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 up-regulates 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.

genes result in a mutually-exclusive mutation pattern across tumor genomes because one 40 alteration is sufficient to activate the specific oncogenic pathway. Based on this 41 assumption, bioinformatic tools have been generated to search for functional redundancy 42 of mutated genes in larger cancer genome data sets 1,2 . 43 In prostate cancer, recurrent gene fusions involving the ERG transcription factor and 44 point mutations in the ubiquitin ligase adaptor SPOP are two truncal mutations that are 45 mutually exclusively distributed across tumor genomes ( Fig. 1a and Supplementary Fig.  46 1a) 3-7 . The underlying cause for this exquisite pattern remains controversial. While earlier 47 reports suggested a functional redundancy between mutant SPOP and ERG based on the 48 revert the growth suppressing function of mutant SPOP in VCaP cells. Indeed, knockdown 95 of ERG by short-hairpin RNA interference decreased the growth of VCaP control cells and 96 of cells over-expressing wild-type SPOP, while it promoted the growth of cells over-97 expressing SPOP-W131G ( Supplementary Fig. 3e). In addition, low doses of the ETS 98 inhibitor YK-4-279 promoted specifically the growth of VCaP cells over-expressing 99 mutant SPOP (Fig. 1f). We noted a similar effect when VCaP cells were co-treated with a 100 small molecule inhibitor of SPOP ( Supplementary Fig. 3f) 16 . In aggregate, the data support 101 an antagonistic relationship between oncogenic activation of ERG and a loss of SPOP 102 function in prostate cancer cells. 103 104

Mutant SPOP-induced androgen receptor signaling antagonizes ERG activity 105
To assess the underlying molecular biology of the antagonistic relationship between 106 ERG-fused and SPOP-mutants tumors, we interrogated the transcriptomes from the TCGA 107 cohort to nominate differences across these tumor subtypes. Indeed, the unbiased principal 108 component analysis (PCA) revealed major differences in the transcriptional output (Fig.  109 2a). The differences were maintained in castration-resistant prostate cancers (CRPC) from 110 the (SU2C) cohort using a single-sample gene-set enrichment analysis approach (Fig. 2b). 111 Furthermore, derived (PDX) models also retained analogue transcriptional differences, as 112 demonstrated by different behavior shown by SPOP-Mutant (LuCaP-78, -147) and ERG-113 Fused (LuCaP-35, -23.1, VCaP cells) models (Fig. 2b). 114 In SPOP-mutant prostate cancer, several dysregulated SPOP substrates (e.g. NCOA3, 115 TRIM24, BET proteins) have been shown to boost the AR pathway leading ultimately to 116 high levels of AR target genes ( Supplementary Fig. 4a) 3,11,17-24 . In contrast, ERG-fused 117 6 cancer cells express typically lower levels of AR target genes as illustrated by the widely 118 adopted AR score ( Supplementary Fig. 4a) 3 . To get more insights into this different 119 behavior, we further performed differential expression analysis between the two tumor 120  Fig. 4b, 5a and 5b). Based on these results, we 132 posited that differential levels of androgen receptor (AR) signaling in SPOP-mutant versus 133 ERG-fused cancers might be at the root of the incompatibility between the driver events. 134 Thus, we analyzed in particular AR-and ERG-related transcription in VCaP cells, and 135 generated a custom signatures using ChIP-seq data and matched RNA-seq samples 136 (Supplementary Table 1) 25 . As expected, SPOP-MTs increased the transcription of genes 137 bound by AR and induced by its ligand dihydrotestosterone (DHT), whereas genes bound 138 by AR and repressed by DHT were further reduced (Fig. 3a, Supplementary Fig. 5c, and  139 Supplementary Table 1). Remarkably, we observed the opposite effect on genes bound 140 only by ERG. Mutant SPOP downregulated ERG-induced genes (e.g. MYC) and 141 upregulated ERG-repressed genes, respectively (Fig. 1d). In line with these findings, gene 142 ontology analysis of AR-ERG co-bound gene signature in VCaP cells indicated that the 143 most striking transcriptional changes were linked to cellular differentiation and cell cycle 144 arrest that are directly induced by DHT and repressed by ERG (e.g. HOXA genes, 145 CDKN1A/p21, Fig. 1d, Fig. 3b, Supplementary Fig. 5d). To reduce the number of genes 146 falling within our custom signatures, we used a particularly restrictive approach and 147 considered as co-bound only genes where AR and ERG binding sites were overlapping for 148 at least 1bp. As a result, some genes (i.e. CDKN1A/p21) which are bound both by AR and 149 ERG in their promoter region, but which bindings do not overlap, are not included in this 150 category despite being bona fide co-bound targets. 