DNMT1 inhibition reactivates CTCF peaks and loops on gene bodies

A. Schematic demonstrating the experimental workflow. K562 and HCT116 cells were treated with DNMT1i for 3 days followed by CTCF ChIP-seq and CTCF HiChIP.

B. Replicate-averaged input-normalized reads within merged peaks for CTCF ChIP-seq in DNMT1i (y-axis) vs. DMSO (x-axis) for K562 cells (left) and HCT116 cells (right). Reactivated peaks are highlighted in red.

C. Example Integrated Genome Browser (IGV) CTCF ChIP-seq tracks ± DNMT1i at the MEG3 locus for K562 (top) and HCT116 (bottom). Reactivated peaks indicated with arrows.

D. Example IGV CTCF ChIP-seq and HiChIP tracks ± DNMT1i at the Protocadherin Alpha (PCDHA) locus (HCT116). Reactivated peaks are indicated below. Red indicates a looping reactivated peak, whereas grey indicates a nonlooping reactivated peak.

E. Boxplot showing the log2(fold-change) in loop strength DNMT1i vs. DMSO (y-axis) for loops with or without at least one anchor overlapping a reactivated peak (x-axis) for K562 (left) and HCT116 (right).

F. Heatmaps depicting log2(mean observed/expected) Hi-C interaction frequency centered on reactivated peaks for DMSO (left) and DNMT1i (right). K562 Hi-C data from Siegenfeld et al., 2022.

G. Proportion of looping and nonlooping peaks with each annotation (y-axis) for both reactivated and all CTCF peaks in K562 in DNMT1i.

In (E), the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

Reactivated CTCF peaks interact with highly-looping partners near nuclear speckles

A. Number of different loops (loops to different partners) the maximally-looping partner peak makes (y-axis) for all vs. reactivated CTCF peaks (x-axis) in DNMT1i for K562 (left) and HCT116 (right).

B. Number of different loops (number of partner peaks) (y-axis) assigned to each peak by rank (x-axis) in DNMT1i for K562. Highly-looping peaks reside above the red dashed line.

C. IGV representation of highly-looping (yellow) and normal-looping CTCF peaks (grey) in K562 DNMT1i-treated cells. CTCF ChIP-seq and HiChIP are shown.

D. Aggregate profile (signal P-value, y-axis) of genomic features over highly-looping peaks (blue) and all other CTCF peaks (grey). CTCF peaks were called in K562 DNMT1i treatment, and histone modifications/ATAC-seq data are from public datasets in K562 (see Methods).

E. Heatmap depicting log2(mean observed/expected) Hi-C interaction frequency centered on highly-looping (left) and all other (right) peaks in the DMSO treatment condition in K562 cells. K562 Hi-C data from Siegenfeld et. al., 2022.

F. Example IGV tracks depicting SON Cut&Tag signal (top), SON TSA-seq normalized counts (middle), CTCF ChIP-seq (bottom), and CTCF HiChIP for a region on chromosome 11 for K562 DMSO.

G. Boxplot showing replicate-averaged DNMT1i SON Cut&Tag signal (RPKM, 20 kb bins, y-axis) in the respective cell type over reactivated peaks vs. all other CTCF peaks called in DNMT1i for K562 and HCT116 cells broken down by whether the CTCF peak is in a loop anchor (x-axis).

H. Same as G, but for highly-looping peaks vs. all other CTCF peaks.

I. Boxplot showing average log2(counts per million (CPM) loop strength) (y-axis) for CTCF HiChIP loops relative to SON Cut&Tag signal decile (x-axis). logCPM defined by Diffloop across all conditions (DMSO and DNMT1i). Loops are segregated into equally sized deciles by Cut&Tag signal in DMSO (RPKM, 20kb bins).

J. Heatmap depicting log2(mean observed/expected) Hi-C interaction frequency centered on CTCF peaks at speckles (denoted by high SON signal, left) and not at speckles (right) in DMSO. K562 Hi-C data from Siegenfeld et. al., 2022.

In (A,G,H,I), the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

DNMT1i-induced gene activation and speckle association depend on CTCF

A. Schematic showing the CTCF degradation experimental workflow. Following pre-treatment with HaloPROTAC3 for 8 hours to degrade HaloTag-CTCF, cells are co-treated with HaloPROTAC3 and DNMT1i for 24 hours. DMSO controls without CTCF depletion and/or DNMT1 inhibition were also included.

B. Western blot showing the depletion of CTCF in HaloTag-CTCF knock-in cell line in a HaloPROTAC3-dependent and DNMT1i-independent manner (330 nM HaloPROTAC3, 10uM DNMT1i)

C. Aggregate profile of replicate-averaged SON Cut&Tag signal (RPKM) over 1-day reactivated CTCF peaks in HaloTag-CTCF HCT116 cells with and without CTCF degradation through HaloPROTAC3.

D. Left: Heatmap showing z-score normalized r-log transformed counts for genes that are differentially expressed in DNMT1i vs. DMSO non-HaloPROTAC3 conditions (Padj <0.05, |log2(fold-change)| > 0.5) clustered by K-means. Number of genes per cluster are as follows: cluster 1: 33, cluster 2: 48, cluster 3: 6 cluster 4: 54. Right: Average log2(fold-change) in CTCF binding in DNMT1i vs. DMSO for all CTCF peaks on genes within the four gene clusters.

