus-specific capture strategies allows much deeper sequencing of a locus of interest, highlighting the importance of resolution for understanding how genomes are folded²². These approaches have revealed novel features like micro-compartments and resolved the dynamic changes in topological interactions in response to the depletion of important transcriptional factors²². However, despite their high resolution, such sequence-based capture strategies are limited to a small region in the genome.

Histone post-translational modifications (PTMs) allow targeting sequencing efforts to functionally relevant genomic regions. For example, H3K4me3 marks active promoters²³, H3K4me1 is enriched at enhancers²⁴, and H3K27me3 is associated with Polycomb-bound domains²⁵. Enriching these histone modifications enables the focus on specific chromatin states, thereby enhancing the resolution of key regulatory interactions while reducing the sequencing burden on unrelated genomic regions.

Chromatin immunoprecipitation (ChIP)-based strategies of Hi-C-like protocols, such as HiChIP²⁶, PLAC-seq²⁷, and ChIA-PET²⁸, have proven to be valuable tools to overcome the high burden of sequencing costs. Here, we combine the high-resolution method Micro-C with histone PTM-specific chromatin immunoprecipitation, a hybrid methodology we call Micro-C-ChIP. Because Micro-C leverages MNase as the chromatin fragmenting enzyme, which digests accessible DNA and leaves nucleosomes intact, this strategy is ideal for determining the 3D interactions of genomic regions specifically marked by histone post-translational modifications. Using Micro-C-ChIP, we systematically explored the 3D genome organization of H3K4me3- or H3K27me3-marked chromatin. We benchmarked our approach against existing methods, addressed inherent biases of enrichment-based technologies, and disentangled methodological artifacts from true biological interactions. While we uncovered an extensive promoter-originating connectome in cells at active promoters, we also gained insights into interactions involving bivalently marked promoters.

Results

Micro-C-ChIP maps histone-PTM-specific 3D contacts

To explore chromosome folding across chromatin domains marked with specific PTMs, we established a protocol to efficiently pull down histone mark-specific Micro-C ligation products after in situ proximity-ligation (Fig. 1a). Briefly, nuclei from dually crosslinked cells were MNase-digested, the DNA ends were biotin-labeled, and proximity ligated. Ligated chromatin was sonicated to solubilize the heavily cross-linked chromatin prior to the immunoprecipitation. The optimal conditions, e.g., sonicator, sonication cycles, and concentration of detergent, were selected to release a high fraction of proximity-ligated dinucleosomal-sized DNA fragments into the soluble fraction (Fig. 1b). To benchmark this protocol, we performed Micro-C-ChIP in mESC and hTERT-RPE1. Experiments were performed in technical and biological replicates and merged for further analysis after correlation analysis (Supplementary Fig. 1a). In total, the combined replicates yielded approximately 300 million valid read pairs for both cell types (Supplementary Fig. 1b). Our protocol recovers a high ratio of non-duplicated reads (90% for mESC and 85% for hTERT-RPE1) and resulted in interaction heatmaps correlating with ChIP-seq peaks by visual inspection, confirming the quality of Micro-C-ChIP libraries (Supplementary Fig. 1c, d).

Fig. 1. Micro-C-ChIP detects H3K4me3-specific 3D genome interactions in mESC and hTERT-RPE1 with high signal-to-noise ratio.

a Outline of the Micro-C-ChIP method. The cells are dually crosslinked, nuclei are MNase-digested to mono-nucleosomal fragments, biotinylated (green dots), ligated and ExoIII-treated. Dually crosslinked chromatin was solubilized by sonication and immunoprecipitated against histone mark of interest (purple dots). Antibody-bound ligated fragments are washed, and the DNA was purified and pulled down with streptavidin before sequencing library preparation. b Representative 1.5% Agarose gel electrophoresis of digested chromatin (5U MNase, Lane 1), compared to proximity-ligated sample (Lane 3), and sample pellet (Lane 4) and supernatant (Lane 5) after sonication. Lane 2: NEB 2-log DNA ladder. The ligation and solubilization efficiency were checked for each replicate. c Sequencing statistics of the following datasets: single dataset of bulk Micro-C in mESC (~300 mio reads)², Micro-C-ChIP, MChIP-C³⁰, and HiCHIP²⁹. d Contact map comparison of Micro-C-ChIP and MChIP-C³⁰ at comparable sequencing depth. e Unbalanced interaction heatmaps comparing Micro-C-ChIP (top-right triangle) to bulk Micro-C (bottom-left triangle) at 100 bp resolution¹. Genome coordinates and gene annotations are shown above, and corresponding ChIP-seq tracks (Supplementary Data. 1) are shown below. The Micro-C-ChIP data shown throughout this paper were pooled from two biological replicates (Supplementary Fig. 1b). The contact intensity scales are shown next to the maps here, and for all other contact maps. f As in (e), but input-based normalization is applied to Micro-C-ChIP and ICE is used for Micro-C. Scaling factors computed for the corresponding input (bulk Micro-C) were used for Micro-C-ChIP contact matrix normalization. g, h Top panel: input-normalized mESC (g) and hTERT-RPE1 (h) Micro-C-ChIP interaction heatmaps color-coded based on the sites of corresponding 1D ChIP-seq peaks. The H3K4me3-enriched sites are visualized with dark blue and non-enriched sites with gray throughout the manuscript. Bottom panel: 4C-like interaction profiles for mESC (g) and hTERT-RPE1 (h) generated using Micro-C-ChIP (orange) and corresponding Micro-C datasets (black; mESC from Hsieh et al., hTERT-RPE1 input). Normalized interactions emerging from different selected viewpoints corresponding to H3K4me3 peaks for the same genomic region. i Pileup plots of H3K4me3 Micro-C-ChIP (200 bp resolution) at putative enhancer-promoter (E-P) (left) and promoter-promoter (P-P) (right) interactions within ±10 kbp. Source data are provided as a Source Data file.

ChIP-based strategies for MNase-digested proximity ligation products were proposed by others^29,30. To benchmark Micro-C-ChIP (mESC, H3K4me3) against MChIP-C and HiChIP protocols, we compared the frequency of short (< 5000 bp), long (> 5000 bp), and trans reads (Fig. 1c). Here, the fraction of “informative reads,” as defined by Golov et al., is maintained in Micro-C-ChIP (42%), compared to genome-wide Micro-C (37%). In contrast, other protocols deplete this important fraction (i.e., 4% in MChIP-C). We explain this by differences in the protocols. MChIP-C omits the biotin enrichment step, which has been shown to enhance the abundance of proximity-ligated DNA products over non-cleaved dinucleosomes (< 5000 bp)³⁰. HiChIP performs proximity ligation after ChIP enrichment, potentially leading to non-specific ligation between molecules. In contrast, this protocol performs in situ proximity ligation before immunoprecipitation, thus preserving true 3D interactions throughout the protocol. Furthermore, we compared Micro-C-ChIP in hTERT-RPE1 to a subset with similar sequencing depth of MChIP-C by visual inspection and observed a stronger patterning at H3K4me3 enriched sites (Fig. 1d). In summary, compared to other protocols, Micro-C-ChIP is superior in enriching histone PTM-specific reads.

To further evaluate the performance of Micro-C-ChIP, we visually compared contact heatmaps for Micro-C-ChIP with a deeply sequenced Micro-C mESC dataset (~3 billion reads) from Hsieh et al.¹ (Fig. 1e). Notably, despite the much lower sequencing depth, Micro-C-ChIP detects the structural features of bulk Micro-C with high definition. Unlike bulk technologies, such as Micro-C and Hi-C, Micro-C-ChIP cannot be normalized by conventional methods like ICE because it assumes equal coverage across genomic regions³¹. This assumption does not hold for enrichment-based methods, where coverage varies inherently. We implemented a tailored normalization strategy to enable a direct comparison at a normalized level. Applying ICE normalization to Micro-C-ChIP reduces signal variations, resulting in a dataset resembling a low-sequenced bulk Micro-C dataset. Having this in mind, we introduced input-based normalization, similar to 1D ChIP-seq experiments (Fig. 1f). We leveraged the corresponding bulk Micro-C as an input, using its scaling factors (referred to as “weight”) for plotting Micro-C-ChIP contact matrices (Fig. 1f, upper panel, Supplementary Fig. 1e, f). Input normalization allows us to account for biases inherent to chromatin accessibility, sequencing, and experimental artifacts, ensuring that observed interactions reflect true protein-mediated enrichment rather than general chromatin features.

Micro-C-ChIP matrices show strong interaction signals at regions marked by the respective histone mark from published ChIP-seq data (Fig. 1e–g, Supplementary Fig. 1d, e), confirming specific enrichment. For example, precise, narrow H3K4me3 ChIP-peaks at promoter regions translated into fine stripes in 3D space, forming a grid-like structure in pluripotent (mESC) and differentiated cells (hTERT-RPE1) (Fig. 1h, Supplementary Fig. 1f). To further assess if these interactions are true 3D genome folding features, we generated 4C-like plots using H3K4me3 peaks as viewpoints (Fig.1g, h). At these sites, bulk and ChIP-enriched interaction profiles should be similar. Indeed, the signals from bulk Micro-C and Micro-C-ChIP are comparable, supporting that Micro-C-ChIP detects genuine 3D contacts.

Enrichment based on the epigenetic signature provides high-resolution insights into genome organization at low sequencing depth. However, genome-wide heatmaps can be challenging to interpret because of the intentional enrichment bias for PTM-associated regions. To indicate this bias in genome-wide interaction heatmaps, we visualized the matrices by color-coding specific sites identified as peaks in ChIP-seq, presenting interacting regions as distinct viewpoints rather than as genome-wide data (Fig. 1g, h, upper panel, Supplementary Fig. 1e, f).

