Decoding 3D chromatin architecture reveals distinct enhancer classes underlying hierarchical gene regulation in prostate cancer Authors: Huan Cao, Zexun Wu, Baixi Ji, Seolyn Yang, Leonardo Gonzalez-Smith, Andrew Vu, Suhn K. Rhie Published: 2026 ABSTRACT BackgroundThe transcription process is controlled by non-coding regulatory elements, more than 70% of which are putative enhancers. These enhancers comprise over 600,000 regions and are marked by histone modifications. However, the mechanisms by which altered enhancers in cancer cooperate within the three-dimensional chromatin architecture to drive oncogenic programs remain poorly understood.ResultsBy integrating 201 H3K27ac ChIP-seq datasets from prostate, we identify 3,216 high-confidence prostate cancer-specific putative enhancers. Ultra-high-resolution chromatin interaction profiling by Region Capture Micro-C at a representative chr6q24.1 locus reveals that these enhancers form cancer-specific, highly nested interactions with promoters that coalesce into a multi-connected hub absent in normal prostate cells. CRISPR/Cas9 perturbations of these enhancers, examined one by one, distinguish enhancer classes within the hub. Deletion of a central enhancer collapses hub-wide enhancer activities and architecture, leading to the downregulation of target genes, impaired proliferation, and reduced clonogenic growth. In contrast, deletion of a redundant enhancer results in minimal transcriptional changes, as neighboring enhancers rescue cancer signaling through compensatory architectural rewiring that strengthens alternative enhancer-promoter interactions. We also observe that FOXA1, a pioneer transcription factor activated in prostate cancer, directly binds to these enhancers and regulates distinct enhancer classes, leading to varying degrees of chromatin accessibility and gene expression changes.ConclusionsThese findings suggest that enhancers function in a coordinated manner, forming multi-connected cancer-specific chromatin interaction hubs, with distinct enhancer classes contributing differently to gene regulation. This study advances our ability to modulate gene expression in a cell type-specific manner, opening new avenues for precision therapies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13059-026-04178-9. Background In the United States, prostate cancer is the most prevalent cancer in men, accounting for 30% of all diagnoses [1]. While studies have identified driver genetic alterations such as mutations, fusions, and copy number changes in prostate cancer [2], epigenetic alterations have emerged as critical players in cancer initiation and progression [3]. Among them, changes in the activity of regulatory elements that control transcription are major drivers of carcinogenesis. For example, promoters of tumor suppressor genes are often epigenetically silenced, whereas promoters of oncogenes are activated in cancer cells compared to normal cells [4, 5]. In addition, enhancer reprogramming plays a prominent role in reshaping the transcriptional landscape by regulating lineage- [6–9] and cancer-specific genes [10–13]. Therefore, gaining mechanistic insights into aberrant enhancer activation is essential for understanding the molecular basis of prostate tumorigenesis. The ENCODE project identified candidate regulatory elements, revealing that over 600,000 regions possess enhancer-like signatures in the human genome, more than tenfold greater than the number of promoter-like and insulator-like regions [14]. A comparison of regulatory elements revealed that enhancer activities are more cell type- and individual-specific than promoters and insulators [15–17], due to binding of transcription factors (TFs) crucial for defining cell identity [18, 19]. Technological advances now enable precise enhancer editing [20, 21], leading to recent FDA-approved therapies that allow targeted regulation of cell-type specific gene expression [22, 23]. Despite the crucial role of enhancers and their potential for precision medicine, a comprehensive understanding of the mechanisms by which enhancers exert their regulatory functions is greatly needed. However, identifying target genes of enhancers is challenging. Enhancers are distal regulatory elements located far from transcription start sites (TSSs), so they can regulate not only the nearest genes but also distant genes by forming chromatin loops with the promoters of their target genes [24]. Moreover, a single enhancer has been reported to regulate multiple genes [15–17]. Importantly, the mechanisms by which enhancers regulate the expression of genes critical for prostate carcinogenesis within the context of chromatin architecture remain poorly understood. In prostate cancer, we previously performed Hi-C and observed cancer-specific enhancer-promoter (E-P) interactions near the androgen receptor (AR) gene locus, a critical driver gene overexpressed in prostate cancer [10]. However, conventional Hi-C predominantly detects chromatin loops anchored by the architectural cohesin complex proteins such as CTCF, which define topologically associating domains (TADs) [25], whereas the number of detected E-P loops remains relatively small compared to the total repertoire of active enhancers [26]. These limitations highlight the need for higher-resolution approaches capable of capturing enhancer-centered chromatin interactions to uncover transcriptional dysregulation during prostate tumorigenesis. To address these limitations, Region Capture Micro-C (RCMC), a new method that combines nucleosome-resolution contact mapping of Micro-C with targeted capture, was recently developed to enable ultra-high-resolution interrogation of locus-specific chromatin architecture [27]. Nevertheless, functional studies endogenously characterizing enhancers in prostate cancer and their roles in three-dimensional (3D) epigenomics remain very limited. In this study, leveraging recent advances in molecular assays and applying pioneering bioinformatic methods, we identified and characterized prostate cancer-specific enhancers (PCEs) clustered within defined genomic regions. Using ultra-high-resolution chromatin interaction mapping, we uncovered cancer-specific, multi-connected E-P hubs centered on PCEs. Functional perturbations of individual PCEs through CRISPR/Cas9-mediated deletion, followed by multi-omic analyses, revealed distinct regulatory roles of enhancers within the hub. We found that PCEs form interconnected chromatin loops and preferentially regulate other PCEs, rather than enhancers shared with normal prostate. Individual PCEs differed in their contributions to regulating local chromatin architecture, enhancer activity, and target gene expression. Some PCEs, which we termed central enhancers, were essential for maintaining the activity of other PCEs, stabilizing E-P loops, and sustaining transcriptional upregulation in prostate cancer. Others were functionally redundant and rescuable by activation of neighboring PCEs, revealing a cooperative and hierarchical organization of enhancers within the hub. Furthermore, perturbation of the central PCE reduced chromatin accessibility and disrupted FOXA1 binding, indicating its role in regulating distinct enhancer classes and leading to varying degrees of changes in the epigenome and transcriptome. Results Integration of epigenomic datasets identifies prostate cancer-specific enhancers (PCEs) To identify enhancers associated with prostate tumorigenesis (Fig. 1A), we systematically processed and analyzed 165 H3K27ac ChIP-seq datasets from prostate tumor tissue samples and 36 from normal prostate tissue samples across 3 publicly available studies (GSE120738 [28], GSE130408 [29], GSE161948 [30]; Additional file 1: Table S1A). Peak calling yielded a median of 123,859 (SD = 36,001) peaks per tumor and 113,918 (SD = 29,396) per normal sample, with no statistically significant difference in peak numbers (Mann–Whitney P = 0.37). By assessing the differences in the genome-wide H3K27ac landscape between prostate tumor and normal prostate tissues, we classified putative enhancers into three categories (Fig. 1A, B): common enhancers (CEs) across conditions, normal prostate-specific enhancers (NPEs) that are activated in the normal prostate, and prostate cancer-specific enhancers (PCEs) that are only activated in the prostate tumor. PCEs are of particular interest due to their association with cancer-specific activation.Fig. 1Prostate cancer-specific enhancers (PCEs) are clustered within specific topologically associating domains (TADs) that contain multiple upregulated genes activated and exhibit increased chromatin interactions among enhancers and promoters in prostate cancer. A Schematic illustration of three enhancer types identified from this study: common enhancers (CEs, green), normal prostate-specific enhancers (NPEs, blue), and prostate cancer-specific enhancers (PCEs, red). Inactive enhancers and genes are shown in gray. B H3K27ac ChIP-seq signals at the top 1,000 CEs, NPEs, and PCEs in three representative matched normal prostate (dark blue) and prostate cancer (pink) cases in the Baca et al. dataset [27]. C Schematic overview of PCE identification and prioritization using normal and prostate cancer tissues and cell lines. D Bar plot showing the distribution of PCEs across topologically associating domains (TADs). Bar height indicates the frequency of TADs based on the number of PCEs inside, with the number labeled above each bar. E, F Box plot comparing the number of upregulated genes in 22Rv1 (cancer) cells compared to RWPE-1 (normal) cells (|Fold Change|> 1.5, FDR < 0.05) in TADs based on the number of PCEs inside. The Wilcoxon rank-sum test or Kruskal–Wallis test P values are indicated as follows: **** P < 0.0001; * P < 0.05. G, H Box plot comparing the number of enhancer/promoter (E/P) interactions in 22Rv1 cells in TADs based on the number of PCEs inside. The Wilcoxon rank-sum test or Kruskal–Wallis test P values are indicated as follows: **** P < 0.0001; ** P < 0.01. I Schematic illustrating the components used to calculate TAD impact score. J Genome-wide ranking of TADs based on the impact score. Top 5 ranked TADs are annotated with example overexpressed genes in 22Rv1 cells compared to RWPE-1 cells, while the chr6q24.1 TAD is shown in red Prostate cancer-specific enhancers (PCEs) are clustered within specific topologically associating domains (TADs) that contain multiple upregulated genes activated and exhibit increased chromatin interactions among enhancers and promoters in prostate cancer. A Schematic illustration of three enhancer types identified from this study: common enhancers (CEs, green), normal prostate-specific enhancers (NPEs, blue), and prostate cancer-specific enhancers (PCEs, red). Inactive enhancers and genes are shown in gray. B H3K27ac ChIP-seq signals at the top 1,000 CEs, NPEs, and PCEs in three representative matched normal prostate (dark blue) and prostate cancer (pink) cases in the Baca et al. dataset [27]. C Schematic overview of PCE identification and prioritization using normal and prostate cancer tissues and cell lines. D Bar plot showing the distribution of PCEs across topologically associating domains (TADs). Bar height indicates the frequency of TADs based on the number of PCEs inside, with the number labeled above each bar. E, F Box plot comparing the number of upregulated genes in 22Rv1 (cancer) cells compared to RWPE-1 (normal) cells (|Fold Change|> 1.5, FDR < 0.05) in TADs based on the number of PCEs inside. The Wilcoxon rank-sum test or Kruskal–Wallis test P values are indicated as follows: **** P < 0.0001; * P < 0.05. G, H Box plot comparing the number of enhancer/promoter (E/P) interactions in 22Rv1 cells in TADs based on the number of PCEs inside. The Wilcoxon rank-sum test or Kruskal–Wallis test P values are indicated as follows: **** P < 0.0001; ** P < 0.01. I Schematic illustrating the components used to calculate TAD impact score. J Genome-wide ranking of