E. Boxplot showing normalized SON TSA-seq signal (20 kb bins, y-axis, 4D Nucleome 4DNFIBY8G6RZ) in wildtype HCT116 cells over genes within the different clusters identified in D.

For (E), the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

Acute disruption of nuclear speckles alters gene expression without disrupting CTCF

A. Schematic illustrating use of the dTAG system to acutely deplete SON and SRRM2, thus disrupting nuclear speckles.

B. Immunofluorescence for SON and SRRM2 for K562 speckle knock-in cells treated with dTAG-13 for 6 or 12 hours to deplete SON and SRRM2. Right: per-nucleus mean fluorescence following immunofluorescence for SON/SRRM2 double dTAG knock-in K562 cells treated with dTAG-13 for 6 or 12 hours or DMSO for 12 hours.

C. Replicate-averaged input normalized tags between CTCF ChIP-seq in 6-hour dTAG-13 treatment (x-axis) vs. DMSO (y-axis) for K562 speckle dTAG knock-in cells.

D. Boxplot showing log2(fold-change) in loop strength for CTCF HiChIP loops in speckle knock-in K562 cells treated with dTAG-13 vs. DMSO (y-axis) for 6 hours relative to SON Cut&Tag signal (x-axis). Loops are segregated into equally sized deciles by Cut&Tag signal with 10 representing the decile closest to speckles (RPKM, 20kb bins).

E. Model for methylation-mediated insulation of genes from regulatory elements near nuclear speckles

For boxplots, the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

DNMT1 inhibition reactivates CTCF peaks and loops on gene bodies

A. Dose-response curve in K562 (left) and HCT116 cells (right) of percent growth (y-axis) in varying doses of DNMT1i treatment (x-axis) relative to DMSO-treated control cells after 3 days of treatment. Data represent mean ± s.e.m.

B. Percent methylation in wildtype cells across CpG sites within 300 bp CTCF peak regions for reactivated peaks and all other CTCF peaks.

C. Percent methylation status of CTCF motifs underlying reactivated peaks clustered by K-means. Number of CTCF sites per cluster: K562 (cluster 1: 139, cluster 2: 351, cluster 3: 221, cluster 4: 738); HCT116 (cluster 1: 405, cluster 2: 1005, cluster 3: 746, cluster 4: 2249). Bisulfite data for HCT116 from GEO GSM3317488, K562 from ENCODE ENCFF459XNY

D. Proportion of looping and nonlooping peaks with each annotation (y-axis) for both reactivated and all CTCF peaks in HCT116 in DNMT1i.

E. log2(fold-change) ratio of the fractions of each annotation within looping vs. nonlooping reactivated peaks. Ratio of bars from Fig 1G, S1D.

F. Number of different loops each CTCF peak makes (number of interaction partners) (y-axis) for reactivated K562 (left) and reactivated HCT116 (right) peaks in DNMT1i by peak annotation (x-axis).

For (B,F), the interquartile range (IQR) is depicted by the violin with the median represented by the center dot. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

Reactivated CTCF peaks interact with highly-looping partners near nuclear speckles

A. Number of different partner peaks the maximally-looping partner interacts with (y-axis) for all K562 (left) and all HCT116 (right) peaks in DNMT1i by peak annotation (x-axis).

B. Number of different loops (partner peaks) (y-axis) assigned to each peak by rank (x-axis) in DNMT1i for HCT116. Highly-looping peaks reside above the red dashed line.

C. Boxplot showing the log2(fold-change) in loop strength DNMT1i vs. DMSO (y-axis) for loops connecting reactivated peaks to highly-looping partners vs all other loops (x-axis) for K562 and HCT116.

D. Histogram showing density (y-axis) of genomic distances (x-axis) of CTCF highly-looping peaks (top) and all other CTCF peaks (bottom) in DNMT1i-treated K562 cells from stripe anchors identified from DNMT1i-treated K562 Hi-C data. Median distance is shown with a solid black line. K562 LIMe-Hi-C data from Siegenfeld et. al., 2022.

E. Proportion of all, reactivated, and highly-looping CTCF K562 DNMT1i peaks in each published subcompartment (y-axis) assigned by SNIPER for wildtype K562 cells.

F. Boxplot showing replicate-averaged SON Cut&Tag signal (RPKM, 20 kb bins, y-axis) after DNMT1i treatment in the respective cell type over reactivated peaks vs. all other CTCF peaks called in DNMT1i treatment for K562 and HCT116 cells. Same as 2G, but not broken into looping categories.

G. Boxplots showing SON TSA-seq normalized counts (y-axis) for reactivated vs. non-reactivated CTCF DNMT1i peaks for K562 and HCT116 (left) broken down by whether the CTCF peak is in a loop anchor.