H3K4me3 is known to be an active promoter mark, and H3K4me3-based interactions can be used as a proxy for promoter-originating interactions¹. To validate the enrichment of Micro-C-ChIP libraries at expected chromatin domains, we computed the intersections of promoters (P-P) as well as promoters with putative enhancers (E-P) (Fig. 1i). We plotted the observed/expected Micro-C-ChIP signal for both cell types at the intersection sites. Promoter intersections show a focal signal intensity at the center of the interaction pileup plot and stripes emerging from the promoter sites, generating a cross pattern. To validate the enrichment of Micro-C-ChIP libraries at expected chromatin domains, we computed the intersections of promoters (P-P) as well as promoters with putative enhancers (E-P) (Fig. 1i). We plotted the observed/expected Micro-C-ChIP signal for both cell types at the intersection sites. As expected, both the mESC and hTERT-RPE1 datasets show a strong focal signal intensity at the center of the P-P plot, confirming the enrichment of promoter-specific interactions.

In bulk Micro-C experiments (mESC), P-P intersections show a cross-like pattern in pileups because stripes emerge from both promoters. In contrast, E-P intersections show the stripe originating only from promoters and not enhancers^1,4. We also observe this pattern in H3K4me3 Micro-C-ChIP in mESC (Fig. 1i). Interestingly, in hTERT-RPE1, unlike in mESC, H3K4me3 Micro-C-ChIP shows a cross-like pattern at both putative P-P and E-P intersection sites. The discrepancy in the E-P interaction pileups between the two cell types can be explained by higher enhancer activity and H3K4me3 accumulation. Presumably, this makes enhancers more potent loop extrusion barriers. We conclude that Micro-C-ChIP yields histone modification-specific proximity-ligation libraries with a high signal-to-noise ratio at moderate sequencing depth and reproducibly captures 3D organization at domains of interest in different cell types.

Micro-C-ChIP contacts are not driven by ChIP enrichment bias

In this data, P-P interactions appear stronger than E-P interactions. This finding is comparable with published bulk Micro-C data. However, H3K4me3 marks P-P sites at both anchors and E-P sites are only marked by H3K4me3 at the promoter site. One concern is that Micro-C-ChIP could preferentially enrich P-P because proximity ligates Micro-C samples have more epitopes (H3K4me3) than E-P ligation products (Fig. 2a). To address this, we reconstituted chromatin with defined histone PTM levels (H3K27me3) in vitro. Here, we mix H3K27me3-modified and unmodified histone octamers at different ratios and assemble chromatin on a barcoded 2xWidom601 DNA fragment. This strategy enables us to control the abundance of methylation in dinucleosomes, thereby determining a ChIP-enrichment bias during immunoprecipitation. The probability of a Widome601 position being methylated or not follows a binomial distribution with the fraction of H3K27me3/H3K27me0 being the binominal coefficient (Fig. 2b). Thus, for a pair of Widome601 positioning sequenced to carry none, one or two H3K27me3 methylated histones can be predicted by the mixing ratios of H3K27me3 and me0 histone octamers prior chromatin reconstitution. For example, in a reconstitution where 50% of histone octamers are modified, the probability of assembling dinucleosomes with two H3K27me3-histones would be 0.25 (50% × 50%), dinucleosomes with one H3K27me3- and one unmodified histone 0.5 (50% × 50% for either nucleosome) or dinucleosomes with no methylated histones 0.25 (50% × 50% unmodified). Importantly, because only the modified nucleosomes are expected to be immunoprecipitated, we can compute the expected ratio of reads from a 100% compared to a 50% H3K27me3 / H3K27me0 ratio assembly. For example, if the ratio is 0.5, we expect a 75% fraction of dinucleosomes to have one or both nucleosomes methylated, and a 25% fraction with both nucleosomes unmethylated. If the observed pulldown efficiency is lower than 0.75, it could indicate that dinucleosomes with only one modified histone octamer are not efficiently captured, further supporting the notion of a potential bias toward sites marked by histone PTMs at both anchors.

Fig. 2. Enrichment biases in Micro-C-ChIP: validation through in vitro pulldown and chromatin architecture perturbation.

a Schematic of potential ChIP enrichment bias toward interactions with histone marks (purple dot) at both over one anchor. b Probability calculation of either both, one or no nucleosomes carrying the modification at different methylation degrees (100%, 50%, and 25%). c Expected and observed relative pulldown efficiency of different methylation degrees quantified by Western blot. Expected was determined as in (b). Two independent pools were generated, each composed of three reconstitutions, and Micro-C-ChIP was performed in duplicates for each pool. Observed values represent the mean of two replicates, calculated by dividing read counts mapped to each barcode by total reads and normalizing to H3. Values for 100% methylation were set to 1, and other values were scaled proportionally. d Illustration of two scenarios of Micro-C-ChIP responsiveness to RAD21 degradation. In the case of H3K4me3-marked interactions arising from bona fide interactions (top triangles), changes of genome architecture followed by RAD21 depletion are captured. In the opposite scenario (bottom triangles), if interactions are artificially enriched due to the presence of a histone mark at both anchors, the genome architecture perturbation is not detected by Micro-C-ChIP. e Unbalanced interaction heatmaps comparing untreated (UT) (bottom-left triangle) and IAA-treated (IAA) (top-right triangle) HCT116-RAD21-mAC H3K4me3 Micro-C-ChIP at 2-kb resolution. ChIP-seq tracks for H3K4me3, RAD21, and CTCF are shown below. f Off-diagonal pileup of H3K4me3-enriched Micro-C-ChIP datasets of untreated (UT) (upper panel) and IAA-treated (IAA) (bottom panel) HCT116-RAD21-mAC (500 bp resolution) at RAD21 (left panel), CTCF (mid panel) and H3K4me3 (right panel) ChIP-seq peak intersections. 1D H3K4me3 and ATAC-seq signal at corresponding peaks. g Venn diagram of stripe anchors identified in Micro-C-ChIP. The fraction of stripes whose anchors overlap is considered as common, and their length distributions are shown for both conditions below. h Scatter plot of stripe lengths in the UT (x-axis) and IAA-treated (y-axis) cells. The stripes were divided into 4 quantiles based on the fold-change of stripe length between conditions. The profile plots (right panel) show H3K4me3 ChIP-seq signals at the stripe anchors of the respective groups. Source data are provided as a Source Data file.

Experimentally, we determined the actual H3K27me3 levels quantitatively by Western blot and computed the expected pulldown efficiencies of each reconstitution (Fig. 2c). We tested two assemblies in two replicates: the first assembly consisted of pooled reconstitutions with 100%, 40%, and 20% H3K27me3, while the second consisted of 100%, 25%, and 10% H3K27me3 assemblies. The samples were subjected to immunoprecipitation using either H3 or H3K27me3 antibodies. Following sequencing, the reads were mapped to the barcode sequences, and the relative pulldown efficiency was quantified (Fig. 2c). Across all reconstitutions of dinucleosomes with varying degrees of H3K27me3 in both replicates, the observed sequencing coverage closely matched the predicted distribution of all H3K27me3 dinucleosomes derivates being immunoprecipitated, demonstrating that Micro-C-ChIP captures all 3D informative proximity-ligated of any H3K27me3 methylation state in an unbiased manner.

Micro-C-ChIP detects genuine genome-architecture features

Next, we addressed whether the observed interactions reflect the genuine underlying 3D organization or are artificial due to the histone mark enrichment at one or both anchors. We performed Micro-C-ChIP in a cell model where chromatin architecture can be disrupted. At the same time, H3K4me3 levels remain unchanged (Fig. 2d). We utilized HCT116-RAD21-mAC³², in which RAD21, a component of the cohesin complex, is degraded upon treatment with indole-3-acetic acid (IAA), resulting in the elimination of loop domains, as previously reported. We performed H3K4me3 Micro-C-ChIP in untreated (UT) and 6-hour IAA-treated (IAA) cells, in two replicates, which were merged after quality assessment to yield approximately 200 million valid reads in each condition.

Micro-C-ChIP captures chromatin architecture perturbations in Rad21-depleted cells. A visual inspection of H3K4me3 Micro-C-ChIP at a single locus showed a striking loss of 3D genomic features upon RAD21 degradation (Fig. 2e). We noted that particularly long-range loops and stripes were affected (Supplementary Fig. 2a, b). Importantly, the distribution of H3K4me3 remains largely unaltered upon RAD21 degradation. This suggests that Micro-C-ChIP identifies alterations in local chromatin folding induced by RAD21 loss, despite a steady level of H3K4me3.

The depletion of RAD21 primarily affects structural CTCF loops, not E-P and P-P interactions. To test if Micro-C-ChIP measures bona-fide 3D genome features, we computed homotypic intersections of CTCF, RAD21, and H3K4me3 sites based on publicly available ChIP-seq datasets. Public data showed that H3K4me3 and ATAC-seq signals are unchanged at all utilized sites. The observed/expected signal at structural sites (RAD21 and CTCF) showed a dramatic reduction of Micro-C-ChIP signal upon RAD21 depletion and a mild effect at promoter intersections (H3K4me3) (Fig. 2f). This confirms that genome-wide changes induced by chromatin folding perturbation, namely the elimination of cohesin-mediated loop domains, are detected by Micro-C-ChIP, although the H3K4me3 level is consistent. This demonstrates that Micro-C-ChIP detects bona fide 3D genome interactions and that the underlying 1D epigenetic landscape does not define the Micro-C-ChIP signal.

