CTCF-RNA interactions orchestrate cell-specific chromatin loop organization

Abstract

CCCTC-binding factor (CTCF) is essential for chromatin organization. CTCF interacts with endogenous RNAs, and deletion of its ZF1 RNA binding region (∆ZF1) disrupts chromatin loops in mouse embryonic stem cells (ESCs). However, the functional significance of CTCF-ZF1 RNA interactions during cell differentiation is unknown. Using an ESC-to-neural progenitor cell (NPC) differentiation model, we show that CTCF-ZF1 is crucial for maintaining cell type-specific chromatin loops. Expression of CTCF-∆ZF1 leads to disrupted loops and dysregulation of genes within these loops, particularly those involved in neuronal development and function. We identified NPC-specific, CTCF-ZF1 interacting RNAs. Truncation of two such coding RNAs, Podxl and Grb10, disrupted chromatin loops in cis, similar to the disruption seen in CTCF-∆ZF1-expressing NPCs. These findings underscore the inherent importance of CTCF-ZF1 RNA interactions in preserving cell-specific genome structure and cellular identity.

Fig. 1.. Acute CTCF depletion and rescue using the AID2 system.

(A) Schematic showing the CTCF-AID2 degron and dox-inducible rescue system generated in ESCs. (B) Western blot of CTCF-AID2 and rescue lines after 24 hours with no treatment or treatment with 5-Ph-IAA or 5-Ph-IAA and dox. (C) CTCF-AID2 and dox-inducible rescue lines were differentiated into NPCs for 2 days. Cells were immunostained with anti-Sox1 antibody and percentages of Sox1+ cells were quantified by flow cytometry analysis. Neither 5-Ph-IAA nor dox was added. Data are represented as means ± SEM, N = 2 biological replicates. (D) Diagram of the experimental approach. ESCs or NPCs were treated with 5-Ph-IAA and dox and then harvested for downstream analysis after 24 hours. Image created in BioRender. Lucero, K. (2025) (https://BioRender.com/hu9z2jx).

Fig. 2.. CTCF loop anchors induced after differentiation are disrupted in the ΔZF1 mutant.

(A and B) APA plots (top) of CTCF loop anchors in (A) ESCs and (B) NPCs, comparing WT and ΔZF1. APA was generated using the combined set of CTCF loop anchors called in both WT and ΔZF1. The numbers indicate the APA score calculated using Juicer. The bottom-left panel is the difference in APA enrichment between ΔZF1 and WT. The bottom-right panel is a box plot quantifying the pixel enrichment of each loop. p-values were determined using the Wilcoxon signed-rank test. Plots represent merged biological replicates (N = 2). (C) Bar plots showing the number of WT CTCF loop anchors and the proportions that are lost or retained in the ΔZF1 mutant. Loops were classified as ΔZF1-lost if they were found to be significant only in WT but not in ΔZF1 (q value cutoff of 0.01) and if contact counts had log₂ fold change ≤ −1 (see Materials and Methods). (D) Bar plot showing the proportions of cell type–specific and common loops among ΔZF1-lost or ΔZF1-retained loops. The numbers indicate the absolute numbers of loops. P values were determined using Fisher’s exact test. (E and F) Micro-C contact heatmaps (top) at the (E) Podxl and (F) Grb10 loci. Arrows point to CTCF-bound loops. Numbers above the arrows indicate the fold enrichment of pixel intensity at the indicated loop anchor relative to the local background, as calculated using Cooltools. (Bottom) Loop anchor annotation and CTCF ChIP-seq track at the gene body of (E) Podxl and (F) Grb10. CTCF motifs with their orientation are annotated. The distance of the CTCF binding sites to the gene TSS is indicated. The data are represented as merged biological replicates (Micro-C N = 2; CTCF ChIP-seq N = 4).

Fig. 3.. ΔZF1 loop loss is not due to changes in CTCF chromatin binding.

