Cohesin prevents cross-domain gene coactivation

Abstract

The contrast between the disruption of genome topology after cohesin loss and the lack of downstream gene expression changes instigates intense debates regarding the structure-function relationship between genome and gene regulation. Here, by analyzing transcriptome and chromatin accessibility at the single-cell level, we discover that, instead of dictating population-wide gene expression levels, cohesin supplies a general function to neutralize stochastic coexpression tendencies of cis-linked genes in single cells. Notably, cohesin loss induces widespread gene coactivation and chromatin co-opening tens of million bases apart in cis. Spatial genome and protein imaging reveals that cohesin prevents gene co-bursting along the chromosome and blocks spatial mixing of transcriptional hubs. Single-molecule imaging shows that cohesin confines the exploration of diverse enhancer and core promoter binding transcriptional regulators. Together, these results support that cohesin arranges nuclear topology to control gene coexpression in single cells.

Fig. 1. Cohesin loss induces cross-domain gene coactivation.

a, Schematic diagram showing that the correlation of two genomic features A and B at the single-cell level may vary drastically even when their average levels remain the same in the cell population. b, The workflow for Smart-SCRB sequencing of control and RAD21-depleted mouse ES cells; Ctrl, control; RT, Reverse Transcription. c, Acute cohesin depletion results in only subtle changes in gene expression at the population level. Each circle represents a detected gene. Average gene expression was quantified by calculating the logarithmic value of averaged sequencing counts of ~400 individual cells. The red line marks the diagonal y = x. d, ACDs were identified based on ATAC peak densities by Gaussian peak fitting and thresholding for each chromosome. In total, 776 ACDs were identified across the genome. Red triangles indicate ATAC peaks. e, Statistics of ΔC_RNA(i,j) per ACD pair after cohesin depletion in chromosomes 1–19 and X. Each circle indicates the value of the differential coexpression coefficient for one ACD pair. The red line indicates the median value, and the dotted line indicates the zero-change line. Only active ACDs (those ACDs with detectable gene expression by Smart-SCRB) were included for analysis. The statistics were derived from 76,943 intrachromosomal ACD pairs with quantifiable values over 20 different chromosomes. f, Heat maps show elevated differential coexpression coefficients per active ACD pair in Chr 2 after cohesin depletion. Source data

Fig. 2. Cohesin loss increases cross-domain chromatin co-opening.

a, A schematic workflow for scATAC-seq (10x Genomics) of control and RAD21-depleted mouse ES cells. b, Acute cohesin loss results in no significant alteration of normalized average counts of ATAC peaks per ACD at the cell population level. Each circle represents one ACD, and empty ACDs without ATAC peaks were omitted from analysis. Normalized average counts of ATAC peaks per ACD were calculated by dividing the accumulated ATAC peak count per ACD by M_Average (the average ATAC peak counts across all ACDs and cells). The red line marks the diagonal y = x. c, Acute cohesin loss increases chromatin coaccessibility per ACD pair globally. Coaccessibility Spearman correlation coefficients per ACD pair (S_ATAC(i, j)) under control (black) and cohesin-depleted (red) conditions were plotted as a function of the genomic distance after a five-point smoothing. Only the data from ACD pairs within the same chromosome were used to generate the plot. Data are presented as mean values ± s.e., and shadow regions indicate s.e.m. d, Heat map showing that acute cohesin loss selectively increases intrachromosome chromatin coaccessibility per ACD pair across 20 chromosomes in the mouse genome. After filtering out cells with low read counts and batch normalization, around 3,000 cells were analyzed for each condition. e, The heat map shows elevated differential coaccessibility per ACD pair within each chromosome. Differential coaccessibility Spearman correlation coefficients per ACD pair were calculated by subtracting the coaccessibility Spearman correlation coefficient value under control conditions (d, top left) from that under cohesin-depleted conditions (d, top right). f, Heat map showing elevated differential coaccessibility per ACD pair in Chr 2. g, Dot plots of differential coaccessibility Spearman correlation coefficients (ΔS_ATAC(i,j)) before and after cohesin depletion for chromosomes 1–19 and X. Every circle indicates the value of the differential coaccessibility Spearman correlation coefficient per ACD pair. The red line indicates the median value, and the dotted line indicates the zero-change line. The statistics were derived from 15,308 intrachromosomal ACD pairs with quantifiable values over 20 different chromosomes. Source data

