Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Jul 10.
Published in final edited form as: Mol Neurobiol. 2023 Nov 18;61(6):3240–3257. doi: 10.1007/s12035-023-03762-5

Atypical Modes of CTCF Binding Facilitate Tissue-Specific and Neuronal Activity-Dependent Gene Expression States

Morgan Crewe 1,2,3, Amir Segev 1,2,3, Richard Rueda 1,2,3, Ram Madabhushi 1,2,3
PMCID: PMC13347739  NIHMSID: NIHMS2185987  PMID: 37979036

Abstract

Multivalent binding of CTCF to variable DNA sequences is thought to underlie its ability to mediate diverse cellular functions. CTCF typically binds a 20 base-pair consensus DNA sequence, but the full diversity of CTCF binding sites (CBS) within the genome has not been interrogated. We assessed CTCF occupancy in cultured cortical neurons and observed surprisingly that ~ 22% of CBS lack the consensus CTCF motif. We report here that sequence diversity at most of these atypical CBS involves degeneracy at specific nucleotide positions within the consensus CTCF motif, which likely affect the binding of CTCF zinc fingers 6 and 7. This mode of atypical CTCF binding defines most CBS at gene promoters, as well as CBS that are dynamically altered during neural differentiation and following neuronal stimulation, revealing how atypical CTCF binding could influence gene activity. Dynamic CBS are distributed both within and outside loop anchors and TAD boundaries, suggesting both looping-dependent and independent roles for CTCF. Finally, we describe a second mode of atypical CTCF binding to DNA sequences that are completely unrelated to the consensus CTCF motif, which are enriched within the bodies of tissue-specific genes. These tissue-specific atypical CBS are also enriched in H3K27ac, which marks cis-regulatory elements within chromatin, including enhancers. Overall, these results indicate how atypical CBS could dynamically regulate gene activity patterns during differentiation, development, and in response to environmental cues.

Keywords: CTCF, CTCF motif, Neuronal activity, Neuronal development, Tissue-specific genes

Introduction

CTCF (CCCTC-binding factor) is a conserved architectural protein that binds to thousands of sites within mammalian genomes [1, 2]. CTCF acts together with the cohesin complex to fold chromosomes into loops and units of frequently self-interacting chromatin segments called topologically associated domains (TADs). According to current models, cohesin uses ATP hydrolysis to bidirectionally reel in and extrude DNA, forming loops, and loop extrusion is halted when cohesin encounters CTCF bound to inwardly oriented CTCF motifs [3-7]. Acute degradation of CTCF results in genome-wide loss of CTCF-anchored loops and TAD insulation, suggesting that CTCF plays a crucial role in chromosome organization [8, 9]. Additionally, CTCF was shown to modulate enhancer-promoter contacts and to function variously as a transcriptional repressor, activator, or epigenetic insulator within specific cellular contexts and genomic sites [10-12]. These studies suggest that understanding the molecular logic behind CTCF dynamics could inform how chromosome structure impacts gene activity states. Yet precisely how CTCF mediates its diverse cellular functions is still unclear.

In vitro studies show mutations in each of CTCF’s 11 different zinc-finger (ZF) domains affect its binding to different target sites, suggesting that CTCF could bind distinct DNA sequences by employing distinct ZFs in vivo [13-18]. Chromatin immunoprecipitation sequencing (ChIP-seq) across mammalian cell types and tissues indicates that a majority of CTCF binding sites (CBS) contain a 20-bp core motif that is engaged by ZFs 4–7, while several upstream and downstream sequences affect CTCF binding affinity by engaging ZFs 1–2 and 8–11 [1, 15, 16, 18]. However, nucleotides underlying CBS can vary significantly from the core motif, and CBS that diverge from the core motif were shown to correlate with cell type–specific and developmentally regulated CTCF occupancy patterns [1, 19, 20]. Despite this, how the diversity of CBS relate to gene regulation and cellular functions has not been systematically analyzed.

Many studies have highlighted crucial roles for CTCF-mediated transcriptional control in cellular differentiation and development, citing changes in CTCF occupancy as a key regulator of cell fate [21]. CTCF is ubiquitously expressed, and germline depletion of CTCF leads to embryonic lethality [2, 22]. CTCF depletion during embryonic development impairs differentiation of NPCs into mature neurons, while CTCF depletion in mature cortical neurons impairs synaptic plasticity and associated learning and memory behaviors, indicating a role for CTCF in both the development and maintenance of neuronal health and function [23-26]. Although there has been considerable progress towards our understanding of the role of CTCF during cellular differentiation and early cellular development, the role of dynamic CTCF binding for facilitating responses to cellular state change in post-mitotic cells remains poorly understood. We hypothesized that CTCF binding in neurons and its dynamics in response to neuronal activity could be relevant to experience-driven gene activity patterns.

In an attempt to test this idea, we performed CTCF ChIP-seq as a function of neuronal stimulation in cultured mouse cortical neurons. Unexpectedly, our analysis revealed a novel atypical mode of CTCF binding at thousands of sites that exhibit degeneracy at specific nucleotides within the 20-bp core CTCF motif. We show that degeneracy at these nucleotides defines most CBS within neuronal gene promoters. Analysis of publicly available CTCF ChIP-seq data revealed a similar atypical mode of CTCF binding across a variety of mouse tissues, indicating that our findings in cultured cortical neurons can be generalized across tissues. Intriguingly, CBS lacking the core CTCF motif are highly overrepresented among tissue-specific CBS and were enriched in the bodies of tissue-specific genes, revealing how binding of CTCF within gene bodies could regulate the expression of tissue-specific genes. We observe that both canonical and atypical CBS co-localize with SMC1, chromosome loop anchors, and TAD boundaries. Whereas the majority of TAD boundaries and loop anchors in mouse embryonic stems cells (mESCs) are bound by both canonical and atypical CBS, neuronal differentiation triggers a dramatic loss of atypical, but not canonical, CBS, indicating how stable chromatin structures could be established during development. Finally, we show that atypical CBS are enriched among CBS that are dynamically altered following neuronal stimulation, are associated with genes encoding for experience-driven neuronal functions, and nucleate more SMC1 compared to canonical CBS. Overall, these results provide new insights into how distinct modes of CTCF binding allow it to simultaneously facilitate stable and dynamic gene activity states and highlight the need for future investigations to consider atypical CBS when interrogating CTCF function.

Methods

Neuronal Cultures and Drug Treatments

Dissociated primary cortical neurons from E16 Swiss-Webster mice (Charles River) were plated at a density of 5 million cells/plate in 10-cm plates. The plates were pre-coated with poly-D-lysine (0.5 mg/ml) overnight at 37 °C, followed by washing twice with dH2O. Neurons were maintained in neurobasal media (GIBCO) supplemented with L-glutamine, penicillin/streptomycin, and B27. N-methyl-D-aspartic acid (NMDA, 50 μM, Sigma M3262) was dissolved in dH20. NMDA was added to the cell culture media to a final concentration of 50 μM for 10 min, after which cells were washed with PBS and recovered for 50 min in drug-free media prior to collection.

