Effects of topological domain disruption on transcriptional regulation are chromatin context dependent

Abstract

Background

Three-dimensional genome organization helps coordinate enhancer-promoter communication while insulating loci from inappropriate regulatory contacts. CTCF and cohesin contribute to this organization by forming topologically associating domains. However, how boundary elements at individual loci influence transcription remains context dependent.

Results

We investigated the conserved topological organization of the mammalian NOTCH1 locus. Across human cell types, NOTCH1 resides within a defined topologically associated domain with CTCF/cohesin occupancy at both 5’ and 3’ boundaries. In human K562 cells, CRISPR-Cas9 deletion of boundary CTCF sites increased transcription of NOTCH1 and the intradomain non-coding transcripts NALT1 and LINC01451. Boundary perturbations impaired proliferation and clonogenic growth. Chromatin conformation profiling revealed defects in domain insulation and a redistribution of regulatory contacts between NOTCH1 promoter and enhancers within the domain. Cross-species analyses showed that domain architecture is conserved in mouse, yet transcriptional and phenotypic effects associated to domain boundary disruption were cell-type specific and correlated with differential chromatin contexts.

Conclusions

CTCF-dependent boundary integrity at the NOTCH1 locus tunes transcriptional output and cellular phenotypes in a chromatin context-dependent manner, supporting a model in which conserved 3D architecture constrains regulatory communication but yields distinct outcomes across cellular states.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13072-026-00660-7.

Introduction

In eukaryotes, genetic information is organized in multiple chromosomes, and it is highly organized inside the nucleus, and within it, important nuclear processes such as replication, DNA repair and transcription take place. We know now that these processes are possible thanks to the controlled compaction of chromatin and its three-dimensional organization within the nucleus [2, 6, 8, 15]. In this sense, the organization of chromatin within the nucleus occurs at different levels, from the clustering of nucleosomes to the arrangement of chromosomes in the nuclear space, and each level of organization is involved in transcriptional regulation [30, 47].

Cis-regulatory elements, such as promoters and enhancers, communicate spatially to maintain proper gene expression, which is partly regulated by the topological organization of the genome, through chromatin folding that allows to physically approach these elements that could be linearly distal [24]. Growing experimental evidence suggests that the establishment of chromatin loops allows not only contacting enhancers with their target promoters, but also creating regions of regulatory isolation, such as contact domains that allow fine control of gene expression of the genes contained within [7].

In mammals, the best characterized proteins involved in the formation of chromatin loops are CTCF and cohesin, which are currently referred to as architectural proteins. Of these proteins, CTCF is the only one that binds directly to DNA via its 11 zinc fingers, and tens of thousands of binding sites have been observed in the mammalian genome, most of which are enriched at so-called borders of topological domain [44]. The elimination of CTCF during the embryonic development of various organisms such as zebrafish and mouse is lethal and its elimination or decrease in different cell types or organs results in defects in proliferation and differentiation [4, 42]. Likewise, it has been described that the CTCF binding sites are frequently mutated in cancer [31].

It has been shown that three-dimensional genome organization can contribute to transcriptional regulation, and that deletion of boundary elements between topological domains can lead to inappropriate gene regulation within or across domains [28, 38]. Acute CTCF depletion via an auxin-inducible degron system in mouse embryonic stem cells (mESCs) can strongly alter chromatin folding (weakening of loops/TAD) and is accompanied by transcriptional changes for a subset of genes [42]. Similarly, perturbation of cohesin subunits profoundly affects chromatin architecture, whereas the magnitude of steady-state gene expression changes is typically limited for most genes [46, 55]. In Drosophila, extensive chromosomal rearrangements that alter TAD organization can have limited effects on gene expression [23]. In mammals, the transcriptional impact of architectural perturbations can also be modest at some loci (e.g., Sox9-Kcnj2 in mouse) [17], although other boundary disruptions and structural variants can drive ectopic enhancer-gene regulation and disease [38]. As well as removing CTCF or cohesin for only 3 h affects the expression of a small number of genes and does not have a large effect on enhancer-promoter interactions, although it does affect the contacts between topological domain boundaries [12, 29]. While previous studies have demonstrated that TAD boundaries function to insulate genes from inappropriate interactions with distal enhancers or silencers, the functional impact of these boundaries on gene expression remains complex and context dependent.

Given this evidence, and since the relationship between transcriptional regulation and the three-dimensional organization of the genome is currently not clear, we decided to address this question using the NOTCH1 gene as a model locus, since we previously demonstrated that in Drosophila the Notch gene is organized in two topological domains and perturbing the boundaries of these domains results in a decrease in their expression levels [3]. Additionally, it has been shown that NOTCH1, the mammalian NOTCH homologue gene, is organized in a topological domain in human cells [28]. NOTCH1 is a highly conserved gene across species, and its expression is tightly controlled due to its pivotal role in signaling pathways that drive cell fate decisions. Dysregulation of NOTCH1 expression is implicated in a range of human diseases, including cancer, where either overexpression or inappropriate activation of NOTCH signaling can contribute to oncogenesis [9, 10]. Additionally, a recent study uncovered a novel mechanism by which SNAI2 and CTCF cooperatively regulate NOTCH1 expression in fusion-negative rhabdomyosarcoma (FN-RMS) through maintaining a sub-topologically associating domain (sub-TAD) boundary [56]. Disrupting this interaction leads to the loss of NOTCH1 expression, reduced self-renewal, and increased myogenic differentiation, highlighting a potential therapeutic target [56]. So, we wondered if there is a relationship between its three-dimensional organization and its transcriptional regulation in mammals.

Here, we investigate the relationship between the topological organization of NOTCH1 and its expression regulation in mouse and human cells. By analyzing the effects of CTCF and cohesin-mediated TAD boundaries on NOTCH1 expression and the expression of adjacent genes, we explore how these architectural proteins contribute to chromatin insulation and transcriptional control. Furthermore, our examination of CTCF-mediated topological organization of the NOTCH1 locus in mouse embryonic stem cells (mESCs) and murine erythroleukemia cells (MEL) highlighted the chromatin context dependency of TAD boundary function. Our findings provide evidence for the functional importance of topological domain boundaries in maintaining proper gene expression and underscore the relevance of 3D genome organization in the regulation of key developmental genes like NOTCH1. We study the relationship between the topological organization of the genome with the regulation of NOTCH1 expression in mouse and human. We found that NOTCH1 is organized in a topological domain enriched with CTCF and cohesin at its boundaries, perturbing these boundaries caused changes in the expression of the gene, as well as in the genes near the domain, even though we did not find drastic changes in the interactions within the domain. These changes in expression were partially explained by a decrease in the insulation of the borders and changes in chromatin marks inside and outside the domain. Interestingly, these changes at the transcriptional level influenced the phenotype of the cells, which correlated with NOTCH1 gene expression levels. Finally, we show that the organization of this locus is conserved in mice, where we also observed changes in the expression levels of the genes in this region, which indicates that the three-dimensional organization of the locus is important to maintain proper transcriptional regulation. Taken together, our data show evidence for the role of topological domain boundaries and their contribution to the regulation of gene expression.

Results

NOTCH1 is organized in a conserved topological domain independent of its transcriptional activity

The Notch receptor is a central regulator of development, mediating the activity of an evolutionarily conserved cell–cell communication pathway that controls cell differentiation and tissue homeostasis in invertebrates and vertebrates [9, 10]. The transcriptional and epigenetic dysregulation of the Notch genes has been extensively documented in different pathologies and causally linked to tumor development in Drosophila, mouse, and humans, underscoring the importance of tightly controlled Notch transcriptional regulation [19, 27, 43, 54, 61].