TADs based on the impact score. Top 5 ranked TADs are annotated with example overexpressed genes in 22Rv1 cells compared to RWPE-1 cells, while the chr6q24.1 TAD is shown in red To systematically identify and prioritize PCEs for downstream functional analysis, we first established a patient-derived enhancer “roadmap” (Fig. 1C, Additional file 2: Fig. S1A). By merging 165 H3K27ac ChIP-seq datasets from prostate tumor tissues (Additional file 1: Table S1A), peaks enriched in at least one sample were identified (n = 426,316, termed prostate tumor tissue consensus peaks) (step 1) (Additional file 2: Fig. S1A). We subsequently integrated them with H3K27ac ChIP-seq datasets from 10 prostate cell lines (step 2), narrowing down the peaks to 82,187 (Additional file 2: Fig. S1A). Further, we performed differential enrichment analysis of H3K27ac ChIP-seq signals using the DiffBind R package [31] on matched normal and tumor tissue samples, revealing regions with cancer-specific enrichment (FDR < 0.05) (step 3). A parallel analysis in normal and cancer cell lines was performed and overlapped with patient-derived regions to reveal consistent cancer-specific enrichment (FDR < 0.05) (step 4). As enhancers are located distal from TSSs, we selected peaks located outside of the ± 2 kb of TSSs, retaining 11,338 peaks (step 5) (Additional file 1: Table S2). Finally, to prioritize candidate regions for downstream investigation, we performed differential enrichment analysis in the 22Rv1 prostate cancer and RWPE-1 normal prostate cell lines, which were optimized for functional characterization in previous studies [10, 32]. This yielded 3,216 PCEs (FDR < 0.05) (step 6) marked by robust cancer-specific H3K27ac ChIP-seq signals (Additional file 2: Fig. S1B), which were used for further analyses (Additional file 1: Table S2). Analysis of PCEs across 165 prostate tumor tissue samples from diverse prostate cancer patients revealed a broad distribution of enhancer frequencies (Additional file 1: Table S2), indicating the presence of both frequently occurring and rare enhancers that reflect inter-patient heterogeneity. PCEs are clustered in topologically associating domains (TADs) linked to transcriptional dysregulation Having defined a high-confidence set of 3,216 PCEs in 22Rv1 prostate cancer cells, we next examined their genomic distribution. Genome-wide overlap with chromosomal cytobands revealed that PCEs are not uniformly distributed across the genome but clustered in specific genomic regions (Additional file 2: Fig. S2A). To further assess their genomic location, we assigned PCEs to topologically associated domains (TADs) called using Hi-C data in 22Rv1 cells [10]. About 50% of the TADs (1,506 out of 2,960) possessed one or more PCEs, notably including a small subset of TADs with a disproportionately high number of PCEs (Fig. 1D, Additional file 1: Table S3A). For instance, less than 5% of the TADs (121 out of 2,960) possess more than 4 PCEs inside, accounting for 24% of the PCEs in total (776 out of 3,216). This indicates that PCEs are clustered in selective chromatin domains, suggesting that their involvement in regulatory programs is specifically associated with prostate cancer. To evaluate the functional impacts of this distribution of PCEs, we examined its association with transcriptomic differences using RNA-seq data from 22Rv1 and RWPE-1 cells. We revealed that TADs containing at least one PCE harbored more upregulated genes in cancer than normal (|Fold Change|> 1.5, FDR < 0.05) relative to TADs lacking PCEs (Wilcoxon Rank Sum P < 0.0001) (Fig. 1E, Additional file 1: Table S3A). Furthermore, stratification of TADs based on the number of PCEs revealed a dosage-dependent relationship between PCE clustering and cancer-specific gene activation: TADs with higher numbers of PCEs showed progressively higher numbers of upregulated genes (Kruskal–Wallis test P < 0.05 for all three comparisons) (Fig. 1F). All these results suggest that the clustering of PCEs plays an important role in transcriptional activation in prostate cancer. Similarly, when we turned to chromatin architecture, we observed a parallel dosage-dependent effect at the level of enhancer-promoter communications; enhancer-enhancer (E-E), enhancer-promoter (E-P), and promoter-promoter (P-P) interactions are collectively referred to as enhancer/promoter (E/P) interactions (Additional file 2: Fig. S2B). On average, TADs containing at least one PCE exhibited significantly more E/P interactions than TADs absent of PCEs (Wilcoxon Rank Sum P < 0.0001) (Fig. 1G). Further categorization of TADs based on PCE count revealed a gradual increase in interaction frequency as PCE number increased (Kruskal–Wallis test P < 0.05 for all three comparisons) (Fig. 1H). Together, these findings suggest that PCE clustering within a TAD coincides with increased chromatin connectivity, reinforcing the model where PCE-rich domains serve as highly transcriptional active and 3D interactive regulatory units in prostate cancer. To assess genome-wide TADs in an integrated manner, we then developed a simple multiplicative model to quantify the combined impact of PCE load, transcriptional activation, and chromatin interaction density among enhancers and promoters (Fig. 1I). This framework was used to identify key PCE-enriched, highly transcriptionally active, and interaction-dense TADs in prostate cancer. Ranking 2,960 TADs based on this score revealed a strongly right-skewed distribution, with a small subset of TADs exhibiting disproportionately high impact scores (Fig. 1J). Notably, the top ranked TADs included overexpressed genes that are positively correlated with prostate cancer progression and therapy resistance as previously reported [33, 34] (Additional file 2: Fig. S2C, Additional file 1: Table S3A). Those regions also contained well-studied prostate cancer genes, including AR [35], along with additional candidate genes not previously linked to prostate cancer (Fig. 1J, Additional file 1: Table S3A). To assess the translational relevance, we evaluated genes associated with the top 20 high-impact TADs regarding pharmacological targeting and identified candidates targetable by existing small-molecule inhibitors (Additional file 1: Table S3B). This finding highlights the importance of integrating enhancer profiles and chromatin interactions, in addition to gene expression, to identify cancer-specific genes. We further examined the top ranked TADs and assessed whether their sizes differed between cancer and normal cells. We found that 413 out of top 500 TADs exhibited different sizes in cancer compared to normal (e.g., TADs in chr16q12.2, chr7p14.1, and chr8q12.1), supporting our previous observation that cancer-specific TADs contribute to transcriptomic changes [10, 36, 37]. However, there were still TADs that did not show size differences between cancer and normal (Additional file 1: Table S3A). For example, a previously uncharacterized locus at chr6q24.1 region emerged as one of the highest-ranking TADs genome-wide (Fig. 1J, Additional file 2: Fig. S2D), where a total of 8 PCEs is clustered in a single TAD with multiple upregulated genes (e.g., ARFGEF3, SMIM28, HEBP2, and NHSL1) and dense E/P interactions in 22Rv1 cells. In contrast to many other top-ranked regions, this chr6q24.1 region also displayed no detectable size differences between RWPE-1 and 22Rv1 cells. Together, these features made the chr6q24.1 locus an ideal model to dissect how clustered PCEs drive cancer-specific regulatory programs independently of large-scale TAD boundary remodeling. PCEs clustered at the chr6q24.1 locus form a cancer-specific, multi-connected chromatin interaction hub To investigate the mechanism underlying PCE clustering, we selected the TAD in the chr6q24.1 region as a representative locus for functional characterization (Fig. 2A-C). H3K27ac ChIP-seq signals revealed a locus-wide increase of enhancer activities in 22Rv1 prostate cancer cells compared to RWPE-1 normal prostate cells (Additional file 2: Fig. S3A, Additional file 1: Table S4), with the majority of enhancers characterized as PCEs (Fig. 2A). At the transcriptomic level, this locus possesses several of the most significantly upregulated genes in prostate cancer cells, including ARFGEF3 (log2FC = 9.06, FDR = 7.13E-177), SMIM28 (log2FC = 8.93, FDR = 2.59E-09), and CITED2 (log2FC = 4.58, FDR = 1.39E-71) (Additional file 2: Fig. S3B, Additional file 1: Table S5). Overall, this locus region exemplified the focal clustering of PCEs with cancer-specific transcriptional activation.Fig. 2Region Capture Micro-C (RCMC) chromatin contact maps reveal cancer-specific, multi-connected PCE interaction hubs at the chr6q24.1 locus. A Region Capture Micro-C (RCMC) chromatin contact maps at chr6:138,000,000–139,500,000 for RWPE-1 normal prostate cells (top, green) and 22Rv1 prostate cancer cells (bottom, blue) at 5 kb resolution. Locations of common enhancers (CEs), prostate cancer-specific enhancers (PCEs), and normal prostate-specific enhancers (NPEs) are shown alongside RefSeq gene annotation tracks. B Interaction network maps showing the hierarchical connectivity of regulatory elements in RWPE-1 and 22Rv1 cells within the TAD marked with a red triangle, with line color and width relative to the interaction strength and node size representing connectivity. C Differential RCMC chromatin contact map between 22Rv1 cells and RWPE-1 cells (blue indicates decreased interactions, red indicates increased interactions). RefSeq gene annotations (black), H3K4me3 (purple) and H3K27ac (brown) ChIP-seq, and RNA-seq (orange) signal tracks for both cell lines (top: RWPE-1, bottom: 22Rv1) are shown below the heatmap. Enhancers characterized as PCEs are also annotated Region Capture Micro-C (RCMC) chromatin contact maps reveal cancer-specific, multi-connected PCE interaction hubs at the chr6q24.1 locus. A Region Capture Micro-C (RCMC) chromatin contact maps at chr6:138,000,000–139,500,000 for RWPE-1 normal prostate cells (top, green) and 22Rv1 prostate cancer cells (bottom, blue) at 5 kb resolution. Locations of common enhancers (CEs), prostate cancer-specific enhancers (PCEs), and normal prostate-specific enhancers (NPEs) are shown alongside RefSeq gene annotation tracks. B Interaction network maps showing the hierarchical connectivity of regulatory elements in RWPE-1 and 22Rv1 cells within the TAD marked with a red triangle, with line color and width relative to the interaction strength and node size representing connectivity. C Differential RCMC chromatin contact map between 22Rv1 cells and RWPE-1 cells (blue indicates decreased interactions, red indicates increased interactions). RefSeq gene annotations (black), H3K4me3 (purple) and H3K27ac (brown) ChIP-seq, and RNA-seq (orange) signal tracks for both cell lines (top: RWPE-1, bottom: 22Rv1) are shown below the heatmap. Enhancers characterized as PCEs are also annotated However, it remains unclear which genes are regulated by PCEs, as enhancers are reported to control genes that are not necessarily their nearest by forming chromatin loops [38, 39]. While existing 3C-based assays such as Hi-C and ChIA-PET have been instrumental in identifying stable chromatin loop extrusions, they often lack the sensitivity to detect dynamic or transient interactions between enhancers and promoters [26, 40]. With lower sequencing depth, Region Capture Micro-C (RCMC) [27] overcame such limitations and detected significantly more chromatin interactions between regulatory elements that were not captured by conventional Hi-C (Additional file 2: Fig. S3C). To reveal chromatin loops connecting PCEs with promoters of target genes, we generated RCMC ultra-high-resolution chromatin contact maps targeting the chr6q24.1 region in RWPE-1 and 22Rv1 cells (Fig. 2A, Additional file 1: Table S1B). Comparison of RCMC contact maps surrounding the TAD in the chr6q24.1 region (red dashed triangle) between cancer and normal cells revealed cancer-specific multi-connected