ERG-Mediated Coregulator Complex Formation Maintains Androgen Receptor Signaling in Prostate Cancer

Recurrent genomic rearrangements involving ETS family genes are a distinguishing feature of human prostate cancers. The TMPRSS2-ERG fusion is the most common of these alterations, occurring in more than 40% of clinically localized prostate cancer cases (1). In this rearrangement, the promoter and 5′ untranslated region of TMPRSS2 are fused to the open reading frame of ERG, resulting in overexpression of the oncogenic ERG protein, either in a full-length or truncated form depending on where the breakpoint occurs. Other less frequent ETS family gene fusions are also found in prostate cancer, including those involving ETV1, ETV4, and ETV5, accounting for approximately 10% of prostate cancer ETS family rearrangements.

Numerous model systems have demonstrated the oncogenic potential of ETS family genes in prostate tumorigenesis. In particular, mouse models overexpressing ERG in combination with other genetic alterations, such as PTEN loss, display rapid onset adenocarcinoma in the mouse prostate (2–4). Interestingly, ERG overexpression alone is not sufficient to drive this transformation. In these models, ERG is able to drastically reprogram the androgen receptor (AR) cistrome and prime prostate tumorigenesis in response to PTEN loss, thus leading to reactivation of an AR transcriptional program that underlies cellular transformation (2). Furthermore, direct interactions of ERG with AR through the ETS domain have been suggested to promote the ability of AR to initially bind DNA in this context (5).

These findings underscore the importance of ERG overexpression in prostate cancer initiation. In addition, disruption of ERG through siRNA or peptidomimetic degraders significantly inhibits the growth of prostate cancer xenografts that endogenously harbor the TMPRSS2-ERG fusion, highlighting the tumors' dependence on ERG and its potential as a clinically relevant therapeutic target (6, 7). However, it is still unknown how ERG maintains this cancer phenotype and the downstream AR signaling pathways that are integral for tumor cell growth and survival. Here, we utilize a murine organoid system to uncover the mechanisms behind prostate cancer's dependence on the ERG oncogene, and how ERG alters and maintains AR signaling at distinct stages of prostate cancer disease progression.

Mouse prostate organoids were derived from wild-type (WT), Pb-Cre4 Pten^flox/flox, and Pb-Cre4 Pten^flox/flox Rosa26-ERG (human ERG transcript) genetically engineered mouse models and maintained in organoid culture conditions as described previously (8, 9). Briefly, murine prostates were digested with Collagenase/Hyaluronidase (Stemcell Technologies, 07912) and then with TrypLE (Gibco). Cells were cultured in suspension for 5–10 days and transferred to 2D collagen-coated plates or 3D Matrigel conditions using established protocols (10). Multiple clones were derived from different organoid lines to validate genetic alterations and protein expression. Cells were used at low passage numbers and were Mycoplasma tested using the MycoAlert Mycoplasma Detection Kit (Lonza).

ERG-knockout (KO) lines were established as described previously (5, 8). Briefly, to KO ERG in Pten^−/− R26-ERG organoids, two pairs of single guide RNA (sgRNA) constructs were cloned into the LentiCRISPRv2 vector obtained from Addgene. Lentiviruses for ERG sgRNAs or vector control were generated in 293T cells using lentiviral packaging vectors. See Supplementary Materials and Methods for further details.

Intact organoid spheres were fixed in 4% paraformaldehyde (Electron Microscopy Sciences) for 30 minutes at room temperature and washed with 70% ethanol. Resultant organoids were embedded in HistoGel (Thermo Fisher Scientific) and paraffin embedded using the Leica ASP6025 Tissue Processor (Leica Biosystems) at the Dana-Farber, Harvard Cancer Center Brigham Histopathology Core (Boston, MA). Five-micron sections were stained on Ventana Discovery XT. Antibodies and dilutions used for IHC were as follows: PTEN 1:100 (Cell Signaling Technology, 138G6 Rabbit mAb 9559) and ERG 1:100 (Recombinant EPR3864 ab92513).

Bulk RNA was isolated using QIAshredder (Qiagen, 79656) and RNeasy Mini Kits (Qiagen, 74106) per the manufacturer's standard protocol. mRNA libraries were generated by the Center for Functional Cancer Epigenetics (at Dana-Faber Cancer Institute, Boston, MA) using 1 μg of total RNA and the Illumina TruSeq stranded mRNA Sample Kit. Libraries were sequenced on the Illumina NextSeq 500 Platform at the Molecular Biology Core Facility (at Dana-Faber Cancer Institute). Differential gene expression and downstream analysis were performed using the Visualization Pipeline for RNA-seq pipeline (11) and DEseq2. Single-cell RNA-sequencing was performed at the Translational Immunogenomics Laboratory (at Dana-Faber Cancer Institute) using the 10x Genomics platform. See Supplementary Materials and Methods for further details. All sequencing data are uploaded to Gene Expression Omnibus (GEO, accession number: GSE157413).