CTCF and SMC1 ChIP-Seq

CTCF and SMC1 ChIP-seq were performed for two biological replicates using a ChIP assay kit (MilliporeSigma) as described by the manufacturer. Briefly, DIV13 primary cortical neurons (~ 15 M cells/replicate) were crosslinked with 1% formaldehyde for 10 min at 37 °C, washed twice with ice-cold PBS, scraped into 1.5-ml tubes, and resuspended in 300 μl SDS lysis buffer (50 mM Tris–HCl pH 8, 10 mM EDTA, 1% SDS, protease and phosphatase inhibitor cocktail). Cells were sonicated to shear DNA to 200–500 bp using a bioruptor (Diagenode; high power, 30 s on/30 s off for 40 cycles). Samples were then centrifuged at 13,000 RPM for 10 min at 4 °C to remove cellular debris, and the supernatant was diluted up to 2 ml in ChIP dilution buffer (16.7 mM Tris–HCl pH 8, 167 mM NaCl, 1.2 mM EDTA, 1.1% Triton X-100, 0.01% SDS). Prior to immunoprecipitation, 1% of each sample was transferred to a clean tube to be used for input normalization. Samples were then incubated with anti-CTCF antibody (1:200, CST, #3418) or anti-SMC1 antibody (1:1000, Bethyl, A300-055A) overnight at 4 °C followed by incubation with ChIP-grade protein A/G magnetic beads (Thermo) for 3 h at 4 °C. Antibody-conjugated beads were washed with low salt buffer (20 mM Tris–HCl pH 8, 1% Triton-X-100, 2 mM EDTA, 0.1% SDS, 150 mM NaCl), high salt buffer (20 mM Tris–HCl pH 8, 1% Triton-X-100, 2 mM EDTA, 0.1% SDS, 500 mM NaCl), LiCl buffer (10 mM Tris–HCl pH 8, 1% IGEPAL-CA630, 1% deoxycholic acid (sodium salt), 1 mM EDTA, 0.25 M LiCl), and twice with TE buffer (50 mM Tris–HCl pH 8, 1 mM EDTA) for 5 min each with agitation at room temperature. Immunoprecipitated CTCF-DNA complexes were eluted from the magnetic beads in 500 μl ChIP elution buffer (1% SDS, 0.1 M NaHCO3) for 30 min at 65 °C. DNA–protein crosslinks for immunoprecipitated samples and input DNA were reversed via incubation with an additional 20 mM NaCl for 4 h at 37 °C, and the residual protein was digested with Proteinase K (NEB) for 1 h at 45 °C. DNA was precipitated using phenol/chloroform extraction and ethanol precipitation, washed with 70% ethanol, and resuspended in 50 μl TE buffer. Samples were then prepared for sequencing using a paired-end library preparation kit compatible with Illumina sequencing instruments (KAPA biosciences) according to the manufacturer’s instructions. Following library amplification, DNA fragment size was assessed via Tapestation (Agilent) to ensure a final library size of 350–450 bp. DNA concentration was quantified by fluorometry (dsDNA high sensitivity Qubit assay, Thermo) to ensure equal DNA concentration for each replicate before replicates were pooled and sequenced on an Illumina Next-Seq 500 instrument using V2.5 reagents to an average depth of 25–35 million reads per replicate.

CTCF and SMC1 ChIP-seq reads were processed in accordance with ENCODE analysis guidelines [27]. Read alignment and all subsequent analysis requiring chromosome length files were performed using GRCm38.p6 assembly (mm10) reference genome. Peaks were called using MACS2 from biological duplicates and pseudo-replicates where each ChIP sample was normalized to input [28]. IDR (irreproducible discovery rate) thresholding was performed on MACS2 peak calls using an IDR threshold of 0.05 [29]. Where indicated, peak scores from MACS2 outputs for peaks that passed the IDR threshold were used to rank CTCF peaks into quartiles from the lowest (Q1) to the highest (Q4) peak score.

Hi-C Analysis

Previously published Hi-C data from mouse embryonic stem cells, neural progenitor cells, and cortical neurons [30] were used to identify topologically associating domains (TADs) and chromatin loops using the Juicer suite [30, 31]. TADs were called using Arrowhead from the Juicer suite (5 kb bins, KR normalized), where 10 kb TAD boundaries were used for downstream analysis. Chromatin loops were called using HiCCUPS with default parameters, where resulting 5 kb loop anchors were used for downstream analysis.

DNA Binding Motif Analysis

The motif finding software STREME from the MEME-ChIP suite was utilized to discover de novo DNA binding motifs enriched within each ChIP-seq dataset [32]. Briefly, peak summits defined by MACS2 peak calling were extended ± 250 bp and then converted to mm10 (GRCm38. p6 assembly) FASTA sequence using bedtools getfasta [33]. FASTA sequences were analyzed via STREME using default parameters. STREME outputs were then submitted for known DNA binding motif analysis via TomTom using the JASPAR 2020 CORE vertebrate motif library [34].

To identify CTCF binding sites that contain the canonical CTCF binding motif, we first utilized PWMScan (default parameters) to identify all canonical CTCF binding sites (JASPAR motif MA0139.1) throughout the mouse genome using the mm10 mouse reference genome [35]. The resulting bed file containing all canonical CTCF binding sites was intersected with IDR thresholded CTCF peaks using bedtools intersect [33]. CTCF peaks that overlap with these sites were considered canonical binding sites, and CTCF peaks that do not overlap with these sites were considered atypical binding sites. This analysis was repeated for consistency using the HOMER motif scanning algorithm, which yielded similar results [36].

Where indicated, the score function of PWMScan (sum occupancy mode, default parameters) was used to calculate scores for each CTCF binding site according to how well the DNA sequence aligns with the canonical CTCF binding motif (JASPAR MA0139.1). Higher scores indicate a higher degree of concordance between the DNA sequence and MA0139.1. PWM scores were used to compare average scores between different categories of CBS or were used to further segregate canonical and atypical CBS into quartiles (Q1 = lowest scores/lowest similarity to MA0139.1, Q4 = highest scores/highest similarity to MA0139.1) for further analysis.

Chromatin State Analysis

Chromatin state enrichment of CTCF peak files generated by MACS2 was analyzed via ChIPseeker using the UCSC mm10 genome annotation database and default parameters [37]. The percentage of each chromatin state as a proportion of total genome coverage was plotted in bar graphs using GraphPad Prism 9. To identify chromatin loops anchored at regulatory elements, ChromHMM was used to segment the mm10 mouse genome into 15 chromatin states based on ENCODE ChIP-seq data from the P0 mouse forebrain (Table 1) for eight histone modifications (H3K27ac, H3K9ac, H3K4me3, H3K4me1, H4K4me1, H3K27me3, H3K9me3, H3K36me3) [38, 39]. Annotation of chromatin states was as described previously. BED files of individual promoter (active, flanking, poised, and weak) and enhancer (strong proximal, strong distal, poised proximal, poised distal, and weak) chromatin states were intersected with intervals corresponding to chromatin loop anchors to identify chromatin loops anchored at regulatory elements.

Table 1.

ENCODE/GEO accessions for previously published data included in the manuscript

P0 mouse tissues (ENCODE)
H3K9me3 ChIP-seq (forebrain) ENCSR093DWU
H3K27me3 ChIP-seq (forebrain) ENCSR070MOK
H3K36me3 ChIP-seq (forebrain) ENCSR069TDC
H3K4me2 ChIP-seq (forebrain) ENCSR906UJW
H3K4me3 ChIP-seq (forebrain) ENCSR258YWW
H3K4me1 ChIP-seq (forebrain) ENCSR465PLB
H3K9ac ChIP-seq (forebrain) ENCSR369RBO
H3K27ac ChIP-seq (forebrain) ENCSR094TTT
CTCF ChIP-seq (forebrain) ENCSR677HXC
CTCF ChIP-seq (midbrain) ENCSR985ZTV
CTCF ChIP-seq (hindbrain) ENCSR150RGT
CTCF ChIP-seq (lung) ENCSR418SBY
CTCF ChIP-seq (heart) ENCSR491NUM
H3K27ac ChIP-seq (heart) ENCSR675HDX
CTCF ChIP-seq (kidney) ENCSR143WOK
CTCF ChIP-seq (liver) ENCSR041SMK
CTCF ChIP-seq (intestine) ENCSR002ZAG
CTCF ChIP-seq (stomach) ENCSR104QEN
iPSC derived embryonic stem cells, neural progenitor cells, and cortical neurons
CTCF ChIP-seq, Hi-C GSE96107
DIV7 mouse primary cortical neurons
H3K27ac ChIP-seq GSE60192
Open in a new tab

Gene Ontology Analysis

CTCF peak files generated by MACS2 were assigned to the nearest known gene via ChIPseeker annotatePeak using the UCSC mm10 genome annotation database and default parameters [37]. Gene ontology for known genes was analyzed via clusterProfiler enrichGO with default parameters and represented as dotplots [40].

Visualization and Statistical Analysis

All aggregate plots and heatmaps were generated from score matrices using deepTools computeMatrix in reference-point mode (center = peak summit, signal averaged over 50 bp bins, upstream and downstream flanking distance noted on x-axis) [41]. Matrices from computeMatrix outputs were plotted using the plotprofile or plotheatmap commands for visualization. Median signals for each bin were calculated using kentUtils bigWigAverageOverBed for significance testing of differences in signal intensities between two groups using Wilcoxon Signed Rank test [42]. Outputs from ChIPseeker chromatin state annotation and data in all bar graphs were plotted using Graphpad Prism 9. Screenshots of signal tracks were generated using the Integrated Genome Browser (IGV, Broad Institute). All figures were compiled and finalized using Adobe Illustrator.

GEO and ENCODE Data

Several CTCF ChIP-seq and Hi-C datasets from ENCODE or other previously published work were used throughout the manuscript, and the complete list is available in Table 1. For each ChIP-seq dataset included in the manuscript, replicated, IDR thresholded peaks were used for analysis.

