Metastatic castration-resistant prostate cancer (mCRPC) remains a significant clinical challenge, necessitating a deeper understanding of its complex genomic architecture. To unravel the intricacies of the 3D genome organization in mCRPC, a comprehensive study, identified by the key identifier 01826, meticulously analyzed 80 metastatic biopsy samples. This research, conducted by the Stand Up 2 Cancer (SU2C) – Prostate Cancer Foundation (PCF) West Coast Dream Team (WCDT) consortium, employed a low-input Hi-C assay [18] to map chromatin interactions with remarkable 10-kb spatial resolution. The patient cohort in study 01826 represented diverse treatment histories, with a majority (77.5%) having received prior androgen signaling inhibitors (ASI) like abiraterone or enzalutamide, while subsets had undergone chemotherapy (15%) or sipileucel-T therapy (15%) (Supplementary Table [1]).
The study 01826 integrated deep Hi-C sequencing data, amassing a median of 1.3 billion reads per sample, with a rich multi-omic dataset. This included deep Whole Genome Sequencing (WGS, n = 76, median depth 129×), RNA-sequencing (RNA-seq, n = 78, median = 113 million reads per sample), deep Whole Genome Bisulfite Sequencing (WGBS, n = 69, median depth 46×), and 5-hydroxymethylcytosine sequencing (5hmC-seq, n = 52, median = 26 million reads per sample). All these analyses were performed on the same cohort of mCRPC samples, providing an unprecedentedly detailed view of the 3D genome and its interplay with other genomic and epigenomic layers in mCRPC (Fig. [1a] and Supplementary Table [1]).
Figure 1: Multi-omic Approach to 3D Genome Analysis in mCRPC Study 01826.
Overview of the integrated multi-omics data in study 01826, highlighting the Hi-C approach combined with WGS, RNA-seq, WGBS and 5hmC-seq for comprehensive 3D genome analysis in metastatic castration-resistant prostate cancer.
a, Schematic representation of the study design for project 01826. b–i, Analysis of A and B compartments in mCRPC genomes from study 01826, showing associations with: genome coverage (n = 80) (b), gene overlap percentage (n = 80) (c), mean gene expression (TPM, n = 78) (d), 5hmC peak overlap (n = 52) (e), PMD overlap (n = 69) (f), HMR overlap (n = 69) (g), mutation overlap (n = 76) (h) and SV overlap (n = 76) (i). Each data point represents an individual mCRPC sample from study 01826. Dotted lines connect A and B compartment values within the same sample. Statistical significance (P values) was determined using a paired, two-sided Wilcoxon rank-sum test. Box plot elements: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range (IQR). Source data for Figure 1 in study 01826 is available [/articles/s41588-024-01826-3#MOESM3).
A/B Compartments and Their Association with Genomic and Epigenomic Alterations in mCRPC: Findings from Study 01826
Study 01826 delved into the functional implications of A and B compartments within the mCRPC genome. A compartments, known for harboring active chromatin loops and facilitating gene transcription, were contrasted with B compartments, characterized by repressive loops. The analysis within study 01826 revealed that A compartments encompass a larger fraction of the genome compared to B compartments in mCRPC (Fig. [1b]; P < 1.0 × 10−10). Consistent with their role in active transcription, A compartments demonstrated a greater overlap with genes and exhibited higher gene expression levels (Fig. [1c,d]; P < 1.0 × 10−10 and P < 1.0 × 10−10, respectively).
Furthermore, study 01826 explored the relationship between A/B compartments and epigenetic modifications. A compartments showed a significant enrichment for 5-hydroxymethylcytosine (5hmC), a marker of active gene regions [6] (Fig. [1e]; P < 1.0 × 10−10). Conversely, B compartments displayed a stronger association with partially methylated domains (PMDs) (Fig. [1f]; P < 1.0 × 10−10), aligning with previous research linking PMDs to B compartments [29, 30] and lamina-associated domains (LADs) [31]. No significant difference was observed in the overlap with hypomethylated regions (HMRs) between A and B compartments (Fig. [1g]; P = 0.1).
The study 01826 also confirmed the known association of PMDs [32] and B compartments [33] with LADs, which organize chromatin at the nuclear periphery and influence replication and transcription timing. Furthermore, consistent with previous research, study 01826 found a lower global mutational burden in A compartments compared to B compartments (Fig. [1h]; P < 1.0 × 10−10), potentially due to increased accessibility of open chromatin in A compartments to DNA repair mechanisms [31, 34, 35, 36]. Similarly, structural variations (SVs) were also less frequent in A compartments (Fig. [1i]; P < 1.0 × 10−10).