151 The dramatic upregulation of this gene set was paralleled by a downregulation of cell 152 cycle genes (e.g. E2F and MYC targets), implying a direct link between the induction 153 AR/ERG co-bound genes, the repression of ERG targets, cell differentiation and the 154 synthetic sick relationship of ERG and mutant SPOP (Fig. 1d, Supplementary Fig. 5a-e). 155 The relationship of AR-and ERG-related custom signatures to the hallmark gene sets are 156 highlighted in Fig. 3c in a two-dimensional network. Moreover, independently generated 157 signatures of senescence-associated transcripts were enriched in VCaP overexpressing 158 SPOP mutants, further corroborating our data of a senescence-induced cell cycle arrest ( 159 Supplementary Fig. 5f) Fig. 6a,b) 17,21 . Over-expression of ΔERG in this setting robustly reverted 163 the induction of signatures related to cell proliferation (e.g. E2F and MYC targets) and AR 164 signaling. Taken together, the data implies a reciprocal incompatibility of mutant SPOP 165 induced AR signaling and the function of the ERG oncogene. 166 Next, we verified if corresponding transcriptional changes were found in clinical tissue 167 samples. Indeed, ERG-regulated genes culled from VCaP cells were up-regulated in ERG-168 fused and down-regulated in SPOP-mutant primary tumors (Fig. 3d, Supplementary Fig.  169 4c) 3 . Importantly, the most striking changes between the two groups were found again in 170 the AR/ERG co-bound gene set in primary prostate cancers ( Supplementary Fig. 4d) 3,6 . 171 The results underscore both the relevance of our cell culture-based data and highlight the 172 transcriptional differences among ERG-and SPOP-driven tumors. 173 174 ZMYND11 is a de novo SPOP substrate 175 Using tandem mass tag (TMT)-based quantitative mass-spectrometry, we set out to 176 search for SPOP substrates that may influence the activity of AR and ERG and thereby 177 may cause to the synthetic sick relationship between mutant SPOP and ERG in VCaP cells 178 overexpressing mutant SPOP (SPOP-MTs; SPOP-Y87C, -F102C, -W131G, Fig. 4a). 179 Because recurrent loss-of-function SPOP mutants impair substrate ubiquitylation and 180 proteasomal degradation, we searched for proteins which expression levels increase 181 without concomitant increase in mRNA levels (Fig. 4b, Supplementary Fig. 7a). Overall, 182 we noted a strong correlation of protein with mRNA expression changes with consistent 183 changes of our AR and ERG custom signatures at the protein level (Fig. 4b Fig. 7d-f), while over-expression of 190 AR was sufficient to decrease cellular growth ( Supplementary Fig. 7g, h). 191 The most striking upregulation was noted for the bromodomain histone reader 192 ZMYND11 (Fig. 4b). In line with a SPOP substrate, wild type SPOP bound and decreased 193 the expression of HA-ZYMND11 in a proteasome-dependent manner (Fig. 4c, d). We 194 found two degron sequences that were required for efficient SPOP-mediated ubiquitylation 195 and protein degradation ( Next, we assessed if ZMYND11 protein upregulation also contributed to the synthetic 203 sick relationship. In support, forced expression of the degron-deficient variants of 204 ZMYND11 (HA-ZMYND11-DMT1/DMT2) was sufficient to diminish the growth of 205 VCaP cells (Fig. 5a, b, Supplementary Fig. 8c), while knockdown of ZMYND11 partially 206 reverted the growth inhibition mediated by mutant SPOP (Fig. 5c). 207 We postulated that ZMYND11 up-regulation could contribute to the synthetic sick 208 relationship by repressing the transcriptional activity of the ERG oncogene or enhancing 209 AR signaling. To this end, expression changes induced by HA-ZMYND11-DMT2 largely 210 overlapped with genes perturbed by mutant SPOP while the opposite was noted when 211 ZMYND11 expression was reduced by RNA interference (Fig. 5d and Supplementary Fig.  212 8d). In comparison to mutant SPOP, AR and ERG target genes were similarly dysregulated 213 by HA-ZMYND11-DMT2 (Fig. 5e). Because the PWWP domain of ZMYND11 has been 214 involved in the regulation of transcription through its ability to bind H3K36me3 histone 215 marks 28 , we tested the contribution of this domain to the overall transcriptional output.

