RFX5 as a regulator of spatial chromatin contacts and gene expression

Abstract

The three-dimensional chromatin architecture is critical for gene regulation, yet factors beyond CCCTC-binding factor (CTCF) and cohesin remain poorly characterized. Here, we identify RFX5 as a novel regulator of chromatin organization. Using CRISPR-mediated RFX5 knockout A375 cells, together with RNA-seq, ChIP-seq, ATAC-seq, QHR-4C, and Hi-C, we demonstrate that RFX5 binds to promoters and enhancers, co-localizes with CTCF, RAD21, and H3K27ac, maintains chromatin accessibility, and preserves chromatin loop strength. RFX5 deletion alters the expression of ~2,000 genes, with strong suppression of cancer-associated genes and oncogenic pathways. Loss of RFX5 reduces CTCF/RAD21 occupancy and promoter accessibility at downregulated genes. Notably, RFX5 acts as an insulator to balance chromatin looping: its absence weakens enhancer–promoter contacts at oncogenic loci while enabling inappropriate long-range enhancer interactions at upregulated genes. Hi-C analysis reveals globally diminished loop strength, with only mild effects on TAD insulation and compartmentalization. These findings establish RFX5 as a key architectural factor that links 3D genome structure to transcriptional programs in cancer and immunity.

RFX5 acts as a novel architectural regulator that strengthens enhancer–promoter loops and functions as an insulator. Its deletion globally weakens chromatin loop strength, suppresses oncogenic gene expression, and modestly affects TAD insulation.

Introduction

The three-dimensional (3D) folding of the mammalian genome provides a fundamental framework for transcriptional regulation and cellular identity. At the megabase scale, chromosomes are segregated into transcriptionally active A and inactive B compartments, which reflect underlying epigenetic states and frequently undergo pathological switching in disease contexts^1–5. At sub-megabase resolution, the genome is further partitioned into topologically associating domains (TADs), within which chromatin loops connect enhancers and promoters to enable precise spatiotemporal control of gene expression^6–9.

At higher resolution, transcription relies on chromatin loops that connect distal enhancers with their target promoters^10,11. These loops are generated by the loop-extruding activity of cohesin, with RAD21 as a core subunit, and are stabilized by CTCF, which binds DNA in an orientation-dependent manner and anchors loop boundaries^12–14. Active enhancer–promoter contacts are frequently associated with histone modifications such as H3K27ac, underscoring their transcriptional competence¹⁵. Collectively, these architectural and epigenetic mechanisms safeguard accurate transcriptional outputs and preserve TAD insulation, thereby preventing inappropriate regulatory crosstalk. Disruption of this architecture, through loss of CTCF binding, cohesin dysfunction, or enhancer inactivation, can trigger transcriptional reprogramming and promote tumorigenesis^16–20.

While CTCF and cohesin are considered canonical architectural regulators, they do not fully account for the diversity of chromatin structures across different cellular contexts. Increasing evidence suggests that additional DNA-binding proteins may cooperate with, or modulate, these architectural factors. For example, transcription factor RFX5 (regulatory factor X5), a winged-helix protein best known for its indispensable role in regulating MHC class I and II gene expression, has recently been implicated in higher-order chromatin control^21–26. Beyond immune regulation, RFX5 has been reported to modulate the expression of the clustered protocadherin (cPcdh) genes, which serve as a paradigm for studying 3D genome regulation during neuronal development²¹. Mechanistically, RFX5 deletion was shown to enhance CTCF and RAD21 binding within the Pcdhα cluster and to increase long-range enhancer–promoter interactions, thereby reshaping local chromatin architecture²¹.

These findings raise the intriguing possibility that RFX5 functions as an architectural regulator analogous to CTCF and cohesin. However, whether RFX5 exerts such functions on a genome-wide scale and how it influences transcriptional programs remain largely unknown. In this study, we investigated the role of RFX5 in shaping higher-order chromatin architecture and regulating gene expression. By combining CRISPR-mediated RFX5 deletion with QHR-4C, RNA-seq, ChIP-seq, ATAC-seq, and Hi-C, we showed that RFX5 functions as an insulator and is indispensable for maintaining enhancer–promoter chromatin loops, TAD insulation, and compartmental integrity. Notably, RFX5 directly controls oncogenic signaling pathways through 3D genome remodeling, establishing it as a previously unrecognized architectural factor in gene regulation.

Results

RFX5 is a regulator for signaling networks central to cancer progression

To investigate the global role of RFX5 in gene regulation, we generated three homozygous RFX5 deletion single-cell clones (Del1, Del2, and Del3) with a pair of sgRNAs targeting the RFX5 coding sequence in the A375 cell line (Supplementary Fig. 1A). Sanger sequencing confirmed the homozygous deletion of RFX5 (Supplementary Fig. 1B). RNA-seq analysis revealed that RFX5 loss altered the expression of approximately 2,000 genes, with a comparable number upregulated and downregulated (Fig. 1a, Supplementary Fig. 2A-C, Supplementary Data 1). Notably, the overlap of downregulated genes across the three clones was substantially higher than that of upregulated genes, suggesting a more consistent repressive role for RFX5 at downregulated loci (Supplementary Fig. 2D).