For chromatin immunoprecipitation (ChIP), cells were cross-linked with 1% formaldehyde for 10 minutes and chromatin was sonicated to 300–500 bp in either SDS-based or Sarkosyl-based lysis buffer. ChIP was carried out using Protein A/G Dynabeads (Thermo Fisher Scientific) and ChIP DNA was purified using the PCR Purification Kit (Qiagen). ChIP followed by high-throughput sequencing (ChIP-seq) libraries were generated using the ThruPLEX DNA-seq Kit (Rubicon Genomics) according to the manufacturer's instructions and standard 8-bp Illumina primers. Libraries were pooled and sequenced on the Illumina NextSeq 500 Platform at the Molecular Biology Core Facility at Dana-Faber Cancer Institute. See Supplementary Materials and Methods for further details. All sequencing data are uploaded to GEO (accession number: GSE157413).

Rapid immunoprecipitation mass spectrometry of endogenous proteins (RIME) was performed as described previously (12). Briefly, approximately 20 × 10⁶ cells were used per condition and were double fixed in 2 mmol/L disuccinimidyl glutarate for 20 minutes and 1% paraformaldehyde for 10 minutes at room temperature. Cells were lysed and sonicated until chromatin was sheared to 300–600 bp, pulled down with antibody and Protein A/G Dynabeads (Thermo Fisher Scientific), and washed/frozen in 100 mmol/L ammonium hydrogen carbonate solution before subsequent enzymatic digestion with trypsin and peptide desalting by solid-phase extraction on C18. See Supplementary Materials and Methods for further details.

The TMPRSS2-ERG fusion frequently co-occurs with PTEN deletion in both primary and metastatic human prostate cancer samples (Supplementary Fig. S1A). To study the effects of ERG overexpression on prostate tumorigenesis and the mechanism behind ERG dependency, we utilized the well-established Pb-Cre4 Pten^flox/flox Rosa26-ERG genetically engineered mouse model, which has previously been shown to develop highly penetrant adenocarcinoma in the mouse prostate upon PTEN loss and ERG overexpression (2). From these mice, organoid cultures were derived from different genetic alterations, which faithfully recapitulate the in vivo mouse tumor histologies at distinct stages of prostate cancer disease progression (Fig. 1A; refs. 8, 9). WT organoids display normal glandular structures, with distinct basal and luminal cell layers and a hollow lumen with cellular infoldings. Pten^−/− organoids lose the lumen structures with an overgrowth of cells that form solid 3D spheres, reminiscent of prostatic intraepithelial neoplasia (PIN)–like lesions that are seen in vivo. Upon combined PTEN loss and ERG overexpression (Pten^−/− R26-ERG), finger-like protrusions that mimic invasive prostate adenocarcinoma are found and the glandular morphology seen in WT organoids becomes less prominent.

These established organoids can be further genetically altered in vitro using CRISPR/Cas9-mediated gene editing. As such, to address the question of prostate cancer ERG dependence, ERG was knocked out (ERG KO) in Pten^−/− R26-ERG adenocarcinoma organoids (5, 8). We validated KO via IHC and observed complete loss of ERG protein expression in these organoids, as well as the absence of PTEN protein expression in Pten^−/−, Pten^−/− R26-ERG, and ERG KO organoids (Supplementary Fig. S1B). Remarkably, the phenotype of ERG KO organoids is drastically altered, with an almost normal, WT-like glandular morphology being reestablished, but with a limited potential to form larger lumen structures (Fig. 1A). Interestingly, while ERG KO organoids still harbor Pten loss, they do not revert to the Pten^−/− organoid phenotype. Together, these data suggest that ERG is not only important in the initiation of prostate carcinogenesis, but that prostate cancer cells are dependent on continuous ERG overexpression to maintain the transformed phenotype.