Wild type SPOP is required for ERG oncogenic function 228
We reasoned that ERG-driven tumors might require wild type SPOP to degrade 229 ZMYND11 and thereby unlock the oncogenic function of ERG. In support, over-230 expression of wild type SPOP increased the 3D growth of mouse prostate epithelial 231 organoids and VCaP cells only when ERG was over-expressed (Fig. 1b, Supplementary 232 Fig. 2c, 10a, b). Remarkably, ERG-fused human tumor tissues displayed also the highest 233 SPOP mRNAs levels (Fig. 6a). Thus, we wondered if ERG itself may directly upregulate 234 SPOP transcription to support its own oncogenic activity. Indeed, mining ERG ChIP-seq 235 data in VCaP cells revealed ERG bindings sites in the promoter region of SPOP 236 ( Supplementary Fig. 10c). Moreover, knockdown of ERG reduced SPOP protein levels in 237 VCaP cells, while forced expression of a ∆ERG led to the upregulation of SPOP mRNA 238 and protein levels in PC3 cells (Fig. 6b, Supplementary Fig. 3e, and 10d). 239 We then asked if the elevated SPOP levels in the context of forced ΔERG expression 240 have a functional impact on the oncogenic activity of ΔERG in the androgen-independent 241 PC3 cells, in which ERG promotes tumor cell invasion 29 . Indeed, the reduction of SPOP 242 levels by RNA interference reduced the ability of ΔERG to invade into matrigel (Fig. 6c).  Supplementary Fig. 12f). Strikingly, the sensitivity to SPOP-i and to high testosterone in 280 vivo correlated well with ERG protein expression levels in the respective ERG-fusion 281 positive cell line and PDX model (Fig. 8g, h). The data suggests a therapeutic opportunity 282 for SPOP inhibition or high-dose androgen therapy in prostate cancers that express high 283 levels of ERG. 284 Conversely, and because SPOP mutant cancers are driven predominantly by androgen 285 signaling and consequently display high-level activation of AR-related transcripts in 286 human tumor tissues, we speculated that these tumors may be particularly susceptible to 287 androgen deprivation or anti-androgen therapies (ADT) ( Supplementary Fig. 4c). Indeed, 288 the prevalence of SPOP mutations in primary tumors -and tumors that had progressed after 289 initial surgery or radiotherapy-is consistently higher as compared to tumors that had 290 become resistant to subsequent ADT (also referred as castration-resistant prostate cancer, 291 CRPC, Supplementary Fig. 13a). In line with the notion that this difference may be related 292 to a better response of SPOP mutant tumors to ADT, SPOP mutant tumor display a trend 293 towards better overall survival despite progressing faster after initial therapy ( Fig. 9a,b). 294 To functionally analyze the response of androgen deprivation or the anti-androgen 295 enzalutamide, we chose to ectopically expressed different SPOP variants and ΔERG in the 296 androgen-dependent human LAPC4 prostate cancer cells that are wild-type for both driver 297 genes. In accordance with the clinical observation, the presence of mutant SPOP (SPOP-298 Y87C, SPOP-W131G) rendered LAPC4 cells more susceptible to either ADT or 299 enzalutamide in comparison to cells expressing control vector (Fig. 9c importance of AR target genes in the context of SPOP mutants and ERG positive tumors 317 17,30,31 . Based on the incompatibility between the two tumor subtypes, our work enabled 318 the development of specific custom signatures related to AR and ERG transcript that are 319 necessary to drive proliferation and tumorigenesis in the context of ERG-positive and 320 SPOP-mutants tumors. In addition, we show that the bromodomain histone reader 321 ZMYND11 is a SPOP substrate implicated downstream of SPOP in the opposing 322 regulation of the ERG and AR pathway in the two tumor subtypes (Fig. 10). The AR and 323 ERG pathways have been previously reported to have a partially antagonistic 324 relationship 31,32 , further corroborating our findings. 325 Because activation of the androgen receptor by androgens represents a key lineage 326 specific oncogenic pathway in prostate cancer, androgen deprivation/antagonization 327 therapies (ADT) remain the uniform treatment modality up to this very day. That said, the 328 responses to ADT are highly variable and may last from a few weeks up to many years. 