Fig. 1. RFX5 regulates genes related to cancer progression.

a Differential expression gene (DEG) analysis of RNA-seq results from RFX5-deficient cells compared to the wild-type (WT) control shows that the expression levels of ~2,000 genes are altered upon RFX5 deletion. Some representative genes associated with cancer progression, including EGFR, VEGFA, IL6, ETS1, AXL, RUNX1, and ANKRD1, are marked with large dots. b RNA-seq results reveal that expression levels of seven representative genes associated with cancer progression are significantly downregulated upon RFX5 deletion. FPKM, fragments per kilobase of exon per million fragments mapped. For each genotype, at least 2 biological replicates were performed. c KEGG analysis of RNA-seq results from RFX5-deficient cells compared to the WT control shows that deletion of RFX5 mainly affects pathways related to cancer progression. Pathways highlighted in red are associated with cancer progression.

RFX5 was originally reported to regulate the expression of the MHC-I and MHC-II genes, so we examined how the expression of these genes was altered in RFX5-deficient cells. RNA-seq analysis confirmed that all three MHC-I genes (HLA-A, HLA-B, and HLA-C) and three members of the MHC-II genes (HLA-DRB1, HLA-DRA, and HLA-DPA1) were expressed in A375 cells (Supplementary Fig. 2E, F). Upon RFX5 deletion, the expression of HLA-A and HLA-B was significantly reduced, whereas all examined MHC-II genes were completely silenced. By contrast, the expression of HLA-C remained unaffected (Supplementary Fig. 2E, F), suggesting that its regulation may be independent of RFX5 and requires further investigation.

Interestingly, beyond immune genes, RNA-seq differential expression analysis revealed that a subset of cancer progression-associated genes, including EGFR, VEGFA, IL6, ETS1, AXL, RUNX1, and ANKRD1, were significantly downregulated upon RFX5 deletion, indicating an important role of RFX5 in regulating oncogenic pathways (Fig. 1a, b). KEGG pathway enrichment analysis further showed that these transcriptional changes converged on major oncogenic signaling cascades, such as PI3K-Akt, MAPK, focal adhesion, and Ras signaling (Fig. 1c). In addition, their downstream pathways HIF-1 signaling and EGFR tyrosine kinase inhibitor resistance were also affected²⁷ (Fig. 1c). Collectively, these findings indicate that RFX5 is required to sustain transcriptional networks driving immune function and cancer progression.

RFX5 co-localizes with CTCF and H3K27ac

To investigate how RFX5 influences gene expression, we performed genome-wide ChIP-seq profiling of RFX5. The results showed that RFX5 is enriched at promoter and enhancer regions, consistent with a role in promoting transcription (Fig. 2a). Integration of RFX5 ChIP-seq with RNA-seq data confirmed that the majority of both up- and downregulated genes were in close contact with RFX5 (Supplementary Fig. 3A). Next, we examined the binding patterns of CTCF, RAD21, and H3K27ac using ChIP-seq. RFX5 binds to approximately 45,000 sites genome-wide. Nearly half of RFX5 peaks overlapped with CTCF, and about one-third colocalized with H3K27ac (Fig. 2b). A smaller fraction coincided with RAD21, but notably, RFX5 tended to occupy the side of RAD21 opposite to CTCF (Fig. 2c), suggesting an orientation-dependent role in loop organization. Together, these findings indicate that RFX5 is strategically positioned at regulatory elements where enhancer activity and 3D genome architecture converge, providing a potential mechanism for its transcriptional control.

Fig. 2. RFX5 co-localizes with architectural proteins and modulates chromatin stability.

a Genome-wide distributions of RFX5. b Co-binding analysis of RFX5, CTCF and active enhancer mark H3K27ac reveals that RFX5 and CTCF frequently co-localize. c Relative binding position analysis of RFX5, RAD21, and CTCF shows that RFX5 and CTCF tend to flank RAD21. d Heatmap of ChIP-seq results showing binding patterns of active enhancer histone mark H3K27ac. e Heatmap of ChIP-seq results showing reduced CTCF binding in RFX5-deficient cells. f Heatmap of ChIP-seq results for RAD21 shows decreased cohesin occupancy in RFX5-deficient cells.

RFX5 influences the binding patterns of CTCF and cohesin

To investigate how RFX5 regulates gene expression, we first assessed whether it influences the enhancer-associated histone mark H3K27ac. ChIP-seq analysis using an H3K27ac-specific antibody revealed comparable enrichment in both RFX5-deficient and WT cells (Fig. 2d), indicating that RFX5 does not regulate transcription by directly modulating enhancer activity.

Given the extensive co-localization of RFX5 with the architectural proteins CTCF and cohesin, we next studied whether RFX5 deletion affects CTCF and cohesin occupancy. ChIP-seq with antibodies against CTCF and RAD21 showed that both CTCF and RAD21 occupancy were consistently and markedly reduced across all three RFX5-deficient cell lines (Fig. 2e, f). Integration of RFX5/ RAD21 ChIP-seq data and RNA-seq results showed no significant differences in RFX5 and cohesin occupancy in up-, downregulated, and unchanged genes (Supplementary Fig. 3B, C). Together, these results suggest that RFX5 modulates CTCF/cohesin binding strength, thereby influencing the stability of chromatin loops essential for transcriptional regulation.