Previously, we demonstrated that the Notch gene in Drosophila is organized into topological domains and that disruption of the domain boundaries alters its transcriptional regulation [3]. In mammals, four Notch paralogs (NOTCH1-4) have been identified, with NOTCH1 being the direct ortholog of Drosophila Notch. To explore the possible role of genome organization in regulating NOTCH gene expression in mammals, we first examined the topological structure of the four Notch paralogs by analyzing high-resolution publicly available Hi-C data (The ENCODE Project Consortium et al., 2020). In human cells, heatmaps of Hi-C data binned at 5 kb resolution show that NOTCH1 is the only Notch paralog that consistently exhibits a clear three-dimensional organization into a topological domain (Fig. S1). Based on these data, we define a 147 kb NOTCH1 topological domain that contains the entire NOTCH1 gene and two lncRNAs, NALT1 and LINC01451 (Fig. 1a,b).

Fig. 1.

NOTCH1 is organized in a conserved topological domain independent of its transcriptional activity. a Hi-C contact maps across the NOTCH1 locus shown for human embryonic stem cells (hESCs), human hematopoietic stem cells (HSCs), macaque fetal brain cells and mouse embryonic stem cells (mESCs). b Hi-C heatmaps in GM12878, K562 and HepG2 cells showing the highly frequent interactions between CTCF binding sites upstream and downstream of NOTCH1 gene. c MicroC heatmap in H1-hESCs showing the interaction between CTCF binding sites S1 and S2 with S4 and S5. Arrowheads point to significant interactions between NOTCH1 promoters and putative enhancers inside the topological domain. A cartoon model of NOTCH1 locus organization in hESCs is shown. d CTCF, RAD21 and H3K27ac enrichment in NOTCH1 locus in several human cell types, enhancers from ChromHMM tracks are shown, NOTCH1 gene expression levels in each cell type are shown at the right of the tracks

We observe that the organization of the NOTCH1 domain is conserved across human cell types and cell lines (Fig. 1). For example, the NOTCH1 domain is detected in human embryonic and hematopoietic stem cells (Fig. 1a), but also in cell lines such as GM12878 and K562 (Fig. 1b). Notably, a corresponding NOTCH1 topological domain is also detected in Hi-C data from macaque and mouse cells (Fig. 1a; Fig. S4a) [33, 37]. Overall, these observations indicate that, as in its Drosophila ortholog, the mammalian NOTCH1 is embedded within a constitutive topological domain.

In human cells, a subset of topological domain boundaries coincides with chromatin loop anchors, where pairs of loci at opposite sides of the domain contact each other more frequently than the intervening sequence [47]. These anchors are visualized in Hi-C heatmaps as high-intensity pixels off the diagonal. In our re-analyzed Hi-C data, the 5′ and 3′ boundaries of the NOTCH1 domain coincide with two such anchors, as evidenced by a single prominent off-diagonal high-intensity pixel (Fig. 1a–c; Fig. S1). Notably, this boundary-to-boundary looped configuration is also observed in Hi-C heatmaps from other mammals and across multiple human and mouse cell types (Fig. 1a–c; Fig. S4a), further supporting the notion that NOTCH1 3D organization is conserved across species.

In mammals, chromatin loops that anchor at domain boundaries form between CTCF-bound sites in a convergent orientation and are stabilized by cohesin [5, 14, 47]. To investigate whether CTCF and cohesin bind NOTCH1 domain boundaries, we examined CTCF and RAD21 occupancy using publicly available ChIP-seq data and integrated these profiles with high-resolution Hi-C and Micro-C contact maps. We observe that CTCF and RAD21, a cohesin subunit, are bound at both the 5' and 3' boundaries of the NOTCH1 domain across human cell types (Fig. 1b, d). Analysis of human Micro-C data enabled us to identify the specific locations of the NOTCH1 domain boundaries and to determine which CTCF sites are more likely to mediate interactions between them. The 5' boundary of the domain is located between the NOTCH1 transcription end site (TES) and the SEC16A transcription start site (TSS) (Fig. 1b–d). This boundary is occupied by CTCF at two prominent sites, each with a motif in the forward orientation (CTCF site 1, S1; CTCF site 2, S2). The 3' boundary is located 85 kb from the NOTCH1 TSS and maps to the intergenic region between the LINC01451 and HSPC324 (Fig. 1b–d). Similar to the 5’ boundary, it shows CTCF binding at three prominent sites, each with a motif in the reverse orientation (CTCF site 3, S3; CTCF site 4, S4; CTCF site 5, S5), but only the CTCF sites S4 and S5 show evidence of physical interactions with the 5’ boundary in Micro-C data (Fig. 1c). Consistent with the architectural features of domain boundaries, we also observe RAD21 binding at all the boundary-associated CTCF sites (Fig. 1b,d). Notably, binding of CTCF and RAD21 was invariant between human cell lines (Fig. 1d), consistent with the presence of the NOTCH1 topological domain across all cell types examined.

A subset of topological domains insulates regulatory landscapes, thereby facilitating specific long-range interactions between enhancers and promoters and supporting specific gene expression [13, 16, 53]. To assess whether the NOTCH1 demarcates a regulatory landscape, we used ChromHMM annotations from the ENCODE project to map candidate enhancers within the NOTCH1 domain across multiple human cell lines. The number of enhancers and the transcriptional level of NOTCH1 differed substantially between cell lines (Fig. 1d). In contrast, the topological domain structure and the binding of CTCF and RAD21 at the boundaries remained invariant (Fig. 1b,d). Notably, physical interactions between regulatory elements within the domain were also detected. For example, in Micro-C data from human embryonic stem cells (hESCs), we observe physical interactions between intra-domain enhancers and the NOTCH1 promoter (Fig. 1c, arrowheads). Collectively, these findings suggest that the NOTCH1 topological domain may contribute to cell-type-specific transcriptional regulation of NOTCH1 by modulating intra-domain enhancer-promoter interactions.

NOTCH1 topological domain boundaries modulate NOTCH1 expression in human cells

Topological domains can influence gene transcription [11, 28, 38]. To assess the role of the human NOTCH1 topological domain in transcriptional regulation, we first perturbed its boundaries indirectly by reducing CTCF protein levels. To this end, we employed an inducible shRNA knockdown (KD) in K562 cells. Because K562 cells derive from a patient with chronic myelogenous leukemia [36] and Notch signaling is implicated in hematologic malignancies [54], this context is relevant to study the effects of NOTCH1 domain disruption. We confirmed successful CTCF knockdown by Western-blot (Fig. 2a). RT–qPCR analysis of NOTCH1 revealed a non-significant trend toward up-regulation in the CTCF KD condition (Fig. 2a). Because nearly complete depletion of CTCF is required to alter domain structure and gene expression [42], this result suggests that partial CTCF reduction was insufficient to perturb NOTCH1 domain boundaries.

Fig. 2.

CTCF binding upstream and downstream of the NOTCH1 topological domain is required to maintain correct NOTCH1 gene expression in human cells. a CTCF knockdown in K562 cells, in top Western-blot against CTCF shows the decrease in CTCF protein, in bottom NOTCH1 expression measured by RT-qPCR in control and after doxycycline treatment. b CTCF binding sites selected for CRISPR-Cas9 targeting to delete the CTCF motifs, which are in a convergent orientation. c ChIP-qPCR in CTCF binding sites showing the enrichment of CTCF and RAD21 at those sites in WT and mutant cells. d Expression levels measured by RT-qPCRs of NOTCH1, NALT1, LINC01451 and neighboring genes INPP5E, SEC16A, EGFL7 and AGPAT2 e In WT versus mutant cells. f Cell proliferation assay (MTT) in WT and mutant cells for three days. g Colony formation assay in K562 WT and mutant cells, in left representative images are shown, in right graphs show the quantification of all colonies in each sample

To directly test the role of human NOTCH1 domain boundaries, we used a previously validated CRISPR–Cas9 strategy [3, 40, 45]. We designed pairs of sgRNAs to remove CTCF sites that interact between the domain boundaries (S2, S4, S5). K562 cells were transfected with plasmids expressing Cas9 and the corresponding sgRNA pairs, and from the edited populations, we isolated homozygous mutant lines at all targeted sites except S1 (Fig. S2a). Overall, we obtained four homozygous mutants targeting NOTCH1 boundaries: one at the 5′ boundary, hΔ5′-CTCF-1 (deletion of S2), and three at the 3′ boundary, hΔ3′-CTCF-1 (ΔS4), hΔ3′-CTCF-2 (ΔS5), and the double mutant hΔ3′-CTCF-1,2 (ΔS4 and ΔS5) (Fig. 2b). Sanger sequencing confirmed the expected deletion sizes (50 bp–4 kb) (Fig. S2a), and motif analysis verified the loss of the CTCF motif in all mutants and confirmed that no novel CTCF sites were created.