Fig. 3. Cohesin prevents gene co-bursting along the chromosome.

a, Active genes (208) across Chr 2 in mouse ES cells are targeted by 30 intron probes per gene. Primary probes for each gene contain four unique docking sites, and each docking site can be hybridized to a readout probe corresponding to nine pseudocolors in each readout round. The identity of each bursting gene can be determined according to the barcode after four readout (RO) rounds. The barcode is designed that the gene can be decoded even with one round of drop out. b, One representative cell with decoded genes (color coded). Composite FISH signals (green) were reconstructed by averaging images from all FISH channels from three hybridizations in RO1. The 3D image was rendered by using VVD-viewer; scale bar, 2 μm. c, Heat map of pairwise differential co-bursting frequencies (top) and differential distances (bottom) for 208 genes in Chr 2. The matrix was calculated by subtracting values in the control condition (3 h) from those in the cohesin loss condition. Zoom-in views of the heat map and representative images for three differentially coactivated genes (red, green and blue color coded) after cohesin loss are presented below; scale bar, 2 μm. d, Bursting frequencies for 208 genes (represented by circles) under control and cohesin loss (3 h) conditions. The bursting frequency per gene is calculated by dividing the total number of detected bursting sites (approximately zero to four in a cell) for the gene by the number of cells (N) analyzed under each condition (see Eq. (5)). The red line marks the diagonal y = x. e,f, Box plot of pairwise differential co-bursting frequencies (e) and differential distances (f) for the indicated conditions. A non-parametric two-sided Wilcoxon test was used for statistical testing. The bottom and top whiskers represent 10% and 90% values, respectively. The box represents the range from the 25th percentile to the 75th percentile, the center line represents the median, and the dotted line indicates the zero-change line. The statistics were derived from 21,528 gene pairs with quantifiable values. Source data

Fig. 4. Cohesin removal leads to spatial rearrangement of Chr 2 intron puncta and coactivation of lineage-specific genes.

a, Transcription bursting sites of 208 active genes from Chr 2 are collectively imaged by using intron primary probes (30 probes per gene) and a global imager probe conjugated to Alexa Fluor 647N. b, Representative 3D isosurface images show transcriptional bursting sites from 208 active genes across Chr 2 before (top) and after (bottom) cohesin loss. Nuclei were counterstained with DAPI (blue); scale bar, 2 μm. c, Isolated representative 3D isosurfaces for typical intron puncta formed by transcription bursting sites of 208 genes across Chr 2 before and after cohesin loss; scale bar, 1 μm. d, Violin plots show the normalized radii of detected Chr 2 intron puncta before and after acute cohesin depletion. Black lines are the median values, and dotted lines are the 25% and 75% quantiles. The measurement was repeated three times, and a non-parametric two-sided Wilcoxon test was used for statistical testing. e, Cohesin loss decreases the physical distance between coactivated lineage-specific genes Gbx2 (ectoderm) and Gpc1 (mesendoderm). Top, genomic positions of Gbx2 and Gpc1 in Chr 1 with Hi-C and ATAC density information. Bottom, representative 3D isosurface images of Gbx2 and Gpc1 bursting sites before and after cohesin loss; scale bar, 2 μm; inset scale bar, 200 nm. f, Cohesin loss does not cause uniform up- or downregulation of bursting fractions for individual genes. g, Cohesin loss elevates normalized co-bursting frequencies of four of five pairs of lineage-specific genes. For f and g, data are presented as mean values ± s.d. The measurement was repeated three times independently, and two-sided Student’s t-tests were used for statistical testing. h, Cohesin loss decreases average physical distances between coactivated lineage-specific genes in cis (five pairs). The number of data points (n) used for statistical analysis for each gene pair is marked at the bottom of the corresponding bar plot. A non-parametric two-sided Wilcoxon test was used for statistical testing. Data are presented as mean values ± s.d. i, Schematic showing chromatin reorganization (top) and gene coactivation (bottom) after cohesin depletion. Source data

Fig. 5. Cohesin separates transcriptional hubs.