RFX5 maintains promoter accessibility at downregulated genes

Since enhancer acetylation remained largely intact, we hypothesized that RFX5 may regulate transcription by modulating chromatin accessibility. ATAC-seq revealed a global decrease in chromatin accessibility in RFX5-deficient cells (Supplementary Fig. 4A). Notably, accessibility at transcription start sites (TSS) of downregulated genes was sharply reduced, whereas TSS accessibility of upregulated genes remained largely unchanged (Supplementary Fig. 4B). Moreover, the number of shared genomic regions showing decreased ATAC-seq signal in the 3 RFX5-deficient single cell clones was approximately three times higher than those showing increased signal (Supplementary Fig. 4C–E). These results indicate that RFX5 preferentially maintains promoter accessibility at genes it positively regulates.

RFX5 functions as barriers to balance chromatin loops

ChIP-seq revealed that RFX5 binds at both promoter and enhancer regions of HLA-DPA1 (Fig. 3a), indicating that RFX5 directly stabilizes these promoter-enhancer loops. To define the role of RFX5 in chromatin loop formation, we first applied QHR-4C using the HLA-DPA1 promoter as an anchor. Loss of RFX5 markedly reduced contacts with its proximal enhancer elements (Fig. 3a). We then extended this analysis to seven oncogenic loci, including EGFR, VEGFA, IL6, ETS1, AXL, RUNX1, and ANKRD1. ChIP-seq confirmed RFX5 binding at both promoter and enhancer regions of these genes (Fig. 3b–d, Supplementary Fig. 5A–D). In each case, QHR-4C revealed weakened promoter-enhancer contacts in RFX5-deficient cells (Fig. 3b–d, Supplementary Fig. 5A–D), indicating an important role of RFX5 in mediating long-range enhancer–promoter chromatin loops.

Fig. 3. RFX5 reinforces chromatin loop formation and stability.

a QHR-4C, using the promoter region of HLA-DPA1 as an anchor shows a significant decrease in chromatin loops between gene promoters and their nearby regulatory regions. b-d QHR-4C profiles of promoter regions of a repertoire of cancer-related genes reveal that chromatin loops between gene promoters and their distal enhancer elements are significantly reduced upon loss of RFX5, suggesting a role for RFX5 in regulating gene expression of EGFR (b), VEGFA (c), and IL6 (d). e Quantification of called chromatin loops identified from Hi-C results indicates that the total number of formed loops is not affected by RFX5 deletion. f Aggregate peak analysis (APA) plot for all loops called in Hi-C results of RFX5-deficient cells versus WT control shows a significant reduction in loop strength upon loss of RFX5. The APA score for each genotype is indicated in the upper-left corner of each panel.

For upregulated genes, such as those encoding cell adhesion molecules and isoforms in the PCDHB cluster, we observed increased expression in RFX5-deficient cells (Supplementary Fig. 6A, B). Wang et al. previously reported that RFX5 deletion in HEC-1-B cells enhances CTCF and RAD21 binding within the PCDHA cluster, leading to increased long-range enhancer–promoter interactions²¹. To address potential similarities or differences, we examined CTCF and RAD21 occupancy in our A375-derived RFX5 knockout clones. Unlike the PCDHA-dominant expression pattern in HEC-1-B cells, A375 cells primarily express PCDHB cluster genes. Consistent with their findings at PCDHA loci, we detected a slight increase in CTCF enrichment at PCDHB regions upon RFX5 loss (Supplementary Fig. 6C). However, we observed no significant change in RAD21 (cohesin) enrichment (Supplementary Fig. 6D), which may reflect cell-type-specific differences in architectural protein dynamics or regulatory context.

In line with this, QHR-4C using promoters of cPCDH genes or cell adhesion molecules as anchors revealed reduced contacts between promoters and proximal RFX5-bound sites, accompanied by increased interactions with distant H3K27ac-enriched enhancers (Supplementary Fig. 7A–F). These observations are consistent with RFX5 acting as an architectural barrier or insulator that restricts inappropriate long-range enhancer contacts while promoting specific proximal loops.

To determine whether this role extends to genome-wide loop formation, we performed Hi-C experiments. Hi-C data showed high reproducibility across replicates (Supplementary Fig. 8A, Supplementary Table 1). Although the total number of chromatin loops (~35,000) was comparable between WT and RFX5-deficient cells, loop strength was globally diminished upon RFX5 loss (Fig. 3e, f, Supplementary Data 2). Consistent with this, 4,642 loops were lost across all three knockout clones, whereas only 869 new loops appeared (Supplementary Fig. 8B). Approximately one-third of both up- and downregulated differentially expressed genes were associated with altered chromatin loops (195 or 31.25% in upregulated genes and  366 or 32.33% in downregulated genes, Supplementary Fig. 8C).