329 Here, we provide functional evidence that pre-existing prostate cancer founder mutations 330 influence the treatment response. Most notably, SPOP mutations promote susceptibility to 331 androgen deprivations therapies. In agreement with our findings, earlier reports have 332 shown underrepresentation of SPOP mutant tumors in cohorts of castration-resistant 333 disease and a more favorable response to the abiraterone and enzalutamide 33,34 . 334 Conversely, we show that the presence of the ERG oncogene increases the 335 susceptibility of tumor cells to high-dose androgen therapy, while cells expressing mutant 336 SPOP remain largely unaffected. This is of clinical interest because testosterone treatment 337 of patients with advanced castration-resistant disease has recently shown to trigger anti-338 tumor responses in around one third of the patients 35 . It is tempting to speculate that these 339 insights may help to discern responders from non-responders. 340 In addition, we provide evidence that the antagonistic relationship between mutant 341 SPOP and ERG may be used towards the development of new therapeutic avenues. More 342 specifically, we show that ERG-driven cancer cells are particularly sensitive to the 343 inhibition of wild-type SPOP using recently developed small molecule inhibitors 16 . Our 344 preclinical data suggests that SPOP inhibition may be effective in clinical settings where 345 ERG is robustly expressed (e.g. neo-adjuvant setting or early metastatic disease). 346 More generally, our results identify another paradigm for antagonistic driver genes in 347 prostate cancer that has recently emerged also for other cancer types [36][37][38]

Identification of ERG and AR related gene signatures 605
We retrieved RNA-seq data from GEO Dataset GSE83652 14 to identify transcriptional 606 perturbations in VCaP cells following treatment with DHT or following silencing of ERG. 607 To this purpose we completely reprocessed samples SRR3713255-57, SRR3713267-72 608 using STAR and DESeq2 as previously described for VCaP cells. In addition, to identify 609 direct targets, we integrated information relative to AR and ERG chromatin binding sites, 610 which we derived from GEO Dataset GSE28950 25 . To maximize the number of peaks and 611 to reduce false negatives, we merged results of experiments performed at different time 612 points, namely 2h and 18h after DHT exposure. De-multiplexed reads were aligned to hg38 613 release of the human reference genome using bwa-mem 54 (0.7.15). MACS 55 (v.2.1.0) was 614 used to perform peak calling procedure using a cutoff FDR q-value of 0.01 and a mappable 615 genome size optimized for hg38 equal to 2.9 gigabases. Downstream analysis was 616 performed in R statistical environment. We identified binding sites overlapping promoters 617 by using bedtools 56 . 618 Promoters were defined as DNA regions ranging from 1500 bp upstream to 500 bp 619 downstream of Transcription Start Sites (TSSs). 620 To discriminate between ERG-and AR-specific transcriptional responses we stratified 621 genes into three main classes: genes whose promoter regions are bound by AR but not by 622 ERG, genes whose promoters are bound by ERG but not by AR, and finally, genes whose 623 promoters are co-bound by both AR and ERG. AR bound only genes were further 624 subdivided into two sets, those being significantly (FDR<0.05) induced following DHT 625 treatment and those being significantly repressed. A similar approach was applied to ERG 626 bound only genes, where genes were subdivided into ERG-induced and ERG-repressed 627 gene-sets, if they were respectively down or up-regulated following ERG silencing. To be 628 more stringent in the definition of AR-specific and ERG-specific signatures, we excluded 629 genes from the ERG-induced set that were also significantly up-regulated following DHT 630 treatment, vice-versa we excluded ERG-repressed genes that were significantly down-631 regulated following DHT-treatment. The same criteria were applied for DHT-specific 632 gene-sets. Finally, defined an additional gene-set (DHT-induced/ERG-repressed) 633 consisting of genes