a, Diagram showing biallelic integration of HaloTag and green fluorescent protein (GFP)–mini-AID (mAID) into endogenous Med6 and Rad21 gene loci (for auxin-induced protein degradation). b, Three-dimensional isosurface reconstruction of MED6 hubs (color coded by 3D volumes). Inlets are MED6 fluorescence images; scale bars, 2 μm. c,d, Violin plots show the size distribution of MED6 (c) and MED1 (d) hubs. Black lines are median values, and dotted lines are 25% and 75% quantiles. A non-parametric two-sided Wilcoxon test was used for statistical testing. e, MED6 hub sizes are inversely correlated with residual RAD21 protein levels in single cells. An F-test indicates that the slope is significantly non-zero with P < 0.0001; AU, arbitrary units. f, Timelapse imaging of MED6 hubs during cohesin depletion; scale bar, 2 μm. g, Quantification of MED6 hub sizes (red) and RAD21 residual levels (gray) in f. For MED6 hub size analysis, the statistics were derived from the top 50 hubs, and data are presented as mean values ± s.e. h, Three-dimensional isosurface rendering of overlaps between MED6 hubs (color coded by 3D volumes) and transcription bursting sites for 208 genes in Chr 2 (magenta); scale bars, 2 μm. i, Representative 3D volume overlaps between MED6 hubs and Chr 2 intron clusters; scale bar, 1 μm. j, Violin plots show the statistics of 3D volume overlaps between MED6 hubs and transcription bursting sites for 208 genes in Chr 2. Black lines are the median values, and dotted lines are 25% and 75% quantiles. The measurements were obtained from 20 cells, and a non-parametric two-sided Wilcoxon test was used for statistical testing. k, Three-dimensional isosurface reconstruction of MED6 hubs and intron-FISH signals showing a representative case that co-bursting gene loci are connected by MED6 hubs after cohesin loss; scale bar, 500 nm; smRNA-FISH, single-molecule RNA-FISH. l, Bar plots for the fraction of co-bursting loci connected by MED6 hubs for paired genes within (shadowed) and across ACDs. The fraction was calculated by dividing the number of co-bursting loci that share a common MED6 hub by the total number of co-bursting loci. The measurement was repeated three times, and two-sided Student’s t-tests were used for statistical testing. Data are presented as mean values ± s.d. Source data

Fig. 6. A role of cohesin in regulating gene coexpression.

a, Transcription factors search for targets in a local interaction hub by a bound-state-dominant mode. Local exploration is proposed to be guided by transient interactions with other co-regulators (proteins or non-coding RNAs) in the hub; TF, transcription factor. b, The diagram illustrates the quantification of the RoC and the calculation of its differential cumulative probability ΔC (ΔC = ΔC_Ctrl – ΔC_RAD21–) over analyzed tracks. For this analysis, both loosely and tightly bound fractions were obtained by using a slower frame rate (20 Hz). c, Differential cumulative probability (ΔC) of the RoC for histone subunits (H2B and H2A.Z) and a broad range of transcriptional regulators (MED1, MED6, BRD4, OCT4 and TBP). Histone subunits H2B and H2A.Z were analyzed as controls. The inset shows curves illustrating the cumulative probability of the RoC for H2B before (black) and after (red) cohesin loss. d, Cohesin loss leads to disruption of chromatin loops and spatial mixing of ACDs and Mediator hubs. As a result, the average distance between distant active genes (red and gray dots) in cis decreases, accompanied by their colocalization into a shared transcriptional hub and an elevated chance for their co-regulation. e, Schematic diagrams illustrating that an increase in cofluctuation of genes A (red) and B (gray; top) would alter gene coexpression correlation in single cells (bottom) without affecting their average levels (blue dots) at the cell population level. Gray dotted lines reflect sampling points of single-cell genomics for the correlation analysis in the bottom plots. Source data

Extended Data Fig. 1. Single-cell transcriptome analysis upon cohesin loss.

a. Box and dot plots of the number of detected genes per cell under control and cohesin-depletion conditions for two biological replicates by using Smart-SCRB technology. The total number of genes detected across the population is annotated vertically. Numbers at the bottom of the chart represent the numbers of single cells analyzed for each condition. For all box charts, upper and lower whiskers represent outlier cut-offs based on the 1.5 interquartile range rule; the box represents the range from 25% to 75% percentile; the center line represents the median. b. The pie plot shows the degrees of global gene expression changes from pooled Smart-SCRB data. c. The size distribution of 776 ACDs across the mouse genome. d. ACDs overlap with active compartments, enhancers and promoters, but are more likely to exclude with inactive compartments, intergenic regions, repressed chromatin, heterochromatin and LADs. The ACD enrichment for each specific element calculated by counting the ratio of ACDs overlapped with that element throughout the genome. Our identified ACDs were aligned with LADs, compartments and ChromHMM-identified genomic regions of mESCs downloaded from previously publications^–. Source data