Further integration of ATAC-seq and Hi-C data revealed distinct chromatin accessibility patterns between loop-dependent and loop-independent genes. Notably, loop-independent genes (those without detectable changes in enhancer–promoter contacts) exhibited significantly higher baseline ATAC-seq signals at promoters in WT cells compared to loop-dependent genes (Supplementary Fig. 8D), suggesting that RFX5 regulates gene expression through at least two partially independent mechanisms: (i) maintenance of chromatin loop strength, which primarily influences loop-dependent DEGs—including many oncogenic loci where loop weakening occurs without major promoter accessibility loss—and (ii) direct or indirect preservation of promoter accessibility, which appears to contribute more substantially to the regulation of loop-independent DEGs. Together, these findings support a multifaceted role for RFX5 in coordinating 3D chromatin organization and local chromatin state to achieve precise transcriptional control.

RFX5 strengthens TAD organization genome-wide

We next examined whether RFX5 influences higher-order chromatin architecture. Using the Hi-C data, we identified 3,432 TADs in WT cells and observed a modest decrease in TAD number in RFX5-deficient cells (Fig. 4a, Supplementary Data 3). Aggregate domain analysis (ADA) and average insulation score profiles showed a modest weakening of TAD strength and boundary insulation upon RFX5 loss (Fig. 4b, c, Supplementary Fig. 9A, B).

Fig. 4. RFX5 deletion leads to modest alterations in topologically associating domains (TAD).

a Quantification of called TADs identified from Hi-C data shows a slight reduction in the number of TADs upon loss of RFX5. b Aggregate domain analysis (ADA) plot for all TADs called from Hi-C results of RFX5-deleted cells and their WT control, showing a modest weakening of TAD strength upon deletion of RFX5. c Average insulation score analysis in WT and RFX5-deficient cells reveals a modest weakening of TAD boundary insulation upon RFX5 deletion. Paired Wilcoxon-test, effect sizes are indicated.

Importantly, regions with reduced RAD21 occupancy did not exhibit significantly different insulation scores between WT and knockout cells (Supplementary Fig. 9C, D), suggesting that RFX5 reinforces TAD boundaries at least partially independently of cohesin occupancy changes.

RFX5 maintains compartmental stability

Finally, we examined whether RFX5 influences large-scale compartmentalization with our Hi-C data. Continuous bins with the same labels of positive or negative were defined as one compartment. Approximately 1,500 compartments were called per cell line, with a comparable proportion of A and B compartments (Fig. 5a, Supplementary Data 4). However, correlation analysis revealed reproducible compartment-level differences between WT and RFX5-deficient cells (R < 0.97, p < 0.001; Fig. 5b). To better understand the compartment-level difference, we calculated localized Pearson correlation matrices for each of the seven cancer-related genes mentioned above and the intensity of the red color represents the strength of positive correlation, reflecting the frequency of homotypic interactions between regions of the same compartment type. Notably, upon RFX5 deletion, regions around these genes interact more frequently, suggesting an enhancement to compartments (54.2 Mb to 56.2 Mb in the EGFR locus, etc., Fig. 5c, Supplementary Fig. 10A–F). Saddle plot analysis also confirmed a global increase in compartment strength across all knockout clones (Supplementary Fig. 11A).

Fig. 5. RFX5 maintains compartmental stability.

a Quantification of called compartments identified from Hi-C data shows that the total number of formed compartments is not influenced by RFX5 deletion. Both compartments A and B account for approximately 50% in WT and RFX5-deficient cells. b A scatterplot comparing compartment features of WT and RFX5-deficient cells shows altered compartmentalization upon loss of RFX5. c Close-up of compartment features around the EGFR locus reveals increased interaction frequency in RFX5-deficient cells. d Quantification of the modified compartment in WT cells upon RFX5 deletion, showing that half of the original compartment is perturbed upon RFX5 deletion. e A schematic model illustrating the architectural roles of RFX5. At oncogenic loci, RFX5 reinforces proximal enhancer–promoter loops that are critical for maintaining high-frequency enhancer–promoter interactions; its deletion significantly weakens these specific contacts, thereby impairing transcriptional activation and contributing to the strong downregulation of cancer-related genes. In contrast, at certain upregulated loci (e.g., cell adhesion molecules and PCDHB cluster genes), RFX5 restricts inappropriate long-range enhancer contacts; its loss permits distal enhancer engagement, potentially allowing ectopic activation through alternative regulatory pathways.

To quantify compartment perturbations, we compared compartment identities between WT and RFX5-deficient cells. Approximately half of WT compartments were modified in RFX5-deficient cells (746 or 49.7% in Del 1,749 or 49.9% in Del 2,736 or 49.0% in Del 3, Fig. 5d). Specifically, the “no change” fraction represents genomic regions that maintained their original compartment identity between WT and deletion clones, whereas the “modified” fraction denotes regions that underwent compartmental switching (e.g., A-to-B or B-to-A transitions).

