Chromatin remodeling restraints oncogenic functions in prostate cancer

doi:10.21203/rs.3.rs-5219856/v1

Primary prostate cancer (PCa) is characterized by multifocal growth and a highly variable clinical course, which is not effectively predicted by prognostic screenings. Innovative strategies for the stratification of primary prostate cancers are still needed. Using prostate biopsies, we analyzed the epigenome of 17 chemo-naïve patients with putative PCa for genome-wide mapping of heterochromatic and euchromatic domains, as well as their three-dimensional (3D) compartmentalization in the cell nucleus. We identified two subgroups of cancer patients with different degrees of chromatin 3D architecture and transcriptome alterations: the LDD (Low Degree of Decompartmentalization) and HDD (High Degree of Decompartmentalization) groups. HDD subtype exhibits an extensive chromatin reorganization that restrains tumor potential, by repressing pathways related to extracellular matrix remodeling and phenotypic plasticity. We derived an 18-genes transcriptional signature that distinguishes HDD from LDD subtype and we confirmed its prognostic relevance across multiple cohorts covering more than 900 prostate cancer patients in total. We propose this transcriptional signature derived from chromatin compartmentalization analysis as a novel prognostic tool that could be adopted at the time of the diagnostic prostate biopsy.

Biological sciences/Genetics/Epigenetics/Gene silencing

Health sciences/Biomarkers/Prognostic markers

Chromatin remodeling

PCa subtypes

heterochromatin

biopsies

prostate cancer prognosis

Prostate cancer (PCa) is the second most common tumor type among males and accounts for 10% of cancer-related deaths ¹. While the majority of PCa cases present as indolent and slow-growing, about 20% of prostate tumors progress to metastatic and lethal forms. Currently, PCa diagnosis relies on histological evaluation of multiple biopsies, usually conducted following elevated prostate-specific antigen (PSA) levels or abnormal digital rectal examination (DRE) ². Despite the use of tumor grading systems, the clinical variability and frequent multifocality of PCa complicate the accurate prediction of cancer outcomes, hindering treatment guidance ³. Consequently, many patients with silent PCa are over-treated ⁴. Although the identification of driver coding mutations in solid tumors has significantly advanced the field of clinical oncology ⁵, the translation of PCa genetic biomarkers into clinical practice has only marginally improved prognosis accuracy ⁶. This limitation underscores the need for novel biomarkers that can more precisely guide clinical decision-making in PCa management.

Emerging evidence suggest that epigenetic information, which regulates transcriptional plasticity, plays a critical role in tumor evolution ⁷. Epigenetic dysfunctions have been shown to initiate carcinogenesis even in the absence of driver mutations ⁸, offering a new perspective on cancer research. ⁹

The three-dimensional (3D) genome architecture, which is hierarchically organized on multiple, highly regulated structural levels in the nucleus, has emerged as crucial for regulating cellular programs ^9,10. Notably, the physical separation between euchromatin, the more accessible and transcriptionally active part of the chromatin, and heterochromatin, which is compacted and gene-poor, is a hallmark of healthy cells. Morphological changes in nuclear organization, referred to as nuclear atypia, still represent a gold standard for diagnosis and staging of various cancers, including prostate cancer ¹¹. A recent study on metastatic prostate cancer patients reported multilevel alterations in chromatin structures associated with poor survival outcomes ¹². However, this work was performed on advanced PCa stages, thus providing limited information for earlier diagnostic samples. The same consideration applies to previous transcriptional studies performed on surgically removed prostates ^13–15.

In this study, we aimed to comprehensively characterize chromatin compartment organization in PCa using our recently developed 4f-SAMMY-seq ¹⁶. We analyzed fresh needle biopsies from PCa patients at their initial clinical presentation. We identified patient-specific changes in chromatin compartments and we extrapolated the underlying genomic dysregulation. Our findings reveal that large-scale chromatin remodeling and transcriptional repression are associated with a protective, antitumoral effect. Based on these insights, we derived a novel gene expression signature to classify this specific PCa subtype. We validated the predictive power of this 18-gene signature across multiple patient cohorts, encompassing more than 900 individuals. This signature demonstrates significant potential as a tool for early prognosis prediction in PCa and could be readily implemented in clinical practice. Questa firma dimostra un potenziale significativo come strumento per la previsione precoce della prognosi nel PCa e potrebbe essere facilmente implementata nella pratica clinica

To characterize chromatin architecture in prostate cancer patients at the time of diagnosis, we collected fresh needle biopsies from chemo-naïve patients undergoing the standard diagnostic procedure. Each patient was punctured to collect 14 needle biopsy cores for histopathological examination (diagnostic biopsies) and 1 to be processed for our research project (research biopsy – Fig. 1a). Each research biopsy was divided into two parts: one-third was used for histopathological and immuno-histological analyses. At the same time, the rest of the tissue sample was enzymatically digested to obtain a cell suspension further split for epigenomic (4f-SAMMY-seq), transcriptomic (RNA-seq) and cellular composition (flow-cytometry) analyses.

Our cohort includes seven non-neoplastic controls (CTR), i.e. patients that turned out to be negative for cancer, and ten primary prostate cancer patients (PCa) with histologically confirmed tumor and Gleason Score (GS) between 3+3 and 5+5 (Fig. 1b). We also report in Fig. 1b the percentage of diagnostic biopsy cores positive for cancer cells (PPC), their distribution in a prostate gland diagram (DPC), and the puncture site of the research biopsy used for this study. All the research biopsies of prostate cancer patients were obtained from punctures adjacent to one or more positive diagnostic biopsy cores (Fig. 1b). In addition to the Gleason Score assigned to the patient in the clinical report (GS-P), we recorded the Gleason Score assigned to the closest diagnostic biopsy (GS-CDB) and the score assigned to the research dedicated biopsy (GS-RDB, Extended Data Fig. 1a). We noted that the histopathological examination on research biopsies did not confirm the presence of tumor cells in four out of ten cases (Extended Data Fig. 1a). However, it must be noted that the Gleason Score assignment was done on a third of the tissue biopsy and that all research biopsy samples express diagnostic PCa tumoral markers like PCA3, HPN and GOLM1^17–19 (Extended Data Fig. 1b). Therefore, we retained all ten PCa samples in our experimental analyses.

Our cohort of PCa patients is relatively homogeneous in terms of age (median 72 years; interquartile range, abbreviated as IQR, 62-89 years) and serum PSA level (median 16 ng/mL, IQR: 7.1-45.5) but heterogeneous in terms of clinical features as percentage of cancer positive biopsy cores (median: 21.4%, IQR: 21.4-39.2), clinical stage (T1c: 60%, T2a-T2b: 20%, T3: 20%) and index lesions on Magnetic Resonance Imaging (MRI) (40% with Prostate Imaging Reporting and Data System (PI-RADS) ≥3 and 60% with PI-RADS 1/2) (Extended Data Table 1). Survival outcomes are not available due to the long clinical course of primary PCa and the short follow-up time since the biopsies were collected (median time from biopsy less than four years).

We also examined tissue samples of our cohort by confocal microscopy. Non-neoplastic controls and PCa patients' cells showed similar nuclear areas but a significant difference in nuclear circularity, with PCa patients’ cells showing nuclei deformations (Extended Data Fig. 1c, d). Staining of the nuclear lamina with the Lamin A/C antibody highlighted a slight depletion of the signal in the nuclear interior, corresponding to a parallel decrease in the Hoechst staining (Extended Data Fig. 1e). Although descriptive, these data suggest a cancer-specific remodeling of the nuclear and genome organization in PCa cells.

When dissociating the needle biopsy tissue samples (size around 2 cm in length, Extended Data Fig. 2a, see Methods), we achieved a viable cell yield in the range of 30,000-80,000 cells per sample (Extended Data Fig. 2b).

As a further control, we examined the cell type composition in a set of 13 prostate biopsies by flow cytometry (Extended Data Fig. 2c-f). We didn't detect significant differences in the relative proportion of epithelial (EpCAM/CD326+, CD45-), leukocyte (EpCAM/CD326-, CD45+) or stromal cells (EpCAM/CD326-, CD45-) between non-neoplastic controls and PCa samples (Extended Data Fig. 2g). All samples showed significant fraction of stromal cells (average 68,8% controls and 70,9% PCa patients), while we found an average of 18,5% and 15,2% of epithelial cells in control and PCa biopsies, respectively.

We applied 4f-SAMMY-seq to examine genome architecture in prostate biopsies¹⁶. This method relies on the sequential isolation and high-throughput sequencing of distinct chromatin fractions separated according to their accessibility and solubility properties, which are expected to correlate with chromatin epigenetic and transcriptional status (Extended Data Fig. 3a). The more soluble fractions (S2S, S2L) are associated with gene activation markers, and the less soluble fractions (S3, S4) are enriched for constitutive heterochromatin.

To confirm the applicability of this technique on fresh prostate biopsies, we first sequenced 4f-SAMMY fractions of non-neoplastic control samples. We examined the DNA sonication and short fragments (S2S) selection to obtain high-quality libraries (Extended Data Fig. 3b). We verified that we obtained sequencing reads with comparable quality across samples (Extended Data Fig. 3c). The sequencing reads distribution profiles for individual fractions from CTR samples show a high level of concordance (Extended Data Fig. 3d), with the highest correlation values for S2L fractions (median: 0.94).