being co-bound by AR and ERG in their promoter region, which were 634 significantly up-regulated following DHT treatment but also significantly upregulated 635 following ERG-silencing. All gene-sets are detailed in Supplementary Table 1 (v.2.1.0) was used to perform peak calling procedure using a cutoff FDR q-value of 0.01 685 and a mappable genome size optimized for hg38 equal to 2.9 gigabases. Downstream 686 analysis was performed in R statistical environment. ChIPseeker 63 was used to annotate 687 peaks and to represent the distribution of ZMYND11 binding sites relative to Transcription 688 Start Sites (TSSs). The R package chipenrich 64 was subsequently used to determine 689 enrichment or depletion of ZMYND11 peaks in regions surrounding TSSs of genes that are 690 included in Hallmarks or custom gene-set collections. Surrounding regions were defined 691 as ranging from 5kb upstream to 5kb downstream of their TSSs (locusdef = 5kb), which is 692 in line with the overall behavior of ZMYND11 binding sites around TSSs (Supplementary 693 Fig. 6f-g). 694 695

Identification of AR-binding sites in primary prostate cancer specimen 696
Publicly available ChIP-Seq data were retrieved from GSE120738 3 . ChIP-seq data were 697 reprocessed as described for ZMYND11 samples. Differential binding affinity of AR 698 between ERG-rearranged and SPOP-mutant tumors was performed using DiffBind database. All spectra were allowed +/-20 ppm mass tolerance for precursor and product 785 ions, 30% minimum matched peak intensity, and "trypsin allow P" enzyme specificity with 786 up to 4 missed cleavages. The fixed modifications were carbamidomethylation at cysteine, 787 and TMT at N-termini and internal lysine residues. Variable modifications included 788 oxidized methionine and N-terminal protein acetylation. Individual spectra were 789 automatically designated as confidently assigned using the Spectrum Mill autovalidation 790 module. Specifically, a target-decoy based false-discovery rate (FDR) scoring threshold 791 criteria via a two-step auto threshold strategy at the spectral and protein levels was used. 792 First, peptide mode was set to allow automatic variable range precursor mass filtering with 793 score thresholds optimized to yield a spectral level FDR of 1 %. A protein polishing 794 autovalidation was applied to further filter the peptide spectrum matches using a target 795 protein-level FDR threshold of 0. Following autovalidation, a protein-protein comparison 796 table was generated, which contained experimental ratios. For all experiments, non-human 797 contaminants and reversed hits were removed. Furthermore, data were filtered to only 798 consider proteins with 2 or more unique peptides and was median normalized. All these data will be made public upon acceptance of the manuscript. 826    Experiments were performed using three replicates for each condition. Enrichments are determined on custom gene-sets of direct androgen receptor (AR) and ERG target genes (Supplemental Table 1). Enrichments and FDR-adjusted p-values are computed with Camera (pre-ranked) b Venn Diagram and heatmap depicting the expression of genes included in the custom gene-set of AR/ERG co-bound genes that are repressed by ERG and induced by DHT in VCaP cells overexpressing SPOP-MTs (SPOP-Y87C, F102C, W131G), SPOP-WT and vector Control. Genes (rows) and samples (columns) were clustered using Euclidean distance. Gene expression values were normalized using variance stabilizing transformation (vst) and subsequently scaled and centered by row prior of clustering. Columns represent average expression of three replicates for each condition. c, Two-dimensional network representing overlaps between the 10 most significantly enriched Hallmark and custom gene-sets, identified when comparing SPOP-MTs (SPOP-Y87C, F102C, W131G) to SPOP-wild type (-WT) overexpressing VCaP cells. Thickness of edges is proportional to the significance of the overlap of the connected nodes measured by Fisher test. Only edges with FDR value <0.05 are shown. Size of nodes is proportional to gene-set enrichment significance and equals to -10 x log10 (FDR). d, Heatmap representing gene-set activity stratified according to tumor subtype, derived from TCGA cohort. For each tumor group, the average value of single sample GSEA scores was considered. Values were scaled and referenced to samples that did not harbor any ETSfusion (ERG, ETV1, and ETV4) or point mutations in SPOP 1 .    