To further delve into this dependency on ERG, we performed transcriptome analysis on bulk RNA-sequencing data from the organoids (Fig. 1B). K-means clustering of the top variable genes across all samples reveals that while ERG deletion alters the global transcriptome of the parental Pten^−/− R26-ERG organoids, the ERG KO samples cluster more closely with the cancer organoids than either the WT or Pten^−/− organoids (Fig. 1B). This suggests that although eliminating ERG expression drastically changes the phenotype of these cells, these organoids do not fully revert to either the Pten^−/− or WT state. It is important to note that ERG deletion does result in a significant proportion of genes adopting a WT-like expression profile, which is also observed in a principal component analysis (Supplementary Fig. S1C), indicating that important changes in downstream gene expression pathways may be driving this altered phenotype. Trp63 expression, a key transcription factor regulating the prostate basal cell lineage, is completely abrogated upon ERG overexpression, but is not fully restored following ERG deletion (Supplementary Fig. S1D). Furthermore, transcriptome pathway analyses of the organoid samples reveal a significant enrichment of a prostate luminal signature in Pten^−/− R26-ERG organoids, and this is also seen in primary human prostate cancers with high ERG expression (Supplementary Fig. S1E; ref. 13). Together, this suggests that ERG and AR may be driving a pro-luminal phenotype, as has previously been shown (14), but that continued ERG expression is not necessary in maintaining this lineage state.

Prior reports have shown that disruption of ETS factors in prostate cancer largely affects downstream AR transcriptional output in both cell line and xenograft models (2, 6, 7). To determine whether a similar change is being observed in the prostate organoids, we performed gene set enrichment analysis (GSEA) on the expression data (Fig. 1C). Pten^−/− R26-ERG organoids exhibit a significant enrichment of an AR-dependent gene set versus the WT organoids, whereas ERG deletion significantly reverses and dampens this AR signature. Furthermore, single-cell transcriptomic profiling reveals that each of these organoids form distinct expression clusters, and importantly, Ar gene expression remains unchanged across the different genetic alterations (Fig. 1D). Expression of Fkbp5, a canonical AR target gene, is markedly and uniformly upregulated in Pten^−/− R26-ERG organoids, but is reduced back to WT expression levels upon ERG deletion (Fig. 1D), and a similar pattern is observed in other AR and ERG targets, including Sgk1 and Nkx3-1 (Supplementary Fig. S1D). Together, this shows that ERG modulates AR signaling pathways during prostate tumorigenesis and that elimination of ERG is able to reduce AR transcriptional output despite persistent AR expression.

To determine how ERG regulates AR transcriptional activity, we next defined the AR cistrome in the prostate organoids using ChIP-seq. Similar to what has been reported previously, ERG overexpression was able to drastically reprogram the AR cistrome in the setting of Pten loss (Fig. 2A; ref. 2). Pten loss alone did not substantially alter global AR binding profiles. However, Pten^−/− R26-ERG organoids harbor 9,817 significantly increased AR binding sites across the genome when compared with WT organoids. More than 96% of these sites occurred at putative intronic or intergenic enhancer regions (Supplementary Fig. S2A). Focusing specifically on these gained AR sites, motif analysis revealed a significant enrichment of AP1, androgen response element (ARE), and ERG motifs, in addition to FOXA1 and HOXB13 motifs (Fig. 2A). This is consistent with what is seen in primary human prostate cancer, whereby AR cistrome reprogramming is largely dependent on FOXA1 pioneering activity and the lineage-specific HOXB13 transcription factor (15).

To better understand how ERG is altering the AR cistrome across all organoid samples, we next overlaid these gained AR sites with ERG binding, H3K27ac active enhancer, and ATAC open chromatin profiles (Fig. 2B). These heatmaps revealed that the gained AR sites significantly overlapped with ERG binding sites in Pten^−/− R26-ERG organoids, which subsequently correlate to increased chromatin accessibility and active enhancer marks. This suggests that ERG is directly promoting AR binding to new, open chromatin sites and helping drive active transcription of AR target genes. Remarkably, upon ERG deletion, which resulted in complete loss of ERG binding at these sites, AR binding remained largely unchanged, with fewer than 10% of sites being lost. In addition, ATAC open chromatin profiles were also unchanged and H3K27ac active enhancer marks showed only a modest decrease at these AR-gained sites, which was confirmed by aggregate binding profiles (Fig. 2C). Together, these data indicate that while ERG is important in initially driving AR cistrome reprogramming, its continued expression is largely dispensable in maintaining AR binding and histone acetylation, despite AR transcriptional activity being significantly dampened upon ERG deletion. This is clearly observed at the Fkbp5 gene locus. In Pten^−/− R26-ERG organoids, we saw an increase in AR, ERG, and H3K27ac peaks at putative enhancer regions compared with WT and Pten^−/− organoids (Fig. 2D). Upon ERG deletion, while ERG binding was completely erased and Fkbp5 gene expression was drastically reduced, AR and H3K27ac binding profiles were unchanged across this locus.