To study euchromatic and heterochromatic domains, we defined the solubility profile as the ratio of high-throughput sequencing reads distribution in S2L over S3 fractions (see Methods), where higher and lower values correspond to regions enriched for open and closed chromatin, respectively. We compared the consensus solubility profiles across all CTR and PCa specimens with epigenomic marks in normal prostate tissue for post-translational histone modifications (histone marks) associated with active (H3K27ac, H3K36me3) and inactive (H3K9me3) chromatin states ²⁰ (Fig. 1c). We observed a high similarity of solubility profiles among CTR samples and a clear concordance with the location of euchromatin marks (H3K27ac and H3K36me3), along with a negative correlation with the heterochromatin mark profile (H3K9me3). These observations were also confirmed in a genome-wide quantification of correlations for individual samples (Fig. 1d).

Instead, in PCa biopsies we observed that the solubility profile is more heterogeneous and its correlation with histone marks is more variable with respect to CTR samples (Fig. 1c, d). Visual inspection of individual solubility profiles showed a patient-specific degree of chromatin alterations (Extended Data Fig. 3e). Moreover, we noticed clear differences among PCa samples, with five out of ten samples losing the regular chromatin compartmentalization, as also highlighted by their correlation with histone marks (Fig. 1d). These differences are remarkably not attributable to the cell composition of the samples (Extended Data Fig. 2g).

Thus, 4f-SAMMY-seq data describe a conserved genome architecture among control biopsies with a separation between euchromatin and heterochromatin. On the other hand, diagnostic early-stage PCa biopsies show a patient-specific chromatin solubility dysregulation that mirrors PCa’s highly heterogenous nature.

To further analyze the chromatin 3D organization, we used the sequencing data of all the 4f-SAMMY-seq fractions to identify active and inactive compartments, named "A" and "B", respectively, as per the convention adopted for Hi-C derived compartments¹⁶ (Fig. 2a, b, see Methods). However, unlike Hi-C, the chromatin compartments inferred from 4f-SAMMY-seq are defined based on their chromatin accessibility and solubility properties¹⁶. CTR samples showed a consistent pattern across all seven individuals with 27% of the entire genome having conserved active A compartment and 39% having conserved inactive B compartment across all CTRs (Fig. 2b, c). Instead, PCa samples showed patient-specific chromatin compartment alterations (Fig. 2b, c). We quantified the genomic regions with a different compartment assignment in each PCa sample with respect to the consensus among CTR samples. We labelled these compartment switches as "A to B" or "B to A" for the regions that in CTR samples were A or B, respectively. Unsupervised clustering based on compartment similarity identified two subgroups of cancer patients (Fig. 2d) that, considering the amount of compartment alterations with respect to CTR samples, we named LDD (Low Degree of Decompartmentalization; average 5% over the entire genome for "A to B" and 5% for "B to A" compartment switches) and HDD (High Degree of Decompartmentalization; average 14% over the entire genome "A to B" and 24% "B to A"). We also noted a consistent intra-group similarity for LDD samples with 0.78 mean Jaccard Index (JI) (variance 0.001). On the other hand, HDD samples show more heterogeneity in chromatin compartmentalization (JI 0.42 mean and 0.012 variance) (Fig. 2d).

In line with patient-specific genome remodeling, we did not find regions with compartment changes concordant across all LDD and HDD tumoral samples compared to CTR consensus. However, we found recurrent "B to A" regions, indicating higher chromatin accessibility compared to controls, concordant across all HDD patients and covering 2% of the entire genome (538 genes) (Extended Data Fig. 4a). Gene ontology analysis highlighted an enrichment of genes associated with Protein Deubiquitination and Regulation of Programmed Cell Death (such as FAS) (Extended Data Fig. 4b). Instead, the HDD-specific "A to B" compartment switch regions, which correspond to reduced chromatin accessibility compared to controls, (0.6% of the genome - 59 genes) included genes involved in PI 3-kinase activities (such as PIK3R5 and PIK3R6) and lipid catabolic processes (such as PNLIP and STS), both of which are known drivers of prostate cancer progression ^21–23.

It should be remarked that the LDD vs HDD groups stratification of PCa patients would not be immediately evident based on known clinically relevant pathophysiological features such as PSA level, Gleason Score, age at diagnosis and number of positive cores (Extended Data Fig. 4c). Furthermore, the analysis of the global transcriptome on prostate samples did not separate so clearly the HDD and LDD subtypes (Extended Data Fig. 4d). We excluded the association of HDD and LDD subtypes with cell composition and immune cell infiltration, estimated by flow-cytometry (Extended Data Fig. 4e) and RNA-seq data deconvolution (see Methods, Extended Data Fig. 4f, g). We also estimated the copy number alterations (CNA) distribution across all samples based on 4f-SAMMY-seq reads coverage (see Methods). The inferred copy-number states are notably similar for both LDD and HDD subtypes (Extended Data Fig. 4h), suggesting that structural alterations do not determine the described subtypes.

We further investigated epigenetic differences between LDD and HDD groups. We compared the histone mark enrichment peaks from normal prostate tissue with the 4f-SAMMY-seq solubility profiles of CTR, LDD and HDD samples. As expected, we found that the solubility profile of CTR samples is higher in ChIP-seq peaks for open histone marks (H3K27ac and H3K36me3) and lower in constitutive heterochromatin (H3K9me3 peaks) (Fig. 2e). The HDD group showed instead an inverted trend, thus confirming a large-scale chromatin architecture remodeling. Interestingly, the LDD subgroup, despite the limited alterations in chromatin compartmentalization (Fig. 2b, c), showed an intermediate solubility pattern between the other two groups, especially for H3K27ac and H3K9me3 domains.

We next examined the functional effects of chromatin compartment reorganization by matched RNA-seq analysis. Comparing transcription profiles between CTR (n=7) and PCa (n=10) samples uncovered 66 differentially expressed genes (DEGs): 22 up-regulated and 44 down-regulated in PCa with respect to CTR (Fig. 3a). The limited number of DEGs is likely due to the heterogeneity of expression across tumor patients' biopsies.

Then, we considered LDD and HDD subtypes separately. We identified only 30 DEGs in LDD comparison to CTR samples: 9 up- and 21 down-regulated. Instead, we found 162 up- and 410 down-regulated genes in HDD compared to CTR samples (Fig. 3a). This suggests that the higher degree of chromatin compartment remodeling is associated with a more extensive gene expression dysregulation. We found 5 DEGs up-regulated and 4 DEGs down-regulated in common among all comparisons of different PCa (Fig. 3b). Seven genes already described in the literature as prostate cancer biomarkers (HPN, BICD1, GOLM1, TARP) and two down-regulated tumor suppressor genes (SERPINB5 and GATA6)^19,24–28.

To further dissect the molecular differences between LDD and HDD subtypes, we performed an enrichment analysis for Gene Ontology (GO) annotations on the list of DEGs in HDD vs CTR. We didn't perform this analysis on LDD vs CTR samples due to the small number of DEGs. Nevertheless, in LDD DEGs, we found the upregulation of two critical genes, PDLIM5 and CAMKK2, involved in cell migration and invasion ^29,30. In the other comparison, we discovered that HDD up-regulated genes are enriched for the "negative regulation of cell motility" class. whereas down-regulated ones include GO classes related to stroma remodeling (Fig. 3c), suggesting decreased functions usually associated with tumor progression. These findings were also confirmed through GeneSet Enrichment Analysis (GSEA) ³¹ (Extended Data Fig. 5a).

We then applied a complementary approach for the functional analysis of transcriptome profiles for both LDD and HDD comparison to CTR samples. We used PROGENy ³², a curated signaling pathway compendium derived from perturbation experiments. This compendium includes weights quantifying the response of each gene to pathway perturbations, thus allowing a weighted analysis of transcriptional response to specific signaling cues (see Methods, Extended Data Fig. 5b). We found a positive enrichment for Androgen pathway activity in both LDD and HDD subgroups. Notably, we found a positive enrichment for TGFbeta activity in LDD, whereas it has negative enrichment in the HDD subgroup.

These results demonstrate that the LDD and HDD subgroups are different in their chromatin architecture and gene expression, thus representing previously undescribed PCa patient subtypes. The HDD chromatin reorganization is also reflected in more prominent chromatin remodeling and transcription repression that confer antitumoral effect.

Then, we explored the possible connections between epigenetic remodeling and functions, analyzing compartment changes and transcription at the individual patient level. By considering patient-specific DEGs and compartment switches (see methods), we observed that the majority of DEGs (90% for LDD; 70% for HDD) are located in domains not involved in a compartment transition (Fig. 3d). This suggests that, as extensively described in other models ^33–35, changes in chromatin architecture are not enough to trigger transcriptional dysfunction, but they establish conditions that enable transcriptional deregulation. A higher proportion of DEGs (30%) were found in compartment switch regions for HDD. Within these genes, we identified a trend where up-regulated genes were associated with "B to A" switch regions, while down-regulated genes were associated with "A to B" compartment switch regions (Fig. 3d, e). GO analysis revealed significant terms only for the latter class (973 downregulated DEGs in HDD “A to B” transitions), with “basal cell of prostate epithelium” emerging as the most enriched class, alongside terms related to “hypoxia” and “glycolysis” (Fig. 3f), confirming an acquired silenced state of tumorigenic genes ³⁶. Transcription factor enrichment analysis identified the Polycomb protein Suz12 (Fig. 3f), a subunit of the well-known Polycomb repressive Complex 2 (PRC2), as a key regulator of the identified genes. Interestingly, 58 among the 139 enriched Suz12 targets were described overexpressed in diverse phases of PCa progression, and 8 (SEMA5A, RET, PDGFRA, TWIST2, CDH13, NTN1, EGFR, DUOX2) were associated with positive regulation of cell motility (term of biological processes gene-set), suggesting that in HDD the acquired “heterochromatin-like” state plays a role in constraining cancer phenotypic plasticity.

