High-resolution CTCF footprinting reveals impact of chromatin state on cohesin extrusion

Abstract

Cohesin-mediated DNA loop extrusion enables gene regulation by distal enhancers through the establishment of chromosome structure and long-range enhancer-promoter interactions. The best characterized cohesin-related structures, such as topologically associating domains (TADs) anchored at convergent CTCF binding sites, represent static conformations. Consequently, loop extrusion dynamics remain poorly understood. To better characterize static and dynamically extruding chromatin loop structures, we use MNase-based 3D genome assays to simultaneously determine CTCF and cohesin localization as well as the 3D contacts they mediate. Here we present CTCF Analyzer (with) Multinomial Estimation (CAMEL), a tool that identifies CTCF footprints at near base-pair resolution in CTCF MNase HiChiP. We also use Region Capture Micro-C to identify a CTCF-adjacent footprint that is attributed to cohesin occupancy. We leverage this substantial advance in resolution to determine that the fully extruded (CTCF-CTCF loop) state is rare genome-wide with locus-specific variation from ~1-10%. We further investigate the impact of chromatin state on loop extrusion dynamics and find that active regulatory elements impede cohesin extrusion. These findings support a model of topological regulation whereby the transient, partially extruded state facilitates enhancer-promoter contacts that can regulate transcription.

Fig. 1. MNase CTCF HiChIP data contains short (~<80 bp) CTCF-protected fragments and longer (~>120 bp) nucleosome-protected fragments.

A Schematic illustrating relationship between short fragments and observed ligations. B Schematic illustrating how the fragment length results from MNase cutting around bound proteins of different sizes. Created in BioRender. Aryee, M. (2025) https://BioRender.com/g98u240C Fragment length distribution for all fragments (top plot) and fragments overlapping occupied CTCF motifs (lower plot). Occupied CTCF motifs are defined here as CTCF motifs within 30 bp of a CTCF ChIP-seq peak summit. D Bar plot quantifying the frequency of different fragment lengths genome-wide and how often each fragment length group overlaps an occupied CTCF motif. Occupied CTCF motifs are defined here as CTCF motifs within 30 bp of a CTCF ChIP-seq peak summit. E Fragment coverage metaplot +/− 500 bp around CTCF binding sites. Schematic below the coverage metaplot illustrates the likely proteins producing these peaks. Schematic created in BioRender. Aryee, M. (2025) https://BioRender.com/g98u240 (F) Plot (E) stratified by fragment length.

Fig. 2. True CTCF binding sites have a bimodal strand-specific distribution centered on the CTCF motif.

A Unfiltered reads +/− 1250 bp around a CTCF binding site located on the negative strand (chr1: 30,779,763 − 30,779,781). The midpoint of the CTCF motif is marked with the symbol “ <”, representing that it is on the negative strand, and a pink line. B Plot (A) filtered to observed ligations (equivalently, short fragments.) C Schematic demonstrating the bimodal read pile-up around a CTCF binding site. Created in BioRender. Aryee, M. (2025) https://BioRender.com/g98u240 (D) Plot (B) as a density plot and zoomed in on the CTCF motif, with quadrant annotations. E Distributions of reads in quadrants for true negative and true positive CTCF binding sites in DNA loop anchors. True positives are defined as CTCF motifs that are the only CTCF motif in a loop anchor and within 30 bp of a CTCF ChIP-seq peak. True negatives are areas of the loop anchors with one CTCF motif that are at least 200 bp from the CTCF motif. 3,984,150 fragments across 4523 loop anchors each containing one CTCF binding site are used to make these boxplots. Boxplots are made with ggplot2::geom_boxplot() and show the 25% quantile (lower bound of box), median (center line), and 75% quantile (upper bound of box). Boxplot whiskers subtract (lower whisker) or add (upper whisker) 1.5 * IQR from the lower bound (lower whisker) or upper bound (upper whisker) of the box. Schematics of the quadrant read pile-up patterns are shown next to the corresponding true positive and true negative boxplots. F CAMEL statistic $\left(\widehat{\alpha }=\frac{\min \left(n_2,n_4\right)}{\max \left(n_1,n_3\right)}\right)$ for plot (D) peaks at the CTCF motif.

Fig. 3. CTCF binding sites identified by CAMEL with single basepair resolution in MNase K562 CTCF HiChIP data.