Extrachromosomal AR DNA Identified by Hi-C in Study 01826
Extrachromosomal circular DNA (ecDNA), often carrying amplified oncogenes, plays a significant role in cancer progression. Study 01826 leveraged Hi-C data to investigate ecDNA in mCRPC, focusing on the Androgen Receptor (AR) gene, a frequently amplified oncogene in this cancer type [4, 11, 12]. The study 01826 found a significant association between the A/B compartment assignment of the AR locus and AR gene expression, even after adjusting for other genomic and epigenomic alterations (P = 1.3 × 10−5 for A compartment association; Fig. [2a]).
However, in some samples from study 01826, the AR locus lacked a clear A or B compartment assignment. These instances coincided with samples exhibiting exceptionally high AR expression and copy number (CN) amplification. As a control, the researchers in study 01826 examined SChLAP1, a prostate cancer-specific long noncoding RNA not typically amplified in mCRPC. SChLAP1 locus assignment to A or B compartments was observed in almost all samples and correlated significantly with expression (P = 0.015 for SChLAP1; Fig. [2b]).
Focusing on samples without A/B compartment assignment at the AR locus, study 01826 analyzed the regional contact frequency sliding-window (RCFS) score. A dramatic decrease in RCFS values around the AR locus was observed in these samples, often encompassing the upstream enhancer region as well (Fig. [2c]). This reduction in contact, combined with AR amplification and elevated expression, strongly suggested ecDNA-mediated oncogene activation [42, 43]. Increased Hi-C interaction between the ends of the putative ecDNA region further supported circularization (Fig. [2d]).
To validate the ecDNA hypothesis, WGS data from study 01826 samples were analyzed using AmpliconArchitect. A significant majority (77%) of samples with low RCFS scores showed evidence of ecDNA by AmpliconArchitect. FISH analysis on selected mCRPC samples from study 01826 further confirmed AR ecDNA positivity in AmpliconArchitect-positive cases. These findings highlight Hi-C as a complementary method to WGS for identifying and characterizing ecDNA, particularly in assessing the 3D conformation of genomic rearrangements.
Figure 2: Characterization of AR ecDNA in mCRPC Samples from Study 01826.
Study 01826 reveals AR ecDNA in mCRPC through integrated Hi-C and WGS analysis, impacting gene expression and treatment response.
a,b, Correlation between gene expression and genomic alterations (CN gain, red; loss, blue) for AR (a) and SChLAP1 (b) in study 01826 samples. Promoter methylation status is also indicated. c, RCFS heatmap from study 01826 samples, showing depletion of local contacts around the AR locus in ecDNA-enriched samples. Sample meta-data includes average RCFS, CN, AR TPM, A/B domain call, and AmpliconArchitect ecDNA cyclical amplification score. Gray tiles indicate missing data. d, Hi-C contact frequency matrices from study 01826, contrasting an ecDNA+ sample (above diagonal) and ecDNA− sample (below diagonal) around the AR locus. White circle highlights increased interaction between ecDNA region ends in the ecDNA+ sample. e, Scatter plot of per-sample AR TPM versus RCFS in study 01826, with red dashed line indicating the RCFS cutoff for ecDNA classification. f,g, Kaplan–Meier overall survival (OS) curves from study 01826, stratified by RCFS (high in f, low in g) and ASI treatment post-biopsy. Interaction P value from Cox regression is shown. HR, Cox hazard ratio; ANOVA, analysis of variance. Source data for Figure 2 in study 01826 is available [/articles/s41588-024-01826-3#MOESM4).
Clinical Implications of AR ecDNA and ASI Therapy Resistance: Findings from Study 01826
Given the association of ecDNA with poor cancer outcomes and the role of AR amplification in ASI resistance, study 01826 investigated the clinical relevance of AR ecDNA in mCRPC patients treated with ASIs. Analysis of clinical data from the study 01826 cohort [44] revealed that patients with typical AR RCFS benefited significantly from ASI treatment post-biopsy (P = 0.0041; Fig. [2f]). However, this benefit was absent in patients with depleted RCFS indicative of AR ecDNA (P = 0.72; Fig. [2g]). The statistically significant interaction between ASI treatment and RCFS (P = 0.0051) in Cox regression suggests that AR locus RCFS, as identified in study 01826, is a predictive biomarker for diminished ASI benefit, independent of AR CN. These findings from study 01826 indicate that mCRPC tumors harboring ecDNA-amplified AR may exhibit resistance to ASI therapy.