To identify a signature distinguishing HDD vs LDD subtypes, we examined the differentially expressed genes in comparing HDD vs LDD transcriptomic profiles.

We identified 101 DEGs: 77 down- and 24 up-regulated genes in HDD with respect to LDD (Fig. 4a). As seen in the HDD vs CTR comparison, we found differential expression of Epithelial-Mesenchymal Transition (EMT) genes (GSEA analysis – Fig. 4b and Extended Data Fig. 5c) as well as differential activity of Androgen and TGFbeta pathways (PROGENy analysis - Extended Data Fig. 5d).

Then, we asked if our signature of 101 genes could provide prognostic information on patients. Thus, we queried the cancer genome atlas (TCGA) for the RNA-seq data of 499 prostate adenocarcinomas (PRAD) ³⁷. We defined a quantitative score named PCI (prostate Compartmentalization Index) to summarize the HDD vs LDD transcriptional signature activity for each patient in this cohort (see Methods). We sorted TCGA samples based on the PCI and defined HDD-like and LDD-like patients based on the PCI score indicating a more prevalent HDD (PCI > 0) or LDD (PCI <0) signature, respectively (Fig. 4c).

Intriguingly, survival analysis for biochemical recurrence-free status (BCR) indicated that patients belonging to the HDD-like cluster show a better prognosis (Fig. 4d), confirming the “protective” chromatin compartment alterations described above. This is a remarkable result because the signature was established using biopsies taken at the time of diagnosis, yet it can predict prognosis in more advanced, surgically removed tumor samples from the TCGA.

TCGA samples, extracted from prostatectomies are expected to have generally higher cellularity and tumor purity, with respect to needle biopsies. We noted a potential association between HDD-like and LDD-like groups and the TCGA sample’s tumor purity scores (Fig. 4c). Nevertheless, multivariate Cox regression analysis, including the tumor purity score in the model confirmed that the HDD-like classification is a valid independent prognostic predictor of biochemical recurrence (Fig. 4e). The HDD-like phenotype is associated with a better outcome (HR=0.5, CI=0.33-0.88, p-value=0.01) on top of the reference clinical features such as pathological (TNM) stage, age at diagnosis and Gleason Score.

We further refined the signature to retain only the genes with expression more significantly associated with good or bad prognosis, finally resulting in the 18 genes refined signature (Extended Data Fig. 6a, b, see Methods). As expected, the reduced 18 genes signature improves the stratification of the TCGA cohort used to refine the signature itself (p-value < 0.0001; Extended Data Fig. 6c), also independently of tumor purity levels (Extended Data Fig. 6d, e).

We then validated the clinical relevance of our refined 18 gene signature across multiple independent prostate cancer patients’ cohorts. We obtained a significant prognostic stratification trend, where HDD-like patients show a better prognosis (Fig. 5a, b, c). These independent cohorts include transcriptomic profiles obtained with RNA-seq and microarrays, thus attesting the applicability of our signature to heterogenous datasets. Overall, our results confirm the validity of chromatin-informed patient stratification and the derived gene expression signature to discriminate patients for prognosis.

Prostate cancer diagnosis, which relies on the histological examination of multiple biopsies, has revealed extensive heterogeneity and multifocality of PCa ². This procedure cannot predict the 20% of prostate cancer cases that progress to metastases, resulting in a high mortality rate. Importantly, most transcriptome-based analyses, which might capture the phenotypic plasticity, are unable to provide early insights into clinical progression. Thus, there is an urgent need for new biomarkers to refine the stratification of prostate cancer patients.

The 3D genome architecture in the cell nucleus is crucial for the epigenetic regulation of transcription, with its reorganization frequently linked to neoplastic conditions ^12,38. In this work, using 4f-SAMMY-seq and transcriptome analysis on a limited number of cells (10,000–50,000 cells), we analyzed the epigenome structure and function of PCa patients-derived biopsies to infer their plasticity (Fig. 1).

We identified two novel subtypes of PCa patients presenting different degrees of chromatin compartments and transcriptional alterations: the HDD and LDD subtypes (Fig. 2). Samples with HDD features display a marked chromatin remodeling that promotes transcriptional repression of cancer progression pathways, suggesting that these chromatin changes may counteract the tumor evolution (Fig. 3). Specifically, by intersecting patient-specific 4f-SAMMY-seq and RNA-seq datasets, we found that in HDD, down-regulated genes located in “A to B” transitions are directly involved in prostate cancer progression, with “basal cell of prostate epithelium” being the most enriched class (Fig. 3f). Since basal cells possess stemness-like properties and are prevalent in advanced, castration-resistant, and metastatic prostate cancers ³⁶, this geneset down-regulation further suggests that the epigenetic state of HDD patients may restrict the phenotypic plasticity of cancer cells. Furthermore, transcription factor enrichment analysis identified Suz12 as a key regulator of over 10% of all HDD down-regulated genes in “A to B” regions, suggesting that PRC2, a key complex in prostate cancer progression ³⁹, may mediate this transcriptional repression.

Comparing HDD and LDD subtypes, we derived a specific expression signature that characterizes the HDD phenotypic state (Fig. 4). We confirmed that the HDD signature is significantly associated with a more favorable prognosis using four independent cohorts covering more than 900 patients with follow-up clinical annotations (Figs. 4–5).

The epigenetic remodeling of epithelial tumor cells is coupled with the loss of cell identity homeostasis through multiple steps, driven by molecular changes that depend on intrinsic and extrinsic signals ⁴⁰. The tumor microenvironment plays a central role in this process, constantly adapting its stroma cell composition and transcriptional programs to favor or counteract tumor progression ^41,42. This is particularly evident for indolent cancers such as the PCa, whose slow progression favors the crosstalk between cancer and neighboring cells. Prostate stromal tissue, constituted by fibroblasts, endothelial cells, smooth muscle cells (SMCs), and immune cells, plays a crucial role in prostate tumorigenesis ^43–45. Intriguingly, transcriptional programs can have divergent trajectories in epithelial or stromal cells. This is the case of the androgen receptor (AR) pathway, whose overexpression in epithelial cells is associated with bad prognosis ⁴⁶ whereas in the stroma it has a protective role by constraining cancer growth ⁴⁴. One of the significant pathways upregulated in HDD samples is the androgen receptor and, working on bulk prostatic tissue comprising 70–80% of stromal cells, we speculate that this detected AR up-regulation may be ascribed to the stroma. Further corroborating this hypothesis, the LDD PCa group correlates with a worse prognosis and its signature includes overexpression of TGFbeta that can induce a resistance to androgen deprivation therapy when expressed in the stroma ⁴⁵.

To develop a molecular signature exploitable in the clinic, we selected genes with prognostic value, ultimately restricting our chromatin-based RNA signature to 18 genes validated across multiple independent cohorts (Fig. 5). Notably, when comparing our HDD signature with the 29 previously identified prostate cancer (PCa) signatures listed in the PCa Database (PCaDB; http://bioinfo.jialab-ucr.org/PCaDB/), only 3 out of the 18 genes were shared, namely ACAP3 e ATG16L2 ⁴⁷ and SCRIB ⁴⁸. This highlights the uniqueness and potential specificity of our gene signature for early stratification of prostate cancer patients.

In summary, this study we introduced a novel experimental approach to investigate the epigenomic landscape and 3D chromatin compartmentalization in prostate biopsies. We identified two novel patient subgroups based on their epigenome profiles. The PCa subtype associated with a favorable prognosis exhibits heterochromatin reorganization that represses tumorigenic pathways. We found that the transcriptional signature derived by this chromatin-informed patient stratification constitutes a novel independent prognostic classifier. We validated the signature across multiple independent cohorts, thus confirming its potential for translatability to assess prognosis at the time of diagnosis.

Prostate tissues cohort, ethics approval and consent to participate

Our cohort includes chemo-naïve patients followed in the Urology Division of Fondazione IRCCS Ca' Granda - Ospedale Maggiore Policlinico (Milan) who underwent the transrectal ultrasound-guided systematic sampling of prostate tissue (TRUS biopsy). This diagnostic procedure was performed as part of routine clinical management following the detection of abnormal digital rectal examination or an elevated PSA blood level. According to the current standard for the detection of PCa, 14 ultrasound-guided biopsy cores (diagnostic biopsies), seven from each side of the prostate gland, were collected from each patient for the clinical diagnosis. During the same procedure, one additional biopsy core to be used for our research project (research biopsy), was taken from a site directly adjacent to one of the diagnostic biopsies. The institutional ethics committee board of IRCCS Foundation Ca’ Granda Ospedale Maggiore Policlinico authorized this study (authorization n 1063). All specimens were obtained after the patients had provided written informed consent following the ethical principles of biomedical research on biospecimens. To ensure optimal recovery of living cells, fresh surgical prostate biopsy specimens were placed in ice-cold saline buffer and directly transported to the laboratory within 1 hour from sample collection. Tissue samples were then immediately processed to preserve the intranuclear genomic and protein architectures of cell nuclei. Samples selected for epigenome studies consist of 17 fresh biopsies divided on the bases of the histology and the spatial distribution of the positive cores in two different groups: 10 PCa biopsies from patients with histologically confirmed prostate cancer and 7 from patients who had no cancer in any biopsy core. All the clinical data were provided by the Urology division preserving the confidentiality of patient personal data.