In bulk proximity ligation experiments, chromosome stripes are a feature of active loop extrusion (LE)^12,13. They often emerge at promoters and CTCF sites. Various tools are established to determine the position and extent of such stripes for 3D genomic datasets. Because Micro-C-ChIP enriches specific loci, stripes emerge from sites in frequent contact with proximal regions. Although we cannot distinguish structural stripes that flank TAD triangles from data-rich sites of triangles, we can leverage stripe callers to quantify the extent of the Micro-C-ChIP stripes. To quantify changes in the extent of interactions emerging from a locus, we identified enrichment stripes using Stripenn³³ at a resolution of 2 kb. As anticipated, fewer stripes were detected following RAD21 degradation (Fig. 2g). To assess the impact of RAD21 degradation on enrichment-stripes quantitatively, we examined how individual enrichment-stripe lengths changed upon RAD21 depletion. The global length of all stripes shifted towards shorter distances upon cohesin removal. To assess this further, we divided the stripes into four equally sized groups stratified by fold change in stripe length. Remarkably, all groups collapse to the distance of around 76 kbp in length after RAD21 depletion (Fig. 2h). This is similar to the distance at which interaction decay curves start to diverge upon RAD21 depletion, indicating that larger distances reflect сohesin-mediated interactions (Supplementary Fig. 2c). Notably, again, the H3K4me3 signal does not change in any of these clusters. With other technologies, the threshold distance of RAD21-dependent interactions could only be extracted from genome-wide, locus-unspecific analysis, such as global interaction decay^34,35 (Fig. 2h). This demonstrates that Micro-C-ChIP has the resolution to profile locus-specific interaction decay on a genome-wide scale.

Although the depletion of RAD21 disturbed most patterns (Supplementary Fig. 2a, b), not all short-range features were lost. We explicitly modeled the expected contact maps for both conditions to test whether the patterns could be explained solely by changes in interaction decay and Micro-C-ChIP enrichment of specific loci. We extracted read occupancy and interaction decay from control and RAD21 depletion conditions and generated predicted maps based on their respective distance-dependent decay profiles (Supplementary Fig. 2d). The predicted maps are nearly identical, indicating that if interaction decay alone dictated contact frequencies, we would not expect significant differences between the two conditions. In contrast to the modeled data, our experiments show apparent differences, with shortened and weakened stripes upon RAD21 depletion, which the decay-only model does not capture. This result demonstrates that the observed changes in patterns are not simply a consequence of generic contact decay effects but instead reflect an active disruption of long-range interactions mediated by cohesin.

H3K4me3 Micro-C-ChIP maps extensive promoter networks

Having confirmed that H3K4me3-marked interactions stem from genuine interactions and are not a product of artifacts, we aimed to better understand the functional role of H3K4me3-specific interactions. We called sites of enriched interactions, apparent as dots in contact maps and inferred to represent chromosome loops, for the H3K4me3 Micro-C-ChIP datasets in mESC and hTERT-RPE1 with the Peakachu pipeline. We detected 72574 loops in mESC and 63085 loops in hTERT-RPE1. Visual inspection showed that most of the called loops align with H3K4me3 ChIP-seq peaks and RNAPII signal in both cell types, suggesting that loops are associated with active transcription (Fig. 3a). In bulk experiments, the computational detection of loops at shorter distances is complicated by increased noise floor. Micro-C detects chromosome loops at shorter distances more efficiently than Hi-C, indicating an improved signal-to-noise ratio¹⁸. To estimate if Micro-C-ChIP improves the detection of short-range interactions, we plotted the distances of called loops as CDF (Fig. 3b). Distances between anchors of detected H3K4me3 Micro-C-ChIP loops are smaller compared to loops called from the deepest sequenced mESC Micro-C dataset available¹. This demonstrates a higher signal-to-noise ratio, allowing the detection of short-range features. Notably, the detection of short-range chromosome interactions is even higher in differentiated hTERT-RPE1.

Fig. 3. Promoter-associated loops prevail in H3K4me3-enriched Micro-C-ChIP datasets.

a Unbalanced interaction heatmaps for mESC (top) and hTERT-RPE1 (bottom) H3K4me3 Micro-C-ChIP at 2.5 kb and 5 kb resolution, respectively. Loops called at 2.5 kb resolution via the Peakachu pipeline are shown on the heatmaps as squares. Loop arcs show enriched interactions. 1D chromatin tracks are shown below the contact maps (Supplementary Data 1). b Cumulative distribution function (CDF) plot of distances between loop anchors in H3K4me3 Micro-C-ChIP in mESC and hTERT-RPE1 and bulk Micro-C (Hsieh et al.¹). c Percentage of H3K4me3-enriched loops classified into the following categories: P-P (promoter-promoter), P-Other, E-P (enhancer-promoter), CTCF-CTCF and Other. CTCF-CTCF loops include only anchors that do not overlap with any regulatory elements. P-P loops are further classified by the presence of CTCF: At one of the anchors (1x CTCF) or both anchors (2x CTCF). The loop categorization is shown for both mESC (left) and hTERT-RPE1 (right). d Interaction pileups at P-P loops (classified in Fig. 3b) for mESC (left) and hTERT-RPE1 (right). Data is shown for a ±100 kb window (100 kb upstream and 100 kb downstream) surrounding loop anchors. e Histogram of loops anchor valencies for mESC and hTERT-RPE1 H3K4me3 Micro-C-ChIP.

To understand which genomic elements are connected by chromosome loops, we classified loop anchors for both cell types into promoter-promoter (P-P), enhancer-promoter (E-P), CTCF-CTCF, promoter-other (P-other), and other interaction sites (Fig. 3c). 90–95% of all identified interaction sites were classified as promoter-associated interactions (P-P, E-P, and P-other), which is expected for the promoter mark-enriched H3K4me3 Micro-C-ChIP libraries. In mESC, 45% of all loops represent P-P interactions, while only 5% are classified as E-P interactions. In hTERT-RPE1 cells, a more significant fraction of detected interactions connects promoters with enhancers (12%), although most interactions remain between promoters (39%). Notably, manually annotated chromosome loops in RCMC experiments showed enrichment of P-P over E-P interactions, demonstrating that detecting P-P interactions is a matter of resolution²².

H3K4me3 interactions show intensive networks between promoters. We validated the presence of strong interactions between promoters in mESC and hTERT-RPE1 by plotting averaged contact signals of identified P-P loops for each cell type (Fig. 3d). Interestingly, 60% of all P-P loops were not associated with detectable CTCF in mESC. In contrast, in hTERT-RPE1 cells, we find that 78% of all P-P interactions were associated with at least one CTCF peak at either anchor site (Fig. 3c). Next, we computed the anchor valency, a measurement of how often a given anchor is engaged in chromosome loops (Fig. 3e). Despite fewer loops being called in hTERT-RPE1, the valency increases in this differentiated cell type. In comparison to Micro-C, we showed that even at lower sequencing depth (300 million reads vs. 3.18 billion filtered reads¹), Micro-C-ChIP could detect many loops with a substantial portion of close-to-diagonal contacts. Moreover, those loops possess distinctive features depending on the cell type. Of note, H3K4me3-enriched loops are P-P contacts, and the anchor valency of promoters largely increases with the transcription rate of both cell types (Supplementary Fig. 3a, b).

Micro-C-ChIP reveals structure of distinct chromatin states

To verify that Micro-C-ChIP can be used against various histone marks, we performed it against the repressive histone modification H3K27me3 in mESC and hTERT-RPE1. H3K27me3 yielded a lower non-duplicated rate (~62% in both cell types) than H3K4me3. The cis vs. trans ratio and distribution of long vs. short-range reads were highly comparable between datasets and histone marks, confirming the overall quality and reproducibility of the Micro-C-ChIP protocol (Supplementary Fig. 1a–d).

First, we validated a successful enrichment of H3K4me3 and H3K27me3 Micro-C-ChIP libraries by pileup analysis at H3K4me3-H3K4me3 and H3K27me3-H3K27me3 intersections from available ChIP-seq peaks for a corresponding cell type (Fig. 4a). We also plotted the 1D Micro-C-ChIP signal for both histone PTMs at both intersection sites, each showing strong focal enrichment at the center, which confirms efficient enrichment and a low noise floor of the enriched interactions (H3K27me3: top panel, H3K4me3: bottom panel). In addition, in mESC, we observe weaker but similar signals at non-corresponding interaction sites, e.g., H3K27me3-Micro-C-ChIP plotted at H3K4me3 intersections (Fig. 4a, bottom-left panel) and vice versa (top-right panel). In contrast to mESC, H3K4me3 is enriched at H3K4me3 sites but absent at H3K27me3 sites, and vice versa, in hTERT-RPE1 cells. We suspect that the overlapping signal reflects bivalent chromatin in mESC, marked by the active H3K4me3 and repressive H3K27me3 histone PTMs.

Fig. 4. Micro-C-ChIP efficiently detects 3D interactions at H3K27me3-marked loci.

a Off-diagonal pileup of H3K27me3 and H3K4me3-enriched Micro-C-ChIP datasets (500 bp resolution) at loci of paired ChIP-seq H3K27me3 (upper panel) and H3K4me3 (lower panel) peaks in mESC (left panel) and hTERT-RPE1 (right panel). The intersected ChIP-seq peaks at a distance between 10 and 300 kb were used. The CPM-normalized signal of H3K27me3 and H3K4me3 Micro-C-ChIP datasets was plotted at the respective ChIP-seq peak sites (bottom panel). b Input-normalized interaction heatmaps comparing H3K27me3 (bottom-left triangle) and H3K4me3 (top-right triangle) Micro-C-ChIP at different resolutions in mESC (left panel) and hTERT-RPE1 (right panel). Genome coordinates are shown above, and ChIP-seq signal tracks for H3K4me3 and H3K27me3 are shown below the contact maps. The H3K27me3-enriched sites are visualized with purple and non-enriched sites with gray throughout the manuscript.

