ERG Orchestrates a Dedifferentiation–Senescence–Inflammation Triad in Prostate Cancer

Abstract

ERG fusions occur in about 50% of prostate cancers and represent early somatic events that co-initiate tumorigenesis. Yet, the molecular programs unleashed by ERG at disease onset remain elusive. In this study, the inducible, temporally controlled expression of ERG enabled the accurate mapping of early ERG-driven processes associated with an apparent decrease in cell fitness across a panel of prostate cells. We observed that ERG triggers senescence and epithelial–mesenchymal transition (EMT), generating distinct cellular states in which EMT and senescence coexist in the same cell or occur independently. ERG-mediated activation of canonical and noncanonical TGFβ signaling differentially regulates EMT and senescence markers. Moreover, ERG drives context-dependent changes in cell identity and promotes an intermediate basal–luminal cell population with stem cell gene expression. Notably, p53 loss enhances ERG-driven expression of cell plasticity markers without affecting senescence. Transcriptomic investigation of ERG-expressing epithelial cells revealed activation of inflammatory signals. Conditioned medium experiments demonstrated paracrine-mediated inflammation in epithelial ERG-negative, stromal, and myeloid cells. Consistently, patient-derived expression data from five independent cohorts and 2,048 men showed activation of TNFA–NFKB and TGFβ signaling in ERG-rearranged tumors. Furthermore, deconvolution analysis indicated a higher infiltration of inflammatory macrophages in ERG-positive tumors. Altogether, we uncovered a previously unrecognized triad of ERG-driven processes—dedifferentiation, senescence, and inflammation—that may underpin its oncogenic potential and shape prostate cancer initiation and therapeutic response.

Implications:

Temporal control of ERG expression in prostate cells enabled accurate mapping of ERG-driven processes, identifying dedifferentiation, senescence, and inflammation as candidate contributors to ERG-dependent tumorigenesis and therapeutic response.

Graphical Abstract

Introduction

Genomic rearrangements involving the ETS family transcription factors are the most common aberrations in prostate cancer (1). In both localized prostate cancer and castration-resistant prostate cancer (CRPC)—an advanced and lethal stage of the disease—fusions of the regulatory region of an androgen receptor (AR)-regulated gene (mainly TMPRSS2) to the ETS member ERG (hereafter defined as T2ERG) drive the aberrant expression of the ERG protein in 40% to 50% of cases (2, 3). ERG fusions are detected in premalignant prostatic intraepithelial neoplasia (PIN) and are considered an early event in prostate cancer tumorigenesis (4–6). Accordingly, in vivo work revealed a causal role of ERG in initiating prostate cancer when combined with other mutations (e.g., PTEN loss or AKT activation) or when expressed at high levels (7–12). Furthermore, studies based on T2ERG-positive CRPC models highlighted the relevance of ERG also in this setting (13–15).

Although the oncogenic potential of ERG is clear (7, 9, 10, 12, 14), how ERG promotes prostate cancer initiation and progression remains elusive, limiting ERG-associated translational potential. Primary prostate cancer is often characterized by a loss of cell plasticity, with expansion of the luminal cell compartment and consequent loss of basal epithelial cells (16). Accordingly, several studies have reported a role of ERG in favoring the AR signaling and the prostate cancer cell luminal lineage (9, 16–22). For instance, ERG restores the expression of the AR transcriptional output in PTEN-deleted models and represses TP63 expression, a known regulator of the basal prostate lineage, through inhibition of a distal enhancer (9, 20). Moreover, ERG overexpression and ERG pathway activation characterize prostate cancer patient samples with luminal features (21, 22).

In contrast with these findings, functional studies showed that ERG disrupts lineage-specific differentiation in prostate cancer by inducing epithelial–mesenchymal transition (EMT; refs. 23–25) and blocks luminal and neuroendocrine (NE) cell differentiation in in vitro and in vivo models (26). Furthermore, ERG represses AR and its signaling by promoting PRMT5-dependent methylation of AR and by inducing a repressive epigenetic program mediated by the H3K27 methyltransferase EZH2 and several histone deacetylases (13, 27).

Given these conflicting observations on the oncogenic activity of ERG, in this study we used an inducible, temporally controlled system in multiple prostate cell lines to compare the transcriptional and biological effects of T2ERG expression. We identified induction of dedifferentiation, senescence, and inflammation as key elements in the ERG oncogenic toolbox uncovering novel features induced by ERG to shape prostate cancer tumorigenesis and response to treatment.

Materials and Methods

Cell culture, transfection, and transduction

22Rv1 (RRID: CVCL_1045), LNCaP-FGC (RRID: CVCL_1379), and THP-1 (RRID: CVCL_0006) were grown in RPMI 1640 medium (Gibco) supplemented with 10% fetal bovine serum (FBS; Gibco), 1% L- glutamine (Gibco), and 1% penicillin–streptomycin (Gibco). DU145 (RRID: CVCL_0105) and HEK293T (RRID: CVCL_0063) were grown in DMEM medium (Gibco) supplemented with 10% FBS, 1% L-glutamine, and 1% penicillin–streptomycin. RWPE-1 (RRID: CVCL_3791) were grown in keratinocyte serum-free medium (Gibco) supplemented with 5 ng/mL human recombinant epidermal growth factor (Gibco), 0.05 mg/mL bovine pituitary extract (Gibco), and 1% penicillin–streptomycin. Cell lines were purchased from the American Type Culture Collection (ATCC), the European Collection of Authenticated Cell Cultures, and/or authenticated using the ATCC short tandem repeat profile as a reference. All cell lines were routinely tested and negative for Mycoplasma.

Human primary prostate fibroblasts were purchased from Cell Biologics (H-6223) and grown in complete fibroblast medium (M2267, Cell Biologics).

For doxycycline (DOX)-inducible overexpression, the gene of interest was cloned into pCW57.1 (#41393, Addgene, RRID: Addgene_41393). For ERG knockdown, short hairpin RNAs (shRNA; Supplementary Table S8) were cloned into pLKO.1-TRC (#10878, Addgene, RRID: Addgene_10878). For PTEN silencing, shRNAs (Supplementary Table S8) were cloned into Tet-pLKO-puro (#21915, Addgene, RRID: Addgene_21915). For knockout (KO) of TP53, CDKN2B, and CDKN1C, single-guide RNAs (sgRNA; Supplementary Table S8; ref. 28) were cloned into optimized lentiCRISPR-V1 (29) or lentiCRISPR-V2 (#52961, Addgene, RRID: Addgene_52961). Genome editing efficiency was evaluated using Tracking of Indels by DEcomposition (30).

For lentiviral vector production, HEK293T cells were transfected using polyethylenimine (408727 Sigma-Aldrich) with 10 μg of plasmid encoding the gene, sgRNA, or shRNA of interest, 6.5 μg psPAX2 (#12260, Addgene, RRID: Addgene_12260) and 3.5 μg pCMV-VSV-G (#8454, Addgene, RRID: Addgene_8454). The medium containing the lentiviral vectors was collected at 48 and 72 hours after transfection, filtered through 0.45-μm PES filters, and mixed. For transduction, 22Rv1, LNCaP, and DU145 were incubated for 24 hours with the lentiviral supernatants in the presence of 8 μg/mL polybrene (sc-134220, Santa Cruz Biotechnology). RWPE-1 were spinoculated for 30 minutes at 1,000 g at 37°C with the lentiviral supernatants containing 8 μg/mL polybrene. Cells were selected by using 2 μg/mL puromycin (ant-pr-1, InvivoGen), 0.5 μg/μL hygromycin (ant-hg-2, InvivoGen), or 0.5 μg/μL Geneticin (G418 Sulfate, 11811031, Gibco) depending on the lentiviral vector used.

For transgene or shRNA induction with 1 μg/mL of DOX (14422, Cayman Chemical) and for treatment with the pharmacologic inhibitors, the culture medium was replaced every 2 days.