chromatin interactions that coincide with PCE locations (Fig. 2A, Additional file 2: Fig. S3D). This highly nested chromatin architecture was characterized by both enhancer-promoter (E-P) interactions, which connect PCEs to gene promoters, and enhancer-enhancer (E-E) interactions between PCEs, forming an interconnected regulatory architecture which we termed a “multi-connected E-P hub” (Fig. 2B). Notably, many of these enhancer-centered interactions occur independently of CTCF binding sites, indicating that the differential interactions within the hub are primarily driven by enhancers. In contrast, this hub structure was largely absent in normal prostate cells (Fig. 2B, C, Additional file 2: Fig. S3D). Importantly, CTCF binding sites across the TAD remained largely conserved between normal and cancer cells; however, reduced CTCF ChIP-seq signals at TAD boundaries were observed in cancer, along with decreased interactions of these sites (Fig. 2C, Additional file 2: Fig. S3D). This may reflect a rebalancing of chromatin architecture to reach equilibrium in response to the increased density of enhancer-driven interactions in cancer cells. The domain located on the right side of this hub, which harbors only a single PCE (blue dashed triangle), exhibited comparatively limited enhancer-promoter interaction changes, unlike the hub region containing PCE clusters. Together, these findings suggest that PCE clusters form cancer-specific chromatin loops that generate a multi-connected hub and may play key roles in driving transcriptional activation. CRISPR/Cas9-mediated deletion of a central PCE disrupts the local enhancer activity and chromatin interactions, leading to transcriptional inactivation As multiple PCEs cluster within the same genomic regions that exhibit transcriptional activation, we hypothesized that clustered PCEs cooperate with each other to regulate gene expression. To dissect the functional contributions of PCEs, we established a workflow to delete PCEs at the chr6q24.1 locus in 22Rv1 prostate cancer cells, one by one. Using a CRISPR/Cas9-mediated deletion system, we generated single-cell clones with homozygous deletion of a targeted PCE by transfecting plasmids with guide RNAs (gRNAs) that flanked each region (Fig. 3A, Additional file 1: Table S6). In parallel, non-targeting control (NTC) clones were generated using Cas9 vectors without gRNAs and cultured under the same conditions.Fig. 3Deletion of central PCE (cPCE) reduces the activity and chromatin interactions of other PCEs. A Diagram illustrating the experimental workflow for generating single-cell clones with CRISPR/Cas9-mediated deletion of PCEs for downstream multi-omic analyses. B Comparison of H3K27ac ChIP-seq signals near the chr6q24.1 PCE hub. H3K27ac ChIP-seq tracks of wildtype RWPE-1 cells (blue), wildtype non-targeting control (NTC) 22Rv1 cells (dark grey), and cPCE deletion (cPCE∆) 22Rv1 cells (red) are shown alongside RefSeq gene annotation tracks. The bottom track shows the relative fold change of H3K27ac ChIP-seq signals in cPCE∆ 22Rv1 cells compared to 22Rv1 NTC cells, where red indicates increased signals in cPCE∆ 22Rv1 cells and blue indicates decreased signals. The deleted PCE, cPCE, is annotated with a red box. C Zoomed-in comparison of H3K27ac ChIP-seq signals at chr6:138,171,140–138,586,762. ChIP-seq tracks of NTC 22Rv1 cells (dark grey) and cPCE∆ 22Rv1 cells (red) are shown with locations of other PCEs (in various colors). The deleted PCE, cPCE, is annotated with a scissor icon, while other PCEs with decreased H3K27ac ChIP-seq signals upon deletion of cPCE are annotated with blue down arrows, with their size representing the fold change. D Box plot comparing the H3K27ac ChIP-seq signals on PCEs (left panel) and non-PCEs (right panel) located in the chr6q24.1 region in NTC 22Rv1 cells (grey) and cPCE∆ 22Rv1 cells (red). The paired t-test P value is indicated as follows: * P < 0.05; ns P ≥ 0.05. E Comparison of RCMC chromatin contact maps between NTC 22Rv1 cells (bottom left) and cPCE∆ 22Rv1 cells (top right) at 1 kb resolution in the zoomed-in region, chr6:138,171,140–138,586,762. PCEs with decreased H3K27ac ChIP-seq signals upon deletion of cPCE are labeled with letters (A, B, C (also known as cPCE), D, E). Interactions among PCEs are highlighted by yellow circles on the contact maps. The relative chromatin interaction frequency bar (white to red) is shown on the right. F Zoomed-in chromatin interaction heatmaps (top panel) and loops (bottom panel) of each PCE pair in NTC 22Rv1 cells and cPCE∆ 22Rv1 cells. Statistical significance (FDR) of FitHiC-called loops is shown as color intensity (low: white, high: green), and chromatin contact counts are shown as circle size. G Schematics illustrating the essential role of the central PCE (cPCE) in coordinating chromatin architecture and the activity of other PCEs Deletion of central PCE (cPCE) reduces the activity and chromatin interactions of other PCEs. A Diagram illustrating the experimental workflow for generating single-cell clones with CRISPR/Cas9-mediated deletion of PCEs for downstream multi-omic analyses. B Comparison of H3K27ac ChIP-seq signals near the chr6q24.1 PCE hub. H3K27ac ChIP-seq tracks of wildtype RWPE-1 cells (blue), wildtype non-targeting control (NTC) 22Rv1 cells (dark grey), and cPCE deletion (cPCE∆) 22Rv1 cells (red) are shown alongside RefSeq gene annotation tracks. The bottom track shows the relative fold change of H3K27ac ChIP-seq signals in cPCE∆ 22Rv1 cells compared to 22Rv1 NTC cells, where red indicates increased signals in cPCE∆ 22Rv1 cells and blue indicates decreased signals. The deleted PCE, cPCE, is annotated with a red box. C Zoomed-in comparison of H3K27ac ChIP-seq signals at chr6:138,171,140–138,586,762. ChIP-seq tracks of NTC 22Rv1 cells (dark grey) and cPCE∆ 22Rv1 cells (red) are shown with locations of other PCEs (in various colors). The deleted PCE, cPCE, is annotated with a scissor icon, while other PCEs with decreased H3K27ac ChIP-seq signals upon deletion of cPCE are annotated with blue down arrows, with their size representing the fold change. D Box plot comparing the H3K27ac ChIP-seq signals on PCEs (left panel) and non-PCEs (right panel) located in the chr6q24.1 region in NTC 22Rv1 cells (grey) and cPCE∆ 22Rv1 cells (red). The paired t-test P value is indicated as follows: * P < 0.05; ns P ≥ 0.05. E Comparison of RCMC chromatin contact maps between NTC 22Rv1 cells (bottom left) and cPCE∆ 22Rv1 cells (top right) at 1 kb resolution in the zoomed-in region, chr6:138,171,140–138,586,762. PCEs with decreased H3K27ac ChIP-seq signals upon deletion of cPCE are labeled with letters (A, B, C (also known as cPCE), D, E). Interactions among PCEs are highlighted by yellow circles on the contact maps. The relative chromatin interaction frequency bar (white to red) is shown on the right. F Zoomed-in chromatin interaction heatmaps (top panel) and loops (bottom panel) of each PCE pair in NTC 22Rv1 cells and cPCE∆ 22Rv1 cells. Statistical significance (FDR) of FitHiC-called loops is shown as color intensity (low: white, high: green), and chromatin contact counts are shown as circle size. G Schematics illustrating the essential role of the central PCE (cPCE) in coordinating chromatin architecture and the activity of other PCEs We first deleted a PCE (chr6:138,386,367–138,388,367), which is located in the intergenic region and displays strong cancer-specific enhancer activity with a high H3K27ac ChIP-seq signal in 22Rv1 cells (highlighted in the red box in Fig. 3B). Upon deletion of this PCE (i.e., cPCE∆), H3K27ac ChIP-seq revealed a locus-wide decrease in enhancer activities compared to NTC cells (Fig. 3B, Additional file 1: Table S7), with the most pronounced losses of signals at neighboring PCEs (Fig. 3C, Additional file 2: Fig. S4A). Quantitative analysis identified that the effect was significantly stronger at enhancers that were characterized as PCEs than at non-PCEs in the chr6q24.1 TAD region (Fig. 3D), suggesting that this PCE plays a central role in maintaining cancer-specific enhancer activation. Thus, we termed this enhancer a central PCE (cPCE). To determine whether these changes in enhancer activity were linked to altered chromatin topology, we generated and compared the RCMC map of cPCE deletion cells (cPCE∆) to NTC (wildtype) cells (Fig. 3E). Notably, the cPCE and the PCEs with reduced enhancer activities (i.e., Enhancer A, B, D, E in Fig. 3E) formed a strong interconnected architecture in NTC cells (Fig. 3E). Upon deletion of cPCE, the PCE-PCE interactions that were observed in NTC cells were significantly disrupted or remarkably weakened (Fig. 3E, F). We also observed a significant reduction in cancer-specific E/P interactions at the locus, especially PCE-promoter interactions on the weakened PCEs (Additional file 2: Fig. S4B, C). This was coupled with significantly decreased expression of the PCE-targeting genes such as SMIM28, HEBP2, and NHSL1 (Additional file 2: Fig. S4D). These results indicate that cPCE plays an essential role in maintaining the cancer-specific chromatin architecture involving PCEs, leading to transcriptional activation at the chr6q24.1 locus. Importantly, the disruption of the interactions among enhancers and promoters observed upon deletion was not due to a large-scale collapse of chromatin architecture of the TAD. CTCF-anchored loops called from RCMC contact maps displayed slightly increased interactions among CTCF binding sites at TAD boundaries upon cPCE deletion (Additional file 2: Fig. S5A). Interestingly, we observed that some of these CTCF binding sites modestly increased after the cPCE deletion, possibly to prevent collapse of chromatin architecture and to preserve enhancer interactions among these sites, thereby maintaining a dynamic equilibrium to maintain gene expression (Additional file 2: Fig. S5B). These findings indicate that rather than altering higher-order domain boundaries, the cPCE coordinates fine-scale regulatory loops, highlighting a layer of enhancer-driven chromatin organization that coexists with the structural framework established by CTCF. This is consistent with previous finding that acute depletion of CTCF disrupts TADs but leaves fine-scale regulatory loops intact [41]. Together, these results support a model where the cPCE functions as a core regulatory organizer to maintain enhancer activity, stabilize chromatin structure, and enable coordinated gene activation in prostate cancer (Fig. 3G). A redundant PCE is dispensable due to compensatory rewiring by neighboring rescuing PCEs While the cPCE was required to maintain the local enhancer activity and chromatin interactions, we next asked whether all PCEs within the same cluster exert a similar regulatory impact. We selected another PCE with the highest H3K27ac ChIP-seq signal, located 160 kb upstream of cPCE in the intron of the ARFGEF3 gene (chr6:138,226,499–138,228,499). After deleting this PCE and subsequently performing H3K27ac ChIP-seq, we noted that loss of this PCE did not reduce the activity of other enhancers, including PCEs, unlike what we observed with the cPCE deletion (Fig. 4A, B, Additional file 2: Fig. S6A, Additional file 1: Table S7). Instead, we observed an overall increase in activities at certain neighboring PCEs (Fig. 4B), suggesting a potential compensatory mechanism. Therefore, we termed this enhancer a redundant PCE (rPCE).Fig. 4Deletion of a redundant PCE (rPCE) triggers compensatory neighboring PCE activation and chromatin looping to target genes. A Comparison of H3K27ac ChIP-seq signals near the chr6q24.1 PCE hub. H3K27ac ChIP-seq tracks of NTC 22Rv1 cells (dark gray) and rPCE deletion (rPCE∆) 22Rv1 cells (blue) are shown alongside RefSeq gene annotation tracks. The bottom track shows the relative fold change of H3K27ac ChIP-seq signals in rPCE∆ 22Rv1 cells compared to NTC 22Rv1 cells, where red indicates increased signals in rPCE∆ 22Rv1 cells and blue indicates decreased signals. B Zoomed-in