H3K27me3 and H3K4me3 chromatin marks are binary in hTERT-RPE1. We visualized Micro-C-ChIP contact matrices across relevant PTM-marked regions, focusing on genomic loci carrying both histone marks (Fig. 4b). In mESC, H3K27me3 Micro-C-ChIP displays stripe- and block-like pattern, while in hTERT-RPE1, broad regions of inactive chromatin are more prevalent (Fig. 4b, Supplementary Fig. 4). In contrast to mESC, H3K27me3 in hTERT-RPE1 displays little to no evidence for stripes or flares, reflecting the loss of punctate H3K27me3 and preservation of broad H3K27me3 domains that occur as pluripotent cells differentiate. Conversely, H3K4me3 Micro-C-ChIP is characterized by a precise grid-like structure consisting of stripes and dots stemming from the punctate localization pattern of this modification at promoters.

Importantly, in line with the observations made at the metagene level (Fig. 4b), we confirm that Micro-C-ChIP libraries in hTERT-RPE1 yield highly distinct patterns for each mark with almost no overlap (Fig. 4b, right panel). From this, we conclude that Micro-C-ChIP effectively captures chromatin interactions specific to each histone mark, with minimal cross-mark enrichment in hTERT-RPE1. However, in mESC, the overlapping signal between H3K4me3- and H3K27me3-marked interactions suggests the presence of bivalent chromatin, where active and repressive marks coexist at the same loci, potentially influencing chromatin organization and transcriptional regulation.

Transcription is not required for stripe formation in mESC

The formation of chromosome loops and stripes are characteristic features of the active loop extrusion process³⁶. Computational modeling proposes that RNAPII functions as a moving barrier for loop extrusion⁵. Experimental data support this model and show that stripes and chromosome loops scale with transcriptional activity^3,4. Visual inspections of H3K27me3 and H3K4me3 interaction heatmaps show stripes for both marks in mESC and only for H3K4me3 in hTERT-RPE1 (Figs. 1g, h, 3a, 4b, Supplementary Fig. 1e, f). Putative P-P contacts were recently shown to be unaffected by RNAPII depletion⁴, which contrasts with previous studies¹. Thus, the effect of transcription on promoter connectome is unclear.

Leveraging the ability to annotate P-P interactions from 3D data, we investigated how P-P contacts were identified from the H3K4me3 dataset (Fig. 3d) and scaled with transcription, stratifying them by RNAPII level (Fig. 5a, b). We find that focal P-P contact intensities are comparable between promoters with different RNAPII levels, although the stripe intensities decrease with a reduction in RNAPII level. To test if these P-P contacts also include bivalent promoters, we plotted H3K27me3 Micro-C-ChIP data at these sites. We observe an overlap between H3K27me3 and H3K4me3 signal at RNAPII-depleted sites in mESC (Fig. 5a). This effect is more prominent in mESC than in hTERT-RPE1 (Fig. 5b). This supports the finding that bivalent promoters, such as the Hox gene cluster, engage in more extensive P-P networks in mESC³⁷.

Fig. 5. H3K27me3 Micro-C-ChIP reveals 3D signatures of active promoters at repressed TSS exclusively in undifferentiated cells.

a H3K27me3 (top) and H3K4me3 (bottom) Micro-C-ChIP interaction pileups at P-P loops from Fig. 3d stratified by RNAPII occupancy at loop anchors into three quantiles for ±20 kb for mESC. b Same as in a, but for hTERT-RPE1. c TSS-centered pileups of H3K27me3 (top) and H3K4me3 (bottom) Micro-C-ChIP in mESC stratified by RNAPII occupancy into five quantiles. The sixth panel represents cumulative interactions of TSS. d Same as (c), but for hTERT-RPE1.

Next, we analyzed the chromatin structure at all promoters, sorted by RNAPII signal, and plotted transcription start site (TSS)-centered on-diagonal pileups (Fig. 5c, d). In agreement with previous work¹, the H3K4me3 Micro-C-ChIP signal strengthens with the increase of RNAPII occupancy in both cell types. However, the promoter-originating stripes are present even in the bottom-quantile RNAPII occupancy in mESC. In contrast, H3K27me3-enriched Micro-C-ChIP data show a gradual decrease in interaction signal with increasing RNAPII occupancy in mESC, whereas, in hTERT-RPE1, no H3K27me3-associated promoter-stripes are detectable in any cluster. Notably, the finding that H3K27me3 data show loop extrusion patterns in mESC but not in hTERT-RPE1 cells demonstrates that the detection of loop-extrusion patterns is not a methodological artifact of Micro-C-ChIP at H3K27me3-enriched promoters. Our data suggest that H3K27me3 is part of a different chromatin architecture in mESC compared to hTERT-RPE1, and loop-extrusion features, such as stripes and chromosome loops, accompany this architecture.

Micro-C-ChIP reveals promoter networks at bivalent loci

To get further insight into the commonalities of promoter-originating interaction distances in H3K4me3 and H3K27me3 Micro-C-ChIP datasets in mESC, we again leveraged Stripenn to call enrichment stripes. Scaled pileup analysis at enrichment stripe anchors shows that H3K4me3 and H3K27me3 Micro-C-ChIP signals are enriched at opposing anchors (Fig. 6a), arguing that bivalent regions show loop-extrusion features. In contrast, H3K4me3-called stripes from hTERT-RPE1 cells did not show any enrichment in H3K27me3 Micro-C-ChIP data (Fig. 6a, bottom panel), and stripes called from H3K27me3 data using Stripenn yielded only a marginal number of stripes (Supplementary Fig.4a), consistent with the lack of a stripe pattern upon visual inspection. To further dissect the bivalent nature of stripes common for both histone marks, we classified stripe anchors of the same orientation (5’ in this instance) identified in H3K27me3 and H3K4me3 Micro-C-ChIP in mESC (Fig. 6b). The majority of stripes emerge from promoters. Of 2170 H3K27me3 stripes, more than 50% are bivalently marked promoters. In contrast, only 24% of H3K4me3 stripes are bivalent, and H3K4me3 uniquely labels the majority. This suggests that H3K27me3-marked enrichment stripes emerge mostly from bivalent chromatin, whereas H3K4me3-marked enrichment stripes primarily emerge from active promoters.

Fig. 6. Chromatin organization in mESC is characterized by bivalent promoter-connectome.

a Rescaled pileup plots of 5’-and 3’- stripes at the mESC H3K27me3 (left) and H3K4me3 (right) and hTERT-RPE1 H3K4me3 (bottom panel) Micro-C-ChIP averaged contact maps. The average intensity of interactions across stripes that were detected in the H3K27me3 and H3K4me3 Micro-C-ChIP datasets is shown at the corresponding and non-corresponding heatmaps. The black arrows point at the stripes that were called at the non-corresponding datasets in mESC. b H3K27me3 and H3K4me3 stripe anchors were classified based on overlap with TSS locations (±500 bp). TSS have been categorized according to their 1D epigenetic status: with both H3K27me3 and H3K4me3 peaks (bivalent), with only H3K27me3 peaks, with only H3K4me3 peaks or without any histone marks. The fraction of H3K27me3 stripes that did not overlap with TSS was intersected with H3K27me3 peaks without TSS. c Interaction heatmaps of H3K4me3 (bottom triangle) and H3K27me3 (top triangle) in mESC at 1-kb resolution. d Interaction heatmap of H3K4me3 (bottom triangle) and H3K27me3 (top triangle) in hTERT-RPE1 at 5-kb resolution. Snapshots at (c) show common 3D organization pattern for datasets enriched for the different histone marks in the pluripotent cells, while interactions in hTERT-RPE1 are unique for both histone marks (e). 1D chromatin tracks are shown below the contact maps. Examples of bivalent regions are highlighted. e On-diagonal pileup plots of TSS classified as in (b). Aggregated contact maps are shown for H3K27me3 (upper panel) and H3K4me3 (bottom panel) Micro-C-ChIP. TSS overlapping with both H3K27me3 and H3K4me3 ChIP-seq peaks have been further stratified by the METTL14 signal.

The architecture of bivalent chromatin remains a topic of debate³⁸. Initially, bivalent promoters in Drosophila melanogaster were found to be associated with paused RNAPII, leading to the idea that bivalent marks and paused RNAPII position promoters to respond to developmental cues readily^39,40. However, careful GRO- and ChIP-seq experiments show that bivalent promoters at developmental genes are depleted of RNAPII⁴¹. Co-ChIP and Re-ChIP technologies allow for identifying the co-occurrence of H3K4me3 and H3K27me3^42–44. Recent Re-ChIP experiments report around 8000 genes enriched for H3K4me3 and H3K27me3 at the same nucleosomes⁴⁴. In addition, it was recently reported that bivalent regions in mESC are enriched in the RNA-binding protein METTL14, which regulates PRC2 and KDM5B localization to bivalent domains⁴⁵. Indeed, visual inspection of mESC datasets confirms the enrichment of METTL14 at H3K4me3 and H3K27me3 shared regions (Fig. 6c). Furthermore, in contrast to pluripotent mouse cells, hTERT-RPE1 Micro-C-ChIP displays distinct patterns for histone marks and the absence of bivalent chromatin regions (Fig. 6d).

To investigate the architecture of bivalent promoters, we intersected TSS with H3K27me3 and H3K4me3 peaks and grouped them into three classes: H3K27me3 only, H3K27me and H3K4me3, and H3K4me3 only (Fig. 6e). We observed that the stripe signal for H3K27me3 Micro-C-ChIP is substantially stronger at promoters that are also marked by H3K27me3 and H3K4me3 compared to promoters that are only marked by H3K27me3 (Fig. 6e, Supplementary Fig. 4b). In contrast, the H3K4me3 Micro-C-ChIP signal is stronger for promoters that are only marked by H3K4me3 compared to promoters with both marks (Fig. 6e), presumably because these sites are associated with ongoing transcription.