Furthermore, we analyzed eigenvector (PC1) values specifically at genomic regions with downregulated RAD21 binding. In these RAD21-depleted regions, eigenvector values shifted toward B-compartment identity in the RFX5 knockout clones compared to WT cells (Supplementary Fig. 11B). This shift indicates a loss of active “A-compartment” features due to RAD21 downregulation, which in turn enhances intrinsic phase separation and strengthens compartmentalization globally, independent of cohesin’s loop-extruding effects (Supplementary Fig. 11B).

Discussion

Considerable progress has been made in understanding how distal enhancers communicate with target promoters through precise spatial chromatin organization. It is now widely accepted that chromatin loops and topologically associating domains (TADs) facilitate accurate enhancer–promoter interactions, while CTCF and cohesin serve as core architectural proteins that establish and maintain these structures^28–30. However, the full repertoire of factors that fine-tune 3D genome architecture in a context-dependent manner remains incompletely defined. Emerging evidence points to additional transcription factors that cooperate with or modulate canonical architectural proteins to regulate chromatin folding and gene expression^16,31–33.

In this study, we identify RFX5 as a previously unrecognized architectural regulator that reinforces enhancer–promoter chromatin loops, stabilizes TAD boundaries, and preserves compartmental stability. RFX5 depletion in A375 melanoma cells altered the expression of approximately 2,000 genes, with a particularly strong suppressive effect on cancer-associated genes (EGFR, VEGFA, IL6, ETS1, AXL, RUNX1, and ANKRD1). These genes converge on pivotal pathways such as PI3K-Akt, MAPK, and HIF-1 signaling, positioning RFX5 as a central regulator of pro-tumorigenic networks^34–40. The transcriptional changes were accompanied by genome-wide weakening of chromatin loop strength and modest disruption of TAD insulation, without major alterations in enhancer acetylation (H3K27ac). Notably, RFX5 loss reduced CTCF and RAD21 occupancy at many sites, suggesting that RFX5 helps stabilize these canonical architectural proteins—possibly by facilitating their recruitment, stabilizing their DNA binding, or influencing cohesin loop extrusion dynamics.

A central finding is that RFX5 functions as an architectural barrier or insulator to balance chromatin contacts. At oncogenic loci, RFX5 reinforces proximal enhancer–promoter loops that are critical for maintaining high-frequency enhancer–promoter interactions; its deletion significantly weakens these specific contacts, as shown by QHR-4C and global Hi-C analysis. This reduction in loop strength likely decreases the spatial proximity and contact frequency between enhancers and promoters, thereby impairing transcriptional activation and contributing to the strong downregulation of cancer-related genes (Fig. 5e). In contrast, at certain upregulated loci (e.g., cell adhesion molecules and PCDHB cluster genes), RFX5 restricts inappropriate long-range enhancer contacts; its loss permits distal enhancer engagement, potentially allowing ectopic activation through alternative regulatory pathways (Fig. 5e). This bidirectional effect on loop usage—strengthening productive proximal loops while insulating against aberrant distal ones—underscores RFX5’s role in maintaining precise enhancer–promoter specificity.

Separately, RFX5 also influences chromatin accessibility, but this appears to operate in a more targeted or context-specific manner. ATAC-seq revealed a global decrease in chromatin accessibility upon RFX5 deletion, with reductions at transcription start sites (TSS) particularly evident among downregulated genes (Supplementary Fig. 4A–E). Integration with Hi-C data further revealed distinct patterns: loop-independent genes (those without detectable changes in enhancer–promoter contacts) exhibited significantly higher baseline ATAC-seq signals at promoters in WT cells compared to loop-dependent genes (Supplementary Fig. 8D).

This dissociation suggests that RFX5’s architectural role in maintaining chromatin loop stability can exert dominant effects on transcription in certain gene cohorts—such as many oncogenic loci where loop weakening occurs without substantial promoter accessibility loss—while its influence on promoter accessibility likely contributes more substantially to the regulation of loop-independent downregulated genes. These findings underscore the multifaceted nature of RFX5, which coordinates 3D chromatin organization and local chromatin state through partially independent mechanisms to achieve precise transcriptional control.

Our Hi-C analysis revealed that loss of RFX5 leads to a modest reduction in TAD number and a subtle weakening of TAD strength and boundary insulation. Although the paired comparisons of insulation scores reached high statistical significance, the effect sizes were modest, consistent with the nearly overlapping insulation profiles. We note that our Hi-C analysis was performed with two WT biological replicates and three independent RFX5-knockout clones. Although statistically detectable, the observed modest changes in TAD boundary insulation and TAD strength therefore await further in-depth confirmation in future studies employing larger numbers of biological replicates or complementary approaches.

Approximately two-thirds of the differentially expressed genes (DEGs) (~68%) appear loop-independent on the basis of our Hi-C analysis. While RFX5 loss sharply reduces promoter accessibility at downregulated genes (Supplementary Fig. 4b), TSS accessibility at upregulated genes remains largely unchanged. Thus, the upregulation of loop-independent genes cannot be explained solely by alterations in chromatin looping or local accessibility. These observations suggest the involvement of additional, indirect mechanisms. For instance, downregulation of key repressors or altered competition among transcription factors at shared binding sites may contribute to the observed upregulation. Cell-type-specific differences in RFX5’s insulator activity (e.g., at the PCDHB cluster) may further modulate these effects. We acknowledge this as a limitation of the present study. Future work combining RFX5 degron systems, TF co-occupancy analyses, or targeted perturbation of candidate intermediary factors will be required to fully elucidate these indirect pathways.

