Dual functions of SPOP and ERG dictate androgen therapy responses in prostate cancer

Driver genes with a mutually exclusive mutation pattern across tumor genomes are thought to have overlapping roles in tumorigenesis. In contrast, we show here that mutually exclusive prostate cancer driver alterations involving the ERG transcription factor and the ubiquitin ligase adaptor SPOP are synthetic sick. At the molecular level, the incompatible cancer pathways are driven by opposing functions in SPOP. ERG upregulates wild type SPOP to dampen androgen receptor (AR) signaling and sustain ERG activity through degradation of the bromodomain histone reader ZMYND11. Conversely, SPOP-mutant tumors stabilize ZMYND11 to repress ERG-function and enable oncogenic androgen receptor signaling. This dichotomy regulates the response to therapeutic interventions in the AR pathway. While mutant SPOP renders tumor cells susceptible to androgen deprivation therapies, ERG promotes sensitivity to high-dose androgen therapy and pharmacological inhibition of wild type SPOP. More generally, these results define a distinct class of antagonistic cancer drivers and a blueprint toward their therapeutic exploitation.

Normal cells transform into cancer cells by the acquisition of genetic aberrations in so called driver genes. In some instances, the functional redundancy of mutations in different genes results in a mutually exclusive mutation pattern across tumor genomes because one alteration is sufficient to activate the specific oncogenic pathway. Based on this assumption, bioinformatic tools have been generated to search for functional redundancy of mutated genes in larger cancer genome datasets1,2.

In prostate cancer, recurrent gene fusions involving the ERG transcription factor and point mutations in the ubiquitin ligase adaptor SPOP are two truncal mutations that are mutually exclusively distributed across tumor genomes (Fig. 1a and Supplementary Fig. 1a)3,4,5,6,7,8,9. The underlying cause for this exquisite pattern remains controversial. While earlier reports suggested a functional redundancy between mutant SPOP and ERG based on the finding that mutant SPOP stabilizes the ERG oncoprotein10,11, more recent studies challenge this view by showing descriptive evidence for divergence in tumorigenesis3,12.

Here, we show that SPOP- and ERG-mutant cancer subtypes are driven by antagonistic tumorigenic pathways involving fundamentally different roles of SPOP and androgen receptor signaling levels. The divergence in tumorigenesis is associated with specific susceptibilities to SPOP inhibition and perturbation in the androgen receptor pathway.

Activation of the ERG oncogene and missense mutation in SPOP are synthetic sick

To shed light on these recurrent driver genes’ functional relationship, we assessed the impact of SPOP mutations and ERG activation on the cellular growth of mouse prostate epithelial organoids. To do so, we first established and validated the organoids by the presence of multilayered structure with the expression of CK5 and CK8 in basal and luminal cells, respectively (Supplementary Fig. 1b). In agreement with recent reports, lentiviral-transduced point mutants of SPOP (SPOP-Y87C, SPOP-W131G) or a truncated version of ERG, which typically results from gene fusion with androgen-regulated genes in prostate cancer (ΔERG, amino acids 33–486), promoted cell growth (Fig. 1b and Supplementary Fig. 1c)13,14,15,16. While SPOP mutant organoids displayed a round shape, ERG’s over-expression gave rise to characteristic finger-like protrusions. Surprisingly, the joint expression of both drivers considerably diminished cell growth and reduced finger-like protrusions, implying a synthetic sick relationship between the two genetic alterations. Cytological follow-up analysis revealed reduced proliferation evidenced by reduced Ki-67 and increased p16, P-HP1y positivity, and vacuolization of the cytoplasm compatible with senescence induction (Supplementary Fig. 1d).

