An androgen receptor switch underlies lineage infidelity in treatment-resistant prostate cancer

Abstract

Cancers adapt to increasingly potent targeted therapies by reprogramming their phenotype. Here we investigated such a phenomenon in prostate cancer, in which tumours can escape epithelial lineage confinement and transition to a high-plasticity state as an adaptive response to potent androgen receptor (AR) antagonism. We found that AR activity can be maintained as tumours adopt alternative lineage identities, with changes in chromatin architecture guiding AR transcriptional rerouting. The epigenetic regulator enhancer of zeste homologue 2 (EZH2) co-occupies the reprogrammed AR cistrome to transcriptionally modulate stem cell and neuronal gene networks-granting privileges associated with both fates. This function of EZH2 was associated with T350 phosphorylation and establishment of a non-canonical polycomb subcomplex. Our study provides mechanistic insights into the plasticity of the lineage-infidelity state governed by AR reprogramming that enabled us to redirect cell fate by modulating EZH2 and AR, highlighting the clinical potential of reversing resistance phenotypes.

Extended Data Fig. 1 |. Characterization of the enzalutamide resistance model.

(a) Schematic depicting generation of the ENZ-driven resistance model. (b) Tumour volume and serum PSA of PSA+ (T49) and PSA− (T42) ENZ-resistant tumours at time following ENZ treatment. (c) Immunoblot of AR and PSA in cell lines derived from CRPC (16D^CRPC) and ENZ-resistant AR-driven (49F^ENZR) and lineage plastic (42D^ENZR, 42F^ENZR) tumours. (d) Frequency of activating AR F876L mutation in CRPC and ENZ-resistant cell lines. (e) PCA of the global transcriptome in the indicated cell lines. (f) Significantly enriched (p < 0.05) gene ontology pathways in CRPC and ENZ-resistant cell lines ranked by normalized enrichment score. The following keywords were used to define functional categories: Plasticity (morphogenesis, plasticity, differentiation, mesenchymal); Neuronal (cerebral, axon, synap, neuro); Migration (chemotaxis, migration); Hormone (androgen, hormone); Translation. (g) Expression of ‘Core 9’ embryonic stem cell genes and neuronal lineage markers in the indicated cell lines, reported relative to LNCaP. (h) ASC scores in the indicated prostate cancer cell lines (n = 3) and AR+/NE+ and AR−/NE+ patient tumours from Aggarwal et al. Statistical analysis was performed using a two-tailed unpaired t-test. Error bars represent mean ± SD. (i) Transcript expression of RB1 and TP53 in cell lines and SU2C patient samples with wild-type RB1/TP53 (SU2C^WT) or biallelic RB1 and TP53 deletion (SU2C^RB1/TP53). An RB1/TP53 signature score was applied to cell lines and tumours (higher score indicative of functional RB1/TP53 loss). (j) Immunoblot of pRB1-S780 in the indicated cell lines. (k) Partial least squares discriminant analysis (PLS-DA) of global transcriptome separates AR+/NE+ and AR−/NE+ patient tumours from the Labrecque et al cohort (AR+/NE+, n = 11; AR−/NE+, n = 11; GEO: GSE126078). RNA-seq data from cell lines were projected on the PLS-DA plot. Probability ellipse=95% confidence. (l) Spheroid formation quantified at 8 days following seeding of single cells from the indicated cell lines (mean ± SD; two-tailed unpaired t-test, n = 3). Phase contrast images are shown. Scale bar, 100 μm. (