To identify bivalent promoters more stringently, we sorted promoters with overlapping H3K4me3 and H3K27me3 peaks by the METTL14 signal (Fig. 6e). It is noticeable that the interaction signal of H3K27me3 Micro-C-ChIP scales with the METTL14 signal, but the H3K4me3 signal remains similar. Visual inspection of mESC datasets complements the results of the 2D pileup analysis from Fig. 6e (Fig. 6c). It shows that bivalent TSS regions are characterized by common stripes for both H3K4me3 and H3K27me3 Micro-C-ChIP, overlapping ChIP-seq peak regions, enrichment of METTL14 and absence of RNAPII and ongoing transcription. Importantly, hTERT-RPE1 Micro-C-ChIP displays unique features for both histone marks and, therefore, no bivalency pattern is observed either via visual inspection or TSS-centered 2D pileup analysis (Fig. 6d, e, Supplementary Fig. 4c).

Micro-C-ChIP reveals cell-type-specific genome folding. The epigenetic landscape varies between different cell types, resulting in distinct chromatin organization patterns. Enriching for H3K4me3 and H3K27me3 histone marks showed the spatial organization of active and inactive chromatin. Remarkably, in pluripotent cells, inactive promoters engage frequently in interactions with distal genomic elements, similar to active promoters. These are, in particular, bivalent promoters marked by H3K4me3 and H3K27me3 and emerging enrichment stripes indicating a larger interaction span. In summary, Micro-C-ChIP is sensitive to cell-type-specific 3D genome organization, demonstrated by the different structural features associated with H3K27me3.

Ultrastructural chromatin features revealed by Micro-C-ChIP

Micro-C-ChIP captures ultrastructural chromatin features at CTCF sites comparable with the Micro-C at a fraction of reads. We computed off-diagonal pileups for CTCF-CTCF interaction pairs with convergent CTCF-motif orientations for mESC H3K4me3 Micro-C-ChIP and Micro-C data (Fig. 7a). Despite 10x lower sequencing depth in Micro-C-ChIP compared to Micro-C, composite CTCF-loops are resolved at nucleosome resolution (10 bp/pixel). H3K4me3-Micro-ChIP enriches for interactions adjacent to CTCF sites, whereas Micro-C detects interactions between all positioned nucleosomes flanking CTCF sites. Notably, CTCF interactions with tandem and divergent CTCF motif orientations display drastically reduced interaction frequencies to convergent sites as previously shown¹⁵ (Supplementary Fig. 5), arguing that the detected interactions reflect true interactions in 3D space and not artificial ChIP-enrichment products. Strikingly, Micro-C-ChIP detects an interaction feature that is scarcely detected in Micro-C. Two almost continuous stripes flank each CTCF site closely and span across the 2 kb window from top to bottom and left to right. We hypothesize that these are CTCF-H3K4me3 footprints, where CTCF protects a 50–80 bp fragment in addition to the 147-bp nucleosomal footprint⁴⁶. The size selection step in Micro-C by gel electrophoresis depletes these fragments.

Fig. 7. Micro-C-ChIP captures ultrastructural features of chromatin organization with nucleosome resolution.

a Off-diagonal pileup of convergent CTCF-CTCF loops (at 10–100 kb apart) for the Micro-C¹ and H3K4me3 Micro-C-ChIP dataset at 10 bp resolution. b Off-diagonal pileup of CTCF-CTCF loops with convergent CTCF motifs grouped by read orientation (F – forward, R – reverse). Schematic visualization demonstrates ligation events between interacting nucleosomes (green and purple circles) flanking the CTCF loop (yellow circles). Red dashed line represents the ligation event. Black arrows depict sequencing orientation. Blue arrow points towards enriched nucleosome interactions.

Read orientations in Micro-C-ChIP unravel a distinct loop-extrusion signature at CTCF sites. Next, we leveraged the high read density at CTCF sites in our Micro-C-ChIP data. We stratified the interactions by read orientations (Fig. 7b). Mapped DNA read orientations are particular for nucleosomes directly flanking the CTCF sites and originate preferentially from CTCF-protected fragments. Notably, the sequencing read originates from the opposite of the proximity-ligated ends. In all ligation scenarios, mapped reads of two interacting nucleosomes point towards the CTCF loop, indicating that ligation events occur between the DNA ends that extend outside of the CTCF-nucleosome footprint. This suggests that proximity ligation between interacting nucleosomes is not random and is presumably influenced by CTCF and cohesin. The strongest enriched peaks stem from CTCF adjacent sites, which in size-selected Micro-C, could include CTCF-CTCF only interaction pairs. However, Micro-C-ChIP data, due to the enrichment of histone PTM-specific antibody (H3K4me3 in this case), suggest the involvement of a nucleosome. Here, only CTCF-H3K4me3 protected molecules and not +1 or −1 nucleosomes (n + 1 and n − 1, respectively) flanking the CTCF site (also marked by H3K4me3) show a stripe emerging from CTCF sites, suggesting that loop extrusion emerges from one point close to CTCF and presumably only one cohesin per CTCF site, which is in agreement with recent high-resolution imaging analysis⁴⁷.

Discussion

This manuscript presents a high-resolution histone PTM-specific enrichment strategy to measure 3-dimensional mammalian genome organization. Compared to bulk Micro-C, Micro-C-ChIP detects more short-range interactions, indicating an improved signal-to-noise ratio. Furthermore, Micro-C-ChIP yields a significantly higher fraction of informative 3D contacts than other Micro-C-based ChIP-enrichment strategies. At the level of individual genes, Micro-C-ChIP detects interactions that were previously only accessible through capture strategies, such as RCMC²² or Tiled-MCC⁴⁸, or resource-intensive sequencing experiments. In contrast to RCMC and Tiled-MCC, Micro-C-ChIP is not limited to individual genomic regions but profiles 3D chromatin interactions across the entire genome and, similarly to bulk Micro-C^1,21, can capture ultrastructural architectural features of chromatin (Fig. 7a). There are several features worth noting in the context of the interactions between regions of subnucleosomal size aligned at the CTCF sites. Due to MNase use and histone mark enrichment used in this study, we cannot exclude the nucleosomal nature of those elements. Previous studies have suggested the presence of “fragile” nucleosomes at CTCF sites and TSS in mESC^47,49–51. The binding of CTCF and pioneer factors has been reported to increase susceptibility to MNase^49,52,53; therefore, this could explain the conversion of mildly protected nucleosomal fragments into subnucleosomal fragments. This supports the ability of Micro-C-ChIP to detect interactions at a superior resolution. Overall, due to its inherent use of MNase as the enzyme to fragment genomes, Micro-C-ChIP is ideal for unraveling histone mark-specific 3D genome interactions.

Despite the range of advantages introduced by Micro-C-ChIP, there are some limitations in all similar enrichment-based chromatin conformation technologies. Conventional ICE normalization, which accounts for regions represented differently in Hi-C and Micro-C, cannot be applied to Micro-C-ChIP due to the non-uniform coverage biases inherent to enrichment-based methods. Therefore, we used input-based normalization, namely, scaling factors computed for bulk Micro-C. Unlike ICE³¹, which assumes uniform coverage, input normalization preserves enriched interactions while adjusting for local sequencing biases using a reference dataset, which is bulk Micro-C. Importantly, Micro-C-ChIP signal depends on the abundance of histone marks that should be considered when interpreting changes in 3D genome organization.

To further control the inherent biases of enrichment-based strategies, which only capture interactions within predefined histone-marked regions rather than genome-wide, we applied color-coding of relevant sites corresponding to the peak locations to provide a precise reference for interpretation.

To distinguish whether Micro-C-ChIP reflects true structural features or artifacts, we performed Micro-C-ChIP in RAD21-depleted cells, confirming the detection of true 3D genome features. Here, RAD21 depletion led to the loss of most H3K4me3-enriched interactions, reducing interactions to distances independent of RAD21. Despite extensive evidence supporting the role of chromatin architectural proteins in E-P/P-P interactions^14,16,54,55, recent high-resolution studies report minimal alterations in the E-P/P-P connectome following their depletion^2,22,48. Our data indicate that enrichment stripes are significantly affected, particularly those spanning longer distances. This effect is not simply a consequence of generic interaction decay but instead reflects an active disruption of long-range interactions mediated by cohesin (Supplementary Fig. 2c). This finding aligns with recent evidence highlighting the role of cohesin in maintaining long-range chromatin interactions^56–60.

In contrast, cohesin-independent H3K4me3-marked interactions persist (Fig. 2e, f), consistent with recent observations that cis-regulatory interactions (i.e., E-P interactions) remain primarily intact following acute RAD21 depletion^2,22. Notably, the immunoprecipitation process can be biased towards interactions where both anchors carry the target PTM (i.e., P-P contacts). We mitigated this bias through an in vitro dinucleosome pulldown experiment, demonstrating that single-modified sites are captured with the same efficiency as dually modified sites. This data shows that enriched proximity ligation samples detect bona fide 3D genome interactions and display no or little immunoprecipitation biases.

We consider enrichment stripes as distances at which an enriched region interacts with the rest of the genome, which allows us to quantify the locus-specific interaction range. In our datasets, distinguishing the architectural stripes—caused by a one-sided halt or pause in loop extrusion—from enrichment stripes marking the edges of interaction domains is challenging. Regardless of their precise origin, these patterns can be leveraged to assess enriched interactions quantitatively (Figs. 2g, h, 6a, b).

Here, we show that transcriptionally inactive, bivalently marked promoters display long enrichment stripes. It is demonstrated that the strength of stripe features correlates with transcription rates and RNAPII occupancy at promoters^1,61. We demonstrate that the H3K27me3-marked promoters depleted for RNAPII display stripe formation in mESC. This argues that loop-extrusion occurs at TSS independent of transcription. In differentiated hTERT-RPE1 cells, H3K27me3 stripes are not detectable, even at silenced H3K27me3-enriched promoters. This confirms that promoter-associated stripes are not an artifact of Micro-C-ChIP and that transcriptionally inactive promoter-associated stripes are a feature of mESC. It is intriguing to speculate that bivalent marking preserves a more active-like promoter architecture while maintaining a repressed state.