Our Hi-C results also revealed that loss of RFX5 leads to some modification in the higher-order structure of compartments. Nearly half of the called compartments in WT control were at least partly modified in RFX5-deleted cells, which differs from that previously reported for transcription factors such as CTCF. However, the biological consequences of these compartment-scale changes remain to be elucidated.

RFX5 is known for its ability to regulate the expression of both MHC-II and MHC-I genes. Our RNA-seq results also revealed significant downregulation of MHC-II genes as well as most of MHC-I genes except for HLA-C, which is not influenced by RFX5 deletion. Meanwhile, the regulation of other MHC-I genes (HLA-A, HLA-B) is also not as significant as that of MHC-II genes (HLA-DRB1, HLA-DRA and HLA-DPA) in A375 cells. The basis for this differential regulation remains unclear and warrants further investigation.

Methods

Cell culture

Human A375 cells were cultured in DMEM medium (Hyclone), supplemented with 10% (v/v) FBS (VivaCell) and 10 U/ml (1×) Penicillin-Streptomycin (Gibco). Cells were cultured at 37 °C in a 5% CO₂ humidified incubator and were passaged every three days.

CRISPR screening of RFX5-deleted homozygous single-cell clones

To generate homozygous RFX5 deletion single cell clones, WT A375 cells were transiently transfected with Cas9 and sgRNA plasmids using Lipofectamin 3000 kit (Thermofisher, L3000015) according to the manufacturer’s instructions. For a 6-well plate, one Cas9 plasmid (1.5 μg) and two sgRNA plasmids (each 0.75 μg) were used for transfection. The sgRNA plasmids contain a puromycin-resistant gene. 48 h after transfection, the culture medium was changed to complete DMEM medium supplemented with 1 μg/ml puromycin and cultured for another 5 days to kill cells that failed to be transfected. Afterwards, puromycin was washed away, and cells were suspended in a single clone solution and plated into 96-well plates. Two weeks later, wells with just one cell clone could be marked and genotyped by PCR and Sanger sequencing. All oligonucleotides are listed in Supplementary Table 2.

RNA-seq experiments

RNA-seq experiments were performed as previously described with some modifications. Briefly, cells at ~80% confluency were collected into Trizol reagents (ThermoFisher) for RNA-seq experiments. Total RNA was isolated and precipitated with trichloromethane and isopropyl alcohol, followed by washing with 70% ethanol. For each RNA-seq experiment, 1 μg of total RNA was used to generate mRNA using the mRNA capture bead kit (Vazyme, N403) following the manufacturer’s instructions. The extracted mRNA was then washed and heat fragmented to an average size of ~150–250 bp using the RNA-seq library Prep kit (Vazyme, NR606). Fragmented RNA was then reverse-transcribed, followed by synthesis of second-strand cDNA. Afterwards, the cDNA was purified with Ampure XP Beads (Beckman). The purified cDNA was then end-repaired and phosphorylated at the 5’ end. Adaptors were then added to both ends of the cDNA, and redundant adaptors were removed using Ampure XP Beads (Beckman). The processed cDNA was finally amplified by PCR to generate sequencing libraries. All RNA-seq experiments were performed with at least 2 biological replicates. deletion and their WT control cells were experimented in the same round. The RNA-seq libraries were sequenced using an Illumina Novaseq X plus platform at 2 × 150 bp mode.

ChIP-seq experiments

ChIP-seq experiments were performed as previously described with some modifications. Briefly, for one experiment, a 10 cm dish of cells at ~80% confluency was collected and cross-linked with 1% formaldehyde in 10% FBS/PBS, followed by a wash with PBS supplemented with 1× proteinase inhibitor. Cells were then resuspended in lysis buffer and sonicated to ~100–10,000 bp using the SCIENTZ Sonication system (10% energy, with working time 10 s and resting time of 50 s, 10 cycles). The sonicated samples were then centrifuged to remove insoluble debris and precleaned with protein-A agarose beads (Millipore, 16–157) for 1 h with slow rotation to remove non-specific binding. Samples were then incubated with antibodies against CTCF (CST, 3418S), RAD21 (Abcam, Ab992), RFX5 (Rockland, 200401194) and H3K27ac (CST, 8173S) overnight on a rotator at 4 °C to form antibody-protein-DNA complex (Supplementary Table 3). The complex was formulated with protein A-agarose beads and cleaned with low salt buffer, high salt buffer, no salt buffer, LiCl buffer and TE buffer in turn, and finally eluted into elution buffer (1% SDS, 0.1 M NaHCO₃). Reverse cross-link and DNA purification were then performed, and the samples were prepared into libraries using the VAHTS Universal DNA Library Prep Kit for Illumina (Vazyme, ND610) following the manufacturer’s instructions. The ChIP-seq libraries were sequenced using an Illumina Novaseq X plus platform at 2 × 150 bp mode.