To validate the loss of CTCF binding at the deleted regions, we performed ChIP–qPCR and observed a > 90% reduction in CTCF occupancy at each deleted site, indicating a near-complete loss of binding (Fig. 2c). Interestingly, the 3′-boundary sites S4 and S5 appeared to influence CTCF recruitment to each other. In the hΔ3′-CTCF-1 mutant, which deletes S4, CTCF binding at S5 was also significantly reduced, while in the hΔ3′-CTCF-2 mutant, which deletes S5, CTCF binding was reduced at S4 (Fig. 2c). This reciprocal reduction indicates that CTCF occupancy at S4 depends on the presence of S5, and vice versa. None of the other boundary CTCF sites showed comparable cross-dependence. This finding is consistent with previous observations of interdependent CTCF binding at clustered sites [40]. According to the loop extrusion model, chromatin loops at domain boundaries are stabilized when cohesin accumulates at CTCF-bound anchors, and the loss of CTCF at these anchors is expected to reduce cohesin retention, thereby weakening domain boundaries [20, 51]. Consistent with this, ChIP-qPCR for the RAD21 subunit also showed a concomitant decrease in cohesin binding at the deleted sites (Fig. 2c).

To assess the consequences of boundary disruption on NOTCH1 transcription, we measured gene expression by RT-qPCR in wild-type and mutant cell lines (Fig. 2d,e). All boundary mutants showed increased NOTCH1 expression, ranging from 2- to 3.5-fold induction in the single-site deletions to up to 4-fold in the double 3′-boundary mutant (hΔ3′-CTCF-1,2) (Fig. 2d). Notably, the direction of change observed upon targeted boundary deletion mirrors the non-significant up-regulation trend observed in the CTCF KD (Fig. 2a). Because genes within a topological domain can be co-regulated, we also examined the two lncRNAs located inside the NOTCH1 domain, NALT1 and LINC01451. Both genes were strongly induced in all mutants and were more responsive than NOTCH1 itself. For example, in the double 3′-boundary mutant, NALT1 and LINC01451 showed a 10- and 15-fold increase in expression above wild-type levels, respectively (Fig. 2d). Therefore, deletion of boundary CTCF sites, particularly at the 3′ boundary, leads to coordinated, progressively stronger upregulation of all three genes within the NOTCH1 domain.

Since disruption of domain boundaries can lead to the transcriptional dysregulation of genes outside the domain [13, 28], we next quantified expression of two upstream genes (INPP5E, SEC16A) and two downstream genes (EGFL7, AGPAT2) (Fig. 2e). INPP5E, EGFL7, and AGPAT2 were largely unaffected in the mutants. In contrast, SEC16A, located immediately upstream of the 5′ boundary, was selectively downregulated in the 3′ boundary mutants, most notably in hΔ3′-CTCF-1,2, where its expression dropped to 40% of the wild-type (Fig. 2e). These data suggest that disruption of CTCF binding sites at the NOTCH1 domain boundaries leads to gene dysregulation, likely through changes in the topological organization of this region.

The disruption of the NOTCH1 domain causes a decrease in cell proliferation and the ability to form colonies

Notch signaling has been implicated in the control of proliferation and differentiation in myeloid cells, including K562 cells, where NOTCH1 has been reported to act as a tumor suppressor that limits cell proliferation [1, 62]. Given the strong transcriptional up-regulation of NOTCH1 and its intradomain lncRNAs in the boundary mutants, we asked whether changes in cellular proliferation also accompany these alterations. To address this, we measured proliferation using an MTT assay. We observed that all boundary mutants showed a significant reduction in proliferation relative to wild-type cells, with the 3′ double mutant (hΔ3′-CTCF-1,2) displaying the most potent inhibition (Fig. 2f).

Because Notch signaling has also been shown to modulate the colony-forming potential of K562 cells [57], we next asked whether boundary disruption affected clonogenic growth. We examined the colony-forming capacity of the 5′ boundary mutant (hΔ5′-CTCF-1) and the 3′ double mutant (hΔ3′-CTCF-1,2) in semisolid medium. The boundary mutants formed smaller colonies than the wild-type (Fig. 2g). Notably, the 3’ double mutant (hΔ3′-CTCF-1,2) mutant displayed the most potent effect, showing a near depletion of big colonies accompanied by a marked increase in small colonies (Fig. 2g). To relate these phenotypes to the activity of the canonical Notch pathway, we measured expression of the downstream target genes HES1 and HEY1 by RT–qPCR but did not detect significant changes compared with wild-type cells (Fig. S2b). Taken together, these results indicate that boundary disruption impairs the proliferative and clonogenic capacity of K562 cells.

Deletion of CTCF binding sites rewires chromatin interactions at the NOTCH1 locus

Disruption of CTCF-bound domain boundaries can alter genome organization [32, 40, 48]. Therefore, to determine whether loss of CTCF binding at NOTCH1 boundaries alters local chromatin architecture, we performed 4C-seq in K562 cells and analyzed two biological replicates per genotype. We designed three viewpoints to capture complementary aspects of the NOTCH1 domain organization: the NOTCH1 promoter, to identify changes in enhancer-promoter communication; the 5′ boundary CTCF site S2, to capture changes in interactions made by the upstream domain border; and a 3′ boundary viewpoint encompassing the CTCF S4 and S5 sites, to capture the interaction profile of the downstream boundary (Fig. 3a; Fig. S3a).

Fig. 3.

Loss of CTCF affects the three-dimensional organization of the NOTCH1 domain. (a) 4C-seq profiles in K562 wild-type (WT) and mutant cells from three viewpoints: CTCF S2 (left), NOTCH1 TSS (middle), and CTCF S4 (right). The viewpoint is highlighted in blue; orange boxes mark the windows selected for zoom-ins and quantification. Below the 4C-seq profiles, CTCF and H3K27ac signals in K562 cells provide chromatin context. Zoom-in panels show overlaid, normalized 4C-seq signals for WT and four mutants (hΔ5’-CTCF-1, hΔ3’-CTCF-1, hΔ3’-CTCF-2, hΔ3’-CTCF-1,2) for viewpoint CTCF S2 (b), NOTCH1 TSS (c) and CTCF S4 (d). Boxplots report normalized counts within the highlighted windows; significance for each mutant versus WT is annotated above the boxes

In wild-type cells, all three viewpoints showed preferential interactions within the NOTCH1 contact domain (Fig. 3a; Fig. S3a). To validate our 4C-seq assay, we generated virtual 4C profiles for each viewpoint using K562 cells Hi-C data. The virtual 4C profiles are remarkably similar to our 4C-seq data from wild-type cells, validating the assay (Fig. S3a). Visual inspection of the wild-type profiles revealed that the domain boundaries preferentially interact with each other and with two putative enhancers upstream of the NOTCH1 promoter at − 42 kb and − 70 kb (the latter overlapping LINC01451), although less frequently than the contacts between the boundaries (Fig. 3a left and right; Fig. S3a,b). For the NOTCH1 promoter, we observe that it preferentially interacts with genomic regions within the domain, further supporting the idea that domain boundaries insulate the NOTCH1 region. Notably, the NOTCH1 promoter shows four areas of enhanced interaction within the domain: the two domain boundaries, a potential intragenic enhancer (+ 33 kb from the NOTCH1 TSS), and the − 42 kb enhancer (Fig. 3a center; Fig. S3a,b). To further strengthen our observations, we employed peakC to identify regions of significantly higher interactions for each viewpoint [22]. In wild-type cells, the boundary viewpoints formed a statistically significant contact with the opposite boundary, consistent with a dominant boundary-to-boundary loop anchoring the NOTCH1 domain (Fig. S3a). In contrast, the NOTCH1 promoter viewpoint showed statistically significant interactions with the 5′ boundary, with intragenic enhancers (+ 33 kb), and with two upstream enhancers at − 42 kb and − 70 kb (the latter overlapping LINC01451) (Fig. S3a).