We wondered if the observed synthetic sick relationship also applied to established cancer cells from advanced, castration-resistant metastatic disease. Forced expression of mutant SPOP (SPOP-Y87C, SPOP-W131G) promoted 3D growth of ERG fusion-negative LAPC-4 human prostate cancer cells (Supplementary Fig. 2a). The oncogenic effect was paralleled by an increase in the expression of the oncogenic transcription factors MYC and HOXB13 and a decrease in the cell cycle inhibitor p21 as seen in an organoid line derived from SpopF133V-mutant transgenic mice (Supplementary Fig. 2a, b)13. In contrast, we observed the opposite phenotypic and molecular changes in VCaP human prostate cancer cells harboring the recurrent TMPRSS2-ERG fusion (Fig. 1c, d & Supplementary Fig. 2c, d). In this setting, mutant SPOP (SPOP-Y87C, -F102C, -W131G, -F133S) dramatically decreased cancer cells’ proliferation in culture and the growth of xenograft tumor models in vivo. Similarly, to the mouse prostate organoids, possible induction of senescence was evidenced by increased senescence-associated β-galactosidase (SA-β-gal)-positive cells and upregulation of p21 and GDF15 protein levels. In line with this, the transfer of conditioned medium from VCaP cells expressing mutant SPOP (SPOP-Y87C, SPOP-W131G) also reduced the proliferation of parental VCaP cells, indicating a possible contribution of senescence-associated secretory phenotype (SASP) and suggesting that senescence could be one of the possible biological pathways involved in the synthetic sick relationship (Fig. 1d and Supplementary Fig. 2f, j).

Conversely, forced expression of ΔERG significantly reduced the growth of SPOP-Y83C mutant LuCaP-147 patient-derived xenograft (PDX) cancer cells in vivo and in culture (Fig. 1eand Supplementary Fig. 3a, b)17, adding orthogonal support for a synthetic sick relationship between mutant SPOP and ΔERG in advanced prostate cancer. Besides, the over-expression of MYC promoted cancer cell growth in both VCaP and LuCaP-147 cells (Supplementary Fig. 3c, d). The latter finding may suggest that the over-expression system per se is not the underlying cause of the synthetic sick relationship mentioned above.

Next, we wondered if genetic or pharmacologic suppression of ERG signaling may revert the growth suppressing function of mutant SPOP in VCaP cells. Indeed, knockdown of ERG by short-hairpin RNA interference decreased the growth of VCaP control cells and of cells over-expressing wild-type SPOP, while it promoted the growth of cells over-expressing SPOP-W131G (Supplementary Fig. 3e). In addition, low doses of the ETS inhibitor YK-4-279 promoted specifically the growth of VCaP cells overexpressing mutant SPOP (Fig. 1f). We noted a similar effect when VCaP cells were co-treated with a small molecule inhibitor of SPOP (Supplementary Fig. 3f)18. In aggregate, the data support an antagonistic relationship between oncogenic activation of ERG and a loss of SPOP function in prostate cancer cells.

Mutant SPOP-induced androgen receptor signaling antagonizes ERG activity

To assess the underlying molecular biology of the antagonistic relationship between ERG-fused and SPOP-mutants tumors, we interrogated the transcriptomes from the TCGA cohort to nominate differences across these tumor subtypes. Indeed, the unbiased principal component analysis (PCA) revealed significant differences in the transcriptional output (Fig. 2a). The differences were maintained in castration-resistant prostate cancers (CRPC) from the (SU2C) cohort using a single-sample gene-set enrichment analysis approach (Fig. 2b and Supplementary Data 1). Furthermore, derived (PDX) models also retained analog transcriptional differences, as demonstrated by different behavior shown by SPOP-Mutant (LuCaP-78, −147) and ERG-Fused (LuCaP-35, −23.1, VCaP cells) models (Fig. 2b and Supplementary Data 1).

In SPOP-mutant prostate cancer, several dysregulated SPOP substrates (e.g., NCOA3, TRIM24, BET proteins) have been shown to boost the AR pathway leading ultimately to high levels of AR target genes (Supplementary Fig. 4a)3,13,19,20,21,22,23,24,25,26. In contrast, ERG-fused cancer cells typically express lower levels of AR target genes as illustrated by the widely adopted AR score (Supplementary Fig. 4a)3. We further performed differential expression analysis between the two tumor subtypes to get more insights into this different behavior. Gene-set enrichment analysis resulted in a clear upregulation of the canonical androgen response pathway in SPOP-mutant vs. ERG-fused tumors, as defined by the respective Hallmark gene-set, curated by the Molecular Signature Database (MSigDB) (Supplementary Fig. 4b). In line with the divergence of transcriptome profile identified in the PCA plot of Fig. 2a, a clear division between SPOP-mutant and ERG-fused tumors was also reported in their respective cistrome counterpart25. By re-analyzing ChIP-Seq data, we could determine that most differentially bound regions between both tumor types are characterized by increased AR-binding in SPOP-mutant tumors (Fig. 2c and Supplementary Data 2).