Global H3K27ac analysis (not limited to AR-gained sites) revealed that in ERG KO organoids, the active enhancer profiles were drastically shifted to a less cancer-like and more normal prostate-like phenotype (Supplementary Fig. S2B). This implies that while disruption of the ERG oncogene is not altering AR binding itself, the downstream effects and modulation of AR transcriptional activity may be reversing the cancer phenotype. Of the small number of AR binding sites that were altered upon ERG deletion, AR lost sites had decreased ATAC and H3K27ac profiles and were enriched for ERG and AP1, but not ARE motifs (Supplementary Fig. S2C). Conversely, in AR-gained sites upon ERG deletion, there was a significant enrichment of ARE and RUNX, but not ERG motifs (Supplementary Fig. S2C). This suggests that ERG may have AR-dependent and AR-independent effects on transcription, as has previously been described (16). Furthermore, global analysis of ATAC open chromatin profiles across all organoid lines showed only a modest change upon ERG deletion, although ERG motifs did appear in both gained and lost ATAC sites, suggesting that ERG plays a role in regulating these open chromatin states (Supplementary Fig. S2D).

Previous work has shown that ERG can directly bind AR through an AR-interacting motif in the ETS domain and can activate AR's ability to bind DNA, independent of ERG's own DNA binding (5). This provides a potential molecular mechanism for ERG's ability to reprogram the AR cistrome. Our results suggest that while ERG is sufficient to drive AR cistrome reprogramming in the context of prostate carcinogenesis, its continued expression is not necessary in maintaining AR binding or subsequent acetylation of active enhancer histone marks. Nevertheless, ERG is required to maintain AR-dependent transcription, indicating an alternative mechanism of modulating AR activity that is not dependent on the level of AR binding to DNA.

Having determined that AR binding is largely unchanged upon ERG deletion, we hypothesized that AR transcriptional activity is being modulated through changes in DNA-bound AR complexes upon ERG disruption. To evaluate this, we used a mass spectrometry–based proteomics approach termed RIME, which gives an accurate readout of the AR interactome (12). Performing RIME for endogenous AR protein in Pten^−/− R26-ERG and ERG KO organoids, we initially observed an overall decrease in the number of AR bound peptides upon ERG deletion (Fig. 3A). This is validated when visualizing normalized intensity distributions across the organoid samples, where we see a dropout of peptide intensity scores in ERG KO organoids (Supplementary Fig. S3A). This suggests that ERG, a known AR cofactor, is important in maintaining the integrity of DNA-bound AR complexes.

Focusing on the proteins that are preferentially enriched at AR bound complexes in Pten^−/− R26-ERG organoids compared with ERG KO, we observed that ERG itself is one of the most highly enriched proteins, which acts as an internal positive control for this data (Fig. 3A). Other important AR cofactors also emerged, including NCOA3 (SRC-3), TRIM24, and NCOR2, all of which are known to modulate AR transcriptional activity (17–20). In addition, we observed proteins involved in transcriptional control, such as POLR2B and SPT5, being highly enriched in Pten^−/− R26-ERG samples (21, 22). Importantly, the AR protein itself and the histone acetyltransferase, EP300, which is known to be involved in enhancer activation and H3K27 acetylation, were not significantly altered between Pten^−/− R26-ERG and ERG KO organoids (23, 24). Overall, these data add support to the model that ERG disruption is not altering AR binding or histone acetylation at these enhancer sites. However, loss of crucial AR coregulators and components involved in transcription, including RNA polymerase II (RNA Pol II) and elongation factors, is preventing effective transcription of AR target genes upon ERG deletion.

We next individually validated the normalized protein intensities for each of these factors (Supplementary Fig. S3B). Again, while AR protein intensities were unchanged across samples, AR cofactors (NCOA3, TRIM24, and NCOR2) and core transcriptional machinery components (POLR2B and SPT5) had significantly reduced intensities upon ERG deletion. Furthermore, we plotted high-confidence protein hits in Pten^−/− R26-ERG and ERG KO organoids and saw a strong positive correlation between replicates, with the same AR cofactors and transcriptional components being enriched (Supplementary Fig. S3C). Interestingly, in this analysis, NR2F1 (COUP-TF1) was enriched in ERG KO versus Pten^−/− R26-ERG organoids, although this did not reach statistical significance. NR2F1 has previously been shown to be involved in resistance to antiandrogens and may imply a mechanism behind continued cell survival in the setting of ERG deletion (25). Finally, we analyzed RNA Pol II and NCOA3 chromatin binding profiles via ChIP-seq and saw a drastic loss of binding of these factors at AR sites upon ERG deletion, down to levels observed in Pten^−/− organoids (Fig. 3B; Supplementary Fig. S3D). We also observed that these key proteins, including NCOA3 and POLR2B, did not change their expression between the organoid samples, and thus, differences in expression levels could not account for these changes observed in the proteomics and cistrome analyses (Fig. 3C). These results further validate the proteomics data and confirm that ERG deletion leads to a loss of coregulators and transcriptional machinery at AR bound sites.