ATAC-seq experiments

ATAC-seq experiments were performed as previously described. Briefly, 10⁶ cells were used for one experiment. Cells were washed twice with 1 ml of cold PBS and spined down at low rotations. The cell pellets were resuspended and incubated in cold lysis buffer to release nuclei, followed by centrifugation to discard supernatant. The nuclei were then fragmented and tagged with transposase. After transposition, DNA products were purified, and libraries were constructed using the Hyperactive ATAC-seq library prep kit for illumina (Vazyme, TD711). The ATAC-seq libraries were sequenced using an Illumina Novaseq X plus platform at 2 × 150 bp mode.

Quantitative high-resolution chromatin conformation capture copy (QHR-4C)

The QHR-4C experiments were performed as previously described. Briefly, for one experiment, a 10 cm dish of cells at ~80% confluency was collected, cross-linked with 2% formaldehyde and quenched by adding excess pre-chilled glycine. The cells were then collected with a centrifuge and permeabilized with 4C permeabilization buffer supplemented with 1 × proteinase inhibitor. Afterwards, the cells were digested with DpnII for a total of 16 h, followed by in-nuclei ligation for 20 h. The ligated samples were then reverse cross-linked and purified, finally eluted into distilled water.

The purified ligation products were then fragmented to ~200–1000 bp using the SCIENTZ Sonication system (10% energy, with working time 10 s and resting time of 50 s, 2 cycles). Fragmented products were then linear-amplified to formulate target sequences in a single strand that are tagged with 5’ biotin primers. Target sequences were captured with Streptavidin Magnetic Beads (ThermoFisher, 65001) according to the manufacturer’s instructions. The adaptor was then ligated to the 3’ end of the tagged DNA. Finally, the QHR-4C libraries were generated by PCR amplification on beads with a pair of PCR primers that suit the Illumina sequencing criteria. Samples of the same viewpoint with different barcodes or index were purified in one column. The PCR products were purified with a PCR purification kit (Qiagen). All the QHR-4C libraries were pooled and sequenced on an Illumina Novaseq X plus platform at 2 × 150 bp mode.

In situ Hi-C experiments

We constructed Hi-C libraries from two biological replicates of WT cells and three independent homozygous single-cell clones of RFX5 deletion. Hi-C experiments were performed as previously described. Briefly, ~2 × 10⁷ of cells were collected, cross-linked with 2% formaldehyde, quenched by adding excess pre-chilled glycine, and finally cleaned with PBS. The cross-linked samples were then suspended and lysed in lysis buffer, followed by sufficient digestion with MboI. Digested products were then heat-deactivated, end-repaired and tagged with biotin-14-dATP (Thermofisher, 19524016). Fragments ligation was then performed, followed by reverse cross-linking and purification of DNA.

Ligation products were then sonicated using the SCIENTZ Sonication system (10% energy, with working time 10 s and resting time of 50 s, 2 cycles), and size selection was performed to generate fragments with a size of ~300–500 bp. Afterwards, Sonicated products tagged with biotin were pulled down with Streptavidin Magnetic Beads (ThermoFisher, 65001) according to the manufacturer’s instructions. The captured products were then end-repaired, 3’ d (A) tailed on beads, followed by adaptor ligation. The Hi-C library was finally constructed by PCR amplification. All the QHR-4C libraries were pooled and sequenced on an Illumina platform at 2 × 150 bp mode.

High-throughput sequencing data analysis

For RNA-seq data, reads were aligned to the human genome (hg38_tran) using Hisat2 (version 2.2.1) to generate SAM files and converted into BAM files using samtools (version 1.2.1). The FPKM value was calculated using a BAM file as input with the Cufflinks software (version 2.2.1). The gene count matrix was generated using featureCounts (version 2.0.3). DEG was calculated with DESeq2 and clustered with clusterProfiler (version 4.14.6).

For ChIP-seq data, reads were aligned to the human genome (UCSC hg19) using bowtie2 (version 2.5.4). SAM files were then converted into BAM files and indexed. Peaks were finally called with macs2 (version 2.2.9.1) with BAM files as input. The heatmap matrix was generated using computeMatrix (version 3.5.4) and illustrated with plotHeatmap software.

For QHR-4C data, Viewpoint primer sequences were first removed and duplicated paired-end reads caused by PCR were then removed by FastUniq (version 1.1). Only the unique reads were used to align to the human genomes using the Bowtie2 software. SAM files were then converted into BAM files, and the reads per million (RPM) value was calculated using the r3Cseq program (version 3.20) in the R package (4.3.3).