comparison of H3K27ac ChIP-seq signals at chr6:138,190,918–138,451,108. ChIP-seq tracks of NTC 22Rv1 cells (dark grey) and rPCE∆ 22Rv1 cells (blue) are shown with locations of other PCEs (in various colors in the same way as the previous figures). The deleted PCE, rPCE, is annotated with a scissor icon, while other PCEs with increased H3K27ac signals are annotated with red upward arrows. C Comparison of chromatin contact strength between rescuing PCEs (in colors) and gene promoters in rPCE∆ 22Rv1 cells versus NTC 22Rv1 cells. The change in chromatin contact counts is shown by the size of the dot, and the direction is indicated by color (blue: decreased, red: increased). D Comparison of RCMC chromatin contact maps between NTC 22Rv1 cells (bottom left) and rPCE∆ 22Rv1 cells (top right) at 1 kb resolution in the zoomed-in region, chr6:138,190,918–138,451,108. Locations of HEBP2 and PCEs are shown above the contact maps. The rPCE (blue) and the resPCE (green, Additional file 2: Fig. S6) are indicated with arrows. The interactions between the resPCE and the promoter of HEBP2 are highlighted by yellow circles on the contact maps. The relative chromatin interaction frequency bar (white to red) is shown on the right. E Zoomed-in chromatin interaction heatmaps (left) and loops (right) between the resPCE and the promoter of HEBP2 in NTC 22Rv1 cells and rPCE∆ 22Rv1cells. Statistical significance (FDR) of FitHiC-called loops is shown as color intensity (low: white, high: green), and chromatin contact counts are shown as circle size. F Schematics illustrating the compensatory role of a neighboring rescuing PCE (resPCE) in chromatin architecture upon deletion of a redundant PCE (rPCE) Deletion of a redundant PCE (rPCE) triggers compensatory neighboring PCE activation and chromatin looping to target genes. A Comparison of H3K27ac ChIP-seq signals near the chr6q24.1 PCE hub. H3K27ac ChIP-seq tracks of NTC 22Rv1 cells (dark gray) and rPCE deletion (rPCE∆) 22Rv1 cells (blue) are shown alongside RefSeq gene annotation tracks. The bottom track shows the relative fold change of H3K27ac ChIP-seq signals in rPCE∆ 22Rv1 cells compared to NTC 22Rv1 cells, where red indicates increased signals in rPCE∆ 22Rv1 cells and blue indicates decreased signals. B Zoomed-in comparison of H3K27ac ChIP-seq signals at chr6:138,190,918–138,451,108. ChIP-seq tracks of NTC 22Rv1 cells (dark grey) and rPCE∆ 22Rv1 cells (blue) are shown with locations of other PCEs (in various colors in the same way as the previous figures). The deleted PCE, rPCE, is annotated with a scissor icon, while other PCEs with increased H3K27ac signals are annotated with red upward arrows. C Comparison of chromatin contact strength between rescuing PCEs (in colors) and gene promoters in rPCE∆ 22Rv1 cells versus NTC 22Rv1 cells. The change in chromatin contact counts is shown by the size of the dot, and the direction is indicated by color (blue: decreased, red: increased). D Comparison of RCMC chromatin contact maps between NTC 22Rv1 cells (bottom left) and rPCE∆ 22Rv1 cells (top right) at 1 kb resolution in the zoomed-in region, chr6:138,190,918–138,451,108. Locations of HEBP2 and PCEs are shown above the contact maps. The rPCE (blue) and the resPCE (green, Additional file 2: Fig. S6) are indicated with arrows. The interactions between the resPCE and the promoter of HEBP2 are highlighted by yellow circles on the contact maps. The relative chromatin interaction frequency bar (white to red) is shown on the right. E Zoomed-in chromatin interaction heatmaps (left) and loops (right) between the resPCE and the promoter of HEBP2 in NTC 22Rv1 cells and rPCE∆ 22Rv1cells. Statistical significance (FDR) of FitHiC-called loops is shown as color intensity (low: white, high: green), and chromatin contact counts are shown as circle size. F Schematics illustrating the compensatory role of a neighboring rescuing PCE (resPCE) in chromatin architecture upon deletion of a redundant PCE (rPCE) Furthermore, RCMC maps comparing NTC and rPCE deletion cells (rPCE∆) revealed that the E/P interactions remained largely intact overall (Additional file 2: Fig. S6B), particularly those surrounding the deletion. These findings suggested that, unlike cPCE, rPCE is not essential for maintaining enhancer activity or regulatory chromatin architecture at the locus. When we examined these chromatin interactions closely, we strikingly identified strengthened chromatin contacts between gene promoters and neighboring enhancers, including cPCE and two other ones that showed modestly increased activity upon rPCE deletion (Fig. 4B, C, Additional file 2: Fig. S6B). We termed these neighboring PCEs rescuing PCEs (resPCEs). For example, a resPCE located 30 kb downstream of the rPCE (chr6:138,256,617–138,258,617) formed a new loop to the promoter of HEBP2 (Fig. 4D). Quantitative loop comparison analysis revealed that this E-P contact was strengthened by a 4-fold increase upon rPCE deletion (Fig. 4E), suggesting that the regulatory architecture had undergone compensatory rewiring to preserve transcriptional upregulation of target genes. Consequently, we observed no statistically significant changes in gene expression upon rPCE deletion, including the previously looped genes SMIM28, HEBP2, and NHSL1 (Additional file 2: Fig. S6C). Together, these results demonstrate that a class of PCEs may play supportive or redundant roles within the cluster, and that deletion of this rPCE can induce compensatory rewiring of neighboring PCEs to maintain gene regulation by altering enhancer activity and chromatin interactions (Fig. 4F). Distinct enhancer classes hierarchically influence gene regulation within a multi-connected hub Next, we compared the transcriptomic changes following deletions of individual PCEs at the chr6q24.1 locus to identify the target genes of PCEs using RNA-seq (Additional file 1: Table S8). Deletion of cPCE caused significant downregulation of multiple genes such as SMIM28 (log2FC = −6.89, FDR = 1.24E-07), HEBP2 (log2FC = −1.48, FDR = 6.00E-58), and NHSL1 (log2FC = −1.27, FDR = 4.33E-19), whereas deletion of rPCE had no significant effect on downregulation of these genes, showing distinct regulatory impacts of these two classes of enhancers. When we turned to the deletion of the previously identified resPCE (Fig. 4D), we observed a transcriptional response that closely mirrored the deactivation caused by cPCE deletion: SMIM28 (log2FC = −1.70, FDR = 2.12E-05), HEBP2 (log2FC = −1.32, FDR = 1.53E-133), and NHSL1 (log2FC = −2.19, FDR = 1.86E-176) were all significantly downregulated (Fig. 5A). This similarity suggests that resPCE may function within the same class as cPCE and exert the comparable impact on enhancer activity and chromatin architecture at the locus. Consistent with this hypothesis, H3K27ac ChIP-seq profile following resPCE deletion revealed reduced local enhancer activity with the most notable decreases at neighboring PCEs (Additional file 2: Fig. S7A, B), although the magnitude of reduction was less pronounced compared to cPCE deletion. Together, these findings suggest that transcriptional and chromatin structural outcomes from deletion can distinguish the class of functionally essential enhancers from the redundant one, and that enhancers operate with hierarchical regulatory powers, at least within the locus examined.Fig. 5Genes linked to prostate cancer phenotypes show altered expression upon the deletion of PCEs in a multi-connected hub. A Scatter plot comparing gene expression changes upon enhancer deletion and chromatin contact counts between the enhancer and the corresponding gene promoter (different shapes) at the chr6q24.1 PCE hub in cPCE∆ (red), rPCE∆ (blue), and resPCE∆ (green) 22Rv1 cells compared to NTC 22Rv1 cells. The grey region annotates non-significant changes. B Interaction network maps showing the hierarchical connectivity of enhancers and their target gene promoters in 22Rv1 cells, with line width representing chromatin interaction strength. C Bar plot comparing NHSL1 and HEBP2 expression levels in single PCE deletion cells, including cPCE (red) and resPCE (green), and in double knockouts (DKO) of resPCE and cPCE cells (yellow), compared to NTC cells (grey) normalized to GAPDH expression from RT-qPCR results. P values are indicated as follows: *** P < 0.001. D Cell proliferation assay using NTC (grey circles), cPCE∆ (red squares), resPCE∆ (green diamonds), and DKO (yellow triangles) 22Rv1 cells. Data points represent the mean of three biological replicates. Standard deviation is shown as error bars. E (Left) Representative colony formation images of NTC, cPCE∆, resPCE∆, and DKO cells after 12 days of incubation. (Right) Bar plot comparing the relative colony size of NTC (grey), cPCE∆ (red), resPCE∆ (green), and DKO (yellow) cells. P values are indicated as follows: *** P < 0.001; ** P < 0.01; * P < 0.05; ns P ≥ 0.05 Genes linked to prostate cancer phenotypes show altered expression upon the deletion of PCEs in a multi-connected hub. A Scatter plot comparing gene expression changes upon enhancer deletion and chromatin contact counts between the enhancer and the corresponding gene promoter (different shapes) at the chr6q24.1 PCE hub in cPCE∆ (red), rPCE∆ (blue), and resPCE∆ (green) 22Rv1 cells compared to NTC 22Rv1 cells. The grey region annotates non-significant changes. B Interaction network maps showing the hierarchical connectivity of enhancers and their target gene promoters in 22Rv1 cells, with line width representing chromatin interaction strength. C Bar plot comparing NHSL1 and HEBP2 expression levels in single PCE deletion cells, including cPCE (red) and resPCE (green), and in double knockouts (DKO) of resPCE and cPCE cells (yellow), compared to NTC cells (grey) normalized to GAPDH expression from RT-qPCR results. P values are indicated as follows: *** P < 0.001. D Cell proliferation assay using NTC (grey circles), cPCE∆ (red squares), resPCE∆ (green diamonds), and DKO (yellow triangles) 22Rv1 cells. Data points represent the mean of three biological replicates. Standard deviation is shown as error bars. E (Left) Representative colony formation images of NTC, cPCE∆, resPCE∆, and DKO cells after 12 days of incubation. (Right) Bar plot comparing the relative colony size of NTC (grey), cPCE∆ (red), resPCE∆ (green), and DKO (yellow) cells. P values are indicated as follows: *** P < 0.001; ** P < 0.01; * P < 0.05; ns P ≥ 0.05 To better understand the basis of the functional hierarchy between enhancer classes, we compared the chromatin interaction profiles of all enhancers to the common target genes (Fig. 5A), including SMIM28, NHSL1, and HEBP2. Unsurprisingly, cPCE exerted its transcriptional regulation through connections to all promoters to some degree, however, we found that direct E-P looping alone is not sufficient to predict transcriptomic regulatory impact. Although rPCE formed direct loops with all promoters (Fig. 5A, B), including NHSL1 (normalized chromatin contact reads = 12, FDR = 8.02E-20) and HEBP2 (normalized chromatin contact reads = 84, FDR = 1.15E-218), its deletion did not affect gene expression due to compensatory rewiring of neighboring enhancers. On the contrary, resPCE displayed intermediate to low interactions with given promoters, especially to NHSL1 (Fig. 5A, B). Instead, it formed E-E loops across the hub, including connections to three other PCEs that make significant connections to the NHSL1 promoter (Fig. 5B). This further suggests that enhancers can regulate genes indirectly through cooperative E-E interactions across the hub, resulting in a more modest transcriptional effect compared to directly engaged cPCEs. Next, we generated double-deletion PCE clones lacking both the cPCE and resPCE to investigate how PCEs cooperate by further examining their impact on the expression of NHSL1 and HEBP2, which have not been previously studied in prostate cancer. While resPCE and cPCE deletion showed significant yet similar reduction in NHSL1 and HEBP2 expression, the