A Heatmap of log2(min/max) as a function of distance between CAMEL peak center and CTCF motif center within loop anchors. Only CTCF motifs that are unique within a loop anchor and within 30 bp of a CTCF ChIP-seq peak are used. B Precision recall curve for true negative and true positive CTCF binding sites in DNA loop anchors. True positives are defined as in (A). True negatives are areas of the loop anchors in (A) that are at least 200 bp from the one CTCF motif. Precision is calculated as TP / (TP + FP), recall is calculated as TP / (TP + FN). C CAMEL statistic $\log 2\left(\widehat{\alpha }\right)=\log 2\left(\frac{\min \left(n_2,n_4\right)}{\max \left(n_1,n_3\right)}\right)$ density plots using the same set of true positives and true negatives as (B). D Histogram with 1 bp bin size depicting CAMEL resolution for all peaks genome-wide (not just in loop anchors). E Motif occurrence in ChIP-seq and CAMEL peak centers genome-wide. Motif occurrence is calculated as % peak centers within 20 bp of CTCF motif. Only peak centers within 150 bp of a CTCF motif are used for this figure. F 30 bp sequences centered on genome-wide CAMEL peak centers produce a de novo motif (top) that matches the core JASPAR CTCF motif (bottom). De novo motif is made using STREME.

Fig. 4. Cohesin footprint observed in MNase HiChIP and experimentally confirmed in mESC RCMC.

A High and low CTCF occupancy motifs. For each occupancy level, CTCF ChIP-seq and all fragments overlapping the CTCF motif are depicted, along with the corresponding fragment length histogram. B The fraction of reads overlapping a CBS (+) that are short (<120 bp) strongly correlates with the strength of its CTCF ChIP-seq signal. The fraction of short reads (<120 bp) was calculated for each CBS, plotted in a density, and colored by absence of CTCF ChIP-seq peak or CTCF ChIP-seq signalValue quartile. C K562 CTCF MNase HiChIP 2D density for short (<120 bp) left fragments overlapping CAMEL-identified CBS (+) (N = 10,906 CBS). The fragment start and fragment end are aligned relative to the start of the 35 bp CTCF motif (MA1930.1), so that “0” in the plots corresponds to the start of the 35 bp CTCF motif. D Annotated 35 bp JASPAR CTCF motif (MA1930.1). E mESC RCMC 2D density for short (<120 bp) left fragments overlapping CTCF motifs within 30 bp of a mESC CTCF ChIP-seq peak (N = 65 CBS). The fragment start and fragment end are aligned relative to the start of the 35 bp CTCF motif (MA1930.1), so that “0” in the plots corresponds to the start of the 35 bp CTCF motif. F Schematic illustrating the most prevalent fragment type observed in the mESC RCMC perturbations shown in (E). G Schematic showing the three classified types of fragments relative to the location of the CTCF motif. The interaction partners of these three fragment types are graphed in (H, I). H Short TF-sized fragments (<120 bp) with an extended fragment end have a noticeably larger bump in density of long range interactions (>10 kb) compared to TF-sized short fragments without an extended fragment end and nucleosome-sized long fragments (>120 bp). I P(S) curve for fragments depicted in (H).

Fig. 5. Cohesin extrudes further through quiescent regions than active regions.

A Most CTCF-mediated looping contacts do not reflect the fully extruded state. Estimate is obtained using left fragments that overlap CAMEL identified CBS (+) and have an interaction length greater than 10 kb with start and end at least 5 bp from motif start and end, length <120, and extended fragment end. For each CBS with at least 50 long-range TF-protected fragments overlapping the motif, % convergent is calculated as the number of interaction partners overlapping CTCF (-) motifs / total number of fragments at motif. Because this estimate is conditional on CTCF binding at the anchor, we divide estimates by two to account for the ~50% occupancy of CTCF. B Depiction of how 1 MB regions downstream of CBS were annotated using ChromHMM. Density (C) and P(S) curves (D) for chromatin state clusters shown in (Supplementary Fig. 6A), filtered to the top 20%. Chromatin annotations making up each cluster are added together and quantiles are obtained to determine fragments in the top 20% of active chromatin, quiescent chromatin, and bivalent / polycomb chromatin. Left fragments that overlap CAMEL identified CBS (+) and have an interaction length greater than 10 kb with start and end at least 5 bp from motif start and end, length <120, and extended fragment end are used. E Ridge plots for the bottom 10% quantile (“Low”) and top 10% quantile (“High”) of H3K27ac bp and number of RNAPII binding sites. ChIP-seq from ENCODE was used to annotate 1 MB downstream of left fragments overlapping CBS (+) for this figure. Left fragments that overlap CAMEL identified CBS (+) and have an interaction length greater than 10 kb with start and end at least 5 bp from motif start and end, length <120, and extended fragment end are used. Plots are labeled with the average log10 interaction length. F Diagram illustrating differences in extrusion rates between active and quiescent chromatin states, with numbers obtained from Supplementary Fig. 6B.

Fig. 6. Schematic of supported model whereby single promoter-proximal CTCF sites enable an enrichment of enhancer-promoter contacts.

Left: contact map representation. Right: physical loop representation of left panel.