Next, we analyzed the transcriptome changes of the VCaP cells overexpressing SPOP mutants (SPOP-MTs; SPOP-Y87C, -F102C, -F133S). The unbiased hallmark analysis showed a dramatic increase in the androgen response, recapitulating the changes identified in primary prostate cancer (Supplementary Figs. 4b and 5a, b). Based on these results, we posited that differential levels of androgen receptor (AR) signaling in SPOP-mutant vs. ERG-fused cancers might be at the root of the incompatibility between the driver events. Thus, we analyzed, in particular, AR- and ERG-related transcription in VCaP cells and generated custom signatures using ChIP-seq data and matched RNA-seq samples (Supplementary Data 3)27. As expected, SPOP-MTs increased the transcription of genes bound by AR and induced by its ligand dihydrotestosterone (DHT), whereas genes bound by AR and repressed by DHT were further reduced (Fig. 3a, Supplementary Fig. 5c, and Supplementary Data 3–5). Remarkably, we observed the opposite effect on genes bound only by ERG. Mutant SPOP downregulated ERG-induced genes (e.g., MYC) and upregulated ERG-repressed genes, respectively (Fig. 1d). In line with these findings, gene ontology analysis of AR-ERG co-bound gene signature in VCaP cells indicated that the most striking transcriptional changes were linked to cellular differentiation and cell cycle arrest that are directly induced by DHT and repressed by ERG (e.g., HOXA genes, CDKN1A/p21, Fig. 1d, Fig. 3b, and Supplementary Fig. 5d). To reduce the number of genes falling within our custom signatures, we used a particularly restrictive approach and considered co-bound only genes where AR and ERG-binding sites overlapped for at least 1 bp. As a result, some genes (e.g., CDKN1A/p21) that are bound both by AR and ERG in their promoter region, but bindings of which do not overlap are not included in this category despite being bona fide co-bound targets.

The dramatic upregulation of this gene set was paralleled by downregulation of cell cycle genes (e.g., E2F and MYC targets), implying a direct link between the induction AR/ERG co-bound genes, the repression of ERG targets, cell differentiation, and the synthetic sick relationship of ERG and mutant SPOP (Fig. 1d, Supplementary Fig. 5a–e, and Supplementary Data 6,7). The relationship of AR- and ERG-related custom signatures to the hallmark gene sets are highlighted in Fig. 3c in a two-dimensional network. Moreover, independently generated signatures of senescence-associated transcripts were enriched in VCaP overexpressing SPOP mutants, further corroborating our data of a senescence-induced cell cycle arrest (Supplementary Fig. 5f)28,29.

Conversely, we assessed the consequence of ERG overexpression in LNCaP cells under low DHT levels where mutant SPOP triggers AR signaling and tumor growth (Supplementary Fig. 6a, b and Supplementary Data 8 and 9)19,23. Over-expression of ΔERG in this setting robustly reverted the induction of signatures related to cell proliferation (e.g., E2F and MYC targets) and AR signaling. Taken together, the data imply a mutual incompatibility of mutant SPOP-induced AR signaling and the function of the ERG oncogene.

Next, we verified if corresponding transcriptional changes were found in clinical tissue samples. Indeed, ERG-regulated genes culled from VCaP cells were upregulated in ERG-fused and downregulated in SPOP-mutant primary tumors (Fig. 3d, Supplementary Fig. 6c, and Supplementary Data 10, 11)3. Notably, the most striking changes between the two groups were found again in the AR/ERG co-bound gene set in primary prostate cancers (Supplementary Fig. 6d and Supplementary Data 12, 13)3,6. The results underscore both the relevance of our cell culture-based data and highlight the transcriptional differences among ERG- and SPOP-driven tumors.

ZMYND11 is a de novo SPOP substrate

Using tandem mass tag (TMT)-based quantitative mass spectrometry, we set out to search for SPOP substrates that may influence the activity of AR and ERG and thereby may cause to the synthetic sick relationship between mutant SPOP and ERG in VCaP cells overexpressing mutant SPOP (SPOP-MTs; SPOP-Y87C, -F102C, -W131G, Fig. 4a and Supplementary Data 14). As recurrent loss-of-function SPOP mutants impair substrate ubiquitylation and proteasomal degradation, we searched for proteins the expression levels of which increase without a concomitant increase in mRNA levels (Fig. 4b, Supplementary Fig. 7a, and Supplementary Data 15). Overall, we noted a strong correlation of protein with mRNA expression changes with consistent changes of our AR and ERG custom signatures at the protein level (Fig. 4b and Supplementary Fig. 7b, c). In addition, we found a marked upregulation of the known SPOP substrate and AR activator TRIM24 at the protein level (Figs. 1d and 4a; Supplementary Data 14)22,23 and subsequently assessed if TRIM24 and more generally AR is implicated in the synthetic sick relationship between mutant SPOP and ERG. Indeed, knockdown of TRIM24 by two short-hairpin RNAs partially reverted the growth inhibition mediated by mutant-SPOP in VCaP cells and reduced AR signaling (Supplementary Fig. 7d–f), while over-expression of AR was sufficient to decrease cellular growth (Supplementary Fig. 7g, h).