(A and B) DiffBind MA plot (Deseq2 normalized) of differentially called peaks comparing ΔZF1 versus WT in (A) ESCs and (B) NPCs. Adjusted P value cutoff: ≤ 0.05; log₂ fold change cutoff: ≥ 1, ≤ −1; N = 4 biological replicates. (C and D) Scatterplots showing DiffBind DESeq2-normalized counts of CTCF ChIP-seq peaks in WT and ΔZF1 conditions at loop anchors that are lost (C) or retained (D) in ΔZF1. (E and F) (Left) The APA plots per replicate (N = 2) in (E) ESCs and (F) NPCs show the comparison between WT and ΔZF1 for the lost and retained loop subsets. Numbers indicate APA scores. (Right) CTCF ChIP-seq heatmaps comparing CTCF chromatin binding in (E) ESC-WT and ESC-ΔZF1 and (F) NPC-WT and NPC-ΔZF1. Each row is a loop anchor coordinate, and the heatmap is clustered on the basis of whether the anchor is ΔZF1-lost (top cluster) or ΔZF1-retained (bottom cluster). (G) Micro-C contact heatmap (top) and ChIP-seq tracks for CTCF and cohesin subunits, Rad21 and Smc3 (bottom), at a representative Podxl locus. Loop boundaries are highlighted in yellow. Boxes on the contact heatmaps mark loop anchors. Numbers next to each box indicate the fold enrichment of pixel intensity at the corresponding anchor relative to the local background, as calculated with Cooltools. CTCF motifs with their orientation are annotated. The data in (E), (F), and (G) are represented as merged biological replicates (Micro-C N = 2; CTCF ChIP-seq N = 4; Rad21 and Smc3 ChIP-seq N = 2).

Fig. 4.. Dysregulated genes in NPC-ΔZF1 mutant are enriched at disrupted loops.

(A) Deseq2 volcano plots showing gene expression changes in NPC-ΔZF1 compared to NPC-WT (see all in table S4). Adjusted P value cutoff: ≤ 0.05; log₂ fold change cutoff: ≥ 0.5, ≤ −0.5; N = 2 biological replicates. (B) PANTHER GO term analysis of NPC-ΔZF1 dysregulated genes (see all in table S5). Shown are the top 10 terms related to neuronal function. (C) Bar plots showing the percentage of genes in each set (dysregulated, random 1, and random 2) that overlap with CTCF peaks in NPCs (−1 kb from the TSS to TES of genes). Statistical significance was determined using Fisher’s exact test. (D) NPC CTCF ChIP-seq heatmaps at dysregulated, random 1, and random 2 genes. The data are represented as merged biological replicates (N = 4). (E) Quantification of CTCF ChIP-seq reads at the TSS-TES of genes shown in (D). P values were determined using the Wilcoxon signed-rank test. (F) Bar plots showing the percentage of genes from each gene set (dysregulated, random 1, and random 2) that are colocalized within ΔZF1-lost loops in NPCs. P values were determined using Fisher’s exact test. (G) Bar plots showing the number of NPC-ΔZF1-lost anchors that overlapped with promoters and/or enhancers. (H) Example locus with decreased enhancer-promoter loop contact in ΔZF1. (Top) Micro-C contact heatmaps of WT compared to ΔZF1. CTCF anchors are highlighted in yellow. Boxes mark loop anchors; numbers next to them show fold enrichment over the local background, calculated with Cooltools. (Bottom) RNA-seq, CTCF, and H3K27ac ChIP-seq tracks. The upstream anchor is at the Dlx3 promoter, and the downstream anchor is at a putative enhancer marked by H3K27ac (reanalyzed data; see table S6). The presence of CTCF motifs and their orientation is annotated. The data are represented as merged biological replicates (Micro-C N = 2; RNA-seq N = 2; CTCF ChIP-seq N = 4).

Fig. 5.. Decreased CTCF-RNA cross-linking in the ΔZF1 mutant.