Immunoblot of indicated proteins (a) and corresponding 2D proliferation assay (b) of VCaP cancer cells overexpressing HA-ZMNYD11-WT and derived degron-deficient mutants (DMT1/2) (n=3). Correlation between cell viability and ZMYND11 protein expression changes (Prot. Exp. Changes), as quantified by immunoblot in the same cell lines. P values were calculated using Pearson rank correlation. c Fold-change cell viability of VCaP cancer cells over-expressing the indicated SPOP species with and without  a SPOP mRNA expression levels in 333 primary prostate cancer tissues stratified according to the indicated driver mutations 1 . Error bars, mean ± s.d. b SPOP mRNA and protein levels in response to forced expression of ΔERG in PC3 prostate cancer cells by qPCR and immunoblotting, respectively. Error bars, mean + s.e.m. (n=3). P values were determined by unpaired, two-tailed Student's t-test. # P < 0.05; Control versus ΔERG for SPOP expression levels. ***P < 0.001; Control versus ΔERG for ERG expression levels. c Transwell Matrigel invasion assay of PC3 cells with forced expression of ΔERG and knockdown of SPOP using two different short hairpin RNAs. Protein expression of the indicated proteins was assessed in parallel by immunoblotting. Error bars, mean ± s.e.m. (n=3). d Transwell Matrigel invasion assay of PC3 cells with forced expression of ΔERG and HA-ZMYND11-DMT2 and corresponding immunoblot analysis. Error bars, mean ± s.e.m. (n=3). e Analysis of the ΔERG-and HA-ZMYND11-DMT2-induced transcriptional changes in the ERG target genes PLAU and PLAT. All error bars, mean ± s.e.m. P values were determined by one-way ANOVA with multiple comparisons and adjusted using Benjamini-Hochberg post-test (a,c,d,e). NS, not significant. **P < 0.01, ***P < 0.001. Molecular weights are indicated in kilodaltons (kDa).  Prior to DHT treatment, PDX were grown in standard media without DHT. VCaP were starved for 24h in CSS medium (RPMI + 10% charcoal-stripped serum). Cell viability was assessed after 2 weeks. b Tumor growth kinetics with (n = 10) or without (vehicle; n = 10) testosterone treatment in xenografts established from LuCaP-147 (SPOP-Y83C). c Tumor growth kinetics with (n = 4) or without (vehicle; n = 4) testosterone treatment in xenografts established from LuCaP-78 (SPOP-W131G) cells. d Tumor growth kinetics with (n = 6) or without (vehicle; n = 10) testosterone treatment in xenografts Fig. 9 SPOP mutant tumors are particularly susceptible to androgen deprivation therapies (ADT). a Progression-free survival of prostate cancer patients derived from the TCGA-cohort. Curves representing TMPRSS2-ERG rearranged and SPOP-mutant patients are indicated in violet and green, respectively. The area around the curves represents 80% confidence interval. The bar plot in the lower left corner indicates the percentage of SPOPmutant tumors within all patients who were diagnosed with prostate cancer (DIAG) and within the individuals who developed a progression of the disease (PROG). b Overall survival of prostate cancer patients derived from the MSK-IMPACT cohort. Curves representing TMPRSS2-ERG rearranged and SPOP-mutant patients are indicated in violet and green, respectively. The area around the curves represents 80% confidence interval. The bar plot in the lower left corner indicates the percentage of SPOP-mutant tumors within all patients who were diagnosed with prostate cancer (DIAG), within individuals who developed a metastatic progression of the disease (PROG), and within individuals who developed castration-resistant prostate cancer (CRPC). P values for Kaplan-Meier curves were determined using log-rank test. c Enzalutamide sensitivity of LAPC4 cells overexpressing ΔERG or SPOP mutant species (Y87C, W131G). All error bars, mean + s.e.m. P values were determined by unpaired, two-tailed Student's t-test (c), NS, not significant. *P < 0.05, **P < 0.01. Molecular weights are indicated in kilodaltons (kDa).