The most striking upregulation was noted for the bromodomain histone reader ZMYND11 (Fig. 4b). In line with a SPOP substrate, wild type SPOP bound and decreased the expression of HA-ZYMND11 in a proteasome-dependent manner (Fig. 4c, d). We found two degron sequences that were required for efficient SPOP-mediated ubiquitylation and protein degradation (Fig. 4e–g and Supplementary Fig. 8a.). As expected, SPOP mutants failed to bind and adequately ubiquitylate HA-ZMNYD11-WT (Fig. 4h–j)10,11,20,21,22,24,25,26. Finally, we confirmed that the expression of mutant SPOP prolonged the half-life of endogenous ZMYND11 in VCaP cells and upregulated ZMYND11 expression in other prostate cancer cells (Fig. 4k and Supplementary Fig. 8b).

ZMYND11 induces AR signaling pathway and represses ERG activity

Next, we assessed if ZMYND11 protein upregulation also contributed to the synthetic sick relationship. In support, forced expression of the degron-deficient variants of ZMYND11 (HA-ZMYND11-DMT1/DMT2) was sufficient to diminish the growth of VCaP cells (Fig. 5a, band Supplementary Fig. 8c), while knockdown of ZMYND11 partially reverted the growth inhibition mediated by mutant SPOP (Fig. 5c).

We postulated that ZMYND11 upregulation could contribute to the synthetic sick relationship by repressing the ERG oncogene’s transcriptional activity or enhancing AR signaling. To this end, expression changes induced by HA-ZMYND11-DMT2 largely overlapped with genes perturbed by mutant SPOP while the opposite was noted when ZMYND11 expression was reduced by RNA interference (Fig. 5d, Supplementary Fig. 8d, and Supplementary Data 16). In comparison to mutant SPOP, AR and ERG target genes were similarly dysregulated by HA-ZMYND11-DMT2 (Fig. 5e and Supplementary Data 17, 18). As the PWWP domain of ZMYND11 has been involved in the regulation of transcription through its ability to bind H3K36me3 histone marks30, we tested this domain’s contribution to the overall transcriptional output. Indeed, the PCA of VCaP cells over-expressing either HA-ZMYND11-DMT2 or a PWWP domain deficient mutant (W294A) revealed a major contribution of this domain to the ZMYND11 induced transcriptional changes (Supplementary Fig. 8e).

Subsequently, we mapped the genomic occupancy of ZMYND11 in VCaP cells expressing the SPOP-Y87C mutant by chromatin immunoprecipitation sequencing (ChIP-seq) and found an enrichment of ZMYND11-binding sites at promoter regions controlling ERG-induced genes (e.g., MYC,) and AR/ERG co-bound genes (e.g., p21/CDKN1A) (Fig. 5f, g, Supplementary Fig. 9a–e, and Supplementary Data 19–22). The data imply a critical enhancer function of ZMYND11 in boosting AR signaling and repressing ERG signaling downstream of mutant SPOP.

Wild type SPOP is required for ERG oncogenic function

We reasoned that ERG-driven tumors might require wild type SPOP to degrade ZMYND11 and unlock ERG’s oncogenic function. In support, over-expression of wild type SPOP increased the 3D growth of mouse prostate epithelial organoids and VCaP cells only when ERG was over-expressed (Fig. 1b, Supplementary Figs. 2c and 10a, b). Remarkably, ERG-fused human tumor tissues also displayed the highest SPOP mRNAs levels (Fig. 6a). Thus, we wondered if ERG itself may directly upregulate SPOP transcription to support its own oncogenic activity. Indeed, mining ERG ChIP-seq data in VCaP cells revealed ERG bindings sites in the promoter region of SPOP (Supplementary Fig. 10c). Moreover, knockdown of ERG reduced SPOP protein levels in VCaP cells, while forced expression of a ΔERG led to the upregulation of SPOP mRNA and protein levels in PC3 cells (Fig. 6b, Supplementary Figs. 3eand 10d).