Tissues processing

The biopsy specimen was stored at 4-8°C for maximum 3 hours before dissociation, avoiding its freezing. Given the little size of needle biopsy cores (typically 20-30 mg) we empirically determined the digestion condition to achieve the optimal recovery of living cells. Briefly, the biopsy tissue was transferred in a 2 ml microcentrifuge tube and rinse twice with 1 ml of ice-cold sterile PBS 1X. Then, the tissue was cut into small pieces (~1 mm) with autoclaved surgical scissors directly in the microcentrifuge tube. The resulting minced tissue was enzymatically digested by adding 1 ml of prewarm HBSS (Gibco, 14025) containing 200 units of collagenase type I (Life Technologies, 17018-029) per ~10 mg of tissue plus 67 µg DNase I (Sigma-Aldrich, 10104159001). Tissue dissociation was carried out in a water bath at 37 °C shaking vigorously for 10 seconds every 5 minutes. To prevent tissue over-digestion, the best incubation time for each experiment was established by monitoring cell viability, cell debris and aggregates every 10 minutes (usually ranges from 1 to 1.5 hour). After completed digestion cells were washed once by topping up to 2 ml with RPMI + 10% FBS and spun down at 300 g. The cell pellet was resuspended in RPMI + 10% FBS and dispersed by passing through a 75µm cells strainer, followed by an additional wash of the filter with RPMI + 10% FBS. Finally, the cells were spun down at 300 g and resuspended in 1 ml of ice-cold PBS for counting with a hemocytometer.

Histological and immunofluorescence evaluation

One third portion of each research biopsy tissue specimen was embedded in Killik (Bio-Optica, 05-9801), immediately frozen in precooled isopentane (MilliporeSigma, 277258) and stored at -80°. OCT- embedded biopsy cores (one per patient) were serially sectioned with 10 μm thickness in a cryostat at -20°C. Ten slides per patient, containing multiple sections representing distinct regions of the same tissue were prepared in parallel and stored at -80°C. Hematoxylin and Eosin staining (H&E) was performed using H&E Staining Kit (Abcam, ab245880). The H&E-stained slides were reviewed by an expert genitourinary pathologist to assign the Gleason Score according to the International Society of Urinary Pathology grading system. The same pathologist evaluated all specimens (diagnostic and research biopsies) presented in this work.

For immunofluorescence, frozen tissue sections were briefly thawed at room temperature (RT), placed in precooled acetone for 20 minutes at -20°C and washed 3 times in PBS 1X for 2 minutes at RT. To permeabilize tissues the sections were incubated with 0.5% Triton X-100 in PBS 1X for 10 minutes at RT with mild agitation. After three washes in PBS 1X for 2 minutes at RT, coverslips were blocked by incubating with 5% Donkey serum, 3% BSA in PBS 1X for 1 hour at RT. After three washes in PBS 1X for 2 minutes at RT, coverslips were incubated with 1:500 Lamin A/C primary antibody (Santa Cruz sc-6215) 1:500 in blocking solution overnight at 4°C in a humidified chamber, isolating the tissue section by a hydrophobic pen. The following day, sections were washed 3 times in PBS 1X for 2 minutes at RT and incubated with fluorescence-conjugated secondary antibody (Alexa FluorTM 488-conjugated Donkey Anti-goat IgG Invitrogen, A-31571), 1:500 3% BSA in PBS 1X for 1 hour at RT in a dark humidified chamber. After three washes in PBS 1X for 2 minutes at RT, the coverslips were incubated with Hoechst solution (Thermo Fisher Scientific H3570, 1:500) for 10 minutes at RT in the dark, rinsed in PBS 1X and mounted on microscope slides with ProLong antifade mounting media (Thermo Fisher Scientific, P36930).

Images were acquired using a Leica TCS SP5 confocal microscope with a HCX Plan Apo ×63/1.40 objective, with a 5-zooming factor. The NIS-Elements v.5.30 software (Nikon-Lim) was employed to analyze nuclei physical parameters including area and circularity. Hoechst staining facilitated nucleus identification and delineation of the region of interest. Over 5 field of views (FOVs) were acquired and analyzed per sample, totaling the examination of more than 70 nuclei within each sample of the cohort (seven non-neoplastic controls, CTR and ten primary prostate cancer patients, PCa). The mean nuclear area or circularity was calculated for each sample and plotted alongside other samples within the same group on the charts. The analysis of Hoechst and Lamin A/C distribution across nuclei was conducted using Fiji software. Leveraging the plot profile plugin, a 10 µm fixed line was implemented to trace and capture the fluorescence intensity distribution across each nucleus. To standardize comparisons between diverse samples, a normalization step was performed. This involved calibrating the fluorescence intensity of individual pixels along the line to the mean intensity derived from the entire set of pixel lines. For each sample the mean distribution values for both Hoechst and Lamin A/C were calculated and graphically represented alongside other samples within the same group.

Flow cytometry Analysis

To quantify the relative amounts of cell populations in each biopsy, 10,000 cells from the digestion step were stained, acquired on a BD FACSCantoII Flow Cytometer and analyzed with FlowJo software in the INGM FACS facility. To avoid unspecific binding, antibodies were incubated with PBS-BSA 1% for 30 minutes at 4°C. TO-PRO®-3 stain (Thermofisher, T3605) was used to assess cell viability. Tissue resident leukocytes were identified as CD326-/CD45+ (EpCAM-CD326 FITC, B347197; CD45 Pacific Blu Biolegend, 982306); epithelia cells as CD326+/CD45-, and stromal cells as CD326-/CD45-.

4f-SAMMY-seq isolation of chromatin fractions

Chromatin fractionation on tissue-derived single cell suspension was performed with minor adaptations to the protocol described in ¹⁶. Cells were counted, washed in cold PBS and resuspended in 600ul cold cytoskeleton triton buffer, CSK-Tr: 10 mM PIPES pH 6,8; 100 mM NaCl; 1 mM EGTA; 300 mM Sucrose; 3 mM MgCl2; 1X Protease Inhibitor Cocktail (Roche, 04693116001); 1 mM PMSF (Sigma-Aldrich, 93482); 1 mM DTT; 0,5% Triton X-100. After 10 minutes on a rotator at 4°C, samples were centrifugated for 3 minutes at 900g at 4°C and supernatant was collected as S1 fraction. Pellets were washed for 10 minutes on the rotator at 4°C with an additional volume of the same buffer followed by centrifugation for 3 minutes at 900g at 4°C. Chromatin was then digested by using 25 units of DNase I (Invitrogen, AM2222) in 100ul of cytoskeleton CSK buffer: 10 mM PIPES pH 6.8; 100 mM NaCl; 1 mM EGTA; 300 mM Sucrose; 3 mM MgCl2; 1mM PMSF; 1X Protease Inhibitor Cocktail for 60 minutes at 37°C. To stop digestion, ammonium sulphate was added to samples to a final concentration of 250 mM and, after 5 minutes on ice, samples were pelleted at 900g for 3 minutes at 4°C and the supernatant was collected as S2 fraction. After a wash of 10 minutes on a rotator at 4°C with 200ul of CSK buffer followed by centrifugation for 3 minutes at 3000g at 4°C, the pellet was further extracted 10 minutes on a rotator at 4°C in 100ul of CSK-NaCl buffer: 10 mM PIPES pH 6.8; 2 M NaCl; 1 mM EGTA; 300 mM Sucrose; 3 mM MgCl2; 1mM PMSF; 1X Protease Inhibitor Cocktail, centrifuged at 2300 g 3 minutes at 4°C and the supernatant was collected as S3 fraction. Pellets were washed twice for 10 minutes on the rotator at 4°C followed by centrifugation at 3000 g 3 minutes at 4°C with 200ul of CSK-NaCl buffer. Finally, pellets were solubilized in 100ul of 8M urea for 10 minutes at room temperature and labelled as S4 fraction. Fractions were stored at –80°C until DNA extraction.

4f-SAMMY-seq DNA extraction, library preparation and sequencing

Chromatin fractions (S2, S3 and S4) were diluted 1:2 in 1X TE buffer (10mM TrisHCl pH 8.0, 1 mM EDTA) and incubated with 3 U of RNAse cocktail (Ambion, AM2286) at 37° for 90 minutes, followed by 40μg of Proteinase K (Invitrogen, AM2548) at 55° for 150 minutes. Genomic DNA was then isolated using phenol/chloroform (Sigma-Aldrich, 77617) extraction, followed by a back extraction of phenol/chloroform with additional volume of TE1X. DNA was precipitated in 2 volumes of cold ethanol, 0.3M sodium acetate and 20ug glycogen (Ambion AM9510) for 1 hour on dry ice or overnight at -20°C. Dry pellets were resuspended in 50 ul (S2) or 15 ul (S3 and S4) of nuclease-free water and incubated at 4°C overnight. On the next day, S2 was further purified using PCR DNA Purification Kit (Qiagen, 28106) and separated using AMPure XP paramagnetic beads (Beckman Coulter, A63880) with the ratio 0.90/0.95 to obtain smaller fragments conserved as S2S (< 300 bp) and larger fragments labelled as S2L (> 300bp) fractions. Both were resuspended in 20 ul of nuclease-free water and then reduced to 15ul using a centrifugal vacuum concentrator. S2L, S3 and S4 fractions were sonicated in a Covaris M220 focused-ultrasonicator using screw cap microTUBEs (Covaris, 004078) to obtain a smear of DNA fragments peaking at 150-200 bp (water bath 20°C, peak power 30.0, duty factor 20.0, cycles/burst 50, 150 seconds for S2L and 175 seconds for S3 and S4). Fractions were quantified using Qubit 4 fluorometer with Qubit dsDNA HS Assay Kit (Invitrogen, Q32854). Libraries were generated from each sample using NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB, E7645L) and Unique Dual Index NEBNextMultiplex Oligos for Illumina (NEB, E6440S). Libraries were then qualitatively and quantitatively checked on and run on an Agilent 2100 Bioanalyzer using High Sensitivity DNA Kit (Agilent, 5067-4626). Libraries with distinct adapter indexes were then multiplexed and, after cluster generation on FlowCell, sequenced for 50 bases in single or paired-end reads mode on an IlluminaNovaSeq 6000 instrument at the IEO Genomic Unit in Milan.