Results

A Distinct Mode of CTCF Binding in Neurons

To understand the roles of CTCF dynamics in neurons, we performed CTCF chromatin immunoprecipitation and sequencing (ChIP-seq) in cultured mouse cortical neurons under basal conditions and after neuronal stimulation with N-methyl-D-aspartate (NMDA; 50 μM for 10 min, recovery for 50 min). Following library preparation, sequencing, and alignment of the reads to the mm10 reference genome, the MACS2 pipeline was used to call peaks from two biological replicates [28]. Quality control metrics indicated high concordance between replicates under both conditions (Figure S1A). Replicated peaks were further filtered using irreproducible discovery rate (IDR) thresholding according to ENCODE guidelines, yielding 33,820 high-confidence CTCF binding sites (CBS) under basal conditions (Figure S1B) [27, 29]. Analysis of DNA binding motifs enriched at high-confidence CBS using STREME recovered a motif similar to the consensus CTCF motif (JASPAR MA0139.1) within a majority of CBS (22,491/33,820), validating our ChIP-seq results (Figure S1C) [32, 34]. However, a significant fraction of CBS (11,329/33,820) seemed to lack the consensus CTCF motif, suggesting that CTCF could bind to alternative DNA sequences in neurons.

To further investigate the extent to which CTCF binds to sites lacking the consensus CTCF motif (MA0139.1), we first used PWMScan to identify all occurrences of the motif throughout the mouse genome and intersected these intervals with IDR-thresholded CBS [35]. CBS that overlap with one or more occurrences of the consensus CTCF motif were termed canonical CBS, and CBS that did not overlap with any occurrences were termed atypical CBS. This analysis revealed that ~ 22% (7,537/33,820) of CBS in untreated neurons lack significant enrichment for the consensus CTCF motif (Fig. 1A). Similar results were observed when CBS were processed using a distinct motif-finding pipeline in the HOMER suite (Fig. 1A) [36]. Based on these results, we conservatively chose atypical CBS identified by PWMScan for further analysis. Finally, to discern whether CBS lacking the consensus CTCF motif represent non-specific binding events and/or other sequencing artifacts, we quantified the proportion of canonical and atypical CBS prior to IDR thresholding, reasoning that non-specific binding events would be dis-proportionately filtered by IDR thresholding. However, the proportion of canonical and atypical CBS was unaffected by IDR thresholding (Figure S1B). Similarly, we repeated our CTCF ChIP-seq experiments in untreated neurons at twice the sequencing depth and again recovered similar fractions of canonical and atypical CBS (Figure S1D). Together, these results suggest that CBS that are not significantly enriched for the consensus CTCF motif in our experiments represent bona-fide CTCF binding events within the neuronal genome.

Fig. 1.

Fig. 1

Open in a new tab

Defining an atypical mode of CTCF binding in neurons. A PWMScan (left) and HOMER (right) were used to identify all consensus CTCF motif (JASPAR MA0139.1) occurrences in the mouse genome. Intersection of these occurrences with CTCF ChIP-seq peaks was used to identify canonical or atypical CTCF binding sites (CBS) in DIV13 primary cortical neurons, represented as a percentage of all CBS (n = 33,820). B Diagram of the core consensus CTCF binding motif with recently identified upstream and downstream flanking sequences. Green boxes indicate nucleotide sequences/positions that promote high-affinity CTCF binding. Blue box indicates nucleotide sequences/positions that promote high-affinity CTCF binding and specify CTCF binding orientation. Red box indicates nucleotide sequences/positions that decrease CTCF binding affinity. C, D STREME was used to discover DNA biding motifs enriched at CTCF ChIP-seq peaks in DIV13 primary cortical neurons, and TomTom was used to identify DNA binding factors known to bind the identified motif sequences. The top 5 DNA binding motifs enriched at canonical CBS (C), or atypical CBS (D), are shown. Green boxes indicate sequences with homology to previously identified flanking sequences that promote high-affinity CTCF binding. Red boxes indicate sequences with homology to previously identified flanking sequences that decrease CTCF binding affinity

Previous reports suggest that differential utilization of CTCF’s 11 zinc finger (ZF) domains could allow it to bind diverse sequences within the genome and mediate distinct functions at these sites [1, 14-16, 43, 44]. Most commonly, CTCF binds a 20-bp “core” consensus sequence by engaging ZFs 4–7, while ZFs 1–3 and 8–11 bind flanking sequences that increase or decrease CTCF stability, respectively (Fig. 1B) [1, 15, 16, 18]. To understand how CTCF could bind to sites that are not significantly enriched for the consensus CTCF motif, we again utilized STREME to identify DNA binding motifs enriched within canonical and atypical CBS. As expected, a motif matching the consensus CTCF motif (MA0139.1) was again retrieved as the highest enriched DNA binding motif within sites identified as canonical CBS (Fig. 1C). Additionally, previously described upstream and downstream flanking sequences known to affect CTCF stability were also retrieved (Fig. 1B, C) [16, 43]. Intriguingly, a motif that partially matched the consensus CTCF motif was retrieved as the highest enriched DNA binding motif within ~ 76% (5746/7537) of sites identified as atypical CBS, and ~ 14% of such sites were enriched for unrelated AT-rich and GC-rich repeat sequences (Fig. 1D). As observed for canonical CBS, previously described upstream and downstream motifs were also retrieved for atypical CBS (Fig. 1D). No additional motifs corresponding to known DNA binding proteins were enriched alongside either canonical or atypical CBS.

To further interrogate how sequences underlying atypical CBS differ from canonical CBS, the score function in PWMScan was used to assign scores to each canonical and atypical CBS based on their similarity to the consensus motif (higher scores = higher similarity). PWM scores were used to segregate canonical and atypical CBS into quartiles based on their similarity to the consensus CTCF motif (Q1—low similarity; Q4—high similarity). Alignment of the position-weight matrices for the highest enriched motif within each quartile revealed that deviation from the consensus CTCF motif does not reflect random changes at nucleotides within this 20-bp sequence. Rather, deviation from the consensus CTCF motif arises from degeneracy at specific nucleotides that are previously shown to be important for maintaining high CTCF binding affinity, including position 4 and positions 6–9 (Table 2) [15, 16, 20]. These nucleotides are previously shown to be bound by CTCF ZFs 6 and 7, and mutation to these ZFs results in markedly decreased affinity for the consensus CTCF biding motif [14-18]. Atypical Q4 CBS with the highest PWM scores share a high degree of similarity to canonical Q1 CBS and the consensus CTCF motif overall (Table 2). However, the lowest PWM scores corresponding to atypical Q1 and Q2 diverge significantly from the consensus CTCF motif, and less than half of all CBS within these categories are significantly enriched for any DNA binding motif (Table 2). Together, these results describe an alternative mode of CTCF binding at thousands of sites in neurons that is defined by variations in specific residues of the core CTCF motif.

Table 2.

Highest enriched motif found via STREME analysis of canonical and atypical CBS segregated into quartiles according to PWM score (Q1—low similarity; Q4—high similarity to MA0139.1)

graphic file with name nihms-2185987-t0005.jpg
Open in a new tab

Atypical CBS Are Enriched in Gene Promoters

Although CTCF binding at alternative DNA sequences has been reported previously, whether CTCF binding at canonical versus atypical DNA motifs could confer distinct functions is unclear. To address this issue, we characterized the properties of canonical and atypical CBS. Average CTCF ChIP-seq signal intensity was reduced at most atypical CBS compared to canonical CBS (Fig. 2A). To interrogate this further, we ranked CTCF ChIP-seq peaks into quartiles according to MACS2 peak scores (Q1 = lowest peak scores, Q4 = highest peak scores) and observed only a small fraction of atypical CBS within the strongest Q4 peaks (Figure S2A). Similarly, we observed a decrease in CTCF signal intensity concordant with decreased similarity to the consensus CTCF motif across canonical and atypical CBS quartiles ranked according to PWM score (Figure S2B). These results are consistent with previous observations that either mutations at positions 4 and 6–9 of the core CTCF motif, or mutations of the ZFs that bind these nucleotides, reduce CTCF binding affinity and affect its ability to bind in an oriented manner [15, 16, 20]. Together, these observations suggest that CTCF binds atypical CBS with lower affinity and/or stability than canonical CBS.

Fig. 2.