Micro-C-ChIP reveals bivalent chromatin architecture in mESC. The function of bivalent promoter marking in ES cells remains a debate in the field²⁵. Currently, the idea is that bivalent promoters are in a state that allows them to rapidly switch to an active or inactive state depending on environmental cues during differentiation^42,62,63. Our data supports this idea. Previous work shows that bivalent promoters adopt an open chromatin configuration and are highly connected^37,62,64. For more stringent identification of the bivalent promoters, we included METTL14 enrichment at those sites⁴⁵. Here, we show that bivalent promoters display an overall similar architecture to active genes when enriched for H3K4me3 and H3K27me3, including loop-extrusion pattern, i.e., stripes and loops. These coincide with METTL14 binding and are independent of RNAPII binding. Furthermore, we identify similar patterns at bivalent promoters in H3K4me3 and H3K27me3 Micro-C-ChIP experiments. Micro-C measures the interactions between two nucleosomes in 3D space. In the case of TSS stripes at bivalent promoters, promoter-proximal nucleosomes, marked by either H3K4me3 or H3K27me3, interact with most nucleosomes within a given region across the cell population. Therefore, our data shows that H3K27me3-marked nucleosomes engage in open promoter architecture. This supports the model of truly bivalently marked nucleosomes carrying both marks, supported by re-ChIP experiments, instead of a composite of two distinct heterogeneous cell populations^42–44.

Taken together, our study provides an experimental strategy to map histone mark-specific 3D genome organization with a high signal-to-noise ratio. Micro-C-ChIP identifies extensive promoter-promoter interaction networks with moderate sequencing effort compared to bulk and other Micro-C-based immunoprecipitation strategies^1,21,30. We demonstrated that Micro-C-ChIP allows the dissection of histone mark-specific promoter architecture, e.g., at bivalent genes. Micro-C-Chip can be readily adapted for other histone marks. Importantly, the moderate sequencing requirement of Micro-C-ChIP will open opportunities to investigate the dynamics of 3D genome organization, for example, in depletion studies of essential factors of the loop extrusion machinery.

Methods

Experimental procedure

Cell culture and dual crosslinking

For a detailed bench protocol, please see the Supplementary Protocol. Mouse embryonic stem cells (mESC, E14JU cell line with a 129/Ola background, male)⁶⁵ were cultured on gelatin-coated dishes (0.2%) in serum + LIF conditions at 37 °C with 5% CO2. The E14 medium (DMEM-GlutaMAX-pyruvate (Gibco #31966-021) supplemented with FBS (15%; Sigma-Aldrich #F0392), LIF (homemade), 1× nonessential amino acids (Gibco #11140-050), 1× penicillin/streptomycin (Gibco #15140-122) and 2-mercaptoethanol (0.1 µM; Gibco #31350010), sterile filtered) was changed daily and cells were passaged every second day at a 1:8-1:10 ratio.

Primary retinal epithelial hTERT-RPE1 cells (ATCC, #CRL-4000) were grown in DMEM-GlutaMAX-pyruvate containing 10% FBS and 1× penicillin/streptomycin at 37 °C under 5% CO2 in 75 cm² flasks, splitted at 70%–80% confluency.

Human HCT-116-RAD21-mAID-mClover cells³² were cultured in McCoy’s 5 A medium (Gibco #16600082) supplemented with 10% FBS, 100 μg/ml Hygromycin (Gibco #10687010) and 100 μg/ml Geneticin (Gibco #10131027) at 37 °C with 5% CO2. To induce degradation of the AID-tagged RAD21, cells were treated with 500 μM indole-3-acetic acid (IAA) (Sigma Aldrich #I3750-5G-A) for 6 h.

The data presented in this study were generated in two biological replicates per condition. For H3K4me3 Micro-C-ChIP in mESC, and for both H3K4me3 and H3K27me3 Micro-C-ChIP in hTERT-RPE1, two technical replicates were generated for each biological replicate. For H3K27me3 Micro-C-ChIP in mESC, the first replicate had two technical replicates, while the second biological replicate consisted of three technical replicates. H3K4me3 Micro-C-ChIP in HCT-116 RAD21-mAC was performed in two biological replicates, each with a single technical replicate. Each biological replicate corresponds to an independent sample collection involving culturing, crosslinking, aliquoting, and snap-freezing of 100–300 million cells. For downstream Micro-C-ChIP, snap-frozen aliquots of 5 million cells were used. Each technical replicate (one preparative Micro-C-ChIP) was prepared from 5–10 million cells, corresponding to one or two aliquots, depending on chromatin yield.

To perform crosslinking, cells were detached using Trypsin-EDTA (Gibco #25200-056), washed once with 1x DPBS w/o Mg²⁺ and Ca²⁺ (ThermoFisher #14190144), and resuspended in 1x DPBS (1 mio cells/mL) for crosslinking with 1% Formaldehyde for 10 min with rotation at RT. The reaction was quenched by adding 2.5 M glycine to a final concentration of 0.25 M. For the second crosslinking, cells were washed with 1x DPBS and crosslinked with 3 mM EGS (ThermoFisher #21565) in 1x DPBS at 4 mio cells/mL for 40 min with rotation at RT. Crosslinking was quenched with glycine of 0.4 M final concentration for 10 min at RT. Cells were washed with 1x DPBS and flash-frozen in aliquots.

Preparative MNase digestion

5–10 mio frozen cells were used for one preparative Micro-C-ChIP. Cells were resuspended in 1x DPBS with 1x BSA (NEB #B9000S) added prior to resuspension to reduce the stickiness of the cells to the tub walls. Upon 10 min incubation on ice, cells were centrifugated for 5 min at 10,000 x g, washed with 500 μl MB#1 buffer (10 mM Tris-HCl, pH 7.5, 50 mM NaCl, 5 mM MgCl₂, 1 mM CaCl₂, 0.2% IGEPAL CA-630 (Sigma-Aldrich #I8896), 1x protease inhibitor (Halt Protease Inhibitor Cocktail, EDTA-free (100X); ThermoFisher #78425)), collected by centrifugation (10,000 x g, 5 min). The derived nuclei were resuspended in 1000 μl MB#1 and splitted into five 200 μl aliquots, which are equivalent of 1 mio cells. Chromatin was digested with MNase (5 U for 1 mio cells) for 10 min at 37 °C. MNase (Wortington Biochem #LS004798) concentration was selected for each batch based on the prior titration to yield 70-90% mono-nucleosomes. The digestion was stopped by the addition of 0.5 M EGTA (Bioworld #405200081) to a 4 mM final concentration and incubation at 65 °C for 10 min.

DNA end processing and ligation

After MNase digestion, samples were pooled to a 5 mio cell-equivalent for further processing. If more than 5 mio cells are to be used, these samples can be processed in parallel. 5% of the sample was used as an MNase digestion input control.

The chromatin was centrifugated for 5 min at 10,000 x g, washed with 500 μl 1x T4 DNA Ligase Buffer (NEB #B0202A) and collected by centrifugation (10000x g, 5 min). The pellet was resuspended in 90 µL of freshly prepared Micro-C “Master Mix 1” (10 μl 10x T4 DNA Ligase Buffer, 75 μl ddH2O, 5 μl T4 PNK (NEB #M0201L)). After incubation for 15 min at 37 °C with shaking, 10 μl Large Klenow Fragment (NEB #M0210L) was added, and the chromatin was incubated for 15 min at 37 °C with shaking. 100 µL of freshly prepared Micro-C “Master Mix 2” (10 μl 1 mM Biotin-dATP (Jenna Bioscience #NU-835-Bio14-S), 10 μl 1 mM Biotin-dCTP (Jenna Bioscience #NU-809-BioX-S), 1 μl 10 mM mix of dGTP and dTTP (NEB #N0442S, #N0443S), 5 μl 10x T4 DNA Ligase Buffer, 0.25 μl BSA, 23.75 μl ddH2O) was added. After incubation for 45 min at 25 °C with shaking, the enzymatic reaction was quenched by adding 0.5 M EDTA to a final concentration of 30 mM and incubation for 20 min at 65 °C with shaking. The chromatin was centrifugated (10,000 x g, 5 min), washed in 500 μl 1x T4 DNA Ligase Buffer, and collected by centrifugation (10,000 x g, 5 min). The chromatin pellet was resuspended in 500 μL of ligation reaction buffer (50 μl 10x T4 DNA Ligase Buffer, 2.5 µL BSA, 25 µL 400 U/µL T4 DNA Ligase, 422.5 µL ddH2O) and incubated for 2.5 h at RT with slow rotation. After proximity ligation, the chromatin was collected by centrifugation (12000x g, 5 min), resuspended in 200 μl 1x NEBuffer 1 (NEB #B7001S) and 200 U NEB Exonuclease III (NEB #M0206S), and incubated for 15 min at 37 °C to remove biotin from unligated ends.

