Cancer-specific CTCF binding facilitates oncogenic transcriptional dysregulation

Abstract

Background: The three-dimensional genome organization is critical for gene regulation and can malfunction in diseases like cancer. As a key regulator of genome organization, CCCTC-binding factor (CTCF) has been characterized as a DNA-binding protein with important functions in maintaining the topological structure of chromatin and inducing DNA looping. Among the prolific binding sites in the genome, several events with altered CTCF occupancy have been reported as associated with effects in physiology or disease. However, hitherto there is no comprehensive survey of genome-wide CTCF binding patterns across different human cancers.

Results: To dissect functions of CTCF binding, we systematically analyze over 700 CTCF ChIP-seq profiles across human tissues and cancers and identify cancer-specific CTCF binding patterns in six cancer types. We show that cancer-specific lost and gained CTCF binding events are associated with altered chromatin interactions, partially with DNA methylation changes, and rarely with sequence mutations. While lost bindings primarily occur near gene promoters, most gained CTCF binding events exhibit enhancer activities and are induced by oncogenic transcription factors. We validate these findings in T cell acute lymphoblastic leukemia cell lines and patient samples and show that oncogenic NOTCH1 induces specific CTCF binding and they cooperatively activate expression of target genes, indicating transcriptional condensation phenomena.

Conclusions: Specific CTCF binding events occur in human cancers. Cancer-specific CTCF binding can be induced by other transcription factors to regulate oncogenic gene expression. Our results substantiate CTCF binding alteration as a functional epigenomic signature of cancer.

Keywords: 3D genome organization; CTCF; Enhancer; Gene regulation; Integrative analysis; NOTCH1; T cell lymphoblastic leukemia; Transcription factor.

Fig. 1

Identification of cancer-specific CTCF binding patterns in the human genome. a Distribution of coefficient of variation of chromatin accessibility at different genomic features, calculated using DNase-seq data from over 60 cell lines from ENCODE. b Distribution of occupancy score for all 688,429 union CTCF binding sites (blue), and percentage of CTCF sites that contain a CTCF motif at each occupancy score (orange). c Distribution of CTCF binding occupancy score in 8 ChIP-seq datasets for T-ALL cell lines (y-axis) and the occupancy frequency score in the other 763 ChIP-seq datasets (x-axis). Color density in each unit represents the number of CTCF binding sites with designated scores. d CTCF ChIP-seq signals at a 2-kb region surrounding T-ALL_lost (top) and T-ALL_gained (bottom) binding sites in normal CD4+ T cells and the T-ALL cell lines Jurkat and CUTLL1, and SMC3 signals at the same regions in CUTLL1. e Example of CTCF ChIP-seq signals around a T-ALL-specific lost CTCF binding site. f Example of CTCF ChIP-seq signals around a T-ALL-specific gained CTCF binding site. g Number of identified gained (left) and lost (right) CTCF binding sites in each of the 6 cancer types and number of shared sites between each pair of cancer types. Color density of each element represents the level of similarity measured by Jaccard index. h Genomic distribution of identified lost (left) and gained (right) CTCF binding sites in the 6 cancer types. Promoter regions are defined as ± 2 kb from any TSS in the genome. i Differential chromatin accessibility (ATAC-seq) in TCGA patient samples at identified cancer-specific lost (blue), gained (red), and constitutive (gray) CTCF binding sites in each of the 4 cancer types compared to all other TCGA samples. *, p < 0.05, **, p < 0.001, by two-tailed unpaired Student’s t test

Fig. 2

Gained/lost CTCF binding events associate with chromatin dynamics. a, c, e, g Volcano plots showing differential chromatin interaction levels between cancer and normal cells at cancer-specific lost (blue), gained (red), and constitutive (gray) CTCF binding sites, measured by Hi-C. Each point represents the interaction changes between a CTCF binding site and 5-kb bins located within 500 kb from the site. Horizontal dotted line represents p value cutoff of 0.05, by two-tailed paired Student’s t test. b, d, f, h Boxplots showing differential interaction frequencies between cancer and normal matched tissues for each group of CTCF binding sites. *, p < 0.05, **, p < 0.001, by two-tailed unpaired Student’s t test

Fig. 3

Gained/lost CTCF binding events associate with differential gene expression in cancer. a CTCF ChIP-seq signals (x-axis) and gene expression levels (y-axis) for one CTCF site-gene pair in 54 cell types with matched data available. R² is calculated as the association score. Sqrt, Square root; TPM, transcript count per million reads; RPKM, read count per kilobases per million reads. b Schematic of categories of intra-chromatin-domain and inter-chromatin-domain gene-CTCF pairs. c Distribution of highly correlated CTCF-gene pairs (defined as R² > 0.25) as a function of the distance between the CTCF binding site and the gene’s TSS. Pairs located within the same CTCF domain (intra-domain, blue) and across different CTCF domains (inter-domain, gray) are plotted separately. P values were obtained using two-tailed Fisher’s exact test. Dashed line represents P = 0.01. d, e Top: Percentage of highly correlated CTCF-gene pairs in which the gene sits within the same domain as a cancer-specific lost (d) or gained (e) binding site, with constitutive sites as control. “Promoter” refers to genes whose promoter region (TSS ± 2 kb) contains a CTCF binding site from a certain category. “Promoter ctrl” refers to genes whose promoter region contains a constitutive CTCF binding site as the control for cancer-specific gained/lost sites. “Intra-domain” refers to genes whose chromatin domain contains a CTCF binding site. “Domain ctrl” refers to genes whose chromatin domain contains a constitutive CTCF site as the control for those with cancer-specific gained/lost sites. Bottom: Percentage of differentially expressed genes (|log2FC| > 1, FDR < 1e−5) contained within the corresponding group of either Promoter or Intra-domain highly correlated CTCF-gene pairs in the corresponding cancer type. *, p < 0.05, **, p < 0.001, by two-tailed Fisher’s exact test. f, g Percentage of genes that are upregulated (top, log2FC > 1, FDR < 1e−5) or downregulated (bottom, log2FC < − 1, FDR < 1e−5) located in the chromatin domains containing certain group of lost (f) or gained (g) CTCF sites. *, p < 0.05, **, p < 0.001, by two-tailed Fisher’s exact test