3D Gene-Enhancer Interactions and Gene Expression Regulation in mCRPC: Insights from Study 01826
Study 01826 further explored the role of 3D chromatin structure in regulating gene expression by examining interactions between gene promoters and enhancers. Integration of Hi-C data with ChIP-seq data for histone modifications (H3K27ac) and DNA methylation from prior prostate cancer studies [10] allowed researchers in study 01826 to investigate how 3D contacts influence gene expression. As hypothesized, increased chromatin interaction strength between gene promoters and enhancers generally correlated with higher gene expression levels, but only up to a certain point (Fig. [3b]). Beyond this point, gene expression decreased, potentially due to the formation of overly condensed, transcriptionally inactive chromatin. This is supported by the A:B compartment ratio analysis in study 01826, which showed an initial increase in open A compartments with increasing interaction strength, followed by a divergence (Fig. [3c]). These genome-wide analyses from study 01826 underscore the crucial role of 3D genome organization in gene regulation.
Figure 3: Genome-Wide Analysis of Gene-Enhancer Contacts in mCRPC Study 01826.
Study 01826 provides a genome-wide perspective on gene-enhancer interactions and their impact on gene expression in metastatic prostate cancer.
a, Chromosome tracks (10-kb bins) from study 01826 showing (left to right): sum of 3D contact frequencies with enhancers, log-ratio of A/B compartment calls, and average gene expression (TPM). b, 2D density plot from study 01826 illustrating the relationship between gene-enhancer contact frequency and gene expression. c, Scatter plot from study 01826 showing the correlation between A/B compartment log-ratio and gene-enhancer contact frequency. d–f, Recurrent intrachromosomal contacts (green links) between TSS (red dot) of FOXA1 (d), MYC (e) and AR (f) and putative enhancers (blue peaks) in study 01826. Pink peaks indicate t-statistic for association between enhancer hypomethylation and gene expression. Source data for Figure 3 in study 01826 is available [/articles/s41588-024-01826-3#MOESM5).
Study 01826 further investigated specific oncogenes known for enhancer interactions: FOXA1, MYC, and AR [4, 5, 45, 46, 47, 48, 49]. Recurrent contacts between these genes and enhancer regions were observed within megabase-sized regions. For FOXA1, the FOXM1-induced enhancer showed looping, and hypomethylation correlated with gene expression. Multiple other putative enhancers also interacted with FOXA1, with correlation diminishing with distance (Fig. [3d] and Supplementary Fig. [3]). Similar patterns were observed for MYC, with interactions involving PVT1, PCAT1, and prostate cancer risk loci on chromosome 8q24 (Fig. [3e] and Supplementary Fig. [3]). AR also exhibited complex enhancer interactions, including the known upstream enhancer and other distal putative enhancers, with decreasing correlation with distance (Fig. [3f] and Supplementary Fig. [3]). These findings from study 01826 suggest a more intricate cis-regulatory landscape for these key mCRPC oncogenes than previously understood.
The TAD structure, exemplified at the AR locus in study 01826 (Fig. [4]), plays a critical role in organizing these regulatory interactions. Amplification and epigenetic modifications of the AR gene and its enhancer plexus within the same TAD are thought to drive overexpression and therapy resistance [4, 5, 7, 11, 12]. The frequent co-amplification of AR enhancers located both upstream and downstream within the same TAD, as observed in study 01826, may be driven by the selective pressure to maintain physical proximity within the TAD structure, influencing the patterns of structural variations in mCRPC.
Figure 4: Genomic and Epigenomic Landscape Surrounding the AR Locus in mCRPC from Study 01826.
Study 01826 provides a detailed view of the genomic and epigenomic events at the AR locus in mCRPC, highlighting the role of TADs in regulatory element organization.
Multi-track visualization from study 01826 (top to bottom): genes (AR in blue), Hi-C contact frequency, TAD structure, CN variation (WGS), ERG and AR ChIA-PET data, 5hmC peaks (5hmC-seq), HMRs (WGBS), and H3K27ac ChIP-seq. Red dashed lines indicate TAD borders associated with AR locus amplification. Source data for Figure 4 in study 01826 is available [/articles/s41588-024-01826-3#MOESM6).
3D Topology Influences Structural Variation Patterns in mCRPC: Evidence from Study 01826
Structural variations (SVs) are prevalent in prostate cancer, and study 01826 investigated the interplay between 3D genome topology and SV formation. Hi-C data in study 01826 revealed distinct interaction patterns at SV breakpoints identified by WGS (Fig. [5a,b]). In contrast to size-matched background regions, SV breakpoints showed increased contact frequency in Hi-C data, indicating that SVs bring together DNA regions that are spatially proximal.
Study 01826 hypothesized that spatial proximity influences SV formation. Analysis of deletions, duplications, and inversions revealed a significantly higher proportion of real SVs with both breakpoints within the same TAD, compared to the size-matched background (Fig. [5c]; P < 0.001 for deletions and duplications, P = 0.013 for inversions). This suggests that TAD structure may predispose certain genomic regions to SV formation.