Fig. 2

Open in a new tab

Characterizing the properties of canonical and atypical CBS across tissues. A Average CTCF signal intensity surrounding canonical and atypical CBS (left) with quantification of median CTCF signal intensity (right, Wilcoxon signed-rank test, p < 0.0001). B Chromatin state distribution of all, canonical, and atypical CBS in DIV13 cultured mouse cortical neurons across chromatin states annotated by ChIPseeker. C, D Top 10 gene ontology terms associated with genes closest to canonical CBS (C) or atypical CBS (D) found in DIV13 cultured mouse cortical neurons according to clusterProfiler GO. E PWMScan motif finding strategy was utilized to identify canonical and atypical CBS within a variety of P0 mouse tissues using CTCF ChIP-seq data obtained from ENCODE. Percentage of canonical and atypical CBS observed in all P0 mouse tissues (conserved), within each P0 mouse tissue (all), or only found within each P0 mouse tissue (unique) is shown. F Chromatin state distribution of canonical and atypical CBS that are conserved in all P0 tissues or are unique to the P0 mouse heart or P0 mouse kidney across chromatin states annotated by ChIPseeker. G Top 10 gene ontology terms associated with genes closest to CBS that are unique to the P0 mouse heart according to clusterProfiler GO. H Representative IGV genome browser view of CTCF binding at atypical CBS in two heart-specific transcripts, Tnnt2 (top), and Dhod3 (bottom), along with H3K27ac ChIP-seq signal traces from P0 mouse heart

We next utilized ChIPseeker to compare the distribution of canonical and atypical CBS within annotated regions of the mouse genome [37]. Canonical CBS were more broadly distributed across annotated genomic regions, but more than 50% of atypical CBS were found in gene promoters (Fig. 2B). Motifs underlying canonical and atypical CBS were similarly distributed genome-wide, suggesting that the occurrence of the motif sequence does not account for the enrichment of atypical CBS within gene promoters (Figure S2C). These results prompted us to investigate whether CTCF occupancy at promoters is defined by degeneracy at specific residues relative to the consensus motif. As mentioned above, both canonical and atypical CBS displayed progressive degeneracy at positions 4 and 6–9 when ranked and grouped into quartiles according to PWM score (Table 2). Surprisingly, ~ 50% of canonical CBS in promoters were found in canonical Q1, whose motif sequence was more similar to atypical CBS, while only ~ 10% canonical CBS in promoters were found in canonical Q4, which most closely resembled the consensus CTCF motif (Table 2, Figure S2D). By contrast, atypical CBS in promoters were more evenly distributed across atypical quartiles, though even here, more CBS were found in atypical Q1 compared to atypical Q4 (Figure S2D). Together, these results suggest that CTCF largely binds within promoters at sites that diverge from the consensus CTCF motif.

Atypical CBS Associate with Genes that Encode for Tissue-Specific Functions

To identify potential biological functions of canonical and atypical CBS, we utilized clusterProfiler to compare gene ontology for genes located in close proximity to each class of CBS [40]. Genes closest to canonical CBS were enriched for general gene ontology (GO) terms, and the few neuronal GO categories identified here, including “synapse organization” and “axonogenesis,” were also identified by analysis of CBS in non-neuronal tissues (Fig. 2C; and see below). By contrast, atypical CBS were highly enriched for GO terms related to neuronal functions, including “cognition,” “learning or memory,” and “forebrain development” (Fig. 2D), suggesting atypical CBS could play a larger role in cell-type/tissue-specific functions compared to canonical CBS.

To explore whether these observations are unique to cultured mouse cortical neurons or were generalizable across other cell types and tissues, we utilized publicly available ENCODE CTCF ChIP-seq data from a variety of P0 mouse tissues [39] (Table 1). PWMScan was again used to identify canonical and atypical CBS, and the intersect function in bedtools was used to further classify CBS as either tissue-specific or conserved/found in all tissues analyzed [33]. CBS within all P0 tissues partitioned similarly into canonical and atypical CBS as observed in cultured mouse cortical neurons (Fig. 2E). Canonical and atypical CBS found to be conserved in all P0 mouse tissues were enriched for DNA binding motifs that resembled those detected in cultured cortical neurons (Figure S2E), and conserved atypical CBS were highly enriched in gene promoters compared to conserved canonical CBS (Fig. 2F). Together, these results suggest that the alternative mode of CTCF binding within gene promoters described in cultured mouse cortical neurons is highly conserved.

Intriguingly, atypical CBS were also overrepresented among CBS that were unique to each P0 tissue compared to canonical CBS (Fig. 2E), further suggesting that atypical CTCF binding could mediate tissue-specific functions. GO analysis of CBS specific to the P0 heart using clusterProfiler revealed an enrichment of terms related to cardiac muscle development and differentiation (Fig. 2G), and GO analysis of CBS specific to the P0 kidney revealed an enrichment of terms related to urogenital development (Figure S2F). As mentioned above, CBS specific to the P0 kidney are also enriched for neuronal terms such as “synapse organization” and “axonogenesis,” highlighting the ubiquitous nature of these terms even in non-neuronal tissues (Fig. 2C, D, and S2F). These results indicate that tissue-specific CBS associate with genes encoding for tissue-specific functions. To further determine how the distribution of CBS within specific genomic regions relates to tissue-specific functions, we compared the patterns of conserved and tissue-specific CBS across various P0 tissues. Distributions of conserved and tissue-specific CBS showed minor variations across annotated genomic regions except for tissue-specific CBS in the forebrain and hindbrain, which were enriched in gene promoters compared to other tissues (Figure S2G). Following this, conserved and tissue-specific CBS were again grouped into canonical and atypical CBS, and their respective distributions across annotated chromatin states were compared. Tissue-specific atypical CBS were enriched within gene bodies, particularly in introns, and reduced within promoters compared to conserved atypical CBS (Fig. 2F). Similar trends were displayed by canonical CBS (Fig. 2F), although these constituted a minority of tissue-specific CBS (Fig. 2E). Intriguingly, overlaying CTCF ChIP-seq tracks with other epigenomic marks revealed that signals for H3K27ac, which marks cis-regulatory elements, including promoters and enhancers, were enriched at CBS within several tissue-specific genes in the heart (Fig. 2H) [47]. CTCF binding at cis-regulatory regions has been linked to either transcriptional activation or repression in a context-dependent manner [9, 12, 45, 46]. Based on these observations, we explored the disposition of H3K27ac signals at canonical and atypical CBS in the P0 heart greater detail (Figure S2H). These analyses revealed that compared to conserved CBS, both canonical and atypical CBS located within the bodies of tissue-specific genes show an enrichment of H3K27ac signals. Furthermore, within this context, H3K27ac signals were more highly enriched at atypical CBS compared to canonical CBS (Figure S2H). Together, these results support a potential regulatory role for CTCF binding, particularly at atypical CBS, in tissue-specific gene expression.

Finally, we used STREME to assess whether DNA binding motifs enriched within tissue-specific atypical CBS were similar across multiple tissues or displayed tissue-specific sequence variations. Significantly enriched DNA binding motifs retrieved for tissue-specific atypical CBS diverged substantially from the consensus motif, varied between tissues, and were only found within one-third of CBS (Table 3). Instead, most tissue-specific atypical CBS lacked specific DNA binding motifs and were enriched for variable GC-rich and AT-rich sequences. For instance, atypical CBS specific to the P0 heart were enriched for a single short, AT-rich motif (Table 3). Together, these results suggest that although conserved atypical CBS are enriched in promoters and are defined by degeneracy at specific nucleotides of the consensus CTCF motif, tissue-specific atypical CBS are enriched in gene bodies and bind to variable sequences that are largely unrelated to the consensus CTCF motif.

Table 3.

Highest enriched motif found via STREME analysis of atypical CBS that are unique to each P0 mouse tissue

graphic file with name nihms-2185987-t0006.jpg
Open in a new tab

Atypical CBS Anchor Chromatin Loops and TADs Together with Canonical CBS

Tissue-specific CTCF binding is previously shown to be important for facilitating cellular differentiation and development [21]. CTCF exhibits dynamic binding patterns across developmental stages, and CTCF depletion impedes differentiation of multiple cell types, including neurons, due to differential expression of tissue-specific genes [26, 48-50]. The role of CTCF binding in this context is widely attributed to CTCF’s ability to serve as an anchor for cohesin-mediated loop extrusion. In this model, structural maintenance of chromosomes (SMC) complexes, particularly cohesin, use ATP hydrolysis to reel and extrude chromatin and thereby generate DNA loops [5, 6]. Extrusion of chromatin is halted when cohesin encounters CTCF bound to inwardly oriented CTCF motifs [3, 4]. In addition to establishing ubiquitous and stable chromatin structures, CTCF-mediated chromatin looping is shown to promote transcriptional activation by either looping gene promoters together with other regulatory elements, such as enhancers, or by insulating these elements to inhibit transcriptional activation [10-12]. We therefore interrogated whether the ability of CTCF to participate in chromatin looping and facilitate tissue-specific gene expression is dependent on binding at a well-conserved canonical CTCF binding motif.