Solubilization and immunoprecipitation

After ExoIII digestion, chromatin was resuspended in the sonication buffer (0.1% IGEPAL CA-630, 50 mM NaCl, 10 mM Tris–HCl, pH 7.5, 5 mM MgCl₂, 2 mM EDTA, 0.5% SDS (Invitrogen #15553-035)), 1x protease inhibitor) and sonicated for 10 cycles (sonication cycle: 30 s ON, 30 s OFF) using Bioruptor Plus (Diagenode). The samples were centrifugated at 14,000 x g for 10 min at 4 °C and the supernatant containing the soluble chromatin was transferred to a fresh low-binding tube. At this stage, it is possible to pool 5 mio cell-equivalent aliquots, processed in parallel, to increase the immunoprecipitation yield. To control the efficient solubilization and to quantify DNA yield, 5–10% supernatant input was taken and visualized by 1.5% agarose gel electrophoresis. The supernatant was diluted 1:4 with ChIP dilution buffer (1.1% Triton X-100, 16.7 mM Tris-HCl, pH 7.5, 1.2 mM EDTA, 167 mM NaCl, 1x protease inhibitor). Anti-H3K27me3 (Cell Signaling Technology #9733 s) and anti-H3K4me3 (Cell Signaling Technology #9751S) antibodies were added at a concentration of 10 μl of antibody per 10 μg of chromatin (typically 10–20 μl of antibody for 5-10 mio cells). Immunoprecipitation was performed overnight at 4 °C with rotation. Magnetic beads (Pierce ProteinA Magnetic Beads; ThermoFisher #88846) were added and incubated for 2 h at 4 °C. The beads were washed in the following order: twice with low-salt wash buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS, 1x protease inhibitor), twice with high-salt wash buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS, 1x protease inhibitor), once with LiCl wash buffer (10 mM Tris–Cl, pH 8.0, 250 mM LiCl, 1% IGEPAL, 1% Sodium deoxycholate, 1 mM EDTA, 1x protease inhibitor) and twice with 1X TE buffer (10 mM Tris-HCl, pH 8, 1 mM EDTA). After the last wash, antibody-protein-DNA complexes were eluted from the beads and reverse-crosslinked by adding 250 μl Elution buffer (20 mM Tris-HCl, pH 8.0, 10 mM EDTA, 5 mM EGTA, 300 mM NaCl, 1% SDS, Proteinase K, ddH2O) and incubated for at least 2 h at 65 °C. DNA was purified via Phenol:Chloroform: Isoamylalcohol (Invitrogen #15593-031) extraction followed by ethanol precipitation and resuspended in 50 µL 0.1x TE buffer (1 mM Tris-HCl, pH 7.5, 0.1 mM EDTA).

Streptavidin pulldown and on-bead library preparation

10 μl Dynabeads™ MyOne™ Streptavidin C1 beads (Invitrogen #65001) were washed twice with 500 μl 1x TBW (5 mM Tris-HCl, pH 7.5, 0.5 mM EDTA, 1 M NaCl) and resuspended in 150 μl 2x TBW (10 mM Tris-HCl, pH 7.5, 1 mM EDTA, 2 M NaCl). The resuspended beads and 100 µL 0.1x TE were added to the 50 µL sample and incubated with rotation at RT for 20 min. The beads were washed twice with 500 μl 1x TBW, once with 300 μl 0.1x TE buffer, resuspended in 50 μl 0.1x TE buffer and transferred to the PCR tubes. Sequencing libraries were prepared with NEBNext® Ultra II DNA Library Prep Kit for Illumina® (NEB #E7645). Note: because the sample remains attached to beads, all purification and size selection steps were omitted. Instead, the beads were washed twice with 300 μl 1x TBW and once with 300 μl 0.1x TE and resuspended in 20 μl 0.1x TE. PCR master mix (64 µL of H2O, 100 µL of Q5 high-fidelity DNA polymerase (NEB #M0544S), 8 µL of i5 primer, 8 µL of i7 primer (NEBNext Multiplex Oligos for Illumina (Dual Index primers); NEB #E7600S) was added to the sample. The reaction mixture was split into 100 µL aliquots. PCR amplification was performed for 14–16 cycles according to the NEBNext Ultra II DNA library prep kit for Illumina. After the PCR reaction, DNA was purified with SPRI size-selection beads (Beckman Colter #B23319) at a ratio of 1:0.9 according to the manufacturer’s protocol and eluted in 20 µL of 0.1x TE.

Sequencing

The samples were sequenced on an Illumina HiSeq 2000. We used Illumina 50 bp paired-end sequencing to obtain ~200–300 million total reads for each replicate in this study (Supplementary Fig. 1b).

In vitro chromatin reconstitution and pulldown

H3K27me3- and unmodified histone octamers from The Histone Source (Colorado State University) were mixed in different molar ratios to reconstitute dinucleosomes with varying levels of methylation. Six biotinylated DNA fragments were generated by PCR amplification using a forward primer with a biotinylated dTTP incorporated. The template was the pUC19 plasmid encoding 2x Widom601 sequences with six unique base pairs at both ends of each Widom sequence. These plasmids were cloned in-house (Supplementary Data 2). The PCR products were purified using the QIAquick PCR cleanup kit (QIAGEN #28506).

The DNA fragments were chromatinized using a standard salt dialysis method. Briefly, DNA and histones were mixed in a 1:1.3 molar ratio of 601-sequences: histones. The template plasmid was used as “carrier” DNA to prevent overassembly. DNA and histones were mixed in a high-salt buffer (10 mM Tris-HCl, pH 7.6, 1 mM EDTA, 2 M NaCl, 0.05% IPEGAL, 20 µg BSA) and transferred to a Slide-A-Lyzer MINI Dialysis Device, 3.5 K (Thermo Scientific #69550). NaCl was titrated out at RT O/N with a low-salt buffer (10 mM Tris-HCl pH 7.6, 4 mM EDTA, 50 mM NaCl, 0.05% IPEGAL and 4 mM DTT). The quality of assembly was assessed by digestion with BsiWI-HF (NEB #R3553L), which cuts in the Widom601 sequence. DNA from the restriction assay was purified with Phenol: Chloroform and analyzed on a 1% Agarose gel.

Before proceeding with the pulldown, we first evaluated the actual methylation levels quantitatively by Western blot to predict pulldown efficiencies for different methylation degrees (Fig. 2c). A total of 0.6 μg protein per sample was resolved on an SDS-polyacrylamide gel (Invitrogen #NP0335BOX). Proteins were transferred onto a nitrocellulose membrane (Merck #GE10600000), which was afterwards blocked in 5% milk in PBS containing 0.05% Tween-20 (PBS-T) for 1 h before being incubated with primary antibodies overnight at 4 °C. Membranes were cut horizontally at appropriate molecular weights prior to antibody incubation. The antibodies used were the following: H3K27me3 (1:1000; Cell Signaling Technology #9733 s) and H3 antibody (1:5000; Abcam #ab176842). The next day, membranes were washed three times with PBST and incubated with secondary antibodies for 1 h at room temperature. The secondary antibodies used were the following: LI-COR Odyssey IRDye 680RD (1:10,000) or LI-COR Odyssey IRDye 800CW (1:10,000). The region surrounding the expected band size was scanned using the LI-COR Odyssey IR Imager (Biosciences), and the Western blot was quantified using ImageJ.

In total, we used two replicates per one pool of three reconstitutions: the first pool contained samples with 100%, 40%, and 20% methylation, while the second pool included 100%, 25%, and 10% methylation. Each pool was split into two parts and subjected to immunoprecipitation with either H3K27me3 (Cell Signaling Technology #9733 s) or H3 antibody (Abcam #ab176842) overnight at 4 °C with rotation. Washing series, DNA extraction, and library preparation were performed the same way as described for Micro-C-ChIP.

Data analysis

Mapping

Micro-C-ChIP paired-end reads were processed with the distiller pipeline (https://github.com/mirnylab/distiller-nf). Briefly, reads were mapped to the reference genome (mm10 and hg38 as mouse and human reference assemblies, respectively) via bwa mem with flags –SP (v0.7.17). Alignments were parsed, and.pairs files were generated using the pairtools package⁶⁶ (v0.3.0). PCR/optical duplicates were filtered with pairtools dedup function with “-max-mismatch 1”. Valid pairs with high mapping quality scores on both sides (MAPQ > 30) were aggregated into binned matrices of Micro-C-ChIP interactions using the cooler⁶⁷ (v0.9.2) package into multiresolution cooler (mcool) files (100 bp, 200 bp, 500 bp, 1 kb, 2.5 kb, 5 kb, 10 kb, 25 kb, 50 kb, 100 kb, 500 kb, 1 Mb). The reproducibility of Micro-C-ChIP replicates was evaluated by HiCRep⁶⁸ (v0.2.6) for contact maps at 10 kb resolution (Supplementary Fig. 1a).

Visualization and normalization

Contact maps shown in figures were generated with cooltools⁶⁹. Genomic tracks (ChIP-seq, GRO-seq, nascent RNA-seq) were generated with CoolBox⁷⁰ (v0.3.9). The multi-cooler files used for contact map visualization were generated with cooler cload pairs and cooler zoomify from the valid pairs files. The valid pairs of reads were shifted by 73 bp with respect to the read orientation. This was done to account for the approximate locations of the nucleosome dyad axes. Additionally, the valid pairs separated by less than 500 bp were disregarded to remove reads from undigested dinucleosomal contaminants.

Contact matrices were balanced using scaling factors (weight) taken from input (conventional Micro-C normalized with ICE) multi-cooler files. The input-normalized matrices were color-coded based on the sites of the corresponding 1D ChIP-seq peaks. Loci corresponding to H3K27me3- and H3K4me3-enriched sites are marked in purple and dark blue, respectively.

Analysis of publicly available ChIP-seq datasets

All the ChIP-seq datasets that were downloaded as raw fastq files (see Supplementary Data 1) were reanalyzed. Bowtie2⁷¹ (v2.4.2) was used for mapping the fastq files to the hg38 or mm10 reference genome. The signal tracks were produced via deeptools bamCoverage⁷² (v3.5.1) with the parameters “-binSize 20 -normalizeUsing RPKM -smoothLength 60 -extendReads 150 -centerReads”.

1D analysis

We used the TSS locations of the mm10 and hg38 refTSS annotations⁷³ with a ± 500 bp window for defining promoter regions. Enhancers were defined based on the overlap of publicly available ChIP-seq datasets for H3K4me1 (ENCFF440FYE for mESC, GSE176035 for hTERT-RPE1) and H3K27ac (ENCFF194TQD for mESC, GSE113399 for hTERT-RPE1) excluding regions that overlap with promoters. Putative enhancer-promoter (E-P) pairs separated by <100 kb distance were created with a custom R script and putative promoter-promoter (P-P) pairs (80-100 kb distance) were generated with the bioframe⁷⁴ (v0.4.1) pair-by-distance function (Fig. 1i).