For Hi-C data, to ensure the robustness of the results, Hi-C libraries were constructed from two biological replicates of WT cells and three independent homozygous single-cell clones of RFX5 deletion. The HiC-Pro toolkit (Servant, N) was utilized to map raw FASTQ reads to the human reference genome assembly (GRCh37/hg19). Following alignment, the allValidPairs files were converted to hic format files with the hicpro2juicebox script from the HiC-Pro package, allowing subsequent analysis with Juicer tools (Durand, N. C). These hic files were further transformed to the cooler format via hic2cool (v0.8.3), and the contact matrix was normalized to a contact depth of 100 million based on cis-interactions using the Knight-Ruiz (KR) normalization method. Hi-C contact maps for specific genomic regions were generated using the fancplot module from the FAN-C toolkits (Kruse, K.), while differential contact heatmaps were constructed using Seaborn, a Python library.

For compartment analysis, the cis-Eigenvector 1 values and Pearson’s correlation matrix were computed using the Juicer software (Durand, N. C.) at a resolution of 100-Kb. The cis-Eigenvector 1 values were subsequently converted into the bedGraph format for visualization in the UCSC genome browser (Kent, W. J.). Scatterplots with heat density representations were created using the R package ggpointdensity (v0.1.0).

To systematically identify genomic regions undergoing A/B compartmental switching upon RFX5 deletion, genomic bins were initially categorized into A or B compartments based on their PC1 values. To accurately detect state switches like A-to-B or B-to-A transitions upon RFX5 deletion, we compared the compartment intervals between the WT and RFX5 deletion clones using bedtools cluster and a custom Python script. This method precisely identified overlapping genomic coordinates exhibiting an absolute compartment state reversal. Finally, to maintain consistent resolution, these identified switched fragments were mapped back to the original WT compartmental bins using bedtools intersect.

Topologically associating domains (TADs) were identified at a high resolution of 10-Kb using the Hidden Markov Models (HMM) algorithm, as previously described (Dixon, J. R). Briefly, the dense contact matrix at 10-Kb bins was processed using the DI_from_matrix script to calculate the Directionality Index (DI) score. This DI score was then used to predict the hidden states of genomic bins via HMM models. Adjacent bins with the same state were merged into regions. Regions containing fewer than 3 bins or with a median posterior bin probability lower than 0.99 were filtered out using a Perl script. TADs were ultimately defined as regions extending from an upstream boundary to a downstream boundary. The DI score was also converted to the bedGraph format for visualization in the UCSC genome browser.

Chromatin loops were identified on 10-Kb binned contact matrices using Chromosight (v1.6.3) (Matthey-Doret, C), with a minimum genomic distance threshold of 40 Kb for loop detection. Specific loops were further analyzed using the Aggregate Peak Analysis (APA) with cooltools (v0.6.1) (Open2C) and coolpup.py (v1.1.0) (Flyamer, I. M).

Identification of the differential loop

To categorize loops as gained, loop or stable upon RFX5 deletion, we first merged raw loops calling from all WT and RFX5 deletion clones. Specifically, loop anchors within a 20-kb range were combined to eliminate calling-related positional shifts, generating a “union loop set”. The interaction intensity of the union loop set across all samples was quantified using “Chromosight Quantify”. Subsequent statistical analysis identified 4,642 lost loops, 869 gained loops, and 17,314 stable loops in the RFX5 KO clones.

Identification of A/B-compartment switching

To systematically identify genomic regions undergoing A/B compartmental switching upon RFX5 deletion, genomic bins were initially categorized into A or B compartments based on their PC1 values. To accurately detect state switches like A-to-B or B-to-A transitions upon RFX5 deletion, we compared the compartment intervals between the WT and RFX5 deletion clones using bedtools cluster and a custom Python script. This method precisely identified overlapping genomic coordinates exhibiting an absolute compartment state reversal. Finally, to maintain consistent resolution, these identified switched fragments were mapped back to the original WT compartmental bins using bedtools intersect.

Statistics and reproducibility

Data analysis was conducted using software mentioned above in the High-throughput sequencing data analysis section. Data visualization was conducted using Microsoft Excel 2024 and GraphPad Prism version 8. To assess statistical significance, we employed paired Wilcoxon test in Hi-C data analysis. Significance was determined with adjusted p values less than 0.05.

For RNA-seq experiments, two biological replicates were performed for wild-type (WT) cells and three independent homozygous RFX5-knockout clones (Del1, Del2, Del3) were used. For ChIP-seq, ATAC-seq, and QHR-4C experiments, two replicates were used for both WT cells and three RFX5-deletion clones. For Hi-C experiments, two independent biological replicates were performed for both WT cells and three RFX5-knockout clones. No error bars or inferential statistics (including p-values) were calculated or presented for analysis with n = 2. Individual data points are shown where applicable.

Author contributions

Y.W. conceived the research. X.G., assisted by Y.H., X.L., Z.J., and Y.L. did experimental work. Y.Z. analyzed experimental data. Y.W. and X.G. wrote the manuscript with inputs from all authors.

Peer review

Peer review information

Communications Biology thanks Argyris Papantonis who co-reviewed with Andrés Penagos-Puigand and the other, anonymous, reviewer for their contribution to the peer review of this work. Primary Handling Editors: Tobias Goris. A peer review file is available.

Data availability

Source data can be found in the Supplementary Data. The raw high-throughput sequencing data and processed data generated in this study could be accessed from the GEO database under accession code GSE305795 .

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-10279-9.