RNA extraction, library preparation and sequencing

Ten thousand cells retrieved from the digestion step were stabilized in 200µl of 1Thioglycerol/Homogenization Solution of the Maxwell® RSC miRNA Tissue Kit (Promega, AS1460) and stored frozen at -80°C for total RNA automated purification using Maxwell® RSC 48 Instrument (Promega, AS8500). Total RNA was quantified by Qubit 4 fluorometer with Qubit RNA HS Assay Kit (Invitrogen, Q32852) and assessed by Agilent 2100 Bioanalyzer using Agilent RNA 6000 Pico Kit (Agilent, 5067-1513) to inspect RNA integrity. For each sample, 1 ng of total RNA was used to construct strand specific RNA-seq libraries with SMARTer Stranded Total RNA-Seq Kit - Pico Input (Takara, 634487). The yield and quality of the libraries were evaluated on Agilent 2100 Bioanalyzer using High Sensitivity DNA Kit (Agilent, 5067-4626). RNA-seq libraries were sequenced on the Illumina NextSeq™ 550 system at the sequencing facilities of Humanitas or Fondazione IRCCS Ca' Granda-Ospedale Maggiore Policlinico in Milan in paired-ends mode.

ChIP-seq public datasets

The ChIP-seq data of histone post translational modifications (histone marks) for non-neoplastic prostate gland tissue were downloaded from the following datasets available on ENCODE data portal ²⁰: ENCSR763IDK (H3K27ac), ENCSR768PFZ (H3K36me3) and ENCSR133QBG (H3K9me3). These data were downloaded as raw high-throughput sequencing reads (FASTQ) and analyzed as described in the next sections.

ChIP-seq and 4f-SAMMY-seq high-throughput sequencing reads data analyses

High-throughput sequencing reads were trimmed using Trimmomatic (v0.39) ⁵¹ with the following parameters for 4f-SAMMY-seq and ChIP-seq: 2 for seed_mismatch, 30 for palindrome_threshold, 10 for simple_threshold, 3 for leading, 3 for trailing and 4:15 for sliding window. The sequence minimum length threshold of 35 was applied to all datasets. We used the Trimmomatic “TruSeq3-SE.fa” (for single-end reads) and “TruSeq3-PE-2.fa” (for paired-end reads) as clip files. After trimming, reads were aligned using BWA (v0.7.17-r1188) ⁵² setting –k parameter as 2 and using as reference genome the UCSC hg38 genome (only canonical chromosomes were used) and the output saved in BAM file format. We uniformly used and aligned only a single read per DNA fragment for both single-end and paired-end sequencing samples. The PCR duplicates were marked with Picard (v2.22; https://github.com/broadinstitute/picard) MarkDuplicates option, then filtered using Samtools (v1.9) ⁵³. In addition, we filtered all the reads with mapping quality lower than 1. Each sequencing lane was analyzed separately and then merged at the end of the process.

Reads distribution profiles analysis

To compute reads distribution profiles (genomic tracks) for individual fractions of 4f-SAMMY-seq, we used Deeptools (v3.4.3) ⁵⁴ bamCoverage function. For these analyses the genome was binned at 50bp, the reads extended up to 250 bp and Deeptools RPKM normalization method was used. We considered a genome size of 2,701,495,761 bp (value suggested in the Deeptools manual https://deeptools.readthedocs.io/en/latest/content/feature/effectiveGenomeSize.html) and we excluded regions known to be problematic in terms of sequencing reads coverage using the blacklist from the ENCODE portal (https://www.encodeproject.org/files/ENCFF356LFX/).

To compute the genomic tracks for ChIP-seq IP over INPUT enrichment profiles (log₂ normalized ratio) we used the SPP R package (v1.16.0) ⁵⁵ and R statistical environment (v3.5.2). The reads were imported from the BAM files using the “read.bam.tags” function, then filtered using “remove.local.tag.anomalies” and finally the comparisons were performed using the function “get.smoothed.enrichment.mle” setting “tag.shift = 0” and “background.density.scaling = TRUE”. The resulting enrichment signal corresponds to a log2 normalized ratio between the pair of sequencing samples.

For computing the relative comparisons between two 4f-SAMMY-seq fractions (relative enrichment, i.e. log₂ normalized ratio) we used the same procedure described above for ChIP-seq IP over INPUT enrichment profiles. We defined the "solubility profile" as the relative enrichment ratio of 4f-SAMMY-seq sequencing reads distribution along the genome for S2L vs S3 fractions. We used S3 as baseline in this ratio as it is the second most consistent fraction among controls (Extended Data Fig. 3d).

To compute correlations between genomic tracks, we used R (v3.5.2) base function "cor" with “method = Spearman”. The genomic tracks were imported in R using the rtracklayer (v1.42.2) ⁵⁶ library. Then the original genomic track files with 50bp resolution were re-binned by averaging data at 150kb resolution using the function “tileGenome” and the correlation was computed per chromosome. The correlation values obtained for each chromosome were then summarized in one value describing the genome-wide sample correlations through a weighted mean, where the weight of each chromosome corresponds to its length.

Open histone marks (H3K4me1, H3K36me3) peaks were called with MACS (v2.2.9.1) using a broad-cutoff of 0.1 ⁵⁷.

H3K9me3 peaks were defined using the EDD (v1.1.19) software ⁵⁸ with parameters (binsize = 150 Kb and gap penalty = 25) processing the filtered bam files obtained as described above. The “required_fraction_of_informative_bins” parameter was set to 0.98. The blacklisted regions were defined as for the reads distribution profile analyses described above.

The enrichment boxplots shown in Fig. 2e was produced retrieving for each histone mark peak the genomic bins that fall in that genomic region. For each group of patients (CTR, LDD and HDD groups) and for each genomic bin we plot the mean 4f-SAMMY-seq solubility profile across the group. The values used were the quantile normalized tracks (across all samples) S2LvsS3 produced using SPP (see “High-throughput sequencing reads analyses”). Genomic tracks obtained by SPP or Deeptools were saved in bigWig file format.

Genomic tracks visualization

The visualization of genomic tracks was performed with Gviz R library (v1.26.5) ⁵⁹. The track profile was calculated using the function “DataTrack” (the input bigWig file was imported using the function “import” of the rtracklayer library) and plot using the function “plotTracks”. Extended Data Fig. 3e has been created plotting each track as type "polygon" and all the tracks have been set using a window of 500 except for ChIP-seq of H3K27ac where the window has been set to 5000. It is worth remarking that the "window" parameter in the "DataTrack" function is referring to the number of intervals in which the displayed region is divided to plot the profile. As such, a larger number indicate a finer grain definition of the profile. As H3K27ac is known to have sharp peaks, we aimed to plot a finer grain profile. For Fig. 1c the window was set to 1000 and for all the tracks except for ChIP-seq of H3K27ac where we used 5000, i.e. the same value used for the Extended Data Fig. 3e. CTR and PCa samples track overlayed with confidence interval were drawn using at the same the time the types “a” and “confint”. The same settings of histone mark tracks of Fig. 1c were used for Fig. 2b.

Chromatin compartment analysis

Chromatin compartments were calculated using a revised version of the CALDER algorithm (version 1.0 ⁶⁰), as implemented in the original 4f-SAMMY-seq pipeline for the sub-compartment identification ¹⁶. Namely, for each chromosome: i) we calculated the four 4f-SAMMY-seq fractions (S2S, S2L, S3, S4) reads distribution profiles, binned at 150kb and normalized with RPKM (see "Reads distribution profiles analysis" section); ii) for each genomic bin, defined as a vector containing the four RPKM values (one for each fraction), we calculated the Euclidean distance (dist, R stats package, method="euclidean") with all the other bins of the same chromosome. These steps produced an NxN matrix (hereinafter referred to as biochemical similarity matrix), where N is the number of bins for the considered chromosome. Starting from the biochemical similarity matrix, we performed the CALDER procedure as described in ¹⁶, to derive the eigenvector and to reconstruct chromatin compartmentalization. Specifically, we called sub-compartments but limiting their segmentation to the highest level, thus obtaining only 2 compartments corresponding to "A" and "B" compartments (Fig. 2a, b).

The consensus of compartment calls across healthy controls has been produced labeling each genomic bin (150kb) according to its more frequent compartment across samples: e.g. if a bin is labeled as A in at least 4 out of 7 CTR samples, that bin will be defined as A in the consensus of controls, and vice versa for bins labelled as B in at least 4 controls (Fig. 2c). The definition of compartment shifts was based on the concordant or discordant compartment classification for each 150kb genomic bin (Extended Data Fig. 4a). The compartment domains shared across patients as shown in Extended Data Fig. 4a were reported using the library upsetR.