We then asked if the elevated SPOP levels in the context of forced ΔERG expression have a functional impact on the oncogenic activity of ΔERG in the androgen-independent PC3 cells, in which ERG promotes tumor cell invasion31. Indeed, the reduction of SPOP levels by RNA interference reduced the ability of ΔERG to invade into Matrigel (Fig. 6c). Similarly, knockdown of SPOP in VCaP cells reduced cell growth in 3D cell culture and impaired ERG-mediated gene transcription (Supplementary Fig. 10e, f and Supplementary Data 23 and 24). In accordance with the ability of mutant SPOP to repress the function of endogenous wild type SPOP in a dominant-negative manner, the over-expression of mutant SPOP (SPOP-Y87C, -F102C, -W131G, -F133S) phenocopied the effect of SPOP knockdown on ERG-mediated invasion in PC3 cells (Supplementary Fig. 10g, h). In agreement with the established repressive function of ZMYND11 on ERG, we found that over-expression of HA-ZMNYD11-DM2 was sufficient to repress ERG-induced invasion and established target genes in PC3 cells (Fig. 6d, e). Taken together, the data imply the existence of a positive feed-forward loop, in which ΔERG promotes the expression of SPOP to sustain its oncogenic activity.

ERG and mutant SPOP trigger different responses to therapeutic interventions

Based on the differences mentioned above in tumorigenesis, we speculated that ERG or mutant SPOP could also trigger different therapeutic responses. In light of the dependency of ERG-driven tumors on wild type SPOP function, we hypothesized that ERG-fusion-positive cells might be particularly sensitive to pharmacological inhibition of SPOP. We analyzed the response of the SPOP small molecule inhibitor compound 6b (SPOP-i) in ERG-fused, SPOPmutant, and other prostate cancer cell lines and patient-derived xenograft models (PDX)18. The SPOP inhibitor increased the protein but not the mRNA levels of established SPOP substrates and ZMNYD11, while the related inactive analog compound 6c did not (Supplementary Fig. 11a–c). The latter did also not exert any activity in 3D culture models (Supplementary Fig. 11d). In agreement with our previous results, we found that ERG-fused cells (VCaP, LuCaP-23.1, −35) were more sensitive to SPOP-i than ERG-negative cells (22Rv1, LNCaP, PC3), while SPOP mutant cells (LuCaP-78, −147) were particularly insensitive in 3D culture models and in xenograft tumor models in vivo (Fig. 7a–f and Supplementary Fig. 11e). We further validated our results in the mouse prostate epithelial organoids and confirmed the increased sensitivity of ΔERG-expressing cells to SPOP inhibition in this isogenic system (Fig. 7g).

Given the notion that wild type SPOP dampens AR function in the context of ERG to sustain tumor growth, we asked if VCaP cells are particularly susceptible to increased DHT levels. Indeed, exposure to a high-dose of testosterone in vivo or DHT in vitro induced similar molecular changes as for the over-expression of mutant SPOP and greatly suppressed the growth of ERG-fusion-positive cells but not of SPOP mutant cells in vitro and in vivo (Figs. 1c and 8a–f; Supplementary Fig. 12a–e and Supplementary Data 25 and 26). Moreover, signatures of senescence-associated transcripts were also enriched in VCaP cells upon treatment with DHT, further corroborating our data of a senescence-induced cell cycle arrest (Supplementary Fig. 12f). Strikingly, the sensitivity to SPOP-i and to high testosterone in vivo correlated well with ERG protein expression levels in the respective ERG-fusion-positive cell line and PDX model (Fig. 8g, h). The data suggest a therapeutic opportunity for SPOP inhibition or high-dose androgen therapy in prostate cancers that express high levels of ERG.