For conditioned medium (CM) experiments, cells were induced in full medium containing ethanol (EtOH) or DOX for 4 days (medium changed at 2 days). The supernatant was collected, filtered with 0.22-μm filters, and administered to recipient cells, mixing two-third of CM with one-third full medium and normalizing for the number of cells in the dish at the moment of collection. CM was stored at −80°C if not used immediately. RWPE-1 and fibroblasts were treated with CM for 4 days (medium changed at 2 days) before assessing senescence-associated β-galactosidase (SA-B-GAL), vimentin (VIM), and gene expression changes. THP-1 were differentiated into M0 macrophages with 20 ng/mL phorbol 12-myristate 13-acetate (10515491, Thermo Fisher Scientific) for 2 days and then treated with CM for 2 days.

Crystal violet and proliferation assays

Cells were seeded at low density and treated as indicated. For LNCaP, a poly-L-lysine (33225.01, Serva) solution was used to coat the well surface. For crystal violet assay, cells were fixed with 4% formaldehyde solution (252549, Sigma-Aldrich) for 10 minutes and stained with 0.1% crystal violet solution (in 20% EtOH) for 30 minutes at room temperature. For proliferation assay, growth curves were obtained by imaging cells using Incucyte S3 (Sartorius) and assessing confluence with the Incucyte Basic Analyzer package.

Soft agar assay

A solution of 0.7% agarose (CSL-AG500, Cleaver Scientific) in complete growth medium was poured and let solidify in each well. A total of 15,000 cells (six-well format) were mixed in complete growth medium with 0.35% agar and the specific treatments and then placed over the lower 0.7% agar layer. Complete growth medium with specific treatments was added on top. Cells were grown for 28 days, changing the medium twice a week. At the end of the treatment, cells were stained with a 0.1% crystal violet (in 20% EtOH) solution for 30 minutes and washed with 1X PBS (Gibco). Four of five images per well were taken and analyzed with the ImageJ (RRID: SCR_003070) software.

SA-Β-GAL staining

The SA-Β-GAL staining kit (9860, Cell Signaling Technology) was used to identify SA-B-GAL–positive cells following the manufacturer’s instructions. Cells were imaged using the Leica DM IL LED Microscope (Leica Microsystems), and four or five pictures per well for each condition were used to quantify the percentage of SA-B-GAL–positive cells with ImageJ (RRID: SCR_003070).

Flow cytometry analyses

5-Ethynyl-2′-deoxyuridine (EdU) and propidium iodide (PI) fluorescence-activated cell sorting (FACS) was performed using the Click-iT Plus EdU Flow Cytometry Assay Kit (C10635, Thermo Fisher Scientific) with the following modifications. Cells were labeled with 10 μmol/L EdU for 2 hours and then harvested by trypsinization. Afterward, cells were pelleted by centrifugation at 400 g for 5 minutes and washed with 1× PBS. Cells were fixed in ice-cold EtOH and kept at −20°C overnight. The Click-iT reaction was carried out as indicated in the manufacturer’s instructions. Finally, cells were resuspended in 1X PBS containing RNase A and PI, incubated for 30 minutes at 37°C, and analyzed.

For Annexin V/PI FACS, cells were harvested using Accutase (A6964, Sigma-Aldrich) and washed twice with 1× PBS. Cells (2 × 10⁵) were incubated with 1× Annexin V binding buffer (BD Pharmingen) and 5 μL APC–Annexin V (550474, BD Pharmingen, RRID: AB_2868885) for 15 minutes at room temperature in the dark. Afterward, 400 μL 1X Annexin V binding buffer and 5 μL PI were added, and samples were analyzed.

For antibody-based FACS, cells were harvested by trypsinization, resuspended in 1× PBS, fixed in ice-cold EtOH, and kept at −20°C overnight. Cells were washed with 1% bovine serum albumin (BSA)–PBS and incubated with primary antibodies in 1% BSA–PBS for 1 hour at room temperature. After washing, secondary antibodies in 1% BSA–PBS were added to the cells for 30 minutes at room temperature in the dark. Antibodies used for FACS are listed in Supplementary Table S8.

For SA-B-GAL/VIM FACS, cells were harvested by trypsinization, washed once with 1× PBS, and fixed in 4% paraformaldehyde (PFA, 043368.9M, Thermo Fisher Scientific)–PBS for 10 minutes. Cells were washed with 1% BSA–PBS and incubated in 100 μL B-GAL working solution (CellEvent Senescence Green Flow Cytometry Assay Kit, C10840, Thermo Fisher Scientific) for 2 hours at 37°C. After washing cells were permeabilized with 0.25% TritonX-100 in 1% BSA–PBS for 30 minutes at room temperature and stained as previously.

All FACS samples were analyzed on a FACSymphony A1 Cell Analyzer (BD Biosciences) and quantified using the FlowJo Software (version 10.9.0, BD Biosciences, RRID: SCR_008520).

Immunoblotting

Cells were lysed in RIPA buffer (50 mmol/L HEPES pH 7.9, 140 mmol/L NaCl, 1 mmol/L EDTA, 1% Triton-X-100, 0.1% sodium deoxycholate, and 0.1% SDS) with protease and phosphatase inhibitor cocktails (535140, P5726, and P0044, Merck). Protein concentration was determined by using the Pierce BCA Protein Assay Kit (23228, Thermo Fisher Scientific), and the same amount of total proteins for each sample was loaded and run on Bolt Bis-Tris Plus Gels (Invitrogen) or Mini-PROTEAN TGX Stain-Free Protein Gels (Bio-Rad) and transferred to Amersham Hybond polyvinylidene difluoride Western blotting membranes (Cytiva). Membranes were blocked with 5% milk or BSA–TBS supplemented with 0.1% Tween buffer. Primary and secondary antibodies used for immunoblotting are listed in Supplementary Table S8.

Amersham ECL Prime or Select reagents (Cytiva) were used for detection on the UVITec Alliance LD2 system.

Immunohistochemistry and immunofluorescence

Immunohistochemistry for ERG and macrophage markers (CD68 and CD163) in PIN and adenocarcinoma human samples was performed as previously described (31). Antibodies are listed in Supplementary Table S8. Samples were processed in accordance with institutional guidelines at Santa Chiara Hospital under ethics protocol approval #1946.

For immunofluorescence, cells were seeded in PhenoPlate 96-well microplates (6055302, Revvity) and treated as indicated. Cells were fixed with 4% PFA for 15 minutes at room temperature and treated with blocking buffer (3% BSA–PBS supplemented with 0.3% Triton X-100) for 40 minutes after washing with 1× PBS. Primary antibodies were incubated for 1 hour at room temperature in 1% BSA–PBS supplemented with 0.3% Triton X-100. After washing, secondary antibodies in blocking buffer were added to cells for 1 hour at room temperature in the dark. Cells were washed with 1× PBS and incubated with 10 μg/mL Hoechst for 10 minutes. Image collection was performed using the ImageXpress Micro Confocal High-Content Imaging System (Molecular Devices).

RT-qPCR and quantitative chromatin immunoprecipitation

Total cellular RNA was extracted using the RNeasy mini kit (74106, Qiagen) including on-column DNase I digestion (79256, Qiagen). Reverse transcription was carried out using the RevertAid First Strand cDNA Synthesis Kit (K1622, Thermo Fisher Scientific), and cDNA was analyzed with the primers listed in Supplementary Table S8 and the KAPA SYBR FAST UNI (KK4602, Roche) on the CFX96 Touch Real-Time PCR Detection System (Bio-Rad). The results were normalized to GAPDH expression.