In the hΔ5′-CTCF-1 mutant, peakC analysis of the 4C-seq profile using the CTCF S2 viewpoint still detected a significant interaction with the 3′ boundary, likely maintained via the nearby CTCF S1 site located < 5 kb away (Fig. S3a). However, visual inspection of the 4C-seq profile clearly showed an increased interaction frequency between the CTCF S2 viewpoint and the region upstream of the 5′ domain boundary (Fig. 3a left and Fig. S3a). We therefore quantified normalized contacts between the CTCF S2 viewpoint and a ~ 100 kb region upstream of the domain overlapping the DNLZ and SNAPC4 TSSs. We found a significant increase in contacts in the hΔ5′-CTCF-1 mutant (Fig. 3b-1). This ectopic contact region coincides with CTCF binding and H3K27ac deposition in K562 cells (Fig. 3b-1), indicating that loss of CTCF S2 allows the 5′ boundary region to interact more frequently with upstream regulatory elements. Consistent with this loss of insulation, peakC analysis of hΔ5′-CTCF-1 4C-seq profile using the CTCF S4-S5 viewpoint no longer detected the canonical interaction with the 5′ boundary (Fig. S3a center). Instead, a new significant long-range interaction appeared with a region more than 100 kb upstream of the annotated 5′ boundary, again overlapping the DNLZ/SNAPC4 region. Quantification of normalized contacts between the CTCF S4-S5 viewpoint and this upstream interval confirmed a significant gain of interactions in the hΔ5′-CTCF-1 mutant, most evident at the DNLZ TSS and again restricted to regions occupied by CTCF and H3K27ac (Fig. 3d-1). Therefore, despite the presence of CTCF S1, the loss of CTCF S2 at the 5’ boundary is sufficient to compromise boundary integrity and allow ectopic long-range contacts spanning at least 100–200 kb upstream of the 5′ domain. From the NOTCH1 promoter viewpoint, peakC analysis detected no changes in the set of significant contacts, either within or outside the domain. Still, quantitative analysis of specific intervals revealed an increase in interactions with the exonic region of SEC16A immediately upstream of the 5′ boundary, consistent with perturbed insulation at this domain edge (Fig. 3c-1). Furthermore, disruption of the 5’ boundary appears to affect the NOTCH1 promoter contacts within the domain, as we observed a significant gain of contacts with the + 33 kb and − 70 kb enhancers, with the most significant increase at + 33 kb, and a loss of contacts with the − 42 kb enhancer (Fig. 3c-2,3,4). Together, these observations indicate that disruption of the 5’ boundary by removal of the CTCF S2 results in compromised 5′ insulation, leading to ectopic upstream contacts, and a redistribution of enhancer-promoter interactions favoring the + 33 kb and − 70 kb/LINC01451 elements at the expense of the − 42 kb enhancer.

For all 3′ mutants, peakC still called a significant interaction between the 3′ boundary viewpoint (S4-S5) and the 5′ boundary region, indicating that the NOTCH1 domain is largely maintained and, unlike the 5′ mutant, no additional significant ectopic interactions were detected outside the domain (Fig. 3a center and Fig. S3a, b). Visual inspection of the 4C-seq maps nonetheless revealed two notable changes. First, particularly in the hΔ3′-CTCF-1,2 double mutant, interactions between the domain boundaries appeared strengthened compared to wild-type, as if boundary-boundary looping were reinforced rather than lost, despite the deletion of two major CTCF sites at the 3’ boundary (Fig. 3a and Fig. S3b). Second, the NOTCH1 promoter showed a shift in its interaction pattern: contacts within the gene body and with intragenic enhancers decreased, while contacts with the intergenic region upstream of the promoter, where at least two enhancers (− 42 and -70 kb/LINC01451) are located, became more prominent (Fig. 3a and Fig. S3b). These qualitative observations were supported by quantitative analysis. From the 3′ boundary viewpoint [(Fig. 3a right), all 3′ mutants displayed a significant > 2-fold increase in normalized contacts with the 5′ boundary region (CTCF S2), with the largest increase in mutants lacking the CTCF S5 site (hΔ3′-CTCF-2 and hΔ3′-CTCF-1,2) (Fig. 3a right and Fig. 3d-3). For the NOTCH1 promoter viewpoint, and in contrast to the 5’ mutant, the 3′ mutants did not increase their contacts with the intragenic + 33 kb enhancer (Fig. 3c-2). Instead, they increased interactions with the intergenic enhancers (Fig. 3c-3,4). In particular, all 3′ mutants showed a significant gain of contacts with the − 42 kb and − 70 kb enhancers (Fig. 3c-3,4). Results from PeakC analysis were consistent with this pattern, reporting a loss of significant promoter interactions with intragenic NOTCH1 enhancers and a shift toward upstream regulatory elements in the 3′ mutants (Fig. S3a right). Outside the domain, the 3’ boundary showed a modest increase in interactions with the EGFL7 TSS, located just downstream of the 3’ boundary, suggesting loss of insulation (Fig. 3d-4). Still, these effects were comparatively small relative to the intradomain rewiring and the evident loss of insulation observed in the 5’ mutant. These findings suggest that disruption of the 3′ boundary preserves the overall domain structure, increases contacts between domain boundaries, and redistributes contacts within the domain.

To complement the 4C-seq experiments, we used Orca [63], a deep-learning model that predicts multiscale 3D genome contact maps directly from DNA sequence, to generate sequence-based multiscale contact map predictions for the K562 cell line at the NOTCH1 locus. We modelled the wild-type sequence and the in silico deletions generated in K562 cells (Fig. S3b,c). In the wild-type configuration, Orca recapitulated the focal off-diagonal contact between the 5′ and 3′ boundaries seen in Hi-C, consistent with a boundary-to-boundary loop (Fig. S3b left). For the in silico 5′ boundary deletion (hΔ5′-CTCF-1), the predicted contact map showed a clear gain of interactions with the upstream region and a reduction in the focal boundary-boundary interaction, consistent with ectopic upstream contacts and a weakened boundary loop detected by 4C-seq (Fig. S3b center). For the 3′ double mutant (hΔ3′-CTCF-1,2), Orca predicted increased downstream contacts beyond the 3′ boundary, in partial agreement with our experimental data. Notably, consistent with our experimental observations, it shows a redistribution of contacts, with contacts now localized nearly exclusively to the NOTCH1 domain, as if the deletion increased the domain's overall insulation (Fig. S3b right). The prediction also shows a reduction in boundary focal interaction, which contrasts with our experimental observations, in which, in addition to intradomain contacts, we detect a nearly 2.5-fold increase in contacts between the 3’ region and the 5’ boundary (Fig. 3d-3). Overall, these modelling results support the directional upstream and downstream shifts inferred from 4C-seq. Still, they fail to capture the observed interactions between boundary-boundary contacts, suggesting that this aspect of loop stabilization may depend on chromatin features or protein-RNA interactions beyond the underlying DNA sequence.

Taken together, these data show that targeted perturbation of CTCF binding substantially rewires chromatin interactions at the NOTCH1 locus without abolishing the underlying domain. Deletion of the CTCF S2 compromises 5′ insulation and promotes ectopic upstream contacts while shifting NOTCH1 promoter interactions toward the + 33 and − 70 kb/LINC01451 enhancers and away from the − 42 kb enhancer. In contrast, deletion of the CTCF S4-S5 at the 3′ boundary strengthens boundary-boundary looping, relocating NOTCH1 promoter contacts from intragenic to upstream intergenic enhancers, particularly the − 42 and − 70 kb/LINC01451 elements.