Because atypical CBS show degeneracy at key nucleotides that underlie CTCF’s ability to bind its cognate motifs in an oriented manner, we first explored whether differences in CTCF binding sequence alter the ability of CTCF to participate in chromatin loop formation in mature cortical neurons [16, 51]. To this end, we performed ChIP-seq for the cohesin subunit, SMC1, in cultured mouse cortical neurons. SMC1 signals were enriched at CBS, and the consensus CTCF motif was the highest enriched motif retrieved for SMC1 binding sites (Figure S3A and B). SMC1 signals were enriched at both canonical and atypical CBS (Fig. 3A). However, heatmap analysis revealed a gradual reduction in SMC1 signal intensity as a function of reduced PWM score for the consensus CTCF motif (Fig. 3A). SMC1 signals were also more focally enriched at canonical CBS with high PWM scores, while canonical CBS with low PWM scores and atypical CBS showed more dispersed SMC1 signals in their vicinity (Fig. 3A). To further understand how CTCF occupancy at canonical and atypical CBS relates to its ability to anchor chromatin loops and TADs, publicly available Hi-C data from cultured mouse cortical neurons was analyzed using the JUICER suite to call loops and TADs [30]. CTCF ChIP-seq peaks were then overlaid onto Hi-C maps to identify loops and TADs that were anchored by CTCF. This analysis revealed that ~ 51% of canonical CBS co-localized with chromatin loop anchors (Fig. 3B). Remarkably, ~ 28% of atypical CBS also co-localized with chromatin loop anchors (Fig. 3B). To further understand the nature of looping interactions anchored by canonical and atypical CBS, chromatin loops were classified as cohesin-mediated (structural loops anchored by CTCF and cohesin), gene regulatory (CTCF-bound loop anchors that contain promoter/enhancer-related chromatin marks), or other (including loops anchored by CTCF or enriched for regulatory chromatin marks at only one anchor). Similar proportions of canonical and atypical CBS were enriched at gene regulatory loop anchors (12.4% canonical versus 10.5% atypical). However, more canonical CBS were found to anchor cohesin-mediated loops relative to atypical CBS (15.9% canonical versus 7% atypical; Fig. 3B), and atypical CBS engaged in fewer looping interactions overall (Fig. 3B and C). Together, these results indicate that both canonical and atypical CBS can anchor chromatin loops and TADs. However, structural, cohesin-mediated loops are largely anchored by canonical CBS in cultured mouse cortical neurons.

Fig. 3.

Fig. 3

Open in a new tab

Assessing the role of canonical and atypical CBS in chromatin architecture. A We performed SMC1 ChIP-seq in DIV13 cultured mouse cortical neurons to assess the degree of SMC1 and CTCF co-occupancy in the genome. Average SMC1 signal intensity surrounding canonical and atypical CBS (top left) with quantification of median SMC1 signal intensity (right, Wilcoxon signed-rank test, p < 0.0001) are shown. Additionally, the score function of PWMScan was used to rank canonical and atypical CBS from highest to lowest PWM score (low scores—low similarity; high scores—high similarity to MA0139.1). A heatmap of SMC1 signal intensity within each interval ranked from highest to lowest PWM score is shown (bottom). B Previously published Hi-C data from cortical neurons was used to assess the role of canonical and atypical CBS for chromatin looping. All chromatin loops identified from Hi-C data were segregated into three categories: loops anchored by CTCF and cohesin on both sides (cohesin), loops anchored on both sides by CTCF bound at promoter and/or enhancer-related chromatin marks according to ChromHMM (regulatory; see methods for details), or loops that are anchored by CTCF and/or cohesin on only one side or are anchored at regulatory regions on only one side (other). The percentage of all canonical or atypical CBS that fall within each of these loop categories or do not overlap with any loop anchors (none) is shown. C Fraction of canonical or atypical CBS that participate in one or more chromatin loops in cortical neurons. D Previously published Hi-C and CTCF ChIP-seq data from three stages of neural differentiation—embryonic stem cells (ESCs), neural progenitor cells (NPCs), and cortical neurons (CNs)—were used to assess the composition of topologically associating domain (TAD) boundary pairs in each cell type. The percentage of TADs that are anchored at both boundaries by canonical or atypical CBS, anchored by a canonical CBS and an atypical CBS at each boundary (canonical-atypical), anchored at both boundaries by both canonical and atypical CBS (both), or do not have CTCF bound at either boundary (no CTCF) is shown. E Representative IGV genome browser view of chromatin loops in ESCs (top) and CNs (bottom) anchored at canonical and atypical CBS in each cell type. F Percentage of canonical or atypical CBS that are constitutively bound, gained, or lost across differentiation of ESCs into NPCs or NPCs into CNs. G Percentage of dynamic (gained/lost) canonical or atypical CBS that overlap with loop anchors/TAD boundaries or are found elsewhere in the genome in NPCs or CNs

To further characterize the engagement of canonical and atypical CBS at chromatin loops and TADs, we analyzed publicly available CTCF ChIP-seq and Hi-C data from three stages of mouse neural differentiation—embryonic stem cells (mESCs), neural progenitors (NPCs), and cortical neurons (CNs) [30]. Canonical and atypical CBS in NPCs and CNs distributed similarly to what was observed in our DIV13 cultured mouse cortical neurons, while mESCs contained relatively elevated levels of atypical CBS (Figure S3C). Intriguingly, a majority of TAD boundaries in mESCs (56.2%) contained both canonical and atypical CBS within the same boundary, and neural differentiation resulted in a dramatic reduction of such TAD boundaries (14.7% in NPCs and 12.2% in CNs; Fig. 3D). By contrast, neural differentiation resulted in an increase in TAD boundaries that exclusively contained either canonical or atypical CBS (Fig. 3D). NPCs and CNs also contained elevated fractions of TAD boundaries lacking CBS altogether (Fig. 3D). Finally, a small but significant fraction of TAD boundaries seemed to be anchored by paired canonical/atypical CBS (Fig. 3D). Analysis of CBS at chromatin loop anchors as a function of neural differentiation showed similar trends to those observed at TAD boundaries (Figure S3D). Notably, we observed a similar reduction in atypical CBS at loop anchors upon differentiation of mESCs, and whereas most loops were anchored by canonical CBS, ~ 7–9% of loops were anchored by paired canonical to atypical CBS across neural differentiation (Fig. 3E and S3D).

Consistent with the loss of atypical CBS detected at TAD boundaries and loop anchors (Fig. 3D and S3D), atypical CBS were overrepresented among CBS that exhibited dynamic occupancy across differentiation from mESCs to NPCs and from NPCs to CNs when compared to CBS that were constitutively bound (Fig. 3F). To further explore DNA sequence variations between constitutive and dynamic CBS during differentiation, we again used the score function of PWMScan to score and segregate canonical and atypical CBS within each cell type into quartiles based on their similarity to the consensus CTCF motif (Q1—low similarity; Q4—high similarity). A degeneracy at similar nucleotides (positions 4 and 6–9) within the consensus CTCF motif was observed across quartiles of canonical and atypical CBS as was described for DIV13 cultured mouse cortical neurons (Table 2; data not shown). CBS that were constitutively bound across differentiation were enriched within the highest scored quartile, canonical Q4, indicating that similarity to the consensus CTCF motif correlates with stability of CTCF occupancy across differentiation (Figure S3E). By contrast, dynamic CBS were enriched within the lowest scored quartiles, canonical and atypical Q1 and Q2, particularly for CBS that were gained across differentiation (Figure S3E).

Finally, to understand whether the potential functions of dynamic CTCF occupancy across neural differentiation are primarily associated with changes in chromatin architecture, we quantified the fraction of dynamic CBS that co-localize with TAD boundaries and chromatin loop anchors relative to other genomic sites. Less than half of dynamic CBS co-localized with either chromatin loop anchors or TAD boundaries during differentiation from mESCs to NPCs (38%) or NPCs to CNs (41%; Fig. 3G). Additionally, atypical CBS were equally prevalent among dynamic CBS at loop anchors/TAD boundaries and other genomic sites (Fig. 3G). Together, analysis of CTCF occupancy at TADs and chromatin loops across neural differentiation revealed the following: (1) both canonical and atypical CBS co-localize with SMC1, TAD boundaries, and chromatin loop anchors in all cell types, but the majority of stable chromatin structures are bound by canonical CBS; (2) deviation from the consensus CTCF motif correlates with dynamic CTCF occupancy patterns across neural differentiation wherein differentiation results in the preferential re-localization of atypical CBS; and (3) neural differentiation involves dynamic changes to CBS genome-wide, and not just at TAD boundaries and loop anchors. These results indicate how stable chromatin structures could be established during development and highlight a tissue-specific role for atypical CBS.