Fig. 4

Patterns of differential DNA methylation near cancer-specific lost/gained CTCF sites. a,b ChIP-seq signals and differential DNA methylation levels surrounding specific lost (a) or gained (b) CTCF binding sites in cancer versus normal tissues for each of the 5 cancer types. Pie charts represent numbers of CTCF sites with (light blue) or without (gray) sufficient DNA methylation signals from bisulfite sequencing data. For sites with sufficient DNA methylation signals, heatmaps show CTCF ChIP-seq signals cover 2-kb regions centered at each site. Differential DNA methylation level at a 300 bp region centered at each CTCF binding site was calculated and presented as a waterfall plot. Purple bars represent increased and green bars represent decreased DNA methylation levels (with values in a range from 0 to 100). Rows in corresponding ChIP-seq and DNA methylation plots are ranked identically. c Association between CTCF binding specificity and differential DNA methylation for each cancer type. All CTCF sites with sufficient DNA methylation coverage in each cancer type were ranked based on their differential CTCF binding level in cancer compared to other samples and grouped into 100 equal-count bins. For each bin, the percentages of sites associated with DNA hypomethylation (green), unchanged methylation (gray), and hypermethylation (purple) were stacked in a column. Top color bar represents the median differential binding level of the CTCF sites in a bin, quantified as the T-test statistic

Fig. 5

T-ALL_gained CTCF binding associates with oncogenic NOTCH1 binding and increased chromatin interaction. a BART-predicted transcription factors binding in genomic regions that have increased interaction with T-ALL_gained CTCF sites comparing Jurkat cells with normal CD4⁺ T cells. b Percentage of chromatin domains including different groups of CTCF binding that contain a NOTCH1 binding site or a dynamic NOTCH1 binding site. *, p < 0.05, **, p < 0.001, by two-tailed Fisher’s exact test. c, d T-ALL_gained sites associate with an increased H3K27ac level in Jurkat cells. c Volcano plot showing the differential H3K27ac level between Jurkat cells and normal CD4⁺ T cells measured by ChIP-seq; each point represents a 10-kb region surrounding a CTCF binding site. d Regions containing dynamic NOTCH1 binding sites are highlighted in red. e Example of Hi-C interaction maps and ChIP-seq tracks around a T-ALL_gained CTCF binding site

Fig. 6

T-ALL_gained CTCF binding facilitates oncogenic NOTCH1 transcriptional activity. a Scatter plot of CTCF sites in T-ALL quantifying CTCF level changes in GSI and GSI washout experiment. Differential CTCF ChIP-seq signal (log2 fold change) in GSI washout vs. GSI (y-axis) is plotted against differential CTCF ChIP-seq signal in GSI vs. DMSO (x-axis). Red dots are T-ALL_gained sites. b ATAC-seq levels at T-ALL_gained CTCF sites in Jurkat cells at DMSO, GSI treated for 72 h, and GSI washout for 16 h. *, p < 0.05, **, p < 0.001, by paired two-tailed Student’s t test. c FLAG-NOTCH1 immunopurified proteins from control and NOTCH1-FLAG-expressing CUTLL1 cells were resolved on SDS-PAGE gels and interacting partners are visualized by western blot. IgG was immunopurified as a negative control. IB, immunoblot; IP, immunoprecipitation. d, e ChIP-seq signals for BRG1 (d) and CTCF (e) surrounding constitutive (gray), AML_lost (blue), and AML_gained (red) CTCF binding sites in AML cell line EOL1. Normalized ChIP-seq read counts (RPKM) covering 2-kb regions centered at CTCF binding sites were plotted per 10-bp non-overlapped bins. f Percentage of genes in different groups that are downregulated (log2FC < − 0.26, FDR < 0.001) in shCTCF experiment in CUTLL1. Black: Genes located in the T-ALL_gained-CTCF-containing chromatin domains. Red: Genes located in the T-ALL_gained-CTCF-containing domains that are also upregulated (log2FC > 0.26, FDR < 0.001) in T-ALL compared to normal T cell. *, p < 0.05, **, p < 0.001, by two-tailed Fisher’s exact test. g BART-predicted TFs that target the downregulated genes (log2FC < − 0.58, FDR < 0.01) upon CTCF silencing experiments in CUTLL1. h MA plot showing differential gene expression after shCTCF treatment in CUTLL1. Most NOTCH1 target genes (red) are downregulated. i Differential gene expression between CUTLL1 and normal T cells. Group A: genes located in dynamic-NOTCH1-containing domains. Group B: genes located in domains containing both dynamic-NOTCH1 and T-ALL_gained CTCF binding sites. *, p < 0.05, **, p < 0.001, by two-tailed unpaired Student’s t test

Fig. 7

Schematic model of CTCF facilitated oncogenic transcriptional activation in T-ALL. a Without gained CTCF binding, intracellular NOTCH1 transcriptional complexes recognize RBPJ, the DNA-binding sequence motif, and recruit SWI/SNF / BAF complexes. b With gained CTCF binding, NOTCH1, BAF complexes, and CTCF protein molecules cooperatively alter the chromosome conformation and form a transcriptional condensate (dashed circle) to regulate expression of the target gene