Conversely, and because SPOP mutant cancers are driven predominantly by androgen signaling and consequently display high-level activation of AR-related transcripts in human tumor tissues, we speculated that these tumors might be particularly susceptible to androgen deprivation or antiandrogen therapies (ADT) (Supplementary Fig. 4c). Indeed, the prevalence of SPOP mutations in primary tumors -and tumors that had progressed after initial surgery or radiotherapy- is consistently higher as compared to tumors that had become resistant to subsequent ADT (also referred to as castration-resistant prostate cancer, CRPC, Supplementary Fig. 13a). In line with the notion that this difference may be related to a better response of SPOP mutant tumors to ADT, SPOP mutant tumors display a trend towards better overall survival despite progressing faster after initial therapy (Fig. 9a, b). To functionally analyze androgen deprivation or the antiandrogen enzalutamide response, we chose to ectopically express different SPOP variants and ΔERG in the androgen-dependent human LAPC4 prostate cancer cells that are wild-type for both driver genes. In accordance with the clinical observation, the presence of mutant SPOP (SPOP-Y87C, SPOP-W131G) rendered LAPC4 cells more susceptible to either ADT or enzalutamide in comparison to cells expressing control vector (Fig. 9c and Supplementary Fig. 13b).

In contrast, ΔERG rendered the same cells more resistant to enzalutamide. In line with the previous findings in VCaP and LuCaP-147 cells, ΔERG expression rendered LAPC4 cells susceptible to high levels of DHT, while mutant SPOP had the opposite effect (Supplementary Fig. 13b). Taken together, the different responses to established and experimental therapeutic modalities observed between mutant SPOP and ERG add further credence to their divergent roles of the AR pathway related to tumorigenesis.

Although multiple studies over recent years have uncovered different genetically-defined subtypes of primary prostate cancer, their biological understanding and therapeutic implications remain a largely unexplored territory. Here, we report two diametrically different paths toward tumorigenesis triggered by either highly recurrent missense mutation in SPOP or gene fusion involving the ERG oncogene. Importantly, wild-type SPOP emerges as a critical component that enforces oncogenic ERG signaling in part through dampening AR activity, while mutant SPOP drives tumorigenesis through activation of AR signaling. Moreover, several studies have previously highlighted the importance of AR target genes in the context of SPOP mutants and ERG-positive tumors19,32,33. Based on the incompatibility between the two tumor subtypes, our work enabled the development of specific custom signatures related to AR and ERG transcript that are necessary to drive proliferation and tumorigenesis in the context of ERG-positive and SPOP-mutants tumors. In addition, we show that the bromodomain histone reader ZMYND11 is a SPOP substrate implicated downstream of SPOP in the opposing regulation of the ERG and AR pathway in the two tumor subtypes (Fig. 10). The AR and ERG pathways have been previously reported to have a partially antagonistic relationship33,34, further corroborating our findings.

As activation of the androgen receptor by androgens represents a key lineage specific oncogenic pathway in prostate cancer, androgen deprivation/antagonization therapies (ADT) remain the uniform treatment modality up to this very day. That said, the responses to ADT are highly variable and may last from a few weeks up to many years. Here, we provide functional evidence that pre-existing prostate cancer founder mutations influence the treatment response. Most notably, SPOP mutations promote susceptibility to androgen deprivation therapies. In agreement with our findings, earlier reports have shown the underrepresentation of SPOP mutant tumors in cohorts of castration-resistant disease and a more favorable response to the abiraterone and enzalutamide35,36.

Conversely, we show that the ERG oncogene’s presence increases the susceptibility of tumor cells to high-dose androgen therapy, while cells expressing mutant SPOP remain largely unaffected. This is of clinical interest because testosterone treatment of patients with the advanced castration-resistant disease has recently shown to trigger anti-tumor responses in around one-third of the patients37. It is tempting to speculate that these insights may help to discern responders from non-responders.

In addition, we provide evidence that the antagonistic relationship between mutant SPOP and ERG may be used towards the development of new therapeutic avenues. More specifically, we show that ERG-driven cancer cells are particularly sensitive to the inhibition of wild-type SPOP using recently developed small molecule inhibitors18. Our preclinical data suggest that SPOP inhibition may be effective in clinical settings where ERG is robustly expressed (e.g., neo-adjuvant setting or early metastatic disease).

Our results generally identify another paradigm for antagonistic driver genes in prostate cancer that has recently emerged for other cancer types38,39,40. In analogy to prostate cancer, truncal point mutations in DNMT3A and gene fusions in PML-RARA are mutually exclusive drivers in acute myeloid leukemia (AML). Similar to SPOP, intact DNMT3A has been found to be critical for PML-RARA-driven leukemia (Supplementary Fig. 14a, b)41,42. Importantly, we demonstrate here for prostate cancer that the concept of antagonistic driver genes can be exploited to identify therapeutic opportunities.