The LINC01451 locus is an enhancer element and participates in the transcriptional regulation of the NOTCH1

Our 4C-seq data identified three candidate regulatory enhancers that interact with the NOTCH1 promoter and are differentially engaged in the boundary mutants. The three candidate enhancers are enriched for H3K27ac and are located at + 33 kb, − 42 kb, and − 70 kb relative to the NOTCH1 promoter (Fig. 4a). The + 33 kb enhancer lies within intron 15 of NOTCH1 and is bound by the TF GATA1, essential for erythroid differentiation [60] (Fig. 4a). The − 42 kb enhancer is intergenic, and it is bound by the TF YY1, which has been involved in mediating enhancer-promoter interactions [29, 59] (Fig. 4a). The − 70 kb element overlaps exon 1 of LINC01451 and shows modest binding by CTCF and RAD21 (Fig. 4a). Notably, the three enhancers are occupied by RNA Pol-II, further suggesting they are active enhancers in K562 cells. Consistent with these observations, and our 4C-seq data, published H3K27ac HiChIP data [39] provide independent support for physical interactions between the − 42 kb and − 70 kb enhancers and the NOTCH1 promoter [26] (Fig. 4a).

Fig. 4.

Linc10451 gene participates in the transcriptional regulation of the NOTCH1 gene acting as an enhancer element. a Chromatin landscape in NOTCH1 locus in K562 cells showing enrichment of architectural proteins, histone marks, RNA Pol-II and HiChIP against H3K27ac. b Luciferase assay cloning enhancer − 42 kb in forward and reverse orientations. NOTCH1 and lncRNAs expression levels obtained by RT-qPCRs using CRISPRa (c) and CRISPRi (d) against enhancers − 42 kb and − 70 kb

Because all 3′ boundary mutants showed increased contacts between the NOTCH1 promoter and the − 42 kb and − 70 kb enhancers, together with the strongest upregulation of NOTCH1, we next tested whether these elements act as positive regulators of NOTCH1. For the − 42 kb enhancer, we first assessed its activity in a reporter assay: a 1.8 kb DNA fragment was cloned upstream of a Luciferase reporter driven by an SV40 promoter and transfected into K562 cells; in both orientations, this fragment increased luciferase signal, confirming enhancer activity in this cellular context (Fig. 4b). To test the function of the − 42 kb and − 70 kb elements in their native chromatin environment, we used dCas9-KRAB (CRISPRi) to repress and dCas9-VP64 (CRISPRa) to activate each locus [25]. CRISPRa-mediated activation of either the − 42 kb or − 70 kb enhancers strongly induced LINC01451 and increased NOTCH1 expression, with the strongest induction observed upon activation of the − 70 kb element (Fig. 4c). CRISPRi-mediated silencing of the − 70 kb/LINC01451 element reduced LINC01451 expression and led to a 50% decrease in NOTCH1 transcript levels relative to wild-type, whereas targeting the − 42 kb enhancer had little or no effect on NOTCH1, making − 70 kb/LINC01451 repression the only condition that robustly reduced NOTCH1 expression (Fig. 4d). Notably, NALT1 expression remained unchanged under all CRISPRa/i conditions, despite NALT1 and NOTCH1 being arranged in a head-to-head orientation and therefore sharing part of their upstream intergenic region (Fig. 4a), indicating that the − 70 kb/LINC01451 enhancer preferentially activates the NOTCH1 promoter.

Altogether, our experiments in human K562 cells reveal an asymmetric regulatory logic at the human NOTCH1 topological domain. The 5′ boundary, centered on CTCF S2, primarily acts as an insulator: its deletion compromises 5′ insulation, allows ectopic upstream contacts, and shifts promoter interactions within the domain. In contrast, deletion of the 3′ CTCF sites preserves the domain but paradoxically strengthens the boundary looping interaction and redistributes promoter contacts from intragenic to upstream intergenic enhancers. 4C-seq analyses show that, under these perturbations, enhancer-promoter communication is redistributed toward a small set of intradomain enhancers, most prominently the − 70 kb/LINC01451 element and, to a lesser extent, the − 42 kb enhancer. Reporter assays and CRISPR-based epigenetic editing further demonstrate that these elements, particularly the − 70 kb/LINC01451 locus, act as positive regulators of NOTCH1 transcription in situ. These topological alterations correlate with the coordinated up-regulation of NOTCH1, NALT1, and LINC01451, and are accompanied by reduced proliferation and clonogenic capacity, indicating that the conserved NOTCH1 domain functions as an insulated regulatory unit whose transcriptional output and cellular effects are tuned by boundary integrity and intradomain enhancer activity.

CTCF-mediated genome organization is important in keeping the Notch1 gene locus insulated in mouse cells

Chromatin state and enhancer usage within domains can vary extensively between cell types, leading to context-dependent transcriptional outcomes [7, 49, 52]. In human K562 cells, we observe that the NOTCH1 domain boundaries are relevant for enhancer-promoter communication and transcriptional output. Because the NOTCH1 domain and its CTCF-bound boundaries are conserved across mammals (Fig. 1a, 5a and Fig. S4a), we next asked whether the same architectural organization also impacts transcription and cellular behaviors at the mouse Notch1 locus and whether these effects depend on chromatin context.

Fig. 5.

CTCF-mediated genome organization is important in keeping the Notch1 gene locus insulated in mouse cells. a MicroC heatmap from mESCs in Notch1 locus along with CTCF and H3K27ac signal, in the right zoom in of the contact frequency between CTCF binding sites upstream (mS1) and downstream (mS2, mS3 and mS4). A cartoon model of Notch1 three-dimensional organization of Notch1 locus in mESCs based on microC data. Chromatin landscape in Notch1 locus in mESCs (b) and MEL cells (c), along with the deletions obtained by using CRISPR-Cas9 system to obtain ΔmS1 and ΔmS2-mS4 mutant cells. Expression levels measured by RT-qPCR of Notch1 and neighboring genes in WT and mutant cells in mESCs (d) and MEL cells (e)

To directly assess the impact of chromatin context on the consequences of domain disruption, we focused on mouse embryonic stem cells (mESCs), where Notch1 is expressed in an active chromatin environment, and murine erythroleukemia (MEL) cells, which, like K562 cells, derive from the hematopoietic lineage but in which Notch1 resides in a Polycomb domain and is transcriptionally silent (Fig. 5b,c). Analysis of public Hi-C data from mESCs revealed that the murine Notch1 locus is organized into a discrete contact domain whose boundaries are occupied by one CTCF site at the 5′ boundary (S1) and three sites at the 3′ boundary (S2, S3, S4) (Fig. 5a-c). The arrangement of CTCF occupancy at these boundaries is remarkably similar to that at the human NOTCH1 locus, particularly the clustered organization of CTCF sites at the 3′ boundary. In contrast to the human NOTCH1 domain, the murine domain encompasses only the Notch1 gene (Fig. 5a-c). Across the syntenic region, gene content and organization outside the domain are broadly conserved: Notch1 lies between Sec16a upstream and Egfl7 and Agpat2 downstream, and in both mESCs and MEL cells, the domain-flanking genes are expressed and are marked by active chromatin (Fig. 5b,c).

CTCF binding at Notch1 domain boundaries sites is also conserved in MEL cells (Fig. 5c), despite the absence of detectable Notch1 expression in MEL cells, indicating that boundary occupancy and domain organization are maintained independently of Notch1 transcription. This transcription-independent maintenance of the murine Notch1 domain is consistent with our observations at the human NOTCH1 locus, where the topological domain persists across cell types despite differences in NOTCH1 expression (Fig. 1b,d). Moreover, Hi-C maps across multiple stages of mouse hematopoietic differentiation show a stable Notch1 domain whose boundaries interact and appear as a focal high-intensity pixel away from the diagonal, while CTCF binding at these boundaries is maintained, despite variations in Notch1 expression (Fig. S4a,b).