The clustering of patients in Fig. 2d were based on Jaccard Indexes (JI) of compartments concordance for each pair of patients. The resulting matrix of pairwise JI was grouped by hierarchical clustering based on the Euclidean distance (among rows or columns) and complete linkage using the pheatmap function in the pheatmap R library.

Cell type deconvolution

For cell type deconvolution analysis, we fist normalized for each sample raw count to TPM (Transcript per Million) by using R 3.6.1 environment and applying the following steps. First, read counts for each gene were normalized (divided) by the length of the gene in kilobases (RPK, Reads per kilobase). Then RPK values in a sample are summed and divided by 1.000.000 to obtain the scaling factor. Finally, RPK values for each gene are divided by the scaling factor.

For the deconvolution of the 3 macro populations (immune, epithelia, stroma cells) we defined a custom signature matrix based on single cell expression data of prostate tissue from the Human Proten Atlas (https://www.proteinatlas.org/; V. 21.1) ^61–63. Following the cluster classification of Human Protein Atlas we considered as epithelia the subpopulations annotated as prostatic glandular cells, basal prostatic cells or urothelial cells. We considered as stroma the populations annotated as muscle cells, fibroblast or endothelial cells. We considered as immune cells infiltrate the populations annotated as t-cells or macrophages. Finally, we selected as representative genes for each population only those with a TPM > 10 for the specific population and a TPM < 2 in all the two remaining populations. We finally got a macropopulation signature matrix of 702 genes. We run the deconvolution analysis exploiting this signature matrix on our bulk data by using the relative mode of CIBERSORTX (V.1.0) ⁶⁴ (Extended Data Fig. 4f).

To compute the immunity score on our biopsies we used the software CIBERSORTX (V.1.0) ⁶⁴ applying the previously computed TPM matrix of our bulk RNA (mixture file) on the LM22.txt provided in the CIBERSORTX repository. The LM22 leukocyte signature is based on a matrix of 547 genes discriminating 22 human hematopoietic cell types isolated from PBMC. We enabled batch correction by using the B-mode correction option. We next run the analysis in absolute mode by applying 500 permutations. Each sample in the mixture file gets a score that reflects the absolute proportion of each cell types in LM22 mixture file. We finally computed the absolute leukocyte score: i.e. the sum for each sample of all the 22 scores of leukocytes populations. We compared the absolute leukocyte score for each biopsy across the different groups (CTR, LDD, HDD) (Extended Data Fig. 4g).

Copy Number Analysis (CNA)

For copy number analysis we used paired-end 4f-SAMMY-seq reads by treating fractions from the same sample as independent replicates. Data was processed with the pipeline nf-core/sarek v3.4.0 ^65,66 of the nf-core collection of workflows ⁶⁷, using whole genome sequencing settings, and segmentation was called with CNVkit v0.9.10 ⁶⁸.

Transcriptomic analyses

For RNA-seq data, the overall quality of the sequenced reads was assessed using FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) (V. 0.11.8), then reads were trimmed with Trimmomatic software ⁵¹ (version 0.39) removing the adapters (ILLUMINACLIP:Picov2smart-PE.fa), primer dimers, and low quality bases at the beginning and at the end of the reads (trimmomatic PE phred33 LEADING:3 TRAILING:3 SLIDINGWINDOWD:4:15 MINLEN:36). STAR ⁶⁹ (V. 2.7.0f_0328) was used to index (STAR --runMode genomeGenerate) the Human Genome (GENCODE Release 39, GRCh38 primary assembly genome ⁷⁰) and to align sequenced reads in paired-end mode (--readFiles R1.FASTQ R2.FASTQ) on the indexed reference genome. Multimapping reads and PCR duplicate were marked in the final output (--bamRemoveDuplicatesType Unique) and unaligned reads stored in a different file (--outReadsUnmapped Fastx).

The read counts on genes were calculated using as a reference a GTF file with RefSeq annotation downloaded from UCSC (http://genome.ucsc.edu) stored in the following directory:(http://hgdownload.soe.ucsc.edu/goldenPath/archive/hg38/ncbiRefSeq/109.20190905/hg38.109.20190905.ncbiRefSeq.gtf.gz). This file was further processed to remove non canonical and mitochondrial chromosomes, selected only curated genes (NM, NR) and finally split in protein coding (NM) and noncoding (NR) files. Reads count was performed with HTSeq-count (V. 0.13.5) on bam files (previously generated by STAR) using as a feature the union of all exons in a gene. The type of library was specified with “-s reverse” parameter. The reads that align to more than one position in the reference genome were discarded (htseq-count --non-unique none). The full matrix with raw read counts for each sample were loaded in R 3.6.1 and normalized using DESeq2 ⁷¹ median of ratios. Differential expression analysis was performed with DESeq2V. 1.26 ⁷²) using Wald test and the Benjamini and Hochberg correction for multiple tests, to compute p-values and adjusted p-values, respectively.

Enrichment analyses

Up-regulated genes (with adjusted p-value < 0.05 and Log₂ Fold Change > 1) and Down-regulated genes (with adjusted p-value < 0.05, and Log₂ Fold Change < -1) were used separately to query the gene-list enrichment tool Enrichr (https://maayanlab.cloud/Enrichr/) ^73–75 over the main Gene Ontology classes (Biological Process - BP, Molecular Function - MF, Cellular Component - CC) updated to 2023 taking as final enriched terms only those with Benjamini-Hochberg adjusted p-value < 0.05. Genes within the recurrent compartment switch “A to B” and “B to A” were also queried over the main Gene Ontology classes taking also in this case the enriched terms with Benjamini-Hochberg adjusted p-value < 0.05. GSEA analysis ³¹ was performed in Preranked mode. Starting from raw p-values of all genes we generated a ranked matrix sorted by the score resulting from the following formula: -Log₁₀(p-value) multiplied by the sign of the Log₂ Fold Change. For the gene set references we downloaded H (Hallmark) and C5 (Ontology) GMT files from MSigDB. Finally, we performed the analysis using the parameters "Number of permutations": 1000; "Collapse": No; "Enrichment Statistic": classic; "Max size": 500; "Min size":15. Resulting geneset with a family-wise error rate (FWER) < 0.05 are finally sorted using the Normalized Enrichment Score (NES).

We then inferred pathway activity for the comparison of our tumor subtypes by using the R package decoupleR ⁷⁶ (V. 2.4). We used the get_progeny (organism=human,top=100) function to get pathways from PROGENy (V.1.20) ³² a curated collection of signaling pathways derived from perturbation experiments, where each gene has a specific weight describing its positive or negative response to a given pathway stimulation. The top 100 responsive genes in the compendium ranked by p-value were used for each pathway. Starting from Wald statistic values previously computed for each gene with DESeq2 (LDDvsCTR, HDDvsCTR and HDDvsLDD comparisons) we run the function decouple with parameters statistics=c(“mlm”,”ulm”,”wsum”) and consensus_score=TRUE, to estimate a consensus score across the top performer methods (according to benchmark done by decoupleR developers) in pathway activity prediction. Multivariate linear model (mlm), Univariate linear model (ulm), Weighted Sum (wsum) are recommended by developers of decoupleR since these methods better estimate pathway activity considering the weights associated with pathway related genes.

Patient specific DEGs and Compartment switches

Significant Down-regulated genes have been identified using DESeq2 with the same workflow described above (adjusted p-value < 0.05 and absolute Fold Change value > 1), comparing each individual LDD and HDD patients against the group of 7 CTRs. Patients-specific compartment switches ("A to B" and "B to A") were defined with respect to the consensus of compartment calls across CTRs.

To define number of DEGs included in switch regions for each patient, patient-specific down-regulated DEGs were intersected to patient-specific A to B regions while up-regulated DEGs were intersected to patient-specific B to A regions. Then the gene lists resulting from all the patient-specific intersections of DEGs and compartment switches were merged and the reultin union gene list was used as input for functional classes enrichment analysis with Enrichr (https://maayanlab.cloud/Enrichr/).

TCGA Data

TCGA-PRAD RNA-Seq expression data were downloaded by using gdc-client tool on gdc_manifest.2022-08-08.txt metafile input downloaded from the GDC portal repository (https://portal.gdc.cancer.gov/; V.34.0; Release July, 27, 2022) by selecting from Web interface the TCGA project, prostate and RNA-seq count. The main clinical parameters (Gleason Score, T stage, N stage and age) and info about the samples (primary tumor, normal biopsy and metastatic) were selected from the same Web section of GDC portal repository by downloading clinical and sample-sheet files. ABSOLUTE tumor purity score ⁷⁷ was downloaded from https://gdc.cancer.gov/about-data/publications/pancanatlas. The file is TCGA_mastercalls.abs_tables_JSedit.fixed.txt (Downloaded October ,13, 2022) ⁷⁸. Excluding primary normal biopsies, we obtained from the TCGA repository described above the 500 FPKMUpperQuartile normalized expression counts from prostate primary tumors. The Biochemical Recurrence Free (BCR) status for each patient was obtained from PCaDB (http://bioinfo.jialab-ucr.org/PCaDB/; Release May, 10 2021; eSet V.1.3), downloading the TCGA-PRAD_eSet.RDS object and applying the pData function from Biobase package. We then excluded the only sample with a Gleason Score pattern of 2+4 never described in literature. Despite we performed expression analyses assigning a PCI score on the entire cohort of 499 samples, the BCR data available are 466.