combined perturbation resulted in an even greater decrease compared to either single deletion (Fig. 5C), further indicating that these enhancers cooperate to regulate gene expression. In addition, we performed siRNA knockdown to test whether these target genes regulate each other downstream of the PCE hub and found no evidence of crosstalk between them (Additional file 2: Fig. S6C; Additional file 1: Table S9), further validating our finding that gene downregulation was induced solely by deletion of enhancers. Central PCE class regulates genes with direct impact on cancer phenotypes In parallel to transcriptional changes, we also investigated the functional consequences of PCE perturbations in cancer phenotypes. Cell proliferation and colony formation assays revealed that while resPCE deletion yielded intermediate effects, cPCE deletion and double deletion led to more pronounced reductions in cell proliferation and clonogenic ability (Fig. 5D, E). When we performed Ingenuity Pathway Analysis (IPA) of differentially expressed genes in PCE deletion clones compared to WT (NTC) clones, we found enrichment in pathways related to cell proliferation, tumor growth, and colony formation, with the most pronounced enrichment observed in the cPCE deletion clones (Additional file 2: Fig. S7D, Additional file 1: Table S8). Together, these results indicate that cPCE acts as a central anchor for the multi-connected chromatin hub integrity and cancer cell identity, underscoring its critical role in both cancer-specific transcriptional regulation and cancer phenotypes. FOXA1 regulates distinct PCE classes through its pioneer activity to affect chromatin accessibility To investigate the mechanism underlying PCEs, we first examined chromatin accessibility. More than 95% of PCEs (3,089 out of 3,216) marked by H3K27ac were accessible, as indicated by ATAC-seq peaks in 22Rv1 cells. Differential ATAC-seq signal analysis revealed substantially greater gains in chromatin accessibility from RWPE-1 normal cells to 22Rv1 cancer cells at PCEs (normalized signal difference median = 0.5931) compared with non-PCEs (median = 0.1179) (Mann–Whitney P < 0.0001) (Fig. 6A; Additional file 1: Table S10). A closer examination at the chr6q24.1 locus revealed the same pattern of differential chromatin accessibility. Quantitative analysis supported that PCEs inside the multi-connected chromatin interaction hub have significantly increased chromatin accessibility (Wilcoxon P = 0.0391), whereas no statistically significant changes were observed at non-PCEs (Wilcoxon P > 0.99) (Fig. 6B, C). Together, these observations indicate that a cancer-specific gain in chromatin accessibility is a defining feature of PCEs.Fig. 6cPCE acts as an essential anchor of the PCE hub, sustaining chromatin accessibility and transcription factor binding. A Dot plot showing the mean normalized ATAC-seq signal difference between 22Rv1 and RWPE-1 cells at PCEs vs. non-PCEs located throughout the genome with standard deviations. Mann–Whitney P value is indicated as **** P < 0.0001. B Box plot showing normalized ATAC-seq signals in 22Rv1 and RWPE-1 cells at PCEs (top) and non-PCEs at the chr6q24.1 hub. Kruskal–Wallis test P values are indicated as **** P < 0.0001; * P < 0.05; ns P > 0.05. C Genome browser screenshot near the chr6q24.1 PCE hub. From top to bottom, H3K27ac ChIP-seq in RWPE-1 and 22Rv1 cells, ATAC-seq in RWPE-1 and 22Rv1 cells, FOXA1 ChIP-seq in 22Rv1 cells, and a FOXA1 motif track with colored based on the FOXA1 motif score calculated using the JASPAR database [80] are shown. D Bar plot showing the -log10(FDR) of the top 10 transcription factor (TF) motifs most enriched at PCEs. E Venn diagram showing the percentage of PCEs overlapping with FOXA1 ChIP-seq peaks. F Zoomed-in comparison of H3K27ac ChIP-seq signals (orange), ATAC-seq signals (blue), and FOXA1 ChIP-seq signals (purple) at chr6:138,142,830–138,610,020 in NTC 22Rv1 cells (gray) and cPCE∆ (red) 22Rv1 cells. The deleted PCE, cPCE, is annotated with a scissor icon, while other enhancers with changed signals are annotated with arrows (blue: decreased, red: increased), with arrow size representing the fold change of signal between cPCE∆ 22Rv1 and NTC 22Rv1 cells. G Schematics illustrating the role of the cPCE in maintaining the activity of other PCEs and regulating gene expression via FOXA1 binding and chromatin accessibility changes in the multi-connected hub cPCE acts as an essential anchor of the PCE hub, sustaining chromatin accessibility and transcription factor binding. A Dot plot showing the mean normalized ATAC-seq signal difference between 22Rv1 and RWPE-1 cells at PCEs vs. non-PCEs located throughout the genome with standard deviations. Mann–Whitney P value is indicated as **** P < 0.0001. B Box plot showing normalized ATAC-seq signals in 22Rv1 and RWPE-1 cells at PCEs (top) and non-PCEs at the chr6q24.1 hub. Kruskal–Wallis test P values are indicated as **** P < 0.0001; * P < 0.05; ns P > 0.05. C Genome browser screenshot near the chr6q24.1 PCE hub. From top to bottom, H3K27ac ChIP-seq in RWPE-1 and 22Rv1 cells, ATAC-seq in RWPE-1 and 22Rv1 cells, FOXA1 ChIP-seq in 22Rv1 cells, and a FOXA1 motif track with colored based on the FOXA1 motif score calculated using the JASPAR database [80] are shown. D Bar plot showing the -log10(FDR) of the top 10 transcription factor (TF) motifs most enriched at PCEs. E Venn diagram showing the percentage of PCEs overlapping with FOXA1 ChIP-seq peaks. F Zoomed-in comparison of H3K27ac ChIP-seq signals (orange), ATAC-seq signals (blue), and FOXA1 ChIP-seq signals (purple) at chr6:138,142,830–138,610,020 in NTC 22Rv1 cells (gray) and cPCE∆ (red) 22Rv1 cells. The deleted PCE, cPCE, is annotated with a scissor icon, while other enhancers with changed signals are annotated with arrows (blue: decreased, red: increased), with arrow size representing the fold change of signal between cPCE∆ 22Rv1 and NTC 22Rv1 cells. G Schematics illustrating the role of the cPCE in maintaining the activity of other PCEs and regulating gene expression via FOXA1 binding and chromatin accessibility changes in the multi-connected hub Cell-type specific TFs bind to accessible enhancer regions to regulate genes crucial for cell identity [5]. Therefore, we performed TF motif enrichment analysis at PCEs and revealed that FOXA1, FOX:Ebox, Hnf1, AR, and GRHL2 motifs were the most frequently enriched (Fig. 6D). Next, we used ChIP-seq for FOXA1, the top enriched TF motif at PCEs, and confirmed that the majority of PCEs were indeed bound by FOXA1 (2,723 out of 3,216) (Fig. 6C, E). These findings suggest that FOXA1 may play an essential role in defining PCE classes with different regulatory powers. To test this, we profiled chromatin accessibility and FOXA1 binding in the cPCE deletion cells and compared them with NTC cells. We found that two PCEs (i.e., Enhancers A (also known as rPCE), and E), which were previously found to directly form strong E-E interactions with cPCE, exhibited a statistically significant reduction in ATAC-seq (Enhancer A: FC = 1.01, FDR = 4.22E-11; Enhancer E: FC = −1.57, FDR = 3.86E-7) and FOXA1 ChIP-seq signals (Enhancer A: FC = −0.96, FDR = 1.25E-3; Enhancer E: FC = −2.18, FDR = 7.19E-6) (Fig. 6F, Additional file 1: Tables S10-11). Coupled with the decrease in H3K27ac ChIP-seq signals, these changes indicate that cPCE sustains the enhancer activities of other PCEs through FOXA1 binding and stabilization in the multi-connected hub. Interestingly, we also observed that a non-PCE enhancer outside the cPCE connections (i.e., Enhancer O, chr6:138,468,541–138,470,541) showed increased chromatin accessibility (FC = 0.96, FDR = 1.32E-6) and FOXA1 occupancy (FC = 0.95, FDR = 4.80E-3) upon cPCE deletion (Fig. 6F, Additional file 1: Tables S10-11). This suggests that the loss of the core enhancer anchor drives FOXA1 redistribution to alternative enhancer sites, potentially reshaping the enhancer landscape. Unlike cPCE, rPCE deletion resulted in minimal reductions in chromatin accessibility or FOXA1 occupancy across enhancers within the locus (Additional file 2: Fig. S8A, Additional file 1: Tables S12-13). Instead, all previously identified rescuing enhancers (Fig. 4B) exhibited significantly increased FOXA1 enrichment following rPCE loss (e.g., resPCE: FC = 2.18, FDR = 4.48E-4; B: FC = 1.52, FDR = 8.8E-3; cPCE: FC = 1.89, FDR = 1.8E-2) (Additional file 2: Fig. S7A, Additional file 1: Tables S10-11). These coordinated gains in FOXA1 binding suggest that compensatory rewiring of chromatin interactions accompanies changes in chromatin accessibility and FOXA1 binding (Additional file 2: Fig. S8B). To further assess epigenomic signatures, including chromatin interactions, TF binding, and enhancer activity and frequency, which potentially distinguish cPCE and rPCE, we integrated ChIP-seq datasets from multiple prostate cancer patient cohorts, revealing subtle differences between the two enhancer classes (Additional file 1: Table S12). However, analysis of the FOXA1 motifs enriched in each enhancer class revealed sequence distinctions, where cPCE possesses highly scored, canonical FOXA1 motifs (e.g., GTAAACAG) while rPCE contains motif variants at the core GTAAA element (e.g., GTGAACAT) (Additional file 2: Fig. S8C). To extend these motif-based observations to patient samples, we mapped tissue FOXA1 ChIP-seq onto the matched patient-derived enhancer landscape (Additional file 2: Fig. S8D), where the identified motifs are consistently positioned at or near FOXA1 peak summits (Additional file 1: Table S2). These findings indicate that the distinct motif features observed between enhancer classes are preserved in prostate cancer patient tissue samples and likely contribute to differential FOXA1 engagement in vivo. Together, these findings suggest that cPCE functions as a core regulatory anchor, connecting other PCEs through FOXA1 to form a multi-connected chromatin interaction hub that activates cancer-related genes. FOXA1 engagement was lost upon perturbation of cPCE, leading to the collapse of chromatin accessibility, TF occupancy, and hub architecture of PCEs, ultimately silencing key regulatory programs in prostate cancer cells (Fig. 6G). On the other hand, rPCE also relies on FOXA1 binding to maintain enhancer activity and regulatory connectivity. In contrast to cPCE deletion, removal of an rPCE led to increased FOXA1 binding at neighboring PCEs, directing the program of compensatory rewiring (Additional file 2: Fig. S8B). Discussion In this study, we systematically identified PCE clusters and functionally dissected a previously uncharacterized, PCE-clustered TAD to reveal the regulatory mechanism that drives cancer-specific chromatin interactions and gene expression. First, through integrative analysis of hundreds of H3K27ac ChIP-seq datasets from prostate tissues and cell lines, we identified a high-confidence set of 3,216 PCEs with cancer-specific enhancer activity, which are clustered in defined genomic regions associated with elevated expression of cancer-specific genes and increased E/P interactions. Next, using ultra-high-resolution chromatin interaction mapping of normal prostate and prostate cancer samples, we uncovered highly nested, cancer-specific, multi-connected E-P hubs centered on PCEs. Furthermore, we evaluated the regulatory impact of individual PCEs within the chr6q24.1 TAD using CRISPR/Cas9-mediated enhancer perturbation followed by multi-omic and cancer phenotypic analyses. We found two distinct classes of putative enhancers with different roles in hub-wide chromatin architecture and transcriptional output, revealing a hierarchical enhancer organization characterized by functional cooperation and redundancy and driven by pioneer TF occupancy. How do enhancers located in gene-rich regions regulate transcription? In this study, we observed that PCEs preferentially cluster within gene-rich, transcriptionally active regions. E/P interactions are initially thought to be limited