Extended Data Fig. 2. Quantification of cross-domain gene co-expression.

a. The workflow for calculating differential Spearman correlation matrix (ΔS_RNA) per gene pair (from two ACDs) before and after cohesin depletion from single-cell RNA-seq count matrix (M_RNA). The colormaps included are pure cartoon representations. b. The workflow for calculating differential gene co-expression matrix (ΔC_RNA) per ACD pair before and after cohesin depletion from ΔS in (a).

Extended Data Fig. 3. Cell cycle and Hi-C correlation analysis of cross-domain gene co-expression.

a. Acute cohesin removal increases the average Spearman correlation coefficient (${\overline{S}}_{\mathrm{RNA}}$) per ACD pair globally. ${\overline{S}}_{\mathrm{RNA}}$ under control and cohesin-depletion conditions was plotted as a function of the genomic distance between ACD pair after a five-point smoothing. Only the data from ACD pairs within the same chromosome were used to generate the plot. b. Box plots show the pooled statistics of differential co-expression coefficient (ΔC_RNA(i,j)) per ACD pair throughout the whole genome after cohesin depletion. Differential co-expression coefficient calculated from two repeats with control ES cells was included as a control. The upper and lower whiskers represent maximum and minimum values; the box represents the range from 25% to 75% percentile; the center line represents the median; the dotted line indicates the zero-change line. c. Chromosome-wise differential co-expression coefficient calculated by binning differential co-expression coefficient (ΔC_RNA(i,j)) in cis or in trans. d. Comparison of chromosome-wise differential co-expression coefficients in cis and in trans across all chromosomes. e. Computational assignment of single cells into cell cycle phases by analyzing Smart-SCRB data with cyclone() (R-programmed cell-cycle phase classifier). f. Dot plots show the distributions of differential gene co-expression coefficients before and after cohesin depletion (red dots) for cells in G1, S and G2M cell cycle phases, respectively. The calculated coefficients between two independent control groups (black dots) were used as the control. Red lines indicate the median values, and the blue dotted line indicates the zero-change line. For b and f, two-sided non-parametric Wilcoxon test was used for statistical testing. g. Heatmaps show increased normalized differential Hi-C contact frequencies in Chr 2 after cohesin loss. h. Normalized contact frequency per ACD pair (from Hi-C data) as a function of genomic distance between that ACD pair. i. Average Spearman correlation coefficient (${\overline{S}}_{\mathrm{RNA}}$) per ACD pair as a function of normalized Hi-C contact frequency per ACD pair. The plot was generated by five-point smoothing. For a, h and j, data are presented as mean values ± S.E. and shadow regions indicate S.E. of the mean. Source data

Extended Data Fig. 4. Quantification of cross-domain chromatin co-accessibility and seqFISH validations.

a. The workflow for calculating differential Spearman correlation matrix (ΔS_ATAC) per ACD pair before and after cohesin depletion from single-cell ATAC-seq count outputs. b. Box plots of co-accessibility Spearman correlation coefficients (S_ATAC(i,j)) per ACD pair in cis (within the same chromosome) or trans (from different chromosomes). In the box charts, lower and upper whiskers represent 5%-95% values; the box represents the range from 25% to 75% percentile; the center line represents the median. Dotted line indicates the zero-change line. Non-parametric two-sided Wilcoxon test was used for statistical testing. 15,481 pairs in cis and 279,815 pairs in trans were used for statistical analysis. c. Co-accessibility Spearman correlation coefficient (S_ATAC(i,j)) per ACD pair as a function of normalized Hi-C contact frequency per ACD pair. The plot was generated by five-point smoothing. Data are presented as mean values ± S.E. and shadow regions indicate S.E. of the mean. The slope derived from linear regression of the curve for each condition was labelled below. d. Validating bursting frequencies measured by seqFISH using published nascent RNA-seq data. 199 of 208 genes in Chr2 (probed by seqFISH experiment) with detected nascent RNA-seq counts were used for scatter plot. Each circle represents one gene and the red line is the linear regression line with R² = 0.578. e. Distances between spots of co-bursting gene pairs (10 randomly selected pairs from the pool of 208 active genes in Chr 2) in cis. Each circle represents the distance between one pair of co-bursting spots within a single cell. The dotted line indicates 1 µm cut-off. As shown from the results, there is a specific enrichment of co-bursting spots on the same chromosome with distance < 1 µm. f. Distances between co-bursting gene pairs in trans. The results only showed the uniform distribution within the range between 0 µm and 20 µm, without the concentrated fraction < 1 µm. For e and f, the number of data points (n) used for statistical analysis for each gene pair is marked on top of the corresponding dot plot. Source data