PCI score (Prostate Compartmentalization Index)

Starting from our 101 DEGs of the HDD vs LDD comparison we defined two gene lists with 24 HDD Up-regulated and 77 HDD Down-regulated genes. We subset the TCGA matrix extracting only those genes matching with our full list of 101 genes. We performed a log transformation of TCGA gene expression matrix: Log₂(matrix +1). The log transformed expression of each gene across the 499 samples was standardized into a z-score. For each sample we computed 2 median values on the z-score transformed matrix: the median of the 24 HDD Up-regulated genes (HDD_UP) and the median of the 77 HDD Down-regulated (HDD_DOWN) genes. Finally, we computed the PCI score defined as the difference of median values: PCI=median(HDD_UP) - median(HDD_DOWN). The resulting PCI score with a positive value will indicate that a sample has HDD-like transcriptomic features, whereas a PCI negative value will classify a sample as LDD-like.

Survival Analyses

Univariate and multivariate regression analysis were performed in R using survival and survminer packages. First, we performed a log-rank test to compare survival Kaplan-Meier curves (in terms of BCR STATUS AND BCR TIME) between HDD-like and LDD-like samples. Then we performed a multivariate Cox proportional-hazard model including the following covariates: HDD-LDD status (previously assigned according to the PCI score) of the samples, pathological stage, age at diagnosis, Gleason Score and Tumor purity score. For the multivariate Cox regression analysis, the Gleason Score was included as numeric feature by summing the primary and secondary numeric grades, as reported for TCGA samples.

Signature Refinement

To refine the 101 gene signature, we applied univariate cox regression analysis on each single gene expression values across the TCGA samples, including in the model the BCR STATUS and BCR TIME. The p-values obtained for each gene are then adjusted with Benjamini-Hochberg. Then we selected only those genes whose expression is associated with Hazar Ratio (HR) < 1 or HR > 1 with a adj.p-value < 0.05. We obtained 21 candidate prognostic genes. Unsupervised clustering of these genes can cluster 2 main groups of samples with HDD-like and LDD-like features with high significance in the prognostic status of the two groups (Extended Data Fig. 6a). We further refined the signature by removing 3 genes (NPIPB3, CCNE2, FMO5) that were noisier in the unsupervised clustering heatmap: these are the only genes that cluster in a discordant manner compared to their original association to HDD or LDD phenotype in our dataset (Extended Data Fig. S6a). Finally, we used the resulting refined panel of 18 genes to recompute and validate our PCI score on other independent cohorts of transcriptomic profiles of prostate cancer patients.

Validation

Through the PCaDB (http://bioinfo.jialab-ucr.org/PCaDB/ Release May, 10 2021; eSet V.1.3) we downloaded normalized expression matrices and BCR data of three independent datasets of prostate cancer: Emory (GSE54460) ⁷⁹, Stockholm (GSE70769) ⁸⁰, Belfast (GSE116918) ⁸¹. For each of them we classified samples computing the PCI score based on the refined 18 genes signature.

Data availability

The high-throughput sequencing data generated for this study are available in the database ArrayExpress with accession number ‘E-MTAB-14226’, at the link: https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-14226?key=3d98766f-3ac6-4c07-a4a0-1022bc521bb0. The bioinformatic pipeline along with a step-by-step tutorial explaining how to process 4f-SAMMY-seq data is available on FigShare (https:// doi.org/10.6084/m9.figshare.25437121.v1) and also linked to the Github repository for updates (https://doi.org/10.6084/m9.figshare.25438195).

Authorship

VR designed and performed experiments as well as data analysis. CP and GL designed performed data analysis. FG and RQ performed immunofluorescence (IF) and AF performed the IF quantification. MM performed CNA analysis. ES and EDPS participated in data analysis. MC acquired and analysed FACS data. VV supervised sequencing experiments. EM, GA, FR and EDL provided patient samples and participated to the data interpretation. MMa performed histopathology evaluation. VR, CP and GL wrote the manuscript. FF and CL conceived the study and wrote the manuscript. All authors approved the submitted version.

Conflict of interests

A patent application is being filed for the signature presented in this manuscript by VR, CP, GL, FF and CL.

Acknowledgements

We thank Maria Vivo, Beatrice Bodega, Marina Lusic, Mattia Forcato, Gioacchino Natoli, Martin Schaefer and all members of our laboratories for critical feedback on earlier versions of the manuscript. We thank Claudio Tripodo (IFOM and University of Palermo) and Giorgia Zadra (CNR) for feedback on early phases of the project. For the sequencing we acknowledge support from the Genomics Unit of the European Institute of Oncology (IEO), Department of Experimental Oncology, Humanitas Sequencing facility and Istituto Nazionale Genetica Molecolare (INGM) (Marco Ghilotti). We thank Ilaria Rancati (IFOM cell culture facility) for support with Maxwell instrument. This work was supported by Interomics Flagship Project (CNR) and MFAG (grant #18535) to CL; AIRC Start-Up (grant #16841) to FF; FRRB INTERSTRAT-CAD (grant #CP2_14/2018) to FF; AIRC fellowships n. 26942 to FG and No. 22351 to ES; PIR01_00011 “IBISCo”, PON 2014-2020 to MM. Schematic representations of Figs 1a, b and Extended Data Fig. 3a have been created using BioRender.com.