H. Boxplots showing SON TSA-seq normalized counts (y-axis) for highly-looping vs. normal-looping CTCF DNMT1i peaks for K562 and HCT116 (right). SON TSA-seq normalized counts were analyzed from public data (4DNFIVZSO9RI.bw, 4DNFIBY8G6RZ.bw) in the untreated, wildtype respective cell type (see methods).

I. Density plot showing the correlation between SON Cut&Tag signal (RPKM, 20 kb bins) over a given CTCF peak (y-axis) vs. the number of different loops (partners) assigned to that CTCF peak (x-axis) for K562 DNMT1i treatment.

For A, C, F, G, and H, the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

DNMT1i-induced gene activation and speckle association depend on CTCF

A. Western blot for CTCF levels over a dose/time course of HaloPROTAC3 in HaloTag-CTCF knock-in HCT116 cells. GAPDH is shown as a loading control. 330nM was chosen as the final dose.

B. Volcano plot showing -log10(adjusted P-value) (y-axis) vs. log2(fold-change) in gene expression (x-axis) upon 1 day DNMT1i treatment (left) or 32 hours HaloPROTAC3 treatment (right)in HaloTag-CTCF HCT116 cells. Upregulated genes are highlighted in red, and downregulated genes are highlighted in blue (p-adj < 0.05, |log2fold-change| > 0.5). <5 points per plot are omitted for visualization.

C. Aggregate profile plot of replicate-averaged RPKM normalized CTCF binding in DMSO over the transcription start sites of differential HaloPROTAC3 genes.

D. Boxplot showing expression levels of genes within clusters from Fig 3D in Fragments Per Kilobase per Million mapped fragments (FPKM). Expression in cells treated with vehicle (-HaloPROTAC3, -DNMT1i) is shown.

E. Heatmap of per-CTCF peak log2(fold-change CTCF binding) after 1 day of DNMT1i treatment vs. DMSO control centered over all CTCF peaks overlapping genes in the clusters from Fig 3D. Each gene could have multiple CTCF peaks on it.

F. Boxplot showing log2(foldchange SON Cut&Tag signal DNMT1i vs. DMSO, y axis) over gene bodies for genes in the different clusters from Fig 3D. Foldchange Cut&Tag signal with and without HaloPROTAC3 treatment are shown for each gene cluster. The same genes are shown for the + and – HaloPROTAC3 conditions for comparison.

For boxplots, the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.

Acute disruption of nuclear speckles alters gene expression without disrupting CTCF

A. Separate western blots for SON and SRRM2 in speckle dTAG knock-in K562 cells ± 6 hours of 500 nM dTAG-13 treatment. Vinculin was used as a loading control.

B. Volcano plot showing -log10(adjusted P-value) (y-axis) vs. log2(fold-change) in gene expression (x-axis) upon speckle degradation with 6 hours dTAG-13 treatment. Upregulated genes are highlighted in red, and downregulated genes are highlighted in blue (adjusted p-value <0.05, |log2(fold-change| > 0.25).

C. Proportion (y-axis) of all nonzero (left) and all 6 hour dTAG-13 downregulated (right, padj<0.05) genes in each SON TSA-seq decile (x-axis) from wildtype cells. 10 is the decile closest to speckles.

D. Density plot showing SON Cut&Tag signal in wt K562 cells treated with DMSO (RPKM 20kb bins, y-axis) vs. the change in gene expression (x-axis) for genes that decrease in expression upon dTAG-13 treatment (Spearman R, -0.43).

E. Aggregate profile plot of CTCF ChIP-seq signal (RPKM) in speckle knock-in K562 cells with and without 6-hour speckle degradation (y-axis) over all CTCF peaks (x-axis).

F. Boxplot showing log2(fold-change) loop strength for dTAG-13 treated vs. DMSO treated speckle dTAG knock-in cells relative to TSA-seq decile after 6 or 12 hours of dTAG-13 treatment with matched DMSO controls. Speckle refers to the decile closest to speckles, and non-speckle refers to the deciles 1-9 that are furthest from speckles.

G. Aggregate profile plot of CTCF ChIP-seq signal (RPKM) in speckle dTAG knock-in K562 cells with dTAG-13 vs. DMSO treatment (y-axis) across expressed genes (top) and genes that are downregulated with dTAG-13 treatment (bottom) (x-axis).

H. Boxplot showing log2(fold-change) loop strength for 6 hour dTAG-13 treated vs. DMSO treated speckle dTAG knock-in cells (y-axis) for loops with anchors that do not overlap any genes, overlap only genes that do not change expression, and overlap genes that increase or decrease in expression for at least one anchor (x-axis).

I. Representative SON and SRRM2 immunofluorescence images (green, alexafluor488) in HaloTag-CTCF HCT116 cells with 24 hours of 330 nM HaloPROTAC3 treatment to deplete CTCF or vehicle control. Overlaid on DAPI in blue.

For boxplots, the interquartile range (IQR) is depicted by the box with the median represented by the center line. Outliers are excluded. P values were calculated by a Mann-Whitney test and are annotated as follows: ns: not significant; *: 0.01 <p ≤ 0.05; **: 0.001 <p ≤ 0.01; ***: 0.0001 <p ≤ 0.001; ****: p ≤ 0.0001.