For chromatin immunoprecipitation (ChIP), cells were crosslinked with 1% formaldehyde for 10 minutes at room temperature. After extraction with cell lysis buffer (5 mmol/L PIPES pH 8.0, 85 mmol/L KCl, and 1% Igepal), nuclei were resuspended in lysis buffer (50 mmol/L Tris-Cl pH 8.1, 10 mmol/L EDTA, and 1% SDS). The lysates were sonicated to reach an average DNA fragment size of 100 to 300 bp. Chromatin and antibodies were incubated overnight at 4°C. Dynabeads (Thermo Fisher Scientific) were used to recover the chromatin–antibody complexes. After washing, chromatin was eluted, decrosslinked, and treated with RNase A and proteinase K. The DNA was purified using NucleoSpin Gel and PCR Clean-up Kit (740609250, MACHEREY-NAGEL) and analyzed by qPCR using KAPA SYBR FAST UNI (KK4602, Roche) and the CFX96 Touch Real-Time PCR Detection System (Bio-Rad).

RNA sequencing

Total cellular RNA was extracted using the RNeasy mini kit (74106, Qiagen) including on-column DNase I digestion (79256, Qiagen) and quantified using the Qubit RNA Broad-Range Assay Kit (Q10210, Thermo Fisher Scientific). Poly-A–containing mRNA isolation and library preparation were performed by using the TruSeq Stranded mRNA Library Prep (20020594, Illumina) and the AMPure XP SPRI Reagent (A63881, Beckman Coulter). Qubit dsDNA High-Sensitivity Assay Kit (Q32854, Thermo Fisher Scientific) was used for library quantification. Libraries were sequenced with the HiSeq sequencing system (Illumina) using the HiSeq Rapid SR Kit v2 (100 cycles, Illumina).

Reads were trimmed based on quality using trimmomatic (RRID: SCR_011848; ref. 32) and aligned against the human reference genome hg38 using STAR (version 2.7.2b, RRID: SCR_004463; ref. 33). Differential gene expression analysis was performed using the edgeR package (RRID: SCR_012802; ref. 34).

Gene set enrichment analysis (GSEA; refs. 35, 36) was performed using the GSEAPreranked tool with the C2 and Hallmark collections from the MSigDB (http://www.broadinstitute.org/gsea/msigdb). Selected prostate-specific gene sets were adopted from previous studies (37–41) and inserted in the Hallmark collection for preranked GSEA analysis.

Patient-derived datasets and transcriptomic analysis

Prostate cancer single-cell RNA sequencing (scRNA-seq) data from the Song [six ERG-positive (ERG+) samples; ref. 42] and the Apostolov (24 tumor samples; ref. 43) cohorts were downloaded as .rds and .h5ad files, respectively. Malignant epithelial cells were retained for downstream analyses.

To determine ERG status in the Apostolov cohort, clustering analysis was performed with the Louvain algorithm in the Seurat R package. The optimal resolution (0.1–1) was selected as the value maximizing the median silhouette score computed on PCA embeddings. ERG+ cells were defined as having both an elevated ERG-fusion signature score and ERG gene expression. A reference cluster, defined as the one with the lowest median ERG signature, was used. Clusters were considered ERG+ enriched only if both tests were significant (Wilcoxon rank-sum test with Benjamini–Hochberg correction, FDR < 0.05) and were further subclustered using the same criteria. Patient-level ERG positivity was then assessed using a one-sided binomial test, and patients with FDR < 0.05 were classified as ERG+. A background rate (p), representing the expected fraction of ERG+ cells under the null hypothesis (i.e., ERG+ cells by chance), was set to 1%. To confirm whether patients’ ERG+ cell percentages represent a true biological signal rather than background noise, the test calculates the probability of observing k or more ERG+ cells purely by chance:

$$P\left(X\mathit{\ge }k\right)\mathrm{=}{\mathrm{\sum }}_{i\mathrm{=}k}^n\left(\begin{array}{l}n\\ i\end{array}\right)p^i{\left(1\mathrm{-}p\right)}^{n\mathrm{-}i}$$

In which n is the total number of cells, and k the number of ERG+ cells in a given patient. Sensitivity analysis across background rates (p = 1–50%) demonstrated that patient classification is robust, remaining stable for noise thresholds between 1% and 6%. In total, 8 of 24 patients were classified as ERG+.

Gene expression data of ERG+ (n = 74) and ERG-negative (ERG−; n = 77) samples from the Taylor dataset were obtained from cBioPortal (44, 45).

Bulk RNA-seq data from The Cancer Genome Atlas prostate adenocarcinoma (TCGA-PRAD) cohort were obtained using the recount3 R package (46) and normalized to counts per million (CPM). ERG fusion status was retrieved for a subset of patients (n = 333) via PanCancerAtlas_subtypes() from TCGAbiolinks R package (https://gdc.cancer.gov/content/tcgabiolinks), yielding n = 152 ERG+ and n = 181 ERG− patients. For patients with multiple biopsies, expression values were averaged. To extend ERG classification to unannotated samples, we performed ROC analysis with the pROC R package (47) using the Youden method on patients with known ERG fusion status, identifying 10.64 CPM as a threshold for ERG positivity, leading to 208 ERG+ and 297 ERG− tumors (n = 505) corresponding to 208 ERG+ and 293 ERG− patients (n = 501).

Reference-based immune deconvolution was performed using CIBERSORT (RRID: SCR_016955) and the leukocyte signature matrix (LM22; ref. 48) to estimate cell-type composition. The R script was downloaded from https://cibersortx.stanford.edu/ and ran with default parameters.

ARNEO clinical trial (NCT03080116) transcriptomic data: The Decipher prostate genomic classifier assay, a clinical-grade transcriptome-wide array (Veracyte) was performed on formalin-fixed, paraffin-embedded diagnostic core biopsies (n = 89). Seventy-two biopsy samples passed quality control, and 65 had matched ERG immunohistochemistry staining to determine ERG status. Signature scores were obtained from a compedium of signatures contained in the Decipher GRID (49), where the ERG signature score showed an AUC of 0.96 with ERG immunohistochemistry (Supplementary Fig. S5G). Differential analysis of the Hallmark signature scores was carried out using the limma package (RRID: SCR_010943) in R (23 ERG+ tumors and 42 ERG− tumors). The trial was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Medical Ethics Committee of UZ/KU Leuven. All participants provided written informed consent prior to any study-related procedure.

STAMPEDE (NCT00268476) transcriptomic data: Tumor transcriptome-wide signature score data and ERG continuous/categorical signature scores are as described in (49). Hallmark signature scores were compared between 1,325 ERG+ and ERG− tumors at biopsy, split by metastatic (M0) and nonmetastatic (M1) disease stage at diagnosis. Differential pathway analysis was carried out using the limma library in R.

Statistical analysis

Methods for statistical tests, number of replicates, definition of error bars, and definition of statistical significance are indicated in the figure legends.

Results

ERG reduces the fitness of prostate cells by inducing cell-cycle arrest and senescence

Given that approximately 600,000 new cases of ERG+ prostate cancer are diagnosed worldwide each year (50), we sought to elucidate the impact of ERG overexpression in prostate cells. We used a lentiviral vector encoding for T2ERG (TMPRSS2-Exon1:ERG-Exon4) fusion transcript under the control of a DOX-inducible promoter. In 22Rv1 cells—an AR-positive and ETS fusion–negative cancer cell line—DOX induction led to expected expression changes of canonical ERG target genes after 1 day and resulted in ERG protein levels comparable with those seen in cells with endogenous T2ERG fusion (Supplementary Fig. S1A and S1B). Longer T2ERG expression resulted in a decrease in ERG protein levels and a diminished cell fitness compared with empty or EGFP controls (Fig. 1A and B). Moreover, T2ERG induction also markedly impaired anchorage-independent growth in soft-agar assays (Fig. 1B). The observed phenotype was ERG-dependent, as decreasing ERG levels by shERG_1 and shERG_2 rescued cell fitness, whereas shNTC and shERG_3, targeting the 3′UTR of ERG that is absent in the overexpressed T2ERG, neither reduced ERG levels nor recovered cell fitness of T2ERG-overexpressing cells (Supplementary Fig. S1C).