The Global Cancer Observatory (GCO) https://gco.iarc.fr/en
Mottet N et al (2017) EAU-ESTRO-SIOG Guidelines on Prostate Cancer. Part 1: Screening, Diagnosis, and Local Treatment with Curative Intent. Eur Urol 71:618–629
Hamdy FC et al (2023) Fifteen-Year Outcomes after Monitoring, Surgery, or Radiotherapy for Prostate Cancer. N Engl J Med 388:1547–1558
Loeb S et al (2014) Overdiagnosis and overtreatment of prostate cancer. Eur Urol 65:1046–1055
Trifiletti DM, Sturz VN, Showalter TN, Lobo JM (2017) Towards decision-making using individualized risk estimates for personalized medicine: A systematic review of genomic classifiers of solid tumors. PLoS ONE 12:e0176388
Matulay JT, Wenske S (2018) Genetic signatures on prostate biopsy: clinical implications. Translational Cancer Research; Vol 7, Supplement 6 (July 30, 2018): Translational Cancer Research (Prostate Cancer: Current Understanding and Future Directions)
Flavahan WA, Gaskell E, Bernstein BE (2017) Epigenetic plasticity and the hallmarks of cancer. Science vol. 357 Preprint at https://doi.org/10.1126/science.aal2380
Parreno V et al (2024) Transient loss of Polycomb components induces an epigenetic cancer fate. Nature 629:688–696
Krijger PHL, De Laat W (2016) Regulation of disease-associated gene expression in the 3D genome. Nature Reviews Molecular Cell Biology vol. 17 771–782 Preprint at https://doi.org/10.1038/nrm.2016.138
Willemin A, Szabó D, Pombo A (2024) Epigenetic regulatory layers in the 3D nucleus. Molecular Cell vol. 84 415–428 Preprint at https://doi.org/10.1016/j.molcel.2023.12.032
Fischer AH et al (2010) The cytologic criteria of malignancy. J Cell Biochem 110:795–811
Zhao SG et al (2024) Integrated analyses highlight interactions between the three-dimensional genome and DNA, RNA and epigenomic alterations in metastatic prostate cancer. Nat Genet. 10.1038/s41588-024-01826-3
Klein EA et al (2014) A 17-gene assay to predict prostate cancer aggressiveness in the context of gleason grade heterogeneity, tumor multifocality, and biopsy undersampling. Eur Urol 66:550–560
Erho N et al (2013) Discovery and Validation of a Prostate Cancer Genomic Classifier that Predicts Early Metastasis Following Radical Prostatectomy. PLoS ONE 8
Cuzick J et al (2011) Prognostic value of an RNA expression signature derived from cell cycle proliferation genes in patients with prostate cancer: a retrospective study. Lancet Oncol 12:245–255
Lucini F et al (2024) Biochemical properties of chromatin domains define genome compartmentalization. Nucleic Acids Res. 10.1093/nar/gkae454
Dhanasekaran SM et al (2001) Delineation of prognostic biomarkers in prostate cancer. Nature 412:822–826
Hessels D, Schalken JA (2009) The use of PCA3 in the diagnosis of prostate cancer. Nat Rev Urol 6:255–261
Varambally S et al (2008) Golgi protein GOLM1 is a tissue and urine biomarker of prostate cancer. Neoplasia 10:1285–1294
Consortium EP (2012) An integrated encyclopedia of DNA elements in the human genome. Nature 489:57–74
Sarker D, Reid AHM, Yap TA, de Bono JS (2009) Targeting the PI3K/AKT pathway for the treatment of prostate cancer. Clin Cancer Res 15:4799–4805
Scaglia N, Frontini-López YR, Zadra G (2021) Prostate Cancer Progression: as a Matter of Fats. Front Oncol 11:719865
Ahmad F, Cherukuri MK, Choyke PL (2021) Metabolic reprogramming in prostate cancer. Br J Cancer 125:1185–1196
Wolfgang CD, Essand M, Lee B, Pastan IT -Cell Receptor Chain Alternate Reading Frame Protein (TARP) Expression in Prostate Cancer Cells Leads to an Increased Growth Rate and Induction of Caveolins and Amphiregulin. http://nciarray.nci.nih.gov/
Cocchiola R et al (2019) The induction of Maspin expression by a glucosamine-derivative has an antiproliferative activity in prostate cancer cell lines. Chem Biol Interact 300:63–72
Sun Z, Yan B (2020) Multiple roles and regulatory mechanisms of the transcription factor GATA6 in human cancers. Clinical Genetics vol. 97 64–72 Preprint at https://doi.org/10.1111/cge.13630
Liu S, Wang W, Zhao Y, Liang K, Huang Y (2020) Identification of Potential Key Genes for Pathogenesis and Prognosis in Prostate Cancer by Integrated Analysis of Gene Expression Profiles and the Cancer Genome Atlas. Front Oncol 10
Kelly KA et al (2008) Detection of early prostate cancer using a hepsin-targeted imaging agent. Cancer Res 68:2286–2291
Piao S et al (2022) High Expression of PDLIM2 Predicts a Poor Prognosis in Prostate Cancer and Is Correlated with Epithelial-Mesenchymal Transition and Immune Cell Infiltration. J Immunol Res 2922832 (2022)
Pulliam TL et al (2022) Regulation and role of CAMKK2 in prostate cancer. Nat Rev Urol 19:367–380
Subramanian A et al (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102:15545–15550
Schubert M et al (2018) Perturbation-response genes reveal signaling footprints in cancer gene expression. Nat Commun 9:20
Ghavi-Helm Y et al (2019) Highly rearranged chromosomes reveal uncoupling between genome topology and gene expression. Nat Genet 51:1272–1282
Sebestyén E et al (2020) SAMMY-seq reveals early alteration of heterochromatin and deregulation of bivalent genes in Hutchinson-Gilford Progeria Syndrome. Nat Commun 11:6274
Hug CB, Grimaldi AG, Kruse K, Vaquerizas JM (2017) Chromatin Architecture Emerges during Zygotic Genome Activation Independent of Transcription. Cell 169, 216–228 e19
Zhang D et al (2016) Stem cell and neurogenic gene-expression profiles link prostate basal cells to aggressive prostate cancer. Nat Commun 7
The Molecular Taxonomy (2015) of Primary Prostate Cancer. Cell 163:1011–1025
Johnstone SE et al (2020) Large-Scale Topological Changes Restrain Malignant Progression in Colorectal Cancer. Cell. 10.1016/j.cell.2020.07.030
Venkadakrishnan VB et al Lineage-specific canonical and non-canonical activity of EZH2 in advanced prostate cancer subtypes. 10.1038/s41467-024-51156-5
Bracken CP, Goodall GJ (2022) The many regulators of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol 23:89–90
de Visser KE, Joyce JA (2023) The evolving tumor microenvironment: From cancer initiation to metastatic outgrowth. Cancer Cell vol. 41 374–403 Preprint at https://doi.org/10.1016/j.ccell.2023.02.016
Risom T et al (2022) Transition to invasive breast cancer is associated with progressive changes in the structure and composition of tumor stroma. Cell 185:299–310e18
Pakula H et al (2024) Distinct mesenchymal cell states mediate prostate cancer progression. Nat Commun 15
Liu Y et al (2022) Stromal AR inhibits prostate tumor progression by restraining secretory luminal epithelial cells. Cell Rep 39:110848
Wang H et al (2023) Antiandrogen treatment induces stromal cell reprogramming to promote castration resistance in prostate cancer. Cancer Cell 41:1345–1362e9
Rodriguez-Bravo V et al (2017) The role of GATA2 in lethal prostate cancer aggressiveness. Nat Rev Urol 14:38–48
Li R et al (2021) Extended application of genomic selection to screen multiomics data for prognostic signatures of prostate cancer. Brief Bioinform 22
Ramos-Montoya A et al (2014) HES6 drives a critical AR transcriptional programme to induce castration-resistant prostate cancer through activation of an E2F1-mediated cell cycle network. EMBO Mol Med 6:651–661
Purysko AS, Rosenkrantz AB, Turkbey IB, Macura KJ (2020) Radiographics update: PI-RADS version 2.1—a pictorial update. Radiographics 40:E33–E37
D’Amico AV et al (1998) Biochemical Outcome After Radical Prostatectomy, External Beam Radiation Therapy, or Interstitial Radiation Therapy for Clinically Localized Prostate Cancer. JAMA 280:969–974
Bolger AM, Lohse M, Usadel B (2014) Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30:2114–2120
Li H, Durbin R (2009) Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25:1754–1760
Li H et al (2009) The Sequence Alignment/Map format and SAMtools. Bioinformatics 25:2078–2079
Ramirez F et al (2016) deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res 44:W160–W165
Kharchenko PV, Tolstorukov MY, Park PJ (2008) Design and analysis of ChIP-seq experiments for DNA-binding proteins. Nat Biotechnol 26:1351–1359
Lawrence M, Gentleman R, Carey V (2009) rtracklayer: an R package for interfacing with genome browsers. Bioinformatics 25:1841–1842
Zhang Y et al (2008) Model-based analysis of ChIP-Seq (MACS). Genome Biol 9:R137
Lund E, Oldenburg AR, Collas P (2014) Enriched domain detector: a program for detection of wide genomic enrichment domains robust against local variations. Nucleic Acids Res 42:e92
Hahne F, Ivanek R (2016) Visualizing Genomic Data Using Gviz and Bioconductor. Methods Mol Biol 1418:335–351
Liu Y et al (2021) Systematic inference and comparison of multi-scale chromatin sub-compartments connects spatial organization to cell phenotypes. Nat Commun 12:2439
Uhlén M et al (2015) Proteomics. Tissue-based map of the human proteome. Science 347:1260419
Sjöstedt E et al (2020) An atlas of the protein-coding genes in the human, pig, and mouse brain. Science 367
Karlsson M et al (2021) A single-cell type transcriptomics map of human tissues. Sci Adv 7
Newman AM et al (2019) Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol 37:773–782
Hanssen F et al Scalable and efficient DNA sequencing analysis on different compute infrastructures aiding variant discovery. Tomtebodavägen 23, 75080
Garcia M et al (2020) A portable workflow for whole-genome sequencing analysis of germline and somatic variants. F1000Res 9:63Sarek
Ewels PA et al (2020) The nf-core framework for community-curated bioinformatics pipelines. Nature biotechnology vol. 38 276–278 Preprint at https://doi.org/10.1038/s41587-020-0439-x
Talevich E, Shain AH, Botton T, Bastian BC (2016) CNVkit: Genome-Wide Copy Number Detection and Visualization from Targeted DNA Sequencing. PLoS Comput Biol 12:e1004873
Dobin A et al (2013) STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29:15–21
Schneider VA et al (2017) Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates the enduring quality of the reference assembly. Genome Res 27:849–864
Anders S, Huber W (2010) Differential expression analysis for sequence count data. Genome Biol 11:R106
Love MI, Huber W, Anders S (2014) Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15:550
Xie Z et al (2021) Gene Set Knowledge Discovery with Enrichr. Curr Protoc 1:e90
Chen EY et al (2013) Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics 14:128
Kuleshov MV et al (2016) Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 44:W90–W97
Badia-I-Mompel P et al (2022) decoupleR: ensemble of computational methods to infer biological activities from omics data. Bioinf Adv 2:vbac016
Carter SL et al (2012) Absolute quantification of somatic DNA alterations in human cancer. Nat Biotechnol 30:413–421
Taylor AM et al (2018) Genomic and Functional Approaches to Understanding Cancer Aneuploidy. Cancer Cell 33:676–689e3
Long Q et al (2014) Global transcriptome analysis of formalin-fixed prostate cancer specimens identifies biomarkers of disease recurrence. Cancer Res 74:3228–3237
Ross-Adams H et al (2015) Integration of copy number and transcriptomics provides risk stratification in prostate cancer: A discovery and validation cohort study. EBioMedicine 2:1133–1144
Jain S et al (2018) Validation of a Metastatic Assay using biopsies to improve risk stratification in patients with prostate cancer treated with radical radiation therapy. Ann Oncol 29:215–222

Table 1. Clinicopathological characteristics of the study participants undergoing prostate biopsy

Patient	Histology (GS-P)	Age at time of biopsy (years)	PSA level (ng/ml)	Positive cores (%)	Clinical stage	PI-RADS ⁴⁹	D’Amico risk classification ⁵⁰
CTR 28	Neg	62	3.7	-	-	-	-
CTR 31	Neg	65	7.18	-	-	-	-
CTR 32	Neg	67	3.73	-	-	-	-
CTR 91	Neg	67	8.15	-	-	-	-
CTR 92	Neg	78	4.6	-	-	-	-
CTR 94	Neg	72	8.9	-	-	-	-
CTR 98	Neg	72	20.9	-	-	-	-
PCa 93	3+4	73	8.9	14.2	T1c	2	1 (Intermediate risk)
PCa 73	3+4	65	19	21.4	T1c	5	2 (Intermediate risk)
PCa 75	4+3	82	13	21.4	T1c	2	2 (Intermediate risk)
PCa 87	4+3	72	6.5	28.5	T1c	2	1 (Intermediate risk)
Pca 35	4+4	71	4.88	21.4	T1c	4	4 (High risk)
PCa 2	4+5	89	51	42.8	T2b	2	9 (High risk)
PCa 100	4+5	87	29	21.4	T2a	5	8 (High risk)
PCa 4	4+5	74	487	42.8	T3	4	12 (High risk)
PCa 88	4+5	72	186	42.8	T3	2	12 (High risk)
PCa 33	5+5	78	4	21.4	T1c	5	8 (High risk)

GS-P: Gleason Score for patients’ diagnosis

PSA: Prostate Specific Antigen

PI-RADS: Prostate Imaging Reporting and Data System

Yes there is potential Competing Interest. The authors declare that the PCa signature described in this study is currently under patenting, thus constituting a potential competing interest.

Extendeddata.docx

Chromatin remodeling restraints oncogenic functions in prostate cancer

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1