Stimulus-Driven CTCF Occupancy Changes Are Enriched for Atypical CBS in Neurons

Although previous efforts have also attempted to address the role of CTCF dynamics for cellular differentiation and development, the role of dynamic CTCF binding for cellular state changes that occur in post-mitotic cells, such as neuronal stimulation, has not been thoroughly examined. Several studies have reported cognitive deficits following loss of CTCF in post-mitotic neurons, including impaired synapse formation and deficits in learning and memory [23-25]. CTCF is bound at the promoter regions of many stimulus-responsive genes involved in these processes, and it is hypothesized that CTCF could play an important role in the expression of these genes following neuronal stimulation [23-25]. Whether the ability of CTCF to facilitate these processes relies on stable or dynamic CTCF binding and the extent to which associated changes in chromatin architecture are required for these processes is unclear, as how CTCF occupancy is altered following neuronal activity remains poorly characterized.

To address these issues, we performed CTCF ChIP-seq in cultured cortical neurons that were treated with the glutamate receptor agonist, NMDA (50 μM for 10 min, followed by washout and recovery for 50 min). Comparison of CBS between NMDA-treated neurons and untreated controls revealed changes in CTCF occupancy at thousands of sites—3117 CBS were lost, and 5618 CBS were gained in NMDA-treated neurons compared to controls. To determine how CTCF dynamics under these conditions relates to neuronal functions, GO analysis of dynamic and constitutive CBS was performed using clusterProfiler. Constitutive CBS in NMDA-treated neurons were enriched for terms related to neurogenesis, actin filament organization, and Wnt signaling (Fig. 4A). However, dynamic CBS in NMDA-treated neurons were enriched for terms such as “learning or memory,” “cognition,” and “dendrite development,” indicating association with genes encoding for experience-driven synaptic plasticity and adaptive behaviors (Fig. 4B). Thus, neuronal stimulation is accompanied by CTCF occupancy changes at sites relevant to stimulus-responsive neuronal functions. For instance, dynamic CTCF occupancy at both canonical and non-canonical CBS was observed in the prototypical neuronal activity-induced gene, Bdnf (Fig. 4C). CTCF binding within the Bdnf locus was previously shown to overlap with elevated H3K27ac signals following neuronal stimulation, and loss of CTCF led to reduced induction of Bdnf and other stimulus-responsive genes in neurons [23, 25, 52]. These results suggest that dynamic CTCF occupancy within specific stimulus-responsive genes could activate their transcription.

Fig. 4.

Fig. 4

Open in a new tab

Characterizing stimulus-dependent CTCF occupancy changes in primary cortical neurons. A, B CTCF ChIP-seq was performed in DIV13 cultured mouse cortical neurons stimulated with 50 μM NMDA for 10 min followed by 50-min recovery in drug free media. The top 10 gene ontology terms associated with genes closest to CBS that are constitutively bound (A) or dynamic (B) following NMDA treatment relative to untreated controls are shown. C Representative IGV genome browser view of dynamic (gained/lost) CTCF occupancy at canonical and atypical CBS in the stimulus-responsive transcript Bdnf. D Percentage of canonical or atypical CBS that are constitutively bound in control and NMDA or are gained/lost following NMDA treatment. E The score function of PWMScan was used to calculate average scores for CBS that are constitutively bound in control and NMDA or are gained/lost following NMDA treatment. Higher scores indicate higher similarity to the canonical CTCF motif (MA0139.1). F, G Average SMC1 signal intensity surrounding all canonical and atypical CBS (F) or surrounding dynamic (gained/lost) canonical and atypical CBS (G) in NMDA-treated neurons. Aggregate plot of average SMC1 signal intensity (top left) with quantification of median SMC1 signal intensity (right, Wilcoxon signed-rank test, p < 0.0001) is shown. Additionally, the score function of PWMScan was used to rank canonical and atypical CBS from highest to lowest PWM score (higher scores = higher similarity to the canonical CTCF motif MA0139.1) A heatmap of SMC1 signal intensity within each interval ranked from highest to lowest PWM score is shown (bottom)

Based on our observations described above, we wondered whether dynamic CTCF occupancy patterns in NMDA-treated neurons also involve CTCF binding at atypical CBS. Untreated controls and NMDA-treated neurons showed similar distributions of canonical and atypical CBS (Figure S4A). However, atypical CBS were enriched among CBS that were either gained or lost in NMDA-treated neurons compared to their proportions among constitutive CBS (Fig. 4D). Additionally, average CTCF signal intensity at dynamic CBS was lower than that at constitutive CBS (Figure S4B). To further examine how differences in underlying DNA motifs relate to CTCF dynamics, we computed PWM scores for constitutive and dynamic CBS based on similarity to the consensus CTCF motif. Average PWM scores for dynamic CBS were lower than those for constitutive CBS in NMDA-treated neurons, indicating dynamic CTCF occupancy patterns occur at sites that diverge significantly from the consensus CTCF motif (Fig. 4E). Next, PWM score was used to rank and group constitutive and dynamic CBS into quartiles based on their similarity to the consensus CTCF motif (Q1—low similarity; Q4—high similarity). A degeneracy at similar nucleotides (positions 4 and 6–9) within the consensus CTCF motif was observed across quartiles of canonical and atypical CBS as was described above (Table 2; data not shown). Whereas constitutive canonical and atypical CBS were more evenly distributed across PWM score quartiles, dynamic canonical CBS were more highly enriched in quartiles with lower PWM scores, including Q1 and Q2 (Figure S4C). These results suggest that variations at specific nucleotide positions within the consensus CTCF motif correlate with dynamic changes in their occupancy in response to neuronal stimulation.

Finally, to understand how dynamic CTCF occupancy patterns relate to changes in chromatin architecture, we performed SMC1 ChIP-seq in NMDA-treated neurons. As described above, SMC1 signals were enriched at CBS in untreated neurons but showed a gradual reduction in intensity with decreasing PWM score (Fig. 3A). Following NMDA treatment, we observed a reduction in both the number of high-confidence SMC1 peaks, as well as in SMC1 signal intensity at CBS (Figs. 3A and 4F). SMC1 binding sites that remained in NMDA-treated neurons were enriched in promoter-proximal regions (Figure S4D). SMC1 signals were also retained at dynamic atypical CBS (Figure S4E), suggesting the possibility of a stimulus-driven association between SMC1 and atypical CBS within gene promoters. Indeed, atypical CBS showed relatively elevated SMC1 signals compared to canonical CBS in NMDA-treated neurons (Fig. 4F), and this discrepancy was particularly apparent for dynamic CBS (Fig. 4G). Thus, dynamic atypical CBS preferentially nucleate SMC1 compared to dynamic canonical CBS following neuronal stimulation. These data suggest that although SMC1 occupancy is dynamic following neuronal stimulation, cohesin enrichment is preserved or even strengthened at promoter-proximal atypical CBS. These results suggest that cohesin and CTCF dynamics at atypical CBS could play important roles in neuronal activity-dependent reorganization of chromatin loops and gene activity states.

Discussion

In this study, we profiled CTCF occupancy patterns in cultured mouse cortical neurons as a function of neuronal activity and found that ~ 22% of CBS lacked the consensus CTCF motif. Although the binding of CTCF to variable DNA sequences has been described previously [1, 13, 19, 20], our analysis of multiple cell types and systems that has been lacking in other studies has uncovered new characteristics of these atypical CBS that provide insights into how CTCF could mediate its diverse functions. We observe that divergence from the consensus CTCF motif among CBS is not random but involves degeneracy of specific nucleotides within the 20-bp core motif. These variations likely affect the interaction between CTCF ZFs 6 and 7 and the DNA. Of CTCF’s 11 ZF domains, ZFs 4–7 engage the core CTCF motif [1, 15, 16, 18]. Although mutations in these ZFs markedly reduced CTCF occupancy at canonical CBS, CTCF occupancy at atypical CBS was unaffected [16]. These results suggest that CTCF is recruited/stabilized at atypical CBS through protein–protein interactions with other factors, with some engagement of ZFs 4 and 5.