Figure 1.

ERG reduces the fitness of prostate cells independently of AR and induces cell-cycle arrest and senescence. A, Immunoblot of 22Rv1 cells transduced with EGFP or T2ERG expressing vectors treated with EtOH (−DOX) or 1 μg/mL DOX. B, Crystal violet (left) and soft agar (right) assays in 22Rv1 cells transduced and treated as in A for 8 and 28 days, respectively. For soft agar assay, data are shown as mean ± standard deviation (SD) and statistical significance was evaluated using a two-tailed unpaired t test (n = 4 biological replicates). C, Immunoblot (top) and crystal violet assay (bottom) in LNCaP cells transduced with EGFP, T2ERG, or T2ERG^R367K expressing vectors treated as in A for up to 6 days. D, RT-qPCR analysis (left) of KLK3 (AR target gene) and PLA1A (ERG target gene) and growth curve (right) of LNCaP-T2ERG grown in charcoal-stripped serum (CSS)-containing medium for 7 days and then treated with EtOH, 1 μg/mL DOX to induce T2ERG, and/or 10 nmol/L dihydrotestosterone (DHT) to activate AR. For RT-qPCR, data refer to 4 days after the start of the treatment and are presented relative to the control sample (EtOH-EtOH) as mean ± SD (n = 3 biological replicates). Confluence is presented as mean ± SD. E, Quantification of EdU/PI FACS profiles of cells treated with EtOH or 1 μg/mL DOX. Data are presented as mean ± SD, and statistical significance was evaluated using a two-tailed unpaired t test (n = 3 biological replicates). F, Representative images (left) and quantification (right) of SA-B-GAL staining in cells treated for 5 days (22Rv1) or 4 days (LNCaP and RWPE-1) with EtOH or 1 μg/mL DOX. Data are presented as mean ± SD, and statistical significance was evaluated using a two-tailed unpaired t test comparing the T2ERG DOX and EtOH-treated samples of the different cell lines.

Consistent with results in 22Rv1, ERG decreased the fitness of the AR-positive LNCaP cells (Fig. 1C) compared with EGFP and T2ERG^R367K, a mutant form of ERG with impaired DNA-binding capacity and reduced transcriptional activity (Supplementary Fig. S1D; refs. 51, 52). Similarly, ERG impaired the fitness of the immortalized nontumorigenic RWPE-1 cells (Supplementary Fig. S1E).

We next examined whether ERG-dependent growth reduction relied on the AR signaling. AR-negative DU145 transduced with T2ERG and induced for 14 days displayed decreased ERG protein levels over time and a reduction in fitness (Supplementary Fig. S1F). Likewise, LNCaP-T2ERG cells grown in charcoal-stripped serum-containing medium (to diminish AR activation) and induced with DOX exhibited reduced proliferation compared with control cells (Fig. 1D). The activation of AR signaling through the addition of dihydrotestosterone to the culture media, although generally increased cell proliferation, did not rescue ERG-reduced cell fitness (Fig. 1D).

To characterize the ERG-mediated reduction in cell fitness, we monitored cell-cycle distribution using EdU-PI FACS. Compared with 1-day induction, expression of ERG for 4 or 6 days decreased the proportion of 22Rv1 and LNCaP cells in S-phase, respectively (Fig. 1E; Supplementary Fig. S1G; t test P value = 0.0087 for 22Rv1 and 0.0035 for LNCaP), indicating that ERG induces cell-cycle arrest. Moreover, ERG promoted senescence in 22Rv1-, LNCaP-, and RWPE-1-T2ERG cells (Fig. 1F; t test P value = 0.0199). In contrast, we did not observe ERG-induced apoptosis in Annexin V–PI FACS experiments (Supplementary Fig. S1H).

These data extend previous findings (25, 31, 53) of reduced cell fitness of ERG+ cells in vitro by showing that ERG elicits cell-cycle arrest and senescence to mediate this response.

ERG activates EMT and senescence and generates heterogeneous cell states

To gain further insight into the ERG-induced phenotypic response and assess whether transcriptionally regulated oncogenic processes contribute to the observed reduced proliferation, we performed RNA-seq in LNCaP-T2ERG and LNCaP-T2ERG^R367K treated with EtOH, as control, or DOX for 4 days. ERG induction resulted in the upregulation of 575 genes and downregulation of 74 genes [FDR < 0.01 and log₂ (fold change) FC > 1 or log₂FC < −1, respectively; Fig. 2A; Supplementary Table S1]. By contrast, only 1 gene was differentially expressed in response to T2ERG^R367K (FDR < 0.01 and log₂FC > 1 or log₂FC < −1; Supplementary Fig. S2A; Supplementary Table S1). Integration with published ERG ChIP-seq data (52) identified 530 ERG-bound and regulated genes (Supplementary Fig. S2B). GSEA revealed 143 positively enriched signatures (FDR < 0.05), including two related to senescence (Supplementary Fig. S2C; Supplementary Table S2), consistent with the senescence phenotype detected by SA-B-GAL staining (Fig. 1F). In line with the ERG-induced cell-cycle arrest (Fig. 1E), signatures related to MYC targets, G2M checkpoint genes, and E2F targets were downregulated by ERG (FDR < 0.05; Fig. 2B; Supplementary Table S3). Interestingly, ERG-upregulated genes were enriched for EMT and inflammatory pathways, including TNFA signaling via NFKB, inflammatory response, IFNγ and IFNα responses, and TGFβ signaling (Fig. 2B; Supplementary Table S3). Activation of EMT by ERG was also confirmed in 22Rv1-T2ERG cells by RNA-seq and GSEA (FDR < 0.1; Supplementary Fig. S2D; Supplementary Tables S4 and S5) and in RWPE-1-T2ERG cells (Supplementary Fig. S2E) and may contribute to cell-cycle arrest (54).

Figure 2.

ERG drives EMT and senescence via TGFβ signaling. A, RNA-seq data volcano plot depicting gene regulation and statistical significance upon T2ERG expression in LNCaP cells. Cells were treated with EtOH or 1 μg/mL DOX for 4 days. Red and blue dots represent significantly (FDR <0.01) upregulated [log₂ (DOX+/DOX−) > 1] and downregulated [log₂ (DOX+/DOX−) < −1] genes, respectively (n = 3 biological replicates). B, GSEA using the Hallmark collection and data in A selected for ERG-bound genes. Shown are gene sets significantly (FDR <0.05) positively (red) or negatively (blue) enriched. NES, normalized enrichment score; sign., signaling. C, Immunoblot in cells treated with EtOH or 1 μg/mL DOX for 4 days. GAPDH and ACTB were used as loading controls. l.e., longer exposure. D, Representative SA-B-GAL/VIM FACS profiles (left) and quantification (right) of RWPE-1-T2ERG cells treated with EtOH or 1 μg/mL DOX for 4, 8, and 13 days. Numbers indicate the percentage of cells in each quadrant. Data are presented as mean ± standard error of the mean (SEM; n = 3 biological replicates), and statistical significance was evaluated using a two-tailed unpaired t test. *, P value < 0.05; **, P value < 0.01; ***, P value < 0.001. E, Heatmap for the regulation of TGFβ pathway components in LNCaP-T2ERG used in A and RWPE-1-ERG from (55, 56). Shown are genes with FDR < 0.05 in at least one of the RWPE-1-ERG datasets and concordantly regulated in LNCaP-T2ERG [log₂ (DOX+/DOX−) < −0.2 or >0.2]. aSMAD, activator SMAD; coR, coreceptor; ecAntagonist, extracellular antagonist; Type2R, type 2 receptor. F, Quantitative ChIP experiments for ERG at three TGFβ pathway genes (TGFB2, SMAD3, and TGFBR2), two ERG targets (ARHGDIB and IFIT3), and a control region in RWPE-1-T2ERG cells treated as in A. IgG were used as control. Data are shown as mean ± standard deviation of technical triplicates. G, Immunoblot in RWPE-1 cells treated for 4 days with 1 μg/mL DOX. ACTB was used as loading control. H, Immunoblot in RWPE-1 cells treated for 4 days with EtOH or 1 μg/mL DOX and DMSO, 1 μmol/L A8301 (ALK4/5/7i), 30 nmol/L PD0325901 (MEK1/2i), 10 nmol/L RAPAMYCIN (mTORi), 5 μmol/L TAKINIB (TAK1i), or 10 μmol/L Y27632 (ROCK1/2i). ACTB and GAPDH were used as loading controls. I, Quantification of SA-B-GAL staining in cells used in G and H and evaluated via colorimetric assay (shRNA-transduced cells) or FACS (cells treated with inhibitors). Data are presented as mean ± SEM (n = 3 biological replicates), and statistical significance was evaluated using a two-tailed unpaired t test. **, P value < 0.01; ***, P value < 0.001.