CTCF motifs were obtained from JASPAR (JASPAR motif ID #MA0139.1) and overlapped with the CTCF ChIP-seq peaks (GSE90994 for mESC, GSE176035 for hTERT-RPE1, GSE104888 for HCT116-RAD21-mAC) using bioframe⁷⁴ (v0.4.1). CTCF ChIP-seq peaks that are associated with the strongest motif underlying that peak were then overlapped with the SMC1A peaks (GSE123636 for mESC, GSE146766 for hTERT-RPE1) to obtain cohesin-bound CTCF sites.

We stratified TSS by RNAPII occupancy, calculating the mean RNAPII ChIP-seq signal (GSE52071 for mESC, GSE201982 for hTERT-RPE1) in a ± 500-bp window around all TSS locations and divided them into quantiles (Fig. 5c, d). TSS locations were also stratified based on the overlap with H3K27me3 and/or H3K4me3 peaks (Supplementary Data 1) using bedtools⁷⁵ (v2.30.0) intersect (Fig. 6b, e). To stringently identify bivalent TSS, we further stratified the TSS where H3K27me3 and H3K4me3 peaks overlap by METTl14 occupancy and divided into three quantiles where the top METTl14 quantile corresponds to bivalent promoters (Fig. 6e).

Average signal plots for Micro-C-ChIP datasets were generated with deeptools⁷²(v3.5.1). The ChIP-seq peaks for a corresponding histone mark were used as alignment points. The signal around the alignment points was averaged within a ± 30 kb window. The Micro-C-ChIP signal tracks used for the signal tracks and average signal visualization were generated with bamCoverage using CPM normalization and 20 bp bin size (Fig. 4a, Supplementary Fig. 1d). H3K4me3 ChIP-seq and ATAC-seq signal (Supplementary Data 1) in UT and IAA-treated HCT116-RAD21-mAC was averaged within a ± 10 kb window at the peak coordinates used for off-diagonal pileup plotting (Fig. 2e). For H3K27me3 and H3K4me3 ChIP-seq signal (GSE154379 and ENCFF806NDV, respectively) in mESC, metaplots over TSS, which were stratified based on the overlap with H3K27me3 and/or H3K4me3 peaks, were generated using a ± 5 kb window (Supplementary Fig. 4a).

Loop detection and anchor analysis

The loops in H3K4me3 Micro-C-ChIP datasets from mESC and hTERT-RPE1 were predicted with Peakachu⁷⁶ (v2.2) using a pretrained model (RNAPII ChIA-PET dataset in WTC11 cells) offered in the pipeline. The loops were called at the 2.5- and 5-kb resolution unbalanced contact maps (Fig. 3a).

The loops were visualized using Coolbox⁷⁰. To classify loops, we intersected loop anchors with the promoter, enhancer, or cohesin-bound CTCF regions (Fig. 3b). CTCF-classified anchors do not include regions that overlap promoter or enhancer regions. Anchors which do not overlap any of the three features were defined as “Other”. The promoter-overlapping anchors were further classified based on the overlap with the cohesin-bound CTCF regions.

Anchor valency was used as a proxy of loop engagement of one anchor (Fig. 3e). For this, the genomic coordinates of the left and the right anchors for each loop were concatenated resulting in 1D anchor list. We then used bedtools merge with the parameters “-c 1 -o count” computing the number of merged anchors in a given genomic locus.

The number of interactions per anchor and the lengths of the loops were visualized in R (v 4.1.2) using ggplot2 (v3.4.2) (Fig. 3c, e).

Stripe calling

The stripes were identified with Stripenn³³ (v1.1.65.18) at 2-, 2.5- and 5-kb resolution contact maps (Figs. 2g, h, 6a). The called stripes were classified as 5’- and 3’-stripes having anchors at the 5′- or 3’-end of the stripe domains, respectively.

Pileup analysis

Contact maps used for the pileup plots were unbalanced throughout the manuscript. To account for distance decay in the pileups, all pileups demonstrate aggregated observed-over-expected contact maps where expected interactions for chromosome arms were calculated beforehand. The colorbars to the side from the pileups show the averaged observed-over-expected signal in logarithmic scale.

The off-diagonal pileup analysis was done to calculate Micro-C-ChIP contact aggregation at paired ChIP-seq peaks obtained from the publicly available datasets (Figs. 2e, 4a) using cooltools⁶⁹ (v0.5.1). First, the strongest ChIP-seq peaks were intersected using bioframe⁷⁴ (v0.4.1) to create paired sites separated by a distance between 10 and 300 kb. The paired ChIP-seq peaks were piled up on the center of 500 bp resolution maps around ±20–30 kb window. A 500-bp resolution was used for the pileup plots. Pileup heatmaps computed with cooltools were normalized using percentile-based min-max scaling within each dataset, ensuring comparability across matrices while preserving relative enrichment patterns.

The off-diagonal pileup analysis of P-P loops that were identified in H3K4me3 Micro-C-ChIP and further stratified by RNAPII signal (Figs. 3d, 5a, b) was performed with Coolpup⁷⁷ (v1.1.0) leveraging the advantage of subsetting the snippets. 100-kb flank was used for the pileup analysis of P-P loops for 500-bp resolution Micro-C-ChIP matrices (Fig. 3d) and 20-kb flank for the pileup analysis of P-P loops with anchors stratified by RNAPII level for 100-bp resolution matrices (Fig. 5a, b).

The putative E-P/P-P pairs were centered and piled up at 200-bp resolution matrices around ±10 kb window (Fig. 1i).

The off-diagonal pileups of CTCF-CTCF interactions (10-100 kb apart) (Fig. 7a, b, Supplementary Fig. 5) were computed at high resolution with single bin corresponding to 10 bp.

Instead of aggregating at the intersection of ChIP-seq peaks or loop anchors, we applied the same approach to analyze the genome-wide target-centered contact intensity (Figs. 5c, d, 6a, e, Supplementary Fig. 4b). All on-diagonal (local) pileup maps were computed with Coolpup (v1.1.0). TSS-centered pileups were plotted around a ±200 kb window where the TSS coordinates were stratified based on the RNAPII mean signal (Fig. 5c, d), the overlap with the H3K27me3 and/or H3K4me3 ChIP-seq peaks and METTL14 mean signal (Fig. 6e).

The rescaled local pileup analysis was applied to evaluate average interaction signal across the identified 5’-and 3’-stripes (Fig. 6a) using “-rescale” parameter with rescale flank equal 1 rescaling the pileups to the same shape and size.

In vitro chromatin pulldown data analysis

Read 1 of the paired reads of both H3K27me3 and H3 was mapped to the reference sequences of barcoded 2xWidom601 DNA fragments via bowtie2⁷¹. As different methylation degrees corresponded to a distinct barcode sequence, we could compare the mapping ratios of different methylation degrees. The read counts mapped to each barcode sequence were divided by the total read counts in the sample and normalized by H3 read counts to account for equal representation in the pool. The read counts corresponding to 100% methylation degree reconstitution (carried only methylated histone octamers) were set to 1 as a reference for complete pulldown efficiency. The mapping rates for assemblies with varying methylation degrees were then scaled proportionally. Finally, the normalized mapping rates (relative pulldown efficiency) were compared to the expected values calculated for different methylation degrees as described before.

Virtual 4 C analysis

To generate 4C-like contact profiles, we analyzed chromatin interactions using.pairs files. We defined viewpoints based on ChIP-seq peak coordinates, extracting interaction frequencies within a specified genomic window. Interaction data was filtered to retain contacts spanning at least 400 bp and falling within the viewpoint region. For each viewpoint, we quantified interaction frequencies at each genomic position. Data was binned into larger intervals (bin size = 10). Finally, the interaction profiles were normalized per viewpoint by setting the sum of interactions (area under the curve) to 1 and smoothed using a rolling mean before visualization.

Modeling contact maps based on interaction decay and 1D chromatin signal

For this analysis, the data from.pairs files for UT and IAA samples was used. The genomic region of interest (chr12: 55,774,428 − 55,994,269) was partitioned into 2000 equally sized bins. To model the effect of 1D chromatin signal, the frequencies of occurrence of each bin, which reflect the 1D signal value, were recorded to create a frequency vector for each sample. Row-wise and column-wise interaction frequency matrices were computed by repeating the frequency vector along rows and columns, respectively. These row and column matrices were then combined: in the additive model, the matrix for each sample was created by summing the row and column matrices elementwise; in the multiplicative model, the matrix was generated by multiplying the row and column matrices elementwise. For the expected interaction matrix, a distance-based modeling approach was applied. The expected interactions for each bin pair were computed, filling in the diagonal values of the matrix. This model accounts for the natural decay in interaction frequency with the distance between interacting regions. All matrices were normalized by dividing each element by the total sum of all elements in that matrix. Finally, the 1D chromatin signal matrices (additive and multiplicative) were multiplied by the expected matrix.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.M. and N.K. conceived the project and designed experiments. M.M. performed and analyzed all cell experiments, with input from N.K. M.L.V. performed biochemical chromatin reconstitution experiments and supported M.M. with in vitro Micro-C-ChIP. M.M. and N.K. wrote the manuscript, with input from M.L.V.

Peer review

Peer review information

Nature Communications thanks Hang He and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Sequencing data produced in this study have been deposited in NCBI GEO with the accession code GSE246346. A list of publicly available datasets used in this study is provided in Supplementary Data 1. Source data are provided with this paper.

Code availability

The custom scripts used in this study are available at 10.5281/zenodo.15084044.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-64350-w.