Given our observations in human K562 cells that boundary integrity is critical for NOTCH1 regulation, we next assessed whether deleting the domain boundaries affects Notch1 and neighboring gene expression in mESCs and MEL cells. We engineered two deletions in both mESCs and MEL cells. At the 5′ boundary, we designed a 94 bp deletion (mΔ5′-CTCF) that removes the single CTCF site (S1). At the 3′ boundary, we generated an 8.5 kb deletion (mΔ3′-CTCF) that eliminates all three CTCF sites (S2, S3, and S4) (Fig. 5b,c). For each deletion, we isolated homozygous mutant clones in both mESCs (Fig. S4c) and MEL cells and measured expression of Notch1, Inpp5e, Sec16a, Egfl7, and Agpat2 by RT–qPCR (Fig. 5d,e). In mESCs, deletion of the 5′ boundary significantly reduced Notch1 expression (Fig. 5d). In contrast, disruption of the 3′ boundary produced only a non-significant trend toward downregulation in Notch1, indicating that the 5′ boundary exerts a more substantial influence on Notch1 transcription in this context. In contrast, the expression of Egfl7, which lies just downstream of the domain, was reduced by 40% in both boundary mutants, while the rest of the analyzed genes remained unchanged (Fig. 5d). This pattern contrasts with human K562 cells, where boundary disruption increases NOTCH1 expression and affects SEC16A and AGPAT2 rather than EGFL7 (Fig. 2e). In MEL cells, Notch1 remained undetectable in the boundary mutants, and just the expression of Egfl7 was significantly reduced in the 3’ boundary mutant (Fig. 5e). These findings suggest that, in the mouse, the influence of domain boundaries on Notch1 transcriptional regulation is cell-type-dependent.

Finally, we examined the cellular consequences of boundary deletion. In mESCs, both boundary mutants proliferated faster than wild-type cells (Fig. S4d), whereas in human K562 cells, NOTCH1 boundary disruption reduced proliferation and clonogenic capacity (Fig. 1f,g). In MEL cells, neither boundary deletion altered proliferation nor affected colony number or size compared with wild-type cells (Fig. S4d). These observations indicate that while CTCF boundary loss at the Notch1 locus can modulate proliferation in an active pluripotent context, its impact on cellular behavior is strongly cell-type- and chromatin-context-dependent and does not simply follow from boundary disruption alone.

Discussion

In this study, we investigated the regulatory role of CTCF-mediated topological domain boundaries at the NOTCH1 locus in two mammalian species across different cellular contexts. Our data show that NOTCH1 is embedded within a conserved topological domain in human and mouse, anchored by CTCF and cohesin, and maintained independently of chromatin and transcriptional state. Together with our previous work in Drosophila, where NOTCH1 also resides in a defined topological domain, these findings support the conclusion that NOTCH1 is contained within an architectural unit across evolution and cell types. Such an organization could contribute to the specificity of the regulatory programs that are imposed on NOTCH1 in a chromatin- and context-dependent manner.

In human K562 cells, perturbation of CTCF sites at the boundaries of the NOTCH1 domain revealed an asymmetric regulatory logic at the human NOTCH1 topological domain (Fig. 6). The 5′ boundary, centered on CTCF S2, primarily functions as an insulator, as its deletion compromises 5′ insulation, allows ectopic upstream contacts, and shifts promoter interactions within the domain. In contrast, deletion of the 3′ CTCF sites (S4 and S5) does not result in loss of insulation. Instead, it strengthens the interaction between boundaries, particularly between the 3’ domain boundary region and the 5’ boundary. This also redistributes NOTCH1 promoter contacts from intragenic to upstream intergenic enhancers.

Fig. 6.

Model of CTCF boundary function at the NOTCH1 locus. Schematic representation of the NOTCH1 locus organized within a CTCF-anchored topological domain, with CTCF sites at the 5′ and 3′ boundaries insulating the gene and its intradomain lncRNAs (NALT1 and LINC01451). Deletion of boundary CTCF sites, particularly at the 3′ boundary, leads to increased enhancer-promoter contacts between the NOTCH1 promoter and the − 70 kb/LINC01451 enhancer, resulting in upregulation of NOTCH1, NALT1 and LINC01451, accompanied by reduced cell proliferation and smaller colonies. These findings highlight that the 5′ and 3′ boundaries make distinct contributions to maintaining proper 3D genome organization and gene regulation at this locus

The asymmetric effects observed upon boundary disruption suggest that 5′ and 3′ boundaries contribute differently to the balance between insulation and long-range looping, and that boundary function depends on anchor identity, motif orientation, and local CTCF cluster architecture. In this regard, the CTCF clustered architecture at the 3’ domain boundary could explain our experimental observations. In wild-type K562 cells, the 3’ boundary region contains more CTCF-bound sites near the deletion site, whereas the 5’ region shows a depletion of additional neighboring CTCF binding sites. In such a context, the CTCF S3 site at the 5’ boundary may now be responsible for interaction with the 5’ boundary, whereas in the wild-type condition, the presence of the other CTCF sites (S4-S5) buffers the boundary interactions. This pattern is consistent with the idea that the boundary function is robust to single-site perturbations when anchors are composed of clusters of nearby CTCF motifs. In those configurations, clustered sites behave as partially redundant and competitive barriers, such that removal of one or more motifs redistributes contact probabilities among alternative loops, as predicted by loop extrusion models [20, 51], rather than simply eliminating the original interaction.

Functionally, boundary perturbation increased NOTCH1 expression together with the lncRNAs NALT1 and LINC01451, while genes outside the domain showed minimal changes (Fig. 6). These results support a regulatory role for the NOTCH1 domain boundary in constraining enhancer-promoter communication within the domain, consistent with prior work showing that topological insulation can modulate gene expression by shaping regulatory contact landscapes [40]. We also observed functional effects resulting from the increased NOTCH1 expression, particularly in cell proliferation and colony formation abilities. In K562 cells, deletion of CTCF sites was correlated with decreased proliferation and reduced colony size, potentially linking NOTCH1’s tumor suppressor role in myeloid lineage cells with its regulated expression [9, 10, 43]. These findings suggest that alterations in the 3D genome architecture could impact phenotypes by disrupting coordinated expression of genes within topological domains [18].

In terms of enhancer dynamics, we identified three putative enhancers within the NOTCH1 domain in K562 cells, with the − 70 kb enhancer residing within LINC01451 and playing a central role. This enhancer, when experimentally activated or repressed, significantly influenced NOTCH1 expression, supporting its role as a key regulatory element. Enhancer-promoter contact frequencies and enhancer activity assays confirm that this interaction contributes to the transcriptional control of NOTCH1 and may be modulated by architectural proteins, consistent with recent models of enhancer-promoter looping and their role in fine-tuning gene expression [21, 59]. Because the − 70 kb enhancer region overlaps LINC01451, a potential mechanistic contributor is enhancer-associated transcription (Fig. 6). Enhancer transcription and eRNAs have been implicated in the formation or stabilization of promoter-enhancer loops, and we and others have shown that CTCF binds RNA in ways that modulate its chromatin-binding and architectural functions. Notably, analysis of eCLIP data for CTCF [35] reveals binding interaction of CTCF with RNA molecules at the 3’ boundary. Then, it is therefore plausible that transcription and RNA molecules derived from the − 70 Kb/LINC01451 locus influence long-distance interactions with the 5’ boundary. From an experimental standpoint, separating enhancer transcription from LINC01451 transcription is challenging, and our current experimental design does not allow us to distinguish transcriptionally related influences in enhancer function.