Further analyses showed that ERG induced EMT to varying extents. The nontumorigenic RWPE-1-T2ERG cells exhibited a full EMT phenotype, characterized by the upregulation of the transcription factors ZEB1 and SOX9, the increased levels of VIM and N-cadherin (CDH2), and the concomitant downregulation of the epithelial marker E-cadherin (CDH1; Fig. 2C; Supplementary Fig. S2F). Cancer cells showed an intermediate phenotype: 22Rv1-T2ERG cells upregulated SOX9 and CDH2 without a decrease in CDH1 (Fig. 2C), whereas LNCaP-T2ERG displayed increased expression of ZEB1 (at RNA level) and SOX9 but no changes in other EMT and epithelial markers by immunoblot (Fig. 2C; Supplementary Fig. S2G). FACS analysis showed that 3% of LNCaP cells expressed VIM upon ERG overexpression (Supplementary Fig. S2H), suggesting that only a minority of cells undergoes full EMT, whereas the majority remains in an intermediate state or resists full transition.

To determine whether EMT and senescence occur within the same cells, we assessed activation of both pathways at single-cell resolution. In RWPE1-T2ERG, 16.6% of cells were characterized by expression of both VIM and SA-B-GAL at 4 days after induction (SA-B-GAL+/VIM+ cells), a percentage that increased to about 30% at 8 and 13 days (Fig. 2D; t test P value < 0.05). Instead, about 40% of cells consistently showed only VIM expression (SA-B-GAL−/VIM+ cells) throughout the experiment (t test P value < 0.05).

Leveraging our RNA-seq results, we investigated pathways mediating ERG-induced EMT. Inhibition of TGFβ signaling by A8301 treatment, but not of p38 or NFKB pathways by SB202190 and BAY117082, respectively, reduced the percentage of VIM+ cells and of CDH2 levels (Supplementary Fig. S2I and S2J). In both LNCaP-T2ERG and RWPE-1-ERG (55, 56), ERG upregulated the expression of multiple components of the TGFβ signaling cascade, including the ligand TGFB2, the type 2 receptor TGFBR2, the coreceptor ENG, the transcription factors SMAD2 and SMAD3, as well as AKT3 and MAP2K3, which are involved in noncanonical TGFβ signaling (Fig. 2E; Supplementary Fig. S2K; Supplementary Table S1). Quantitative ChIP analysis showed that TGFB2, SMAD3, and TGFBR2 are bound by ERG (Fig. 2F), indicating direct transcriptional regulation. Furthermore, knockdown of SMAD4, which interacts with SMAD2/3 in the canonical TGFβ signaling, and inhibition of MEK1/2 with PD0325901, which participates in noncanonical TGFβ signaling, decreased ERG-mediated VIM upregulation in RWPE-1 cells (Fig. 2G and H). Inhibition of other noncanonical TGFβ pathway mediators—including mTOR (RAPAMYCIN), TAK1 (TAKINIB), and ROCK1/2 (Y27632)—did not affect VIM levels (Fig. 2H). Supporting the activation of the MEK/ERK pathway by ERG via TGFβ, ERG expression increased ERK1/2 phosphorylation, an effect that was reduced with concomitant A8301 treatment (Fig. 2H). Interestingly, inhibition of MEK1/2 but not depletion of SMAD4 decreased ERG-induced senescence (Fig. 2I). Collectively, these findings indicate that ERG generates heterogeneous cellular states in which senescence and EMT are differentially regulated by canonical and MEK/ERK-mediated noncanonical TGFβ signaling.

ERG promotes dedifferentiation and an intermediate basal–luminal cell population

To determine whether ERG regulates other aspects of cell plasticity and explore how ERG alters prostate cell identity, we turned to transcriptomic signatures enabling prostate cell and prostate cancer classification (37–41). In LNCaP cells, ERG-upregulated genes were significantly enriched for the Lim mammary stem cell upregulated signature, the mesenchymal and stem-like prostate cancer (MSPC) signature, and FGF targets (FDR < 0.05; Fig. 3A), which define a prostate cancer molecular subtype with stem cell–like features (37, 38, 40). Whereas canonical AR target genes showed no significant enrichment, transcripts downregulated in AR-driven prostate cancer (ARPC) were enriched among the ERG-upregulated genes (TCGA androgen response and ARPC down, respectively; Fig. 3A). Prostate luminal and basal cell signatures (41) were upregulated upon T2ERG overexpression (FDR < 0.05, Fig. 3A), suggesting that ERG promotes both cell states. Extending this analysis, we found that ERG also induced genes characteristic of basal, club, and hillock epithelial cells of the benign prostate—which were recently shown to be expressed in double-negative prostate cancer (39)—indicative of multilineage plasticity (FDR < 0.05; Fig. 3A). Similarly, 22Rv1 cells displayed the activation of the Lim mammary stem cell up, the MSPC up, the basal, the club, and the hillock signatures upon T2ERG overexpression (FDR < 0.05; Supplementary Fig. S3A). Conversely, the FGF targets and the luminal signatures were not significantly enriched (FDR < 0.05). Instead, these cells upregulated NE and ARPC down markers and downregulated androgen response and ARPC up genes (Supplementary Fig. S3A), consistent with their underlying NE background (37).

Figure 3.

ERG promotes cell plasticity. A, GSEA with gene sets used for prostate cell and prostate cancer (PCa) classification and comparing the expression profile of induced LNCaP-T2ERG cells (DOX+) vs. control (DOX−). ERG-bound genes were used for the analysis. Red bars indicate significantly (FDR < 0.05) positively enriched gene sets. ARPC, AR-driven PCa; MSPC, mesenchymal and stem-like PCa; NEPC, neuroendocrine PCa; NES, normalized enrichment score. B, Immunoblot in cells treated with EtOH or 1 μg/mL DOX for 4 days. HSP70 and GAPDH were used as loading controls. l.e., longer exposure; Lum, luminal. C, Representative KRT15/KRT18 FACS profiles (left) and quantification (right) of RWPE1-T2ERG and LNCaP-T2ERG cells treated as in B. Data are presented as mean ± standard deviation, and statistical significance was evaluated using a two-tailed unpaired t test (n = 3 biological replicates). Numbers indicate the percentage of cells in each quadrant. D, Immunofluorescence in RWPE-1-T2ERG and LNCaP-T2ERG cells treated for 4 days with 1 μg/mL DOX. E, UMAPs of scRNA-seq data from (42, 43). Shown are 2,187 and 1,583 epithelial malignant ERG+ cells (for Song 2022 and Apostolov 2026, respectively) with yellow dots indicating cells with combined expression [ln (cp10K + 1) > 25th percentile] of KRT18 and KRT15. F, Boxplot of the cells in E showing the average expression of the Lim mammary stem cell upregulated signature. Cells were divided according to KRT18 and KRT15 expression status [positive expression: ln (cp10K + 1) > 25th percentile]. Statistical significance was evaluated using the Wilcoxon-Mann–Whitney test. Numbers below the boxplots indicate the number of cells in each group.