within a TAD, but later studies revealed interactions extending beyond TAD boundaries, highlighting the need for a better understanding of E/P interactions [42–45]. In mammalian cells, gene-rich regions pose a unique challenge for mechanistic dissection of E/P interactions [27, 46]. Unlike gene-poor regions, which typically display simpler regulatory topologies with fewer promoters and more stand-alone enhancer contacts, gene-rich regions harbor multiple promoters in close genomic proximity. This complexity is further amplified in disease states, where gene-rich domains are frequently associated with cell type-specific aberrant enhancer activation, heightened transcriptional demand, and rewired 3D organization. Recent advances in RCMC now allow a closer examination of the hub architecture by substantially improving the resolution and sensitivity of detecting E/P interactions. Despite these technical advances, the functions of individual enhancers within these assemblies remain poorly understood. In this study, by performing RCMC in prostate cancer cells, normal prostate cells, and enhancer-perturbed cells, we revealed that PCEs crosstalk via highly nested 3D interactions, forming cancer-specific, multi-connected E-P hubs, whereas non-PCEs appear to communicate through their own 3D interactions. CTCF plays a well-established role in organizing chromatin architecture through the formation of cohesin-anchored loops that act as TAD boundaries and constrain long-range interactions [47]. These CTCF/cohesin-mediated loops provide the structural scaffold within which E-P communication occurs, and we found many TADs that have different sizes between normal and prostate cancer cells, which likely contribute to changes in gene expression. However, we still found TADs that have similar sizes between normal and prostate cancer cells but exhibit different gene expression patterns with PCE clusters. There is accumulating evidence indicating that not all chromatin interactions rely on CTCF. Many E/P contacts, particularly those within compact gene-rich loci, have their independent mechanism such as hub formation [41, 48–50]. Fine-scale regulatory loops can persist even when CTCF-mediated structures are perturbed, underscoring the existence of a parallel, enhancer-driven model of chromatin organization [41]. Similarly, our RCMC data support the finding of multi-connected E-P hubs at gene-rich loci that are largely independent of the cohesin complex, consistent with previous reports [10, 51, 52]. The comparison of RCMC maps, along with CTCF and H3K27ac ChIP-seq profiles, between normal and cancer cells identified many differential chromatin interactions between enhancers and promoters that are independent of CTCF (Additional file 2: Fig. S3D). Moreover, our study revealed that upon cPCE deletion, hub-wide CTCF binding, and loop architecture remained largely unchanged, but slightly increased at TAD boundaries, which were lower in cancer cells compared to normal cells (Additional file 2: Fig. S5A, B). This indicates that rather than altering higher-order domain boundaries, the cPCE coordinates fine-scale regulatory loops highlighting a layer of enhancer-driven chromatin organization. Furthermore, our findings suggest that increased activity of some CTCF binding sites at TAD boundaries, along with their interactions, likely help preserve chromatin architecture and maintain equilibrium, consistent with recent studies showing that cohesin-mediated chromatin loops are dynamic [53–55]. What genes are regulated by multi-connected E-P hubs in prostate cancer? Similar observations have also been reported in prior studies of enhancers across several biological systems and cancer types, where enhancers were found to coalesce with lineage-defining loci [7, 8, 12, 27, 56, 57] or critical oncogenes [51, 52, 58, 59] to form multi-connected E-P hubs. Other researchers reported the importance of these hubs in gene regulation and cell-type specification [60, 61]. For example, Bunting et al. demonstrated that deletion of B-cell specific hub-forming enhancers at the BCL6 locus resulted in complete loss of germinal centers, underscoring their critical role in defining cell phenotypes [60]. Through this study, we identified previously unrecognized genes critical for prostate cancer by identifying and characterizing PCEs and prostate cancer E-P hubs that regulate genes not only through E-P interactions but also E-E interactions. For example, we found that PCE perturbation alters the expression of NHSL1 and HEBP2 genes, leading to changes in colony formation and cell proliferation. Although NHSL1 has not yet been characterized in prostate cancer, it is known to regulate cell migration [62] and is elevated in breast tumors, where it correlates with tumor relapse and poor survival [63]. Furthermore, despite being upregulated in breast and lung cancers [64], HEBP2 has also not been well studied in prostate cancer, and a recent study showed that HEBP2 enhances cell proliferation in chemoresistant ovarian cancer, highlighting its potential prognostic value [65]. These newly uncovered genes warrant further investigation in prostate cancer and may represent important biomarkers and therapeutic targets. What are enhancer clusters and classes, as defined by integrating chromatin interactions? Enhancer clustering has long been recognized as a major principle of gene regulation in prostate cancer, with super-enhancers (SEs) defined as a dense linear aggregation of H3K27ac-enriched regions near prostate cancer-specific genes such as KLK3 [19, 66]. However, our methods to identify key TADs and hubs with PCE clusters, developed in this study, uses 3D chromatin interactions, making it distinct from the traditional definition of SEs, which is based solely on the linear locations of enhancers and their signals. Furthermore, this SE model primarily emphasizes one-to-one enhancer-gene relationships within a linear genomic framework. In contrast, our model of PCE clusters expands this concept by highlighting spatially coordinated, many-to-many E-P interactions that possibly regulate multiple target genes within a locus. Our findings that PCEs interact with each other while non-PCEs interact among themselves in the locus where PCEs are clustered, together with PCE perturbation data, indicate that enhancer clusters may largely depend on cell-type specific TFs rather than the general transcriptional machinery and co-activators such as the mediator complex [67, 68]. Our study, which identifies and characterizes PCE clusters, represents a structurally and functionally distinct class of enhancer assemblies, providing new insights into the traditional linear SE model. Enhancers have been typically characterized by their histone modifications such as poised and active enhancers, and chromatin accessibility [15, 69], but functional classes of active enhancers defined by their 3D connectivity and transcriptional impact have not yet been identified. By functionally dissecting an enhancer cluster hub, we identified two distinct classes of enhancers at the representative locus: the central PCE (cPCE) is indispensable for maintaining locus-wide enhancer activity and chromatin architecture, while the redundant PCE (rPCE) exhibits limited transcriptional impact, and its deletion could induce the compensatory rewiring. These findings align with recent functional studies utilizing high-resolution chromatin conformation analyses, which have revealed that enhancers exhibit functional diversity even within the same cluster. The regulatory logic is not simply additive in power but instead depends on spatially defined cooperation and hierarchy among individual enhancers. Studies across various systems have identified a mechanism by which essential enhancers maintain chromatin structure and transcriptional activation. Some revealed that deletion of “hub enhancers”, which exhibit high chromatin interaction frequency and CTCF binding enrichment, could cause loss of gene expression [47], while others demonstrated that the “first enhancer” in an interaction chain acts as a gatekeeper for enhancer-promoter communications [70]. On the other hand, a growing body of work highlights redundancy as a property of enhancer architecture. A study in mammalian development has shown that enhancers act redundantly to ensure phenotypic robustness as the combined deletion of enhancers causes severe transcriptional and developmental defects [71], while a recent dissection of the Sox2 enhancer cluster demonstrated that only a few TF-bound sites are strictly required for gene activation [72]. Additionally, our study not only supports a hierarchical and cooperative model of enhancer function but also highlights a previously uncharacterized mode of compensatory enhancer rewiring within a disease-relevant regulatory hub. While this study provides a detailed functional dissection of PCE clusters, several limitations should be addressed. First, our analyses focused on the representative region of a PCE cluster at chr6q24.1. Although this region was selected based on strong H3K27ac signals, robust transcriptional activation in cancer, and prominent differences in chromatin looping between normal and cancer cells, it remains unclear whether the regulatory principles uncovered here are universal to all the PCE clusters. While our genome-wide analyses identify PCEs and their associations with transcriptional activation and chromatin interactions, the hierarchical organization of enhancer classes described here has not yet been validated at additional loci. Future studies will be required to determine the extent to which these principles generalize across the genome. Second, our multi-omic analyses, including ChIP-seq, RNA-seq, ATAC-seq, and RCMC, were performed in bulk populations. Consequently, these data do not resolve cell-to-cell variability and may obscure the heterogeneity among cells. While our observations are consistent with a hub-like organization, alternative models, such as pairwise interactions across heterogeneous cell populations, cannot be excluded. Future integration of single-cell or single-nuclei chromatin and transcriptomic approaches would provide more insights into enhancer heterogeneity and will be required to distinguish between these models. Lastly, although our data implicate FOXA1 binding as a key determinant of enhancer classes, how FOXA1 classifies enhancers and contributes to the recruitment of transcriptional machinery and co-factors remains unresolved. FOXA1 is a pioneer TF known to modulate chromatin accessibility in prostate tumorigenesis [30, 67, 73]. FOXA1 is known to form condensates with other TFs [74, 75], and the degree to which enhancers contribute to condensate formation in prostate cancer can differ across distinct enhancer classes [76–79]. The distinction between these enhancer classes can be difficult to identify using only ChIP-seq, RNA-seq, ATAC-seq, and RCMC. Future work incorporating biochemical and biophysical assays to evaluate FOXA1 activity, along with its interactions with co-factors will help reveal the mechanisms of TF and coactivator dependencies that distinguish cPCEs from rPCEs. Lastly, further perturbation studies examining the effects of enhancer sequence and spatial positioning will be essential for a deeper understanding of E-P interactions. Conclusions Together, our findings identify PCEs and provide mechanistic insights into how hierarchically organized PCE clusters coordinate cancer-specific transcriptional programs in prostate cancer. This work also highlights the importance of spatial enhancer architecture in gene regulation and establishes a framework for functionally dissecting enhancer cooperation in cancer. Moving forward, our framework for enhancer dissection could be applied to other malignancies to uncover enhancer-directed cancer-specific transcriptional activation, facilitating the identification of key genes and interactions that drive cancer pathogenesis. Our study reveals how individual enhancers within clusters cooperate to drive cancer-specific gene expression, offering a new avenue to uncover noncoding cancer drivers missed by traditional analyses. Methods Cell culture 22Rv1 (CRL-2505) and RWPE-1 (CRL-11609) cells were obtained from the American Type Culture Collection (ATCC). RC-77N/E and RC77T/E cells were received from Johng Sik Rhim. Cells were cultured according to the suggested protocols at 37 °C with 5% CO2. 