(A and B) ESCs (A) and NPCs (B) were UV cross-linked, and Flag-Halo–tagged WT or ΔZF1 rescue CTCF was immunoprecipitated using anti-Flag beads. RNA was purified, and the cDNA library was sequenced. Reproducible CLIP peaks were called (see Materials and Methods) in the WT and ΔZF1 mutant. Bar graphs of the number of reproducible peaks are shown (N = 3 to 4 biological replicates). (C and D) The combined set of CLIP peaks in WT and ΔZF1 was used to plot heatmaps of ESC-CTCF CLIP peak enrichment in the non-cross-linking negative control and cross-linked WT and ΔZF1. CLIP peaks are shown for the (C) plus (+) and (D) minus (−) strands. (E and F) Heatmaps of NPC-CTCF CLIP peak enrichment in the non-cross-linking negative control and cross-linked WT and ΔZF1. CLIP peaks are shown for the (E) plus (+) and (F) minus (−) strands.

Fig. 6.. Identification of NPC-specific CTCF-RNA interactions.

(A) Deseq2 volcano plot showing differential CTCF-RNA interactions comparing NPCs versus ESCs (see all in table S7). Each dot is a gene. Total CLIP-seq read counts from the TSS to TES of genes were compared. Adjusted P value cutoff: ≤ 0.05; log₂ fold change cutoff: ≥ 1, ≤ −1; N = 3 to 4 biological replicates. (B) RNA-seq heatmaps of NPC-specific, CLIP-seq annotated genes, comparing ESC-WT, ESC-ΔZF1, NPC-WT, and NPC-ΔZF1. Each row is a gene and is annotated on the basis of whether gene expression in NPC-ΔZF1 is significantly down-regulated compared to WT. (C) Venn diagram of (i) genes up-regulated in NPCs relative to ESCs, (ii) genes colocalized at NPC-specific loop anchors, and (iii) genes annotated to be NPC-specific, CTCF-RNA interactors by CLIP-seq. Genes at the intersection of the three gene lists were considered for functional validation with Podxl and Grb10 being selected. Both genes are annotated in (A) and (B). (D) (Left) APA plots at loop anchors that colocalized with RNAs at the intersection of three gene lists (C). (Right) Difference in APA enrichment between ΔZF1 and WT at these anchors. Data are represented as merged biological replicates (N = 2). (E and F) RNA-seq tracks (top) and CLIP-seq tracks (bottom) at genes of interest: (E) Podxl and (F) Grb10. CLIP-seq tracks are zoomed in for better visualization of nucleotide-level cross-links. The data are represented as merged biological replicates (RNA-seq N = 2; CLIP-seq N = 3 to 4).

Fig. 7.. Truncation of NPC-specific RNAs interacting with CTCF leads to decreased chromatin loops in cis.

(A) Schematic of the CRISPR-Cas9–mediated truncation of target RNAs, Podxl and Grb10. We generated a homozygous in-frame knock-in of T2A-eGFP-SV40pA downstream of the first exon of the coding sequence (CDS) of each gene. (B and C) RNA-seq Deseq2-normalized read counts of (B) Podxl in the WT and Podxl pA mutant and (C) Grb10 in the WT and Grb10 pA mutant. Data are represented as means ± SEM, N = 3 biological replicates. (D) APA plots per replicate (N = 2) of NPC WT CTCF anchors, comparing WT, NPC Podxl pA, and NPC Grb10 pA. Numbers indicate APA scores. (E and F) Micro-C contact heatmaps (top) and RNA-seq tracks and ChIP-seq tracks (CTCF and Smc3) (bottom) at (E) the Podxl locus or (F) the Grb10 locus. NPC WT is compared to either the Podxl pA mutant or the Grb10 pA mutant, respectively. Heatmaps were generated using ICE normalization implemented in cooler. Boxes on the contact heatmaps indicate loop anchors, enlarged for clarity. Numbers next to each box indicate the fold enrichment of pixel intensity at the corresponding anchor relative to the local background, as calculated with Cooltools. Loop boundaries are highlighted in yellow. The presence of CTCF motifs and their orientation is annotated. The data are represented as merged biological replicates (Micro-C N = 2; RNA-seq N = 2; CTCF and Smc3 ChIP-seq N = 2).