Extended Data Fig. 5. seqFISH workflow to spatially resolve actively transcribed genes in Chr 2.

a. Hybridization scheme: 4 readout rounds plus a repeat of the first readout round were performed. 3 hybridization per round with 9 distinct readout probes (P) and 9 pseudocolor (C) per round. In each hybridization, 4 sequential 3D acquisitions (640 nm, 561 nm, 488 nm, 405 nm) were performed to image intron seqFISH signals (AF488, Cy5 and AF647N) and nuclei (DAPI) with blue beads (405 nm). b. seqFISH data analysis and gene decoding pipeline: 1) 3D single-molecule localizations for 4 color channels were performed with FISHQUANT^, for all color channels; 2) xyz drift correction based on localizations for beads in the 405 nm channel; 3) Chromatic correction based on multicolor Tetraspeck beads coated on the coverslip surface; 4) Pooling localizations and gene decoding according to predesigned barcodes; 5) 3D nucleus segmentation (right: random colored masks) based on DAPI stains (left) with a pretrained specialist model using Cellpose 2⁷; 6) Parsing genes into single cells based gene localizations and 3D masks for nuclei. The images were rendered by using VVD-viewer.

Extended Data Fig. 6. Intron-FISH imaging of lineage-specific gene pairs in cis.

a. Information of lineage, genomic location and expression level for five pairs of developmental genes. The expression levels in both conditions were computed by averaging counts from Smart-SCRB measurements. b. Genomic positions of four pairs of representative lineage-specific genes with Hi-C and ATAC density information. The genomic distance between the two genes within each pair is labelled. c. Representative 3D iso-surface images of transcription bursting sites (intron-RNA-FISH) of gene pairs showed in (a) before and after cohesin loss. Scale bar, 2 μm. Inlet scale bar, 1 μm. d. The alteration of co-bursting frequencies between co-bursting gene pairs in trans before and after cohesin depletion. Data are presented as mean values ± standard deviation (S.D.). The measurement was repeated three times and two-sided Student’s t-test was used for statistical testing. n.s., not significant. e. The alteration of distances between co-bursting gene pairs in trans before and after cohesin depletion. Each dot represents one pair of co-bursting genes. Data are presented as mean values ± S.D. The number of data points (n) used for statistical analysis for each gene pair is marked at the bottom of the corresponding bar plot and non-parametric two-sided Wilcoxon test was used for statistical testing. Source data

Extended Data Fig. 7. Tagging endogenous mediator subunits with HaloTag.

a-b. PONDR score charts indicate predicted ordered and disordered regions within MED1 (a) and MED6 (b). c-d. PCR genotyping results showing bi-allelic fusion of HaloTag to MED1 c; N-terminus) and MED6 (d; C-terminus). Genomic DNA from wild-type mouse ES cells was used as the control. e-f. Western blots showing HaloTag-MED1 (e) and MED6-HaloTag (f) protein levels before and after RAD21 depletion by auxin-induced degron system. α-tubulin protein was blotted and used as a loading control. g-h. Western blots show the efficacy of RAD21 degron system in parental cell lines and established MED1 (g) and MED6 (h) knockin cell lines before and after auxin treatment for 6 hours. The normalized RAD21 or RAD21-mAID-eGFP protein level to the loading α-tubulin level for each condition was quantified and shown below each lane. i-j. Fluorescence images showing RAD21-mAID-eGFP (Green) and HaloTag-MED1 (i; Red) or MED6-HaloTag (j; Red) levels without or with the auxin treatment (6 hrs). DNA was counter-stained with DAPI (Blue). Scale bar, 5 μm. For experiments from c to j, the measurement was repeated three times independently with similar results. k. Propidium iodide (PI) staining and flow cytometry analysis of DNA contents (CD4-FITC) from parental RAD21-mAID-eGFP cell line, and established MED1 and MED6 knockin cell lines before and after acute cohesin loss. Source data