22Rv1 cells were cultured in RPMI 1640 (Corning, #10–040-CV) supplemented with 10% FBS (Avantor, #89,510–188) and 1% penicillin/streptomycin. RWPE-1, RC-77N/E, and RC77T/E cells were cultured in the Keratinocyte Serum Free Medium kit (Thermo Fisher, #17,005–042). All cell lines were authenticated at the USC Norris Cancer Center cell culture facility by comparison to the ATCC and/or published genomic criteria for that specific cell line; all cells were documented to be free of mycoplasma. CRISPR/Cas9-mediated enhancer deletions Guide RNAs (gRNAs) targeting upstream and downstream of the candidate enhancers were designed via CHOPCHOP (https://chopchop.cbu.uib.no/) and cloned into pSpCas9(BB)−2A-Puro (PX459) V2.0 plasmid (Addgene, #62,988) following the previously established protocol [32]. The sequences of all gRNAs used in this study can be found in Additional file 1: Table S6. PX459 plasmids (with gRNA or the empty vector) were transfected into 22Rv1 cells (wildtype or single deletion clones) using Lipofectamine LTX with PLUS reagent (Thermo Fisher, #15338100) according to the manufacturer’s instructions. After 48 h of incubation, cells were treated with 2 μg/mL puromycin for 48–72 h until all the un-transfected negative control cells died. The media was replaced with normal complete media, and cells were allowed to recover to show normal morphology and confluency. Selected cells were harvested and seeded into 96-well plates with 1 cell/well to grow into colonies. The expanded single-cell clones were harvested using the Quick-DNA Miniprep Plus Kit (Zymo Research, #D4069) according to the manufacturer’s instructions and then genotyped by PCR and qPCR using primers listed in Additional file 1: Table S6. Briefly, four PCR primers were designed to amplify the upstream cut site, the downstream cut site, and the whole flanking region of a deletion. PCR products were analyzed on a 1.5% agarose gel alongside a 1kb DNA ladder. Clones with a homozygous deletion produced only a band in the flanking region reflecting the deletion, whereas left junction and right junction amplicon bands were obtained from a wild-type chromosome. One set of qPCR primers was designed within the CRISPR/Cas9 cut site to select true homozygous clones during real-time PCR further. Chromatin Immunoprecipitation sequencing (ChIP-seq) RC-77N/E, RC-77 T/E, and 22Rv1 enhancer deletion and NTC cells were used to perform ChIP as previously described [26]. Briefly, cells at 70–90% confluence were cross-linked with 1% formaldehyde for 10 min and quenched with 125 mM glycine for 5 min at room temperature. Cells were collected and lysed on ice with a cell lysis buffer (5 mM PIPES pH 8.0, 85 mM KCl, 1% NP-40) with a douncer. After spin down, nuclei were pelleted and lysed on ice with a nuclei lysis buffer (50 mM Tris pH 8.1, 10 mM EDTA, 1% SDS) to collect chromatin. Fixed chromatin was sonicated to 200–500 bp fragments using the Bioruptor Pico sonicator (Diagenode, #B01060010). For overnight immunoprecipitation, 20 μg of chromatin was used with 10 μL of antibody targeting H3K27ac (Active Motif, #39,133) and CTCF (Cell Signaling Technology, #8758S), while 400 μg of chromatin was used for 20 μL of antibody targeting FOXA1 (Cell Signaling Technology, #53528S). Antibody-bound DNA was then collected with protein A/G magnetic beads (Thermo Fisher, #88,803) and subjected to overnight reverse crosslinking at 67 °C. Pull-down DNA fragments were purified with the Qiagen MinElute columns (Qiagen, #28,004) and made into libraries using the KAPA Hyper Prep kit (Roche, #KK8505) according to the manufacturer’s instructions. ChIP libraries were purified with AMPure XP beads (Beckman Coulter, #A63881). ChIP-seq libraries were sequenced to have at least 20 million read pairs (Additional file 1: Table S1B) using Illumina NovaSeq machines. According to the ENCODE3 ChIP-seq guidelines (https://github.com/ENCODE-DCC/chip-seq-pipeline), we generated two biological replicates of each dataset that passed quality control (QC). Datasets were processed using an in-house pipeline according to the ENCODE3 guidelines. Briefly, BWA [80] was used to align the raw sequencing reads to the human reference genome (hg38). Library complexity was computed to determine the PCR bottleneck coefficients (PBC), non-redundant fraction (NRF), normalized strand cross-correlation coefficient (NSC), and relative strand cross-correlation coefficient (RSC) [81]. PCR duplicates and unaligned reads were removed using Picard (https://broadinstitute.github.io/picard/) and samtools [82]. Uniquely mapped reads were used to call peaks using MACS2 [83]. After peak calling, the irreproducibility discovery rate (IDR) (https://github.com/nboley/idr) was performed to identify the reproducible ChIP-seq peaks, which have high consistency between two biological replicates. ChIP-seq tracks were visualized in IGV [84] or UCSC Genome Browser [85] using the bigWig files that include the adjusted p-value of called peaks. Tracks were automatically scaled based on the highest ChIP-seq signal in the genomic window unless otherwise specified. Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) ATAC-seq libraries were prepared using the Zymo-seq ATAC Library Kit (Zymo Research, #D5458) according to the manufacturer’s instructions. Briefly, 50,000 cells were harvested and lysed on ice to obtain intact nuclei for each replicate, and a Tn5 Transposase was used to tagment the genomic DNA in situ at 37 °C for 30 min at 1,000 rpm. The tagmented DNA fragments were purified and then amplified with ATAC library PCR mix. ATAC-seq libraries were sequenced to have at least 50 million read pairs (Additional file 1: Table S1B) using an Illumina NovaSeq X Plus with paired-end 150 bp reads. ATAC-seq data were processed following the ENCODE3 guidelines (https://github.com/ENCODE-DCC/atac-seq-pipeline). Briefly, raw sequencing reads were trimmed to remove adapter sequences using the Cutadapt program [86]. Trimmed reads were aligned to the human reference genome (hg38) using Bowtie2 [87]. PCR duplicates, low-quality alignments, and mitochondrial reads were removed using Picard (https://broadinstitute.github.io/picard/) and samtools [82]. Library complexity was assessed by calculating the NRF, PBC, and RSC. Uniquely mapped, properly paired reads were used to call accessible chromatin peaks using MACS2 [83]. The irreproducibility discovery rate (IDR) (https://github.com/nboley/idr) was performed to identify the reproducible ATAC-seq peaks, which have high consistency between two biological replicates. ATAC-seq tracks were visualized in IGV [84] or UCSC Genome Browser [85] using the bigWig files that include the adjusted p-value of called peaks. Tracks were automatically scaled based on the highest ATAC-seq signal in the genomic window unless otherwise specified. Differential enrichment analysis of ChIP-seq and ATAC-seq datasets To compute differentially enriched regions, we used the R package DiffBind (v.1.16.3) [31] for differential enrichment analysis of ChIP-seq and ATAC-seq datasets. At least two biological replicates of each condition were used, and the aligned BAM reads and peaks from the ChIP-seq or ATAC-seq datasets were used as inputs. A consensus peak set was first constructed by merging peaks that appeared in at least one replicate per condition. Next, read counts were then quantified across this consensus peak set and normalized using an EdgeR-based library size correction. Differential enrichment was then calculated between experimental groups (e.g., tumor vs. normal, or deletion vs. control), and peaks with FDR < 0.05 and log2 fold change > 0 were considered statistically significant. PCE identification We identified putative prostate cancer-specific enhancers (PCEs) using a patient-centric pipeline where the enhancer roadmap was first established from prostate tissue H3K27ac ChIP-seq datasets from prostate cancer patients and subsequently compared to prostate cell line H3K27ac ChIP-seq datasets (Additional file 2: Fig. S1). Specifically, consensus peaks were identified independently in tumor tissue samples and cancer cell lines using DiffBind (dba.count) and recentered on peak summits (patient cohort: n = 426,316; cell lines: n = 92,419). Only regions overlapping between tissues and cell lines were retained (n = 82,187), ensuring consistency across systems. Differential analysis (tumor vs. normal) was then performed separately in tissues and cell lines using DiffBind (dba.contrast), and cancer-specific regions were identified (FDR < 0.05). Regions consistently enriched in both tissues and cell lines were retained (n = 12,934). Promoter-proximal regions (± 2 kb from transcription start sites) were excluded to yield 11,338 PCEs. To quantify inter-patient heterogeneity, we calculated the frequency of each PCE across 165 prostate tumor tissue samples and classified enhancers as frequent (frequency ≥ 0.7) or rare (frequency < 0.7) (Additional file 1: Table S2). Lastly, to perform functional assays in prostate cell lines, PCEs that are statistically significantly enriched in the 22Rv1 prostate cancer cell line compared with the RWPE-1 normal prostate cell line were identified (FDR < 0.05; n = 3,216 overlaps). Transcription Factor (TF) motif analysis To identify TF motifs enriched in PCEs, the findMotifsGenome.pl script from HOMER [88] was used, and the top 10 TF motifs most enriched at ATAC-seq peaks within PCEs were identified with FDR < 0.05. To calculate motif score, JASPAR motif database was used [89], and the motif with the highest score within the central 500bp of the enhancer region was recorded along with its motif sequence (Additional file 1: Table S2). Region Capture Micro-C (RCMC) Micro-C libraries were made in RWPE-1 and 22Rv1 cells as previously described [27]. Briefly, cells were fixed with 3 mM DSG and 1% formaldehyde. After quenching with Tris buffer, cells were lysed on ice to obtain nuclei. Chromatin was digested in situ with MNase to fragment the genomic DNA at nucleosome resolution. The DNA ends were repaired with biotinylated dNTPs, and then proximal DNA ends were ligated while unligated ends were digested. Chromatin was then reverse crosslinked and digested by Protease K, and DNA was purified by phenol–chloroform extraction and purification columns. 