Figure 5: Structural Variations and 3D Chromatin Organization in mCRPC Study 01826.
Study 01826 demonstrates how 3D genome organization, specifically TADs, influences the patterns of structural variations in metastatic prostate cancer genomes.
a, Hi-C contact frequency signatures from study 01826 for deletions, duplications, and inversions (top) compared to size-matched background (bottom). b, Schematic illustrating Hi-C contact frequency matrix analysis around SV breakpoints in study 01826. c, Percentage of SVs and size-matched background regions with both ends within the same TAD in study 01826 samples. P values from Wilcoxon rank-sum test are indicated. d, Median-centered log-ratio of Hi-C contact frequency comparing mCRPC and benign prostate samples from study 01826 at the TMPRSS2–ERG locus. TMPRSS2–ERG fusion-positive samples (above diagonal) and fusion-negative samples (below diagonal). Green, higher in mCRPC; blue, higher in benign prostate. Source data for Figure 5 in study 01826 is available [/articles/s41588-024-01826-3#MOESM7).
Study 01826 also validated previous findings on the TMPRSS2–ERG gene fusion, a common event in prostate cancer. Consistent with cell line studies [55], study 01826 observed increased contact frequency at the TMPRSS2–ERG locus in both fusion-positive and fusion-negative mCRPC tumors compared to benign prostate samples (Fig. [5d]). This suggests that increased spatial proximity at this locus, even in the absence of the fusion, may be a feature of prostate cancer cells.
TAD Subtypes and Their Clinical Relevance in mCRPC: Discoveries from Study 01826
While TAD structure showed commonalities across samples in study 01826, inter-sample variability was also evident. To investigate this heterogeneity, study 01826 quantified TAD edge density across the genome and identified two distinct mCRPC subgroups: one with broad TADs and another with narrow TADs within broader structures (Fig. [6a–c]). These TAD subtypes were not associated with technical factors or basic tumor characteristics.
However, study 01826 revealed significant differences in methylation and gene expression patterns between TAD subtypes. The narrow TAD subgroup exhibited more HMRs (P = 0.04), higher 5hmC levels (P = 0.023), and higher median gene expression (P = 0.007) (Fig. [6d–f]). Gene set enrichment analysis further revealed enrichment of MYC target gene expression in the narrow TAD subgroup (Fig. [6h]), which also showed a trend towards MYC amplification (P = 0.16; Fig. [6g]). Importantly, study 01826 found that the narrow TAD subtype was associated with worse overall survival (OS) (P = 0.048; Fig. [6i]). These findings suggest that TAD subtypes represent distinct transcriptional phenotypes in mCRPC with prognostic implications.
Figure 6: Identification and Characterization of TAD Subtypes in mCRPC Study 01826.
Study 01826 uncovers two distinct TAD subtypes in mCRPC with differing genomic, epigenomic, and clinical characteristics, highlighting the heterogeneity of 3D genome organization in this cancer.
a, t-SNE visualization from study 01826 defining two TAD architecture subtypes based on TAD edge density. b, Distribution of average TAD size per subtype in study 01826. c, Example TAD structure and contact frequency comparison between broad and narrow TAD subtypes from study 01826. d–g, Distributions between TAD subtypes in study 01826 for HMR proportion (n = 69) (d), 5hmC peak abundance (n = 52) (e), median TPM (n = 78) (f), and MYC CN (n = 76) (g). P values from Wilcoxon rank-sum test are indicated. h, Enriched hallmark pathways in narrow TAD subtype from study 01826 (GSEA). i, Kaplan–Meier OS curves for TAD subtypes in study 01826. NES, normalized enrichment score. Source data for Figure 6 in study 01826 is available [/articles/s41588-024-01826-3#MOESM8).
Conclusion: 3D Genome Architecture as a Key Determinant in mCRPC – Insights from Study 01826
In conclusion, study 01826, through integrative analysis of multi-omic data including Hi-C, WGS, WGBS, RNA-seq, and 5hmC-seq, provides a comprehensive characterization of the 3D genome landscape in metastatic castration-resistant prostate cancer. The study 01826 highlights the critical roles of A/B compartments, ecDNA-mediated AR amplification, gene-enhancer interactions, TAD structure, and TAD subtypes in mCRPC biology and clinical outcomes. The identification of AR ecDNA as a potential mechanism of ASI resistance and the discovery of prognostic TAD subtypes underscore the clinical significance of 3D genome organization in mCRPC. These findings from study 01826 pave the way for future research aimed at leveraging 3D genome information for improved diagnosis, prognosis, and therapeutic strategies in advanced prostate cancer.