Although it is unclear why CBS are defined by degeneracy at specific nucleotides of the core CTCF motif, the need to anchor CTCF at atypical CBS could allow its occupancy to be dynamically regulated in several contexts. For instance, most promoter-proximal CTCF binding occurs at sequences that diverge significantly from the consensus motif, and this feature could allow promoter-enhancer contacts to either be strengthened or weakened by altering CTCF occupancy at specific genes. Our observations that dynamic CBS following neuronal stimulation are also enriched for atypical and promoter-bound CBS further support this idea. Similarly, CBS with the lowest resemblance to the consensus CTCF motif were enriched among dynamic CBS during differentiation of mESCs, in line with previous reports demonstrating degeneracy at positions 4, 6–9, and 18 of the consensus CTCF motif [20]. Precisely how CTCF at atypical CBS is dynamically altered in these scenarios is poorly understood. One possibility is that these changes are governed by the dynamic occupancy of CTCF interacting partners that recruit/stabilize CTCF at atypical CBS. On the other hand, recent single-molecule experiments have revealed that DNA tension can directly affect the occupancy of CTCF and the various outcomes of CTCF-loop collisions, including loop bypass, anchoring, direction switching, and loop shrinkage [53]. In eukaryotic cells, supercoiling generated by transcribing RNA polymerases is a major source of DNA tension and could couple transcriptional changes to dynamic CTCF occupancy patterns [54]. Whereas the effects of DNA tension in these studies were assayed at CBS that closely resemble the consensus motif, our data suggest that testing how DNA tension affects CTCF occupancy and function at sequences within atypical CBS could be particularly informative.

In addition to atypical CBS that carry sequences partially matching the consensus CTCF motif, we identified a second class of atypical CBS that largely contain unrelated DNA sequences and define CTCF occupancy at tissue-specific sites across P0 mouse tissues. Furthermore, compared to conserved atypical CBS, which were enriched within gene promoters, tissue-specific CBS were primarily enriched within putative intragenic regulatory regions of tissue-specific genes. It is unclear whether the intragenic enrichment of CBS is crucial for the expression of tissue-specific genes. However, CTCF clustering within tissue-specific genes could drive their inclusion in phase-separated transcriptional condensates and thereby facilitate their transcription [55, 56].

Many functions of CTCF, including enhancer blocking, insulation of chromatin domains, and gene regulation are attributed to its ability to anchor and stabilize chromatin loops [10, 11]. Although our analysis indicates that atypical CBS are less likely to anchor stable chromatin loops that are detected in Hi-C compared to canonical CBS, we also observe that atypical CBS co-localize with SMC1 and can anchor both cohesin-mediated and gene regulatory loops. In fact, atypical CBS that display dynamic occupancy patterns following neuronal stimulation are able to nucleate more SMC1 compared to canonical CBS. These results raise the possibility that dynamic changes in the occupancy of atypical CBS could trigger stimulus-responsive changes in chromatin topology that in turn establish stimulus-dependent gene activity states. Such a scenario is supported by our observations that dynamic CBS following neuronal stimulation are associated with genes that encode for experience-driven adaptive behaviors, such as learning, memory formation, and cognition. Although acute CTCF depletion in post-mitotic neurons caused deficits in synaptic plasticity and memory formation in mice, relatively few genes were differentially expressed under basal conditions following CTCF depletion [23, 25]. Furthermore, enhancer-promoter interactions and transcriptional patterns were largely unchanged following acute CTCF loss [9, 57]. Based on these observations, we propose that dynamic CTCF occupancy changes may not affect gene expression patterns at short time scales but could instead prime chromatin topological changes that establish gene activity states over longer time scales and influence transcriptional outputs following recurrent neuronal activity. Investigations into how CTCF dynamics relates to chromatin architecture and activity-dependent gene expression at longer time scales could further clarify this model.

Supplementary Material

Supplementary Figures

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s12035-023-03762-5.

Acknowledgements

We would like to thank our colleges in the Madabhushi Lab, Lance Heady, Lahiri Konada, and Ilse Delint-Ramirez, for their helpful feedback on this manuscript.

Funding

This work was supported by NIMH under grant number MH120132 (RM) and CPRIT under grant number RP210041(MC).

Footnotes

Ethics Approval This study was performed in line with the principles outlined in the “Guide for the Care and Use of Laboratory Animals” and in accordance with all federal and state policies. Approval was granted by the UT Southwestern Institutional Care and Use Committee (IACUC).

Competing Interests The authors declare no competing interests.

Data Availability

Original data produced for this manuscript, including raw and processed CTCF and SMC1 ChIP-seq data, are openly available from the NCBI GEO database at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE235972, reference number GSE235972.