250—400 bp dinucleosomal DNA fragments were selected on an agarose gel and end-polished. Streptavidin beads were used to pull down ligated dinucleosomal DNA fragments. DNA fragments bound to streptavidin beads were made into Micro-C libraries using NEBNext Ultra II DNA Library Prep Kit (NEB, #E7645L) and KAPA HiFi HotStart ReadyMix (Roche, #09420398001) following the manufacturer's instructions. Purified Micro-C libraries were pooled together and made into RCMC libraries following the Twist Standard Hybridization Target Enrichment Protocol (Twist Biosciences, #104178). Briefly, the libraries were hybridized to a custom probe panel for the chr6q24.1 loci and bound to streptavidin beads. DNA fragments on the beads were amplified using Equinox Library Amp Mix and purified with AMPure beads. RCMC libraries were sequenced to have at least 50 million read pairs on an Illumina Novaseq X Plus with paired-end 150 bp reads (Additional file 1: Table S1B). In each cell type, two biological replicates were generated. RCMC data were processed following 4DN data portal guidelines (https://data.4dnucleome.org/help/user-guide). Briefly, we first aligned the raw sequencing reads with BWA MEM v0.7.17 [80], and the aligned reads were paired, sorted, and filtered with Pairtools v0.3.0 [90] and converted into pairs files. Next, we normalized paired data to generate matrix files with Juicertools [91] for Juicer (v1.5). Topologically Associating Domain (TAD) analysis Hi-C datasets previously generated in-house [10] (Additional file 1: Table S1A) were processed using 4D Nucleome (4DN) pipeline as previously described [26]. Briefly, we first aligned the raw sequencing reads with BWA MEM v0.7.17 [80], and the aligned reads were paired, sorted, and filtered with Pairtools (v0.3.0) (https://pairtools.readthedocs.io/en/latest) and converted into pairs files. Next, we normalized paired data to generate matrix files with Juicertools [91] for Juicer (v1.5). The hic files were then used to call TADs using domaincaller [92]. TAD size analysis was performed as previously reported [10]. Briefly, if the boundaries of the domains are within 50 kb windows or the domains are overlapped ≥ 80% between 22Rv1 and RWPE-1, we considered that these TADs overlapped. TADs that are specific to 22Rv1 cells are annotated in Additional file 1: Table S3. TAD impact score calculations and ranking To quantify the genome-wide regulatory potential of PCE-enriched regions, we integrated three metrics: (1) Expression, measured as the ratio of statistically significantly upregulated genes in 22Rv1 compared to RWPE-1 within the TAD relative to the total number of genes in the TAD; (2) PCE load, defined as the total number of PCEs within each TAD; and (3) Interaction contact within each TAD, quantified as the number of significant Hi-C derived E/P interactions (FDR < 0.05) that were unique to 22Rv1 compared to RWPE-1. These three components were combined using a multiplicative model:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{Impact Score of a TAD}= \frac{\left(\left|\text{Significantly upregulated genes}\right|+ 1\right)\times \left(\left|\mathrm{PCEs}\right|+ 1\right)\times \left(\left|\text{E/P interactions}\right|\right)+ 1}{\left(\left|\mathrm{Genes}\right|+ 1\right)}$$\end{document}Impact Score of a TAD=Significantly upregulated genes+1×PCEs+1×E/P interactions+1Genes+1 A total of 2,960 TADs were scored and ranked based on their impact score. Genomic coordinates of the top-ranked domains were annotated to the overexpressed genes in 22Rv1 cells in the TAD, and to genes that were significantly and positively correlated with prostate cancer progression (FDR < 0.05, correlation coefficient > 0.5) [33] (Additional file 1: Table S3). Differential chromatin interaction analysis The RCMC chromatin loops were called using FitHiC [93] at 1 kb resolution. Each loop anchor was characterized based on the regulatory elements (promoters, enhancers, or insulator) it overlapped with. For each loop comparison between deletion and NTC cells, the normalized chromatin contact count was used to calculate the percentage change. To generate differential interaction heatmaps, raw interaction counts for the RCMC-captured region (chr6:138,000,000–139,500,000) were obtained from HiC-Pro [94] and normalized to counts per million. These normalized matrices were then used to compute differences between conditions (e.g., tumor vs. normal, or deletion vs. control) and generate differential interaction heatmaps using GENOVA R package [95]. To generate interaction network maps, significant loops called from FitHiC (FDR < 0.05) were identified from RWPE-1 and 22Rv1 wildtype cells, and each loop anchor was characterized based on the CEs, PCEs, NPEs, and promoters of genes at the chr6q24.1 locus. In the interaction matrix, interaction strength was recorded as -log10(FDR) of each loop. Then, R package iGraph [96] was used to plot the interaction network, with line color and width representing interaction strength, and node size representing connectivity. Reverse-Transcription Quantitative PCR (RT-qPCR) Total RNA was extracted from cells using the Direct-zol RNA miniPrep Kit (Zymo Research, # R2050) according to the manufacturer’s instructions. For each biological sample, 1000 ng RNA was reverse transcribed into cDNA using the iScript Reverse Transcription Supermix for RT-qPCR (Biorad, #1708840) according to the manufacturer’s instructions. Quantitative real-time PCR was performed on a Bio-Rad CFX96 Real-Time PCR Detection System. The primer sequences are listed in Additional file 1: Table S6. Data were analyzed by the 2(−ΔΔCt) method and represented as fold changes of gene expression relative to GAPDH. RNA sequencing (RNA-seq) RNA quality was assessed by calculating RIN (RNA Integrity Number) using an Agilent 4200 TapeStation System (Agilent, #G2991BA), and QC-passed RNA was made into RNA-seq libraries using the TruSeq Stranded mRNA kit (Illumina, #20020595) according to the manufacturer’s instructions. Three biological replicates for each sample were prepared. RNA-seq libraries were sequenced to have at least 20 million read pairs on an Illumina Novaseq X Plus with paired-end 150 bp reads (Additional file 1: Table S1B). RNA-seq reads were processed using an in-house RNA-seq analysis pipeline, adapted from the NCI Genomic Data Commons bioinformatic pipeline (https://docs.gdc.cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/). Reads were first trimmed using Trimgalore (https://github.com/FelixKrueger/TrimGalore) to filter out low-quality reads and the adapter sequences. The trimmed reads were then aligned to the human reference genome (hg38) with the annotation GENCODE V29 [97] using STAR [98] to generate BAM files. HTseq [99] was used to quantify the read counts for each gene, with the BAM files as the input. BAM files were converted to tagAlign format to be visualized in IGV [84] or UCSC Genome Browser [85]. Tracks were automatically scaled based on the highest RNA-seq signal in the genomic window unless otherwise specified. Differential gene expression analysis DESeq2 [100] was used to calculate the statistical significance of gene expression difference between experimental and control groups. The p-values were then adjusted to correct multiple comparisons and measure the false discovery rate (FDR). The differentially expressed genes (DEGs) were identified based on FDR and fold change (FDR < 0.05, |Fold Change|> 1.5). Pathway analysis DEGs identified from RNA-seq results in deletion clones were used to perform the Ingenuity Pathway Analysis (IPA) (https://digitalinsights.qiagen.com/IPA). The signaling pathways enriched and DEGs in each category are listed in Additional file 1: Table S8. Small interfering RNA (siRNA) transfection HEBP2 and NHSL1 were knocked down using 50 nM of siRNAs. Non-targeting scrambled control siRNA (Dharmacon, #D-001810–10-05) and siRNAs targeting HEBP2 (Dharmacon, #L-020612–02–0005) and NHSL1 (Dharmacon, #L-032698–00–0005) were purchased from Dharmacon. 22Rv1 wildtype cells were transfected with NTC siRNA or siRNAs targeting HEBP2 or NHSL1 using DharmaFECT1 transfection reagent (Dharmacon, #T-2001–02) according to the manufacturer’s instructions. RNA was subsequently extracted from samples and RT-qPCR was used to confirm the knockdown efficiency as described above. After confirmation of consistent knockdown among three biological replicates, RNA samples were used to perform RNA-seq as described above. Cell proliferation assay 22Rv1 enhancer deletion and NTC cells were seeded at a density of 10,000 cells per well in 24-well plates on Day 0. At each time point, cells from one set of triplicate wells were trypsinized and pelleted. Cells were resuspended in PBS and mixed 1:1 ratio with Trypan Blue (Avantor, #10,181–020). Viable cells were counted using Countess 3FL (Thermo Fisher, #AMQAF2000) twice for each well, and the average was recorded. Experiments were repeated at least 3 times. The final growth curves were generated by plotting the mean cell number (± standard deviation) as a function of time. Cell colony formation assay 22Rv1 enhancer deletion and NTC cells were seeded at a density of 500 cells per well in 12-well plates in triplicate on Day 0. Cells were incubated undisturbed for 12 days to allow colonies to form. On Day 12, cells were fixed with 4% paraformaldehyde for 15 min and stained with 0.5% crystal violet solution for 15 min and then washed under running tap water. Experiments were repeated at least 3 times. Images of each well were analyzed with ImageJ, and the number of colonies for each well was recorded. Supplementary Information Additional file 1: Supplemental Tables. Table S1A. Next generation sequencing datasets used in this study. Table S1B. QC metrics of next generation sequencing datasets generated in this study. Table S2. PCE coordinates. Table S3A. Annotated TAD coordinates in 22Rv1. Table S3B. Prioritized genes and candidate small molecule inhibitors from the top 20 TADs with enriched PCEs. Table S4. Differential H3K27ac ChIP-seq peaks in 22Rv1 vs. RWPE-1 at chr6q24.1. Table S5. Upregulated genes in 22Rv1 vs. RWPE-1 at chr6q24.1. Table S6. Oligos and primers used in this study. Table S7. Differential H3K27ac ChIP-seq peaks in PCE-perturbed cells. Table S8. Differentially expressed genes in PCE-perturbed cells at chr6q24.1 and enriched signaling pathways. Table S9. Differentially expressed genes upon siRNA knockdown. Table S10. Differential ATAC-seq peaks in PCE-perturbed cells at chr6q24.1. Table S11. Differential FOXA1 ChIP-seq peaks in PCE-perturbed cells at chr6q24.1. Table S12. Comparison between cPCE and rPCE.Additional file 2: Supplemental Figures. Fig. S1. Stepwise framework for identifying PCEs through integration of prostate cancer patient cohorts and prostate cell lines. Fig. S2. A subset of PCEs is clustered within TADs. Fig. S3. The chr6q24.1 PCE hub exhibits elevated enhancer activity and gene expression in prostate cancer. Fig. S4. cPCE deletion led to a reduction in the activity of other PCEs and their chromatin interactions. Fig. S5. cPCE deletion slightly increased CTCF-mediated chromatin organization. Fig. S6. rPCE deletion did not result in dramatic changes in enhancer activity and chromatin looping. Fig. S7. Deletion of resPCE led to a reduction in the activity of other PCEs. Fig. S8. Pioneer TF FOXA1 differentially regulates distinct enhancer classes. Additional file 1: Supplemental Tables. Table S1A. Next generation sequencing datasets used in this study. Table S1B. QC metrics of next generation sequencing datasets generated in this study. Table S2. PCE coordinates. Table S3A. Annotated TAD coordinates in 22Rv1. Table S3B. Prioritized genes and candidate small molecule inhibitors from the top 20 TADs with enriched PCEs. Table S4. Differential H3K27ac ChIP-seq peaks in 22Rv1 vs. RWPE-1 at chr6q24.1. Table S5. Upregulated genes in 22Rv1 vs. RWPE-1 at chr6q24.1. Table S6. Oligos and primers used in this study. Table S7. Differential H3K27ac ChIP-seq peaks in PCE-perturbed cells. Table S8. Differentially expressed genes in PCE-perturbed cells at chr6q24.1 and enriched signaling pathways. Table S9. Differentially expressed genes upon siRNA knockdown. Table S10. Differential ATAC-seq peaks in PCE-perturbed cells at chr6q24.1. Table S11. Differential FOXA1 ChIP-seq peaks in PCE-perturbed cells at chr6q24.1. Table S12. Comparison between cPCE and rPCE. Additional file 2: Supplemental Figures. Fig. S1. Stepwise framework for identifying PCEs through integration of prostate cancer patient cohorts and prostate cell lines. Fig. S2. A subset of PCEs is clustered within TADs. Fig. S3. The chr6q24.1 PCE hub exhibits elevated enhancer activity and gene expression in prostate cancer. Fig. S4. cPCE deletion led to a reduction in the activity of other PCEs and their chromatin interactions. Fig. S5. cPCE deletion slightly increased CTCF-mediated chromatin organization. Fig. S6. rPCE deletion did not result in dramatic changes in enhancer activity and chromatin looping. Fig. S7. Deletion of resPCE led to a reduction in the activity of other PCEs. Fig. S8. Pioneer TF FOXA1 differentially regulates distinct enhancer classes.