Extended Data Fig. 8. Mediator hubs colocalize with ACDs.

a. Two-color 3D PALM imaging captures spatial distribution of both accessible chromatin sites (ATAC) (left) and MED1-HaloTag (middle) localizations. Color bars indicate localization densities. Scale bar, 2 μm. See Movie S3 for 3D rotatory rendering. In the lower panel, the cropped localization map indicates that ACDs colocalize with MED1 hubs. Color bars indicate localization densities. Scale bar, 500 nm. b. Quantification of colocalization of accessible chromatin localizations and MED1 (or MED6) localizations by pair cross-correlation function c(r). In the upper panel, schematic shows three different spatial relationship – exclusion, uncorrelated (random permutation) and colocalization – between two localization maps. Data are presented as mean values ± S.E. The experiment was repeated for three times and non-parametric two-sided Wilcoxon test was used for statistical testing. c. One 2D section of two-color 3D ATAC and MED1 PALM images (a) was used for spatial intensity correlation analysis in c. The original 3D localization maps were binned into 100 nm³ cubic to generate 3D image volumes for both channels and one slice was selected for colocalization analysis. Scale bar, 2 μm. One-dimensional intensity correlation analysis was performed for signals from two channels along selected line #1 and #2. The measurement was repeated three times independently with similar results. d. 3D reconstruction shows the overlap between Gpc1 intron-FISH iso-surface and MED6-HaloTag hub iso-surface and the separation between Gpc1 intron-FISH iso-surface and HP1-GFP iso-surface. Scale bar, 2 μm. Quantification of the physical distance between the centroid of Gpc1 intron-FISH signal and MED6 hub iso-surfaces and that between the centroid of Gpc1 intron-FISH intron-FISH singal and the nearest HP1 iso-surface. Non-parametric two-sided Wilcoxon test was used for statistical testing. The statistics were derived from 30 measured data points for each group. Data are presented as mean values ± S.D. in the bar plots. Source data

Extended Data Fig. 9. Single molecule tracking of transcription regulators and histone subunits with the HaloTag technology.

a. Western blots indicate HaloTag fusion protein levels for either stably expressed (OCT4, TBP1 and H2A.Z) or endogenously labelled (BRD4) transcriptional regulators before and after RAD21 depletion. α-tubulin was used as the loading control. The measurement was repeated three times independently with similar results. b. A diagram shows the labeling of HaloTag fusion proteins with JF549 dye. c. A schematic diagram illustrates the procedures for single-molecule imaging, localization, and tracking. Source data

Extended Data Fig. 10. Fast tracking reveals no apparent changes in Histone and TF dynamics after cohesin loss.

a. The diagram shows two-state and three-state fitting of single-particle tracks for histone subunits and transcriptional regulators, respectively. For this analysis, fast single molecule imaging (100 Hz) was used to capture the movement of both diffusive and bound molecules. b-c. Representative fittings of 2 state and 3 state model to jump histograms of histone proteins (b) and diverse transcriptional regulators (c) with variable Δt. Two-state model assumes that TFs alternates between one bound and one diffusive state, whereas three-state model assumes that TFs alternates between one bound, one slow and one fast diffusive state. d. Fast tracking reveals that cohesin loss does not significantly alter apparent bound fractions for histone subunits (H2B and H2A.Z) and a broad range of transcriptional regulators (MED1, MED6, BRD4, OCT4 and TBP). Data are presented as mean values ± S.D. The measurement was repeated for three times and two-sided Student’s t-test was used for statistical testing. n.s., not significant. e. Fast tracking reveals cohesin loss has little effect on apparent diffusion coefficients for histone subunits (H2B and H2A.Z) and a broad range of transcriptional regulators (MED1, MED6, BRD4, OCT4 and TBP) in the bound state. Data are presented as mean values ± S.D. The measurement was repeated for three times and two-sided Student’s t-test was used for statistical testing. f. Slow diffusive fractions for different transcriptional regulators before and after cohesin removal calculated by Spot-On. Data are presented as mean values ± S.D. The measurement was repeated for three times and two-sided Student’s t-test was used for statistical testing. g. Diffusion coefficients for slow diffusive fractions as indicated in (f). Data are presented as mean values ± S.D. The measurement was repeated for three times and two-sided Student’s t-test was used for statistical testing. Source data