Our cross-species analysis reveals that although the Notch1/NOTCH1 domain architecture is highly conserved, the transcriptional and phenotypic effects of boundary disruption are highly context-dependent. In human K562 cells, where the domain contains multiple active enhancers and NOTCH1 is expressed, deleting individual 5′ or 3′ CTCF sites primarily rewires intradomain enhancer-promoter interactions and increases expression of NOTCH1, NALT1, and LINC01451, while decreasing proliferation and clonogenic potential. In this scenario, boundaries seem to influence an active regulatory landscape.

In mice, the same locus exists within different chromatin environments. In mESCs, where Notch1 is active, and the domain is in an open chromatin state, deleting the 5′ boundary notably decreased Notch1 expression, while the 3′ deletion had only a minor impact. Both boundary deletions also reduced Egfl7, a gene just downstream of the domain, without significantly affecting nearby genes. In MEL cells, where Notch1 is silenced and located in a Polycomb-repressed chromatin environment, boundary deletions did not activate Notch1; instead, only Egfl7 was significantly lowered in the 3′ mutant, and there were no noticeable effects on cell growth or colony formation. These results show that the same CTCF-anchored domain can impact gene expression differently depending on promoter activity, enhancer presence, and chromatin state.

Comparisons between humans and mice also highlight differences in how neighboring genes respond to boundary disruption. In K562 cells, SEC16A and AGPAT2 show the most noticeable changes outside the domain, while EGFL7 remains largely unaffected; in contrast, in mouse cells, Egfl7 is consistently downregulated by boundary deletions, whereas Sec16a and Agpat2 remain relatively stable. These findings suggest that CTCF-anchored domains form a conserved architectural framework, with their regulatory output depending on the enhancer landscape and chromatin context rather than being a fixed regulatory entity.

In conclusion, our study underscores the essential role of CTCF domain boundaries in maintaining topological organization and their regulatory impact on gene expression across species. Our results also highlight that the transcriptional effects of topological domain structures are greatly influenced by the species of origin, chromatin context, and cellular identity.

Methods

Cell culture

K562 and MEL cells were grown in ISCOVE and RPMI, respectively, the medium was supplemented with 10% fetal bovine serum (Biowest) and 1% penicillin–streptomycin (Biowest). mESCs were grown in DMEM high glucose (Biowest-L0101-500) supplemented with 15% fetal bovine serum, 1% of non-essential amino acids (Lonza 13-114E), 2-mercaptoethanol 0.1 mM (Sigma M3148), L-Glutamine 2 mM (Gibco 56–85-9), 1% penicillin–streptomycin, LIF, PD0325901 and CHIR99021. Cells were maintained at 37°C and 5% CO₂. K562 (ATCC CCL-243) and mouse embryonic stem cells (mESCs; ATCC SCRC-1002) were obtained from the American Type Culture Collection (ATCC). The MEL cells were obtained from Dr. Gary Felsenfeld (Laboratory of Molecular Biology, National Institute of Diabetes and Digestive and Kidney Diseases, NIH, Bethesda, MD, USA).

Luciferase reporter assay

Putative enhancer regions were amplificated utilizing oligonucleotides listed in Table S1. Oligonucleotides (Sigma, St. Louis, MI, USA) were designed to include: 5'-Two-nucleotide overhangs + BglII motif + DNA specific complementary sequence -3'. Digested and dephosphorylated amplicons were cloned into pGL3-Promoter Luciferase vector (Promega, Madison, MI, USA) previously linearized with BglII enzyme (NEB, Ipswich, MA, USA). Insertion and directionality were screened using double enzymatic digestion. For luciferase activity quantitation, 100,000 K562 cells were seeded on 6-well plates in triplicate for each tested plasmid, then co-transfected with Renilla luciferase vector and pGL3 constructions using Lipofectamine 2000 (Invitrogen, Waltham, MA, USA) according to the manufacturer instructions. Luciferase activity was measured 24 h after transfection using the Dual-Luciferase Reporter Assay kit (Promega) on a Luminometer TD-20 (Turner Designs, San Jose, CA, USA). Internal normalization and relative luciferase activity were performed according to pGL3 vector manufacturer instructions.

Lentiviral production

For lentiviral production in HEK-293 T cells, 1 × 10⁷ cells were transfected in 10 ml of final volume of complete media containing 0.3 M MgCl₂, 2X HEBS (280 mM NaCl, 10 mM KCl, 1.5 mM Na₂HPO₄, 12 mM D-glucose and 50 mM HEPES [pH 7.05]) and a mix of plasmids as follow; 10 μg of the vector of interest, 3 μg of pMD2.G, and 6 μg of psPAX2. Transfection media was replaced by fresh complete media after overnight transfection. The supernatant containing the virus was harvested 48 h post-transfection and centrifuged for 90 min at 27,000 rpm, 4°C. The pellet containing the viral particles was eluted with 1X PBS overnight at 4°C. This lentiviral supernatant was aliquoted and stored at -80°C. 1 × 10⁵ cells per well in 6-well plates were infected with 1 ml of lentiviral supernatant and 2 ml of media supplemented with 8 μg/ml polybrene (Sigma). 24 h after infection, lentiviral media was replaced by fresh complete media supplemented with puromycin. Cells were maintained with the selection agent for 5 days before the experiment analysis.

Cell proliferation assay (MTT)

Briefly, 5 × 10³ cells per well were seeded in a 96-well plate. Cell proliferation was measured by Cell Proliferation Kit (MTT) (Roche) for 4 days according to manufacturer protocol.

RNA isolation and RT-qPCR

Total RNA was isolated using TRIzol Reagent (Invitrogen) according to the manufacturer protocol with minor modifications. Briefly, the cell pellet was resuspended in TRIzol Reagent and incubated at room temperature for 10 min. Chloroform (Invitrogen) was added, incubated at room temperature for 10 min, and centrifugated for 12,000 × g at 4°C. The aqueous phase was resuspended in 2-propanol (Invitrogen) and centrifuged for 10 min at 12,000 × g at 4°C. The RNA pellet was washed twice with 75% ethanol and resuspended in nuclease-free water. RNA was used directly to determine gene abundance by KAPA SYBR FAST One Step kit (KAPA Biosystems) using the StepOne Real-Time PCR System. HPRT was used as an internal control. RT-qPCR data was analyzed by the ΔΔCt method. Significance in gene expression was determined by Student’s t-test by Graphpad Prism 9.0. All the primers used in this study are listed in Table S1.

CRISPR-Cas9, CRISPRa and CRISPRi

For CRISPR-Cas9 mediated deletions, sgRNAs cloned into lentiCRISPRv2 were used to generate lentiviral particles as described above. Desired deletions were analyzed by PCR genotypification. For isolation of mutant cell clones, 1 × 10⁴ pooled cells were serially diluted in a 96-well plate containing 100 µl of complete iscove medium per well. After 2 weeks, single clones were identified by microscopy and expanded for subsequent genotypification by PCR. Deletion of each mutant cell clone was further characterized by cloning PCR fragments obtained from genotypification into pGEM-T Easy (Promega) and confirmed by Sanger sequencing. For CRISPRa and CRISPRi assays, sgRNAs targeting the enhancer element were cloned and used for lentivirus production as described before. For CRISPRa, 1 × 10⁵ K562 cells were transduced with lentivirus carrying lentiMPH-v2 (Addgene plasmid # 92,065). 24 h post-transduction, cells were selected using Hygromycin B (Thermo Fisher Scientific) for 4 days. Next, 1 × 10⁵ of selected cells were infected with lentivirus expressing sgRNAs cloned into lentiSAM v2. 24 h post-transduction, cells were selected using Puromycin (Sigma) for 4 days. For CRISPRi, 1 × 10⁵ K562 cells were transduced with lentivirus-carrying sgRNAs cloned into pLV-hU6-sgRNA hUbC-dCas9-KRAB-T2a-Puro. 24 h post-transduction, cells were selected using Puromycin (Sigma) for 4 days.