In summary, our study utilizing a murine organoid culture system has revealed a novel mechanism underlying ERG dependence in prostate adenocarcinoma. ERG overexpression in the setting of PTEN loss is sufficient to reprogram the AR cistrome and drive an AR transcriptional signature that supports prostate tumorigenesis. Upon ERG deletion, while the AR transcriptional program is suppressed and a reversal of the cancer phenotype is observed, AR binding and cognate H3K27ac enhancer marks are largely unchanged. Proteomics analysis reveals a dramatic loss of AR-associated cofactors and basal transcriptional machinery at DNA-bound AR sites across the genome. As such, we propose a model whereby ERG deletion in established prostate adenocarcinoma results in loss of AR transcriptional activity through disruption of AR coregulator complexes (Fig. 3D).

These data shed light on the molecular mechanisms behind ERG oncogene addiction and its ability to modulate and maintain AR activity in prostate cancer cells. With the potential for ERG to be a relevant clinical target, it also allows us to realize the downstream effects of targeting ERG, and how directly inhibiting AR may result in potentially different mechanisms of therapeutic action. Moving forward, it will be important to better understand the effects of ERG disruption in advanced, castration-resistant cases of the disease, and how to more effectively target this key oncogene for therapeutic intervention in patients with prostate cancer.

N. Shah reports receiving grants and funding support from AACR (AACR–AstraZeneca START Grant Award 19-40-12) and CDMRP (Congressionally Directed Medical Research Programs Prostate Cancer Research Program Award W81XWH-16-1-0409). H. Mohammed reports grants, personal fees, nonfinancial support, and other from OHSU (institutional funds) during the conduct of the study. M. Brown reports grants from AACR (AACR–AstraZeneca START Grant Award 19-40-12), CDMRP (Congressionally Directed Medical Research Programs Prostate Cancer Research Program Award W81XWH-16-1-0409), NIH/NCI (NIH Research Program Project Grants P01CA163227), and NIH [NIH core grants P30CA069533 (OHSU Proteomics Shared Resource)] during the conduct of the study and personal fees from GV20 Therapeutics (serves as a scientific advisor to GV20 Therapeutics), Kronos Bio (serves as a scientific advisor to Kronos Bio), H3 Biomedicine (serves as a scientific advisor to H3 Biomedicine), from GTx, Inc. (served as a scientific advisor to GTx, Inc.), and Aleta Biotherapeutics (served as a consultant to Aleta Biotherapeutics) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

N. Shah: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. N. Kesten: Data curation, formal analysis, methodology, writing-review and editing. A. Font-Tello: Validation, methodology, writing-review and editing. M.E.K. Chang: Data curation, formal analysis, validation, visualization, methodology, writing-review and editing. R. Vadhi: Methodology, writing-review and editing. K. Lim: Methodology, writing-review and editing. M.R. Flory: Methodology, writing-review and editing. P. Cejas: Methodology, writing-review and editing. H. Mohammed: Resources, data curation, formal analysis, supervision, validation, methodology, writing-review and editing. H.W. Long: Resources, supervision, methodology, writing-review and editing. M. Brown: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing-review and editing.

We would like to thank Yu Chen, Wouter Karthaus, Brett Carver, and Charles Sawyers (Memorial Sloan Kettering Cancer Center) for kindly and generously providing the organoid lines that were used for this research. The Center for Functional Cancer Epigenetics, the Molecular Biology Core Facility, the Brigham Histopathology Core, and the Translational Immunogenomics Lab at Dana-Farber Cancer Institute helped with experimental designs, data acquisition, and execution. Mass spectrometric runs and analysis were performed by the Oregon Health & Science University (OHSU) Proteomics Shared Resource Facility. This work was supported by funding and grants from 2019 AACR-AstraZeneca START grant, grant no. 19-40-12-SHAH (to N. Shah); Congressionally Directed Medical Research Programs Prostate Cancer Research Program Award W81XWH-16-1-0409 (to N. Shah); NIH Research Program Project grants P01CA163227 (to M. Brown); and NIH core grants P30CA069533 (to OHSU Proteomics Shared Resource).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.