References

  • 1.Kim TH, Abdullaev ZK, Smith AD et al. (2007) Analysis of the vertebrate insulator protein CTCF-binding sites in the human genome. Cell 128:1231–1245. 10.1016/j.cell.2006.12.048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Chen H, Tian Y, Shu W et al. (2012) Comprehensive identification and annotation of cell type-specific and ubiquitous CTCF-binding sites in the human genome. PLoS ONE 7:e41374. 10.1371/journal.pone.0041374 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Rubio ED, Reiss DJ, Welcsh PL et al. (2008) CTCF physically links cohesin to chromatin. PNAS 105:8309–8314. 10.1073/pnas.0801273105 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Pugacheva EM, Kubo N, Loukinov D et al. (2020) CTCF mediates chromatin looping via N-terminal domain-dependent cohesin retention. PNAS 117:2020–2031. 10.1073/pnas.1911708117 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Davidson IF, Bauer B, Goetz D et al. (2019) DNA loop extrusion by human cohesin. Science 366:1338–1345. 10.1126/science.aaz3418 [DOI] [PubMed] [Google Scholar]
  • 6.Kim Y, Shi Z, Zhang H et al. (2019) Human cohesin compacts DNA by loop extrusion. Science 366:1345–1349. 10.1126/science.aaz4475 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hansen AS (2020) CTCF as a boundary factor for cohesin-mediated loop extrusion: evidence for a multi-step mechanism. Nucleus 11:132–148. 10.1080/19491034.2020.1782024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Nora EP, Goloborodko A, Valton AL et al. (2017) Targeted degradation of CTCF decouples local insulation of chromosome domains from genomic compartmentalization. Cell 169:930–944.e22. 10.1016/j.cell.2017.05.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Aljahani A, Hua P, Karpinska MA et al. (2022) Analysis of sub-kilobase chromatin topology reveals nano-scale regulatory interactions with variable dependence on cohesin and CTCF. Nat Commun 13:1–13. 10.1038/s41467-022-29696-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Braccioli L, De Wit E (2019) CTCF: A Swiss-army knife for genome organization and transcription regulation. Essays Biochem 63:157–165. 10.1042/EBC20180069 [DOI] [PubMed] [Google Scholar]
  • 11.Kim S, Yu NK, Kaang BK (2015) CTCF as a multifunctional protein in genome regulation and gene expression. Exp Mol Med 47:1–5. 10.1038/emm.2015.33 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Kubo N, Ishii H, Xiong X et al. (2021) Promoter-proximal CTCF binding promotes distal enhancer-dependent gene activation. Nat Struct Mol Biol 28:152–161. 10.1038/s41594-020-00539-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Ohlsson R, Renkawitz R, Lobanenkov V (2001) CTCF is a uniquely versatile transcription regulator linked to epigenetics and disease. Trends Genet 17:520–527. 10.1016/S0168-9525(01)02366-6 [DOI] [PubMed] [Google Scholar]
  • 14.Filippova GN, Qi C-F, Ulmer JE et al. (2002) Tumor-associated zinc finger mutations in the CTCF transcription factor selectively alter its DNA-binding specificity. Cancer Res 62:48–52 [PubMed] [Google Scholar]
  • 15.Renda M, Baglivo I, Burgess-Beusse B et al. (2007) Critical DNA binding interactions of the insulator protein CTCF: a small number of zinc fingers mediate strong binding, and a single finger-DNA interaction controls binding at imprinted loci. J Biol Chem 282:33336–33345. 10.1074/jbc.m706213200 [DOI] [PubMed] [Google Scholar]
  • 16.Nakahashi H, Kwon KRK, Resch W et al. (2013) A genome-wide map of CTCF multivalency redefines the CTCF code. Cell Rep 3:1678–1689. 10.1016/j.celrep.2013.04.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Yin M, Wang J, Wang M et al. (2017) Molecular mechanism of directional CTCF recognition of a diverse range of genomic sites. Cell Res 27:1365–1377. 10.1038/cr.2017.131 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Guo J, Li N, Han J et al. (2018) DNA recognition patterns of the multi-zinc-finger protein CTCF: a mutagenesis study. Acta Pharm Sin B 8:900–908. 10.1016/j.apsb.2018.08.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Essien K, Vigneau S, Apreleva S et al. (2009) CTCF binding site classes exhibit distinct evolutionary, genomic, epigenomic and transcriptomic features. Genome Biol 10:1–15. 10.1186/gb-2009-10-11-r131 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Plasschaert RN, Vigneau S, Tempera I et al. (2014) CTCF binding site sequence differences are associated with unique regulatory and functional trends during embryonic stem cell differentiation. Nucleic Acids Res 42:774–789. 10.1093/nar/gkt910 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Arzate-Mejía RG, Recillas-Targa F, Corces VG (2018) Developing in 3D: the role of CTCF in cell differentiation. Development 145:1–16. 10.1242/dev.137729/48856 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Moore JM, Rabaia NA, Smith LE et al. (2012) Loss of maternal CTCF Is associated with peri-implantation lethality of Ctcf null embryos. PLoS ONE 7:e34915. 10.1371/journal.pone.0034915 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Kim S, Yu NK, Shim KW et al. (2018) Remote memory and cortical synaptic plasticity require neuronal CCCTC-binding factor (CTCF). J Neurosci 38:5042–5052. 10.1523/jneurosci.2738-17.2018 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Il CD, Kim M, Kim S et al. (2021) Conditional knock out of transcription factor CTCF in excitatory neurons induces cognitive deficiency. Mol Brain 14:1–4. 10.1186/s13041-020-00716-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Sams DS, Nardone S, Getselter D et al. (2016) Neuronal CTCF is necessary for basal and experience-dependent gene regulation, memory formation, and genomic structure of BDNF and Arc. Cell Rep 17:2418–2430. 10.1016/j.celrep.2016.11.004 [DOI] [PubMed] [Google Scholar]
  • 26.Watson LA, Wang X, Elbert A et al. (2014) Dual effect of CTCF loss on neuroprogenitor differentiation and survival. J Neurosci 34:2860–2870. 10.1523/jneurosci.3769-13.2014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Landt SG, Marinov GK, Kundaje A et al. (2012) ChIP-seq guidelines and practices of the ENCODE and modENCODE consortia. Genome Res 22:1813–1831. 10.1101/gr.136184.111 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Zhang Y, Liu T, Meyer CA et al. (2008) Model-based analysis of ChIP-Seq (MACS). Genome Biol 9:1–9. 10.1186/gb-2008-9-9-r137 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Li Q, Brown JB, Huang H, Bickel PJ (2011) Measuring reproducibility of high-throughput experiments. Ann Appl Stat 5:1752–1779. 10.1214/11-aoas466 [DOI] [Google Scholar]
  • 30.Bonev B, Mendelson Cohen N, Szabo Q et al. (2017) Multiscale 3D genome rewiring during mouse neural development. Cell 171:557–572.e24. 10.1016/j.cell.2017.09.043 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Durand NC, Shamim MS, Machol I et al. (2016) Juicer provides a one-click system for analyzing loop-resolution Hi-C experiments. Cell Syst 3:95–98. 10.1016/j.cels.2016.07.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Ma W, Noble WS, Bailey TL (2014) Motif-based analysis of large nucleotide data sets using MEME-ChIP. Nat Protoc 9:1428–1450. 10.1038/nprot.2014.083 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Quinlan AR, Hall IM (2010) BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26:841–842. 10.1093/bioinformatics/btq033 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Fornes O, Castro-Mondragon JA, Khan A et al. (2020) JASPAR 2020: update of the open-access database of transcription factor binding profiles. Nucleic Acids Res 48:D87–D92. 10.1093/nar/gkz1001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Ambrosini G, Groux R, Bucher P (2018) PWMScan: a fast tool for scanning entire genomes with a position-specific weight matrix. Bioinformatics 34:2483–2484. 10.1093/bioinformatics/bty127 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Fu S, Wang Q, Moore JE et al. (2018) Differential analysis of chromatin accessibility and histone modifications for predicting mouse developmental enhancers. Nucleic Acids Res 46:11184–11201. 10.1093/NAR/GKY753 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Yu G, Wang LG, He QY (2015) ChIPseeker: an R/Bioconductor package for ChIP peak annotation, comparison and visualization. Bioinformatics 31:2382–2383. 10.1093/bioinformatics/btv145 [DOI] [PubMed] [Google Scholar]
  • 38.Ernst J (2017) Kellis M (2017) Chromatin-state discovery and genome annotation with ChromHMM. Nat Protoc 1212(12):2478–2492. 10.1038/nprot.2017.124 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Abascal F, Acosta R, Addleman NJ et al. (2020) Expanded encyclopaedias of DNA elements in the human and mouse genomes. Nature 583:699–710. 10.1038/s41586-020-2493-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wu T, Hu E, Xu S et al. (2021) clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innov 2:100141. 10.1016/j.xinn.2021.100141 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Ramírez F, Dündar F, Diehl S et al. (2014) deepTools: a flexible platform for exploring deep-sequencing data. Nucleic Acids Res 42:W187–W191. 10.1093/NAR/GKU365 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Kuhn RM, Haussler D, James Kent W (2013) The UCSC genome browser and associated tools. Brief Bioinform 14:144–161. 10.1093/bib/bbs038 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Weth O, Renkawitz R (2011) CTCF function is modulated by neighboring DNA binding factors. Biochem Cell Biol 89:459–468. 10.1139/O11-033 [DOI] [PubMed] [Google Scholar]
  • 44.Ohlsson R, Lobanenkov V, Klenova E (2010) Does CTCF mediate between nuclear organization and gene expression? BioEssays 32:37–50. 10.1002/bies.200900118 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Ya Guo A, Xu Q, Canzio D et al. (2015) CRISPR inversion of CTCF sites alters genome topology and enhancer/promoter function. Cell 162:900–910. 10.1016/j.cell.2015.07.038 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Ren G, Jin W, Cui K et al. (2017) CTCF-mediated enhancer-promoter interaction is a critical regulator of cell-to-cell variation of gene expression. Mol Cell 67:1049–1058. 10.1016/j.molcel.2017.08.026 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Song Y, Ahn J, Suh Y et al. (2013) Identification of novel tissue-specific genes by analysis of microarray databases: a human and mouse model. PLoS ONE 8:e64483. 10.1371/JOURNAL.PONE.0064483 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Dubois-Chevalier J, Staels B, Lefebvre P, Eeckhoute J (2015) The ubiquitous transcription factor CTCF promotes lineage-specific epigenomic remodeling and establishment of transcriptional networks driving cell differentiation. Nucleus 6:15–18. 10.1080/19491034.2015.1004258 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Balakrishnan SK, Witcher M, Berggren TW, Emerson BM (2012) Functional and molecular characterization of the role of CTCF in human embryonic stem cell biology. PLoS ONE 7:e42424. 10.1371/journal.pone.0042424 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Phillips-Cremins JE, Sauria MEG, Sanyal A et al. (2013) Architectural protein subclasses shape 3D organization of genomes during lineage commitment. Cell 153:1281–1295. 10.1016/j.cell.2013.04.053 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Nora EP, Caccianini L, Fudenberg G et al. (2020) Molecular basis of CTCF binding polarity in genome folding. Nat Commun 11:1–13. 10.1038/s41467-020-19283-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Beagan JA, Pastuzyn ED, Fernandez LR et al. (2020) Three-dimensional genome restructuring across timescales of activity-induced neuronal gene expression. Nat Neurosci 23:707–717. 10.1038/s41593-020-0634-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Davidson IF, Barth R, Zaczek M et al. (2023) CTCF is a DNA-tension-dependent barrier to cohesin-mediated loop extrusion. Nature 616:822–827. 10.1038/s41586-023-05961-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Corless S, Gilbert N (2016) Effects of DNA supercoiling on chromatin architecture. Biophys Rev 8:245–258. 10.1007/s12551-016-0210-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Lee R, Kang MK, Kim YJ et al. (2022) CTCF-mediated chromatin looping provides a topological framework for the formation of phase-separated transcriptional condensates. Nucleic Acids Res 50:207–226. 10.1093/NAR/GKAB1242 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Chowdhary S, Kainth AS, Paracha S et al. (2022) Inducible transcriptional condensates drive 3D genome reorganization in the heat shock response. Mol Cell 82:4386–4399.e7. 10.1016/j.molcell.2022.10.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Hsieh THS, Cattoglio C, Slobodyanyuk E et al. (2022) Enhancer–promoter interactions and transcription are largely maintained upon acute loss of CTCF, cohesin, WAPL or YY1. Nat Genet 54:1919–1932. 10.1038/s41588-022-01223-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Figures
NIHMS2185987-supplement-Supplementary_Figures.pdf (3.7MB, pdf)

Data Availability Statement

Original data produced for this manuscript, including raw and processed CTCF and SMC1 ChIP-seq data, are openly available from the NCBI GEO database at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE235972, reference number GSE235972.

RESOURCES