Chromatin immunoprecipitation (ChIP)

ChIP for CTCF and RAD21 was performed as described with some modifications [41]. Briefly, 4 × 10⁷ cells were crosslinked in 1% formaldehyde for 10 min at RT. Crosslink quenching was carried out with 0.125 M glycine for 5 min at 4°C. Cells were immediately centrifuged and washed twice with 1X PBS. The cell pellet was then re-suspended in cell lysis buffer (10 mM Tris–HCl [pH 7.5], 10 mM NaCl, 0.3% NP-40, supplemented with proteases inhibitors) and incubated for 30 min at 4°C. Nuclear fractions were isolated by centrifugation and dissolved in nuclear lysis buffer (50 mM Tris–HCl [pH 7.5], 10 mM EDTA, 1% SDS, supplemented with protease inhibitors). Chromatin was fragmented by sonication in a bioruptor for 12 cycles as follows: 30 seg on and 30 seg off. Chromatin fragmentation was evaluated by agarose gel electrophoresis. Insoluble chromatin was discarded by centrifugation at max speed, supernatant was kept on ice. 50 µg of chromatin per each IP was diluted 1:5 with dilution buffer (1% Triton X-100, 2 mM EDTA, 20 mM Tris–HCl, 150 mM NaCl, supplemented with protease inhibitors). Chromatin was precleared by adding 50 µl of blocked protein G/A beads for 2 h. The volume corresponding to 1% of chromatin used in each IP was saved as input. Chromatin was incubated overnight at 4°C with the corresponding primary antibodies. Next, 30 µl of blocked protein G/A beads were added to diluted chromatin and incubated for 2 h at 4°C. Beads were washed as follows: four washes with wash buffer I (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris–HCl, 150 mM NaCl, supplemented with protease inhibitors) and a final wash with wash buffer II (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris–HCl, 500 mM NaCl, supplemented with protease inhibitors). Beads along with the input chromatin were eluted in elution buffer (1% SDS and 100 mM NaHCO₃) and decrosslinking buffer (200 mM Tris–HCl, 400 mM NaCl, 0.4% SDS, 10 mM EDTA), incubated in a rotating wheel for 10 min at RT, and treated with RNase A for 1 h at 37°C and Proteinase K (NEB) for 4 h at 65°C. DNA was retrieved by adding 1:1 Phenol–Chloroform (Invitrogen) and mixed by rotation for 10 min at RT, then centrifuged for 10 min at 12,000 rpm at RT. The aqueous layer was retrieved. DNA was precipitated with 1 M ammonium acetate, glycogen (Roche), and 100% ethanol, for 2 h at − 70°C. The DNA pellet was obtained by centrifugation for 30 min at 12,800 rpm and washed twice with 70% ethanol. The pellet was resuspended in nuclease-free water. For CTCF ChIP-qPCR, purified DNA from ChIP was used for qPCR using iTaq Universal SYBR Green Supermix (Bio-Rad) and oligonucleotides flanking the CTCF-binding motif. Oligonucleotides used for ChIP-qPCRs are listed in Table S1. The antibodies used in this study were: anti-CTCF (5 µg, Millipore #07–729) and anti-IgG (2 µg, Millipore #12–371).

Circular chromosome conformation capture (4C-seq)

Primers to obtain 4C viewpoints and 4C-seq libraries were designed essentially as described by Krijger and collaborators [34], using the Primer designer for 4C viewpoints web tool https://mnlab.uchicago.edu/4Cpd/. For the library preparation process, ten million cells were crosslinked with 2% formaldehyde for 10 min, followed by 5 min quenching with glycine. Cells were washed once with PBS and then incubated for 20 min in lysis buffer (10 mM Tris–HCl pH 8, 10 mM NaCl, 0.2% Igepal) with 1 × protease inhibitors. First, three rounds of digestion were carried out with DpnII (NEB R0543S) at 37°C in a thermomixer at 500 rpm (100 U for 4 h, 100 U overnight, and 100 U for another 4 h); the restriction enzyme was inactivated by heating to 62°C for 20 min while shaking at 500 rpm. Next, the first ligation was performed with 2000 U of T4 DNA ligase (NEB M0202L) at 16°C in 7 mL of Milli-Q water overnight. Samples were then phenol–chloroform extracted, and ethanol precipitated, and the second digestion was performed overnight with 50 U of NlaIII (NEB R0125S) or HindIII (NEB R0104S). The second ligation was performed in 5 mL total with 3000 U of T4 DNA ligase (NEB M0202L). The second ligation material was purified with SPRI beads, and it was quantified with a Qubit dsDNA Assay Kit. The first round of PCR amplification with viewpoint-specific primers (Table S2) was performed with 4 × 50 µL PCR reactions with 200 ng of 4C template using 16 PCR cycles and the Phusion polymerase (NEB M0530L). A second round of PCR with universal primers containing Illumina adapters was performed, and the material was purified with SPRI beads. Two independent libraries per condition were sequenced in an Illumina HiSeq 4000 platform as single-end 150 bp reads. After adapter and quality-based trimming, reads were aligned to the hg19 human genome assembly utilizing pipe4C (version 1.1.6). To call valid chromatin interactions we used peakC [22].

Colony formation assay

Two hundred cells (K562 or MEL cells) were cultured with Methocult H4534 classic without EPO (Stem Cell Technologies) media, which contains ISCOVE, stem cell factor, 2-mercaptoethanol, fetal bovine serum, IL-3 and GM-CSF. Colonies were counted after 14 days.

CTCF knockdown

A previously designed shRNA against CTCF [50] was cloned into the plasmid pLL3.7 which contains a system to induce shRNA expression upon Doxycycline treatment. Lentiviral particles were generated to infect K562 cells as described above. shRNA expression was induced with 1 μg/mL every 24 h for seven days. Knockdown was evaluated by Western-blot.

Orca simulations

Orca, a sequence-based deep learning framework for multiscale genome architecture prediction, was used to generate in silico Hi-C contact maps around the NOTCH1 locus. Orca predicts 3D genome organization directly from DNA sequence across scales and was used here without custom retraining. Predictions were run following the official documentation. Briefly, using the human hg38 reference, we (i) predicted the reference (“WT”) contact maps over 32-Mb windows centered on each viewpoint and (ii) simulated the ΔS2 and ΔS4–S5 mutants as deletions at the corresponding coordinates. We used the standard Orca CLI modes (region for reference, del for deletions) with default settings; no additional preprocessing or model modifications were applied. The outputs are multiscale heatmaps in which pixel intensities represent log fold over a distance-based background, as defined by Orca. Predicted maps were visualized alongside our experimental 4C-seq profiles for qualitative comparison.

Statistical analysis

Data represent the mean ± standard deviation of three biological replicates. Significance was determined by a two-tailed unpaired Student’s t-test with Welch’s correction using GraphPad Prism 9.0.

Author contributions

A.J.C.-C. designed and conducted most of the experiments and computational analyses and wrote the manuscript. D.I.M.-S. performed experiments and data analysis in mouse cells. H.N.N.-M. assisted with CRISPRa and CRISPRi experiments. C.A.P.-A. contributed to 4C-seq data analysis. G.T.-U. assisted with ChIP experiments. G.G. performed Luciferase assays and lentivirus production. R.G.A.-M. designed and supervised most of the experiments and data analyses. F.R.-T. supervised the study and secured funding. All authors have read and approved the final version of the manuscript.

Funding

Consejo Nacional de Humanidades, Ciencias y Tecnologías [Investigación en Ciencia Básica 2017–2018 (A1-S-11844) and Fronteras de la Ciencia 2015 (Proyecto 290)] and Programa de Apoyo a Proyectos de Investigación e Innovación Tecnológica (DGAPA-PAPIIT IN203620 and IN206823).

Data availability

The raw and processed data supporting the findings of this study are publicly available in the Gene Expression Omnibus (GEO) database under accession number GSE296362. This repository includes sequencing files and processed datasets for replication and further analysis. Data can be accessed upon request by contacting Dr. Félix Recillas at frecilla@ifc.unam.mx. Reviewers can access the data using the following token: abunysiqvfqbnaz.

Declarations

Competing interests

The authors declare no competing interests.