At the protein levels, ERG increased the luminal marker KRT18 and the basal marker KRT15 in RWPE-1 and LNCaP cells, whereas 22Rv1 showed only increased KRT15 expression (Fig. 3B). The basal markers KRT14 and KRT5 were not detected in LNCaP-T2ERG cells (Fig. 3B) and RWPE-1-T2ERG showed homogenous expression of KRT5 with and without ERG expression (Supplementary Fig. S3B). Importantly, ERG significantly enhanced coexpression of KRT18 and KRT15 in a subset of cells, denoting the induction of an intermediate basal–luminal cellular phenotype (t test P value = 0.006 for RWPE-1 and 0.001 for LNCaP, Fig. 3C and D). As this intermediate cell population has been implicated in prostate cancer initiation (57–59), our findings suggest that ERG may exert its oncogenic effects by acting as a dedifferentiation factor and promoting this intermediate cell state.

We observed a similar increase in the proportion of KRT18/KRT15 double-positive cells in the MDA prostate cancer 2b (AR-positive) cell line upon ERG expression, whereas the AR-negative stem cell-like cells DU145 and PC-3 showed no such changes (Supplementary Fig. S3C). Growing LNCaP-T2ERG cells in charcoal-stripped serum-containing medium did not abolish the emergence of these double-positive cells (t test P value = 0.0368, Supplementary Fig. S3D), excluding AR involvement in ERG-mediated dedifferentiation.

To determine whether ERG+ luminal–basal intermediate cells are present in clinical samples, we analyzed scRNA-seq data from localized prostate cancer (42). Among ERG-expressing cells, we identified a subpopulation co-expressing KRT18 and KRT15 (0.7% of cells, Fig. 3E; Supplementary Fig. S3E). Importantly, these cells exhibited higher expression of stem cell–related genes compared with single KRT18-positive or double-negative cells (Wilcoxon P value = 0.048 and 2.6E−5, respectively, Fig. 3F). Analysis of a second dataset (43) revealed a higher proportion of KRT18/KRT15 double-positive cells (7.8% of cells) in ERG+ human tumors and confirmed the higher expression of stem cell–related genes in these cells (Fig. 3E and F; Supplementary Fig. S3E). This supports our findings of an intermediate basal–luminal cell population induced by ERG.

ERG activates p53, and p53 loss enhances ERG-induced cell plasticity markers

In prostate cancer, TP53 deletions and mutations frequently co-occur with ETS fusions and cooperate with ETV4 and T2ERG to promote prostate tumorigenesis and cancer progression (11, 60). To check whether p53 activity contributes to the identified ERG-mediated phenotypes, we first examined p53 status. Accumulation of p53 and/or its increased phosphorylation at serine 15 were detected upon ERG induction in RWPE-1, 22Rv1, and LNCaP cells (Fig. 4A). p21 was also upregulated by ERG, further supporting p53 activation (Fig. 4A). ERG expression did not regulate TP53 transcript (Supplementary Tables S1 and S4), whereas it consistently increased γH2A.X levels across all cell lines (Fig. 4A), suggesting that ERG-induced DNA damage may underlie p53 activation. Consistent with this hypothesis, p53 phosphorylation did not increase upon ERG induction in cells treated with inhibitors of ATM (KU60019), ATR (VX970), and p38 (SB202190)—kinases regulating the DNA damage response and downstream signaling (Fig. 4B). Inhibition of other p53 activators, such as DNA-PK (NU7441), mTOR (RAPAMYCIN), and reactive oxygen species (N-acetyl-L-cysteine), did not have such effect (Fig. 4B). Unexpectedly, p21 levels remained elevated in all conditions.

Figure 4.

p53 is activated by ERG and its loss enhances ERG-induced cell plasticity markers without affecting senescence. A, Immunoblot in cells treated with EtOH or 1 μg/mL DOX for 4 days. GAPDH, HSP70, and VINC were used as loading controls. l.e., longer exposure. B, Immunoblot in LNCaP-T2ERG cells treated for 4 days with EtOH or 1 μg/mL DOX and for 3 hours with DMSO as control, 5 μmol/L NU7441 (DNA-PKi), 10 μmol/L KU60019 (ATMi), 500 nmol/L VX970 (ATRi), 5 μmol/L SB202190 (p38i), 20 nmol/L RAPAMYCIN (mTORi), and 2.5 mmol/L NAC (N-Acetyl-L-Cysteine). ACTB was used as loading control. C, Immunoblot (left) and growth curve (right) of LNCaP-T2ERG cells treated as in A. VINC and ACTB were used as loading controls. Relative confluence is presented as mean ± standard error of the mean, and statistical significance was evaluated using a two-tailed unpaired t test (n = 3 biological replicates). *, P value < 0.05; **, P value < 0.01. D, Quantification of SA-B-GAL staining in cells used in C. Experiments were repeated independently twice, leading to similar results. One set of data is presented. E, Immunoblot of LNCaP-T2ERG cells treated as in A. GAPDH and HSP70 were used as loading control.

To directly test p53 dependency in the context of ERG activity, we generated TP53 knockout LNCaP and RWPE-1 cells (Fig. 4C; Supplementary Fig. S4A). TP53 loss led to a strong reduction of p21 levels and to a modest increase in the proliferation of ERG-expressing LNCaP cells (Fig. 4C; Supplementary Fig. S4B). Although TP53 knockout did not affect ERG-induced senescence (Fig. 4D; Supplementary Fig. S4C and S4D), it enhanced the expression of ERG-induced cell plasticity markers (Fig. 4E). Altogether, these results indicate that p53 is partially activated by the DNA damage response induced by ERG activity and that p53 loss might cooperate with ERG by enhancing cell plasticity.

We investigated additional mediators that could be involved in ERG-driven cell-cycle arrest. In addition to CDKN1A (encoding p21), the cell-cycle inhibitors CDKN2B (p15) and CDKN1C (p57) were upregulated in LNCaP cells upon ERG induction, and p15 levels remained elevated also in TP53-depleted cells (Supplementary Table S1; Fig. 4C). Moreover, inactivation of the tumor-suppressor PTEN often co-occurs with ERG fusions, jointly driving prostate cancer initiation (2, 9, 12). We depleted p15 and p57 in LNCaP-T2ERG cells and PTEN in RWPE-1-T2ERG cells and detected no differences in the proliferation of ERG-expressing cells with and without p15, p57, or PTEN (Supplementary Fig. S4E and S4F), indicating that their losses do not rescue the proliferation defect of ERG+ cells.

ERG promotes inflammatory gene expression via cell-autonomous and paracrine mechanisms

Transcriptomic profiling in LNCaP and 22Rv1 indicated that ERG upregulated several inflammatory- and senescence-associated pathways (Fig. 2B; Supplementary Fig. S2D), and secreted factors related to these pathways were also induced by ERG in RWPE-1 (t test P value < 0.05; Fig. 5A). Given that both inflammation and senescence can function through paracrine mechanisms modulating tumor initiation and progression (57, 61), we investigated whether ERG exerted paracrine effects by performing CM experiments. ERG-CM reduced RWPE-1 proliferation (Supplementary Fig. S5A), increased SA-β-GAL+/VIM− and SA-β-GAL+/VIM+ populations (Supplementary Fig. S5B), and upregulated cytokines (IL1B, IL6, and CXCL8), TNFA–NFKB targets, and an IFN target (t test P value < 0.05; Fig. 5B), indicating that the ERG-induced secretome promotes inflammation and senescence.

Figure 5.

Inflammatory gene expression is induced by ERG in vitro and upregulated in ERG+ tumors. A, RT-qPCR analysis in RWPE-1-T2ERG cells treated with EtOH or 1 μg/mL DOX for 4 days. Data are presented relative to the EtOH-treated sample as mean ± standard error of the mean (SEM; n = 3 biological replicates for IGFBP5 and n = 4 biological replicates for other genes). B, RT-qPCR analysis in RWPE-1 cells (left) and human primary prostate fibroblasts (right) grown for 4 days with CM obtained from RWPE-1-T2ERG cells treated with EtOH or DOX (1 μg/mL) for 4 days. Data are presented relative to the EtOH-conditioned sample as mean ± SEM (n = 4 biological replicates for RWPE-1 and n = 3 biological replicates for prostate fibroblasts). C, Overview of the human datasets used in the analysis (left), upset plot of the Hallmark gene sets significantly (FDR < 0.05) up- or downregulated (log₂FC > 0.1 or log₂FC < −0.1, respectively) in ERG+ vs. ERG− tumors (middle), and bubble plot for the gene sets regulated in more than 50% of datasets (right). sc, scRNA-seq; sign., signaling; n.s., not significant. D, Overview of the STAMPEDE dataset (left) and bubble plot (right) for the Hallmark gene sets significantly (FDR < 0.005) upregulated in ERG+ vs. ERG− tumors. M0 and M1 refer to the TNM classification (no distant metastasis and evidence of distant metastasis, respectively). E, Violin plots showing the M1-to-M2 macrophage abundance ratio per patient, as estimated by CIBERSORT deconvolution of TCGA expression data. Statistical significance was evaluated using the Wilcoxon-Mann–Whitney test. F, Immunohistochemistry for CD68 (macrophage marker), CD163 (macrophage marker), and ERG on human prostate cancer (PCa) samples. Staining was performed on serial sections of paraffin-embedded samples. Both samples were classified as Gleason score 3 + 4 and tumor–node–metastasis (TNM) stage M0. Arrowheads indicate macrophage-associated signal. G, RT-qPCR analysis in M0 macrophages treated as in B for 2 days. Data are presented relative to the EtOH-conditioned sample as mean ± standard error of the mean (n = 3 biological replicates). For A, B, and G statistical significance was evaluated using two-tailed unpaired t test. *, P value < 0.05; **, P value < 0.01; ***, P value < 0.001. For C and D, statistical significance was evaluated using Wilcoxon-Mann–Whitney test with FDR correction.

We next reasoned that ERG-induced paracrine effects might also extend to nonepithelial cells. Indeed, human primary prostate fibroblasts exposed to ERG-CM displayed upregulation of IL1A, IL6, and CXCL8 (t test P value < 0.05, Fig. 5B), markers characterizing inflammatory cancer-associated fibroblasts (CAF; ref. 62) and inflammation. By contrast, the ACTA2 and CCN2 myofibroblastic CAF-related genes were increased by TGFβ treatment but not ERG-CM (Fig. 5B; Supplementary Fig. S5C).

To assess whether ERG-induced inflammation persists after ERG withdrawal, we induced ERG expression for 4 days in RWPE-1 cells and subsequently cultured them in DOX-free medium for up to 9 days to allow ERG levels to return to baseline (Supplementary Fig. S5D). Proliferation was restored by day 4 (Supplementary Fig. S5D), senescent cells were largely absent by day 9 (Supplementary Fig. S5E), and inflammatory genes remained elevated at both 4 and 9 days—although a general descending trend was observed—(t test P value < 0.05, Supplementary Fig. S5F), indicating that ERG-induced inflammation can be sustained independently of continued ERG expression, and ERG-induced senescence is reversible.

Next, we investigated whether proinflammatory pathways were activated in ERG+ human tumors. We found upregulation of TGFβ signaling and TNFA signaling via NFKB in ERG rearranged versus nonrearranged tumors (Wilcoxon P value with FDR correction <0.05) in three independent prostate cancer patient cohorts, namely Taylor, et al. 2010 (45), Song, et al. 2022 (42), and 65 high-risk patients with prostate cancer from the ARNEO trial (Fig. 5C; Supplementary Table S6). We also confirmed upregulation of TGFβ in the prostate cancer TCGA dataset (extended version; ref. 2). We next queried gene expression data from 1,325 patients of the STAMPEDE trial (49) stratified by metastatic stage at diagnosis (M0: very-high-risk nonmetastatic; M1: de novo metastatic) and confirmed upregulation of NFKB pathways in M0 ERG+ tumors and of TGFβ in M0 and M1 ERG+ prostate cancer (Wilcoxon P value with FDR correction <0.005, Fig. 5D; Supplementary Table S6).

Consistent with the proinflammatory role of ERG, deconvolution of transcriptomic data showed a higher M1-to-M2 macrophage ratio in ERG+ tumors (Wilcoxon P value = 2.9E−8; Fig. 5E; Supplementary Fig. S5H; Supplementary Table S7). We verified infiltration of macrophages in ERG+ PIN lesions (Supplementary Fig. S5I) and in prostate cancer tissue sections, although the sample size was too modest to test for enrichment (Fig. 5F). Importantly, ERG-CM upregulated M1-associated inflammatory genes and downregulated M2-associated scavenger receptors in M0 macrophages (Fig. 5G).

Taken together, these findings support a model in which ERG modulates inflammatory gene expression across epithelial, stromal, and myeloid cells, potentially sustaining an inflammatory microenvironment.

Discussion

ERG fusions characterize about half of prostate cancers (2, 3). Extensive studies based on mouse models and human genetic data implicated these rearrangements as early events in prostate tumorigenesis, driving cancer initiation when combined with other mutations or expressed at high levels (4, 7–12). Yet, the biological programs directly triggered by ERG in prostate cells, independently of co-occurring aberrations, remain incompletely defined. In our experiments, inducible, temporally controlled expression of ERG enabled the precise mapping of ERG-driven transcriptional and phenotypic changes while minimizing secondary effects due to cellular adaptation or selection. Additionally, the focus on early time points minimizes the impact of the progressive decrease in ERG levels (Fig. 1A and C). In this setting, the observed cell-cycle arrest may reflect a transient stress response to acute, high-level ERG expression. Nonetheless, this effect may still recapitulate the initial cellular response occurring in vivo upon T2ERG fusion formation, in which elevated ERG expression is required to initiate tumorigenesis in the absence of additional aberrations (8).

We observed that ERG activates TGFβ signaling, driving EMT and senescence (Fig. 2E–I). Both SMAD-dependent canonical and MEK/ERK-mediated noncanonical TGFβ signaling pathways have been previously implicated in regulating these processes (63, 64). Our data indicate a pattern of specialization, with SMAD-dependent signaling mainly controlling EMT induction and ERK1/2 regulating both VIM expression and SA-B-GAL activity (Fig. 2G–I). Although the mechanistic basis for this differential regulation remains to be investigated, a partial explanation may lie in ERK1/2-mediated phosphorylation modulating ERG transcriptional activity (56).

p53 is another key player able to control both senescence and cell plasticity (41, 65). We detected increased p53 activity upon ERG induction, partially due to the activation of the DNA damage response (Fig. 4A and B). Interestingly, p53 inactivation did not affect ERG-induced senescence but did increase ERG-induced markers of cell plasticity (Fig. 4D and E). This initial evidence is consistent with previous findings that p53 loss enhances cellular reprogramming in fibroblasts and frequently occurs in advanced prostate cancer in which lineage plasticity is activated as a therapy resistance mechanism (41, 65).

The ability of ERG to induce cell plasticity might have important implications for both prostate cancer tumorigenesis and response to therapy. Recently, Feng and colleagues (66) used lineage tracing in mice to show that ERG+ tumors initiate from a basal–luminal cell population that, upon ERG induction, gives rise to intermediate cells coexpressing basal, luminal, hillock, and club cell markers and characterized by stem-like features. This in vivo evidence aligns with our observation of an intermediate basal–luminal cell population induced by ERG and, together with our human samples analysis results (Fig. 3E and F), strongly supports its functional relevance. Moreover, recent studies reported that T2ERG fusions are overrepresented in patients with high-risk localized prostate cancer nonresponding to neoadjuvant antiandrogen and AR signaling inhibitor therapies (67, 68). It is tempting to speculate that ERG-induced cell plasticity might be relevant in this context. However, it should be noted that ERG expression in prostate cancer is under the control of AR (1)—the primary target of these treatments—and that ERG+ tumor cells are dependent on ERG for survival (15). This suggests a functional trade-off, whereby although ERG can promote cellular plasticity, cells that are highly ERG-dependent may fail to exploit this escape route as suppression of AR signaling compromises ERG expression and cell survival.

Chronic inflammation has been implicated as a driver of prostate cancer initiation and progression (69, 70). Multiple internal cues have been described as sources of prostate inflammation, including senescence, inflammatory myeloid-derived cells, and oncogenic signaling, such as that driven by PTEN loss (57). Here, we showed that ERG activated senescence (Figs. 1F, 2D, and I) and inflammatory gene expression changes (Figs. 2B and 5A; Supplementary Fig. S2D). Moreover, ERG-induced inflammation spread across epithelial cells, fibroblasts, and macrophages in a paracrine fashion (Fig. 5B and G), leading to a potentially chronic inflammatory cycle. Analysis of human samples supported the activation of inflammatory pathways in ERG+ tumors (Fig. 5C and D). The mechanistic details and the role of this inflammation in ERG+ tumors are not currently defined. Multiple lines of evidence support the relevance of ERG-induced inflammatory pathways (71–73). Notably, proliferative inflammatory atrophy (PIA) lesions harbor T2ERG fusions, although at a low frequency (74), and inflammation-associated intermediate club-like cells are observed in ERG-positive PIA transitioning to early adenocarcinoma (75). Furthermore, administration of nonsteroidal anti-inflammatory drugs is associated with a reduced incidence of high-grade adenocarcinoma lesions in a T2ERG-PTEN^flox/flox mouse model and with a decreased risk of developing T2ERG-positive prostate cancer in patients (76, 77).

Senescent cells sustain inflammation and promote immunosuppression and EMT via the senescence-associated secretory phenotype through autocrine and paracrine signaling (78). Senescence is also accompanied by the activation of prosurvival signaling that prevents apoptosis and the emergence of a latent stemness program, which confers tumor-initiating capabilities upon cell-cycle arrest escape (79). Within this framework, the ERG-induced bivalent senescent–EMT cell state we observed (Fig. 2D) might serve as a shelter for ERG+ cells during tumorigenesis, allowing them to persist while accumulating additional tumor-promoting features. More detailed characterization of ERG-induced senescence is warranted to fully elucidate how ERG promotes tumor initiation and progression. Such insights can inform novel therapeutic opportunities based on one-two punch approaches using senolytic agents.

Overall, our study provides new insights into ERG oncogenic activity and indicates that further investigations into how ERG-induced dedifferentiation, inflammation, and senescence are interconnected will shed light on ERG-mediated tumorigenesis and may uncover novel therapeutic vulnerabilities relevant for a large fraction of patients with prostate cancer.

Supplementary Material

Supplementary Table S2

GSEA using the C2 gene set collection on ERG-bound genes in LNCaP-T2ERG cells

Supplementary Table S3

GSEA using the Hallmark gene set collection on ERG-bound genes in LNCaP-T2ERG cells

Supplementary Table S4

Differential gene expression analysis in 22Rv1-T2ERG cells

Supplementary Table S5

GSEA using the Hallmark gene set collection in 22Rv1-T2ERG

Supplementary Table S6

Differential regulation of gene sets in the Hallmarks collection in tumor samples

Supplementary Table S7

Deconvolution with CIBERSORT using TCGA-PRAD transcriptomic data

Supplementary Table S8

Oligonucleotides and antibodies used in the study

Supplementary Figure S1

ERG reduces the fitness of RWPE-1 and DU145 cells but does not induce apoptosis

Supplementary Figure S2

Genes related to EMT and inflammation are induced by ERG in 22Rv1 and RWPE-1 cells

Supplementary Figure S3

Assessment of ERG-induced cell plasticity in different PCa cell lines

Supplementary Figure S4

p53, p15, p57, and PTEN inactivation does not restore proliferation

Supplementary Figure S5

Additional characterization of ERG-induced senescence and inflammatory gene expression

Supplementary Table S1

Differential gene expression analysis in LNCaP-T2ERG and LNCaP-T2ERG-R367K cells, and differential expression of TGF beta pathway components in LNCaP and RWPE-1 cells

Data Availability

The RNA-seq data generated in this study are available in the BioStudies database (http://www.ebi.ac.uk/biostudies) under accession number E-MTAB-15836. The publicly available gene expression data reanalyzed in this study were obtained from the recount3 R package at https://bioconductor.org/packages/recount3 (PRAD-TCGA; ref. 46), the Gene Expression Omnibus under accession number GSE176031 (Song scRNA-seq data; ref. 42), and the CELLxGENE portal (Apostolov scRNA-seq data; ref. 43) at https://cellxgene.cziscience.com/collections/bdac7a53-fe34-4f04-8c46-f9bd5297c099. All other raw data generated in this study are available upon request to the corresponding author.

Authors’ Disclosures

E. Grist reports personal fees and nonfinancial support from Johnson & Johnson and personal fees from Veracyte advisory board outside the submitted work; in addition, E. Grist has a patent for “Decipher assay for use in metastatic disease” issued and a patent for “PTEN inactivity signature” pending. S. Joniau reports grants, personal fees, nonfinancial support, and other support from Johnson & Johnson and Ferring Pharmaceuticals during the conduct of the study. E. Davicioni is an employee of Veracyte, makers of Decipher test. G. Attard reports other support from Amgen, AstraZeneca, Bayer, Merck & Co, Merck Serono Ltd, Sanofi, Sandoz, Agilent, and Artera and grants and other support from Astellas Pharma, Blue Earth Therapeutics, Janssen-Cilag, Novartis, Pfizer, and Veracyte outside the submitted work; in addition, G. Attard has a patent for blood methylation markers (GB1915469.9) licensed and is listed as inventor on docetaxel predictive biomarker (Veracyte) and on the Institute of Cancer Research discoveries’ list of abiraterone acetate, G. Attard personally receives share of revenue by The Institute of Cancer Research. F. Demichelis reports a patent for US Patent (7,718,369; 8211645) licensed. No disclosures were reported by the other authors.

Authors’ Contributions

F. Lorenzin: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration. G. Fracassi: Formal analysis, investigation, visualization, methodology. F. Vannuccini: Data curation, formal analysis, investigation, visualization. M. Ismail: Data curation, formal analysis. Y. Ciani: Data curation, formal analysis, investigation. R. Bevilacqua: Formal analysis, investigation, visualization. P. Gasperini: Investigation. F.G. Carbone: Data curation, investigation. S. Cinalli: Data curation, investigation. E. Grist: Data curation, investigation. W. Devlies: Data curation, investigation. A. Giesen: Data curation, investigation. J. Van Goubergen: Data curation, investigation. F. Claessens: Data curation, investigation. S. Joniau: Data curation, investigation. E. Davicioni: Data curation, investigation. M. Barbareschi: Data curation, investigation. D.S. Rickman: Resources. A. Lunardi: Resources, investigation. G. Attard: Supervision, project administration. A. Zoubeidi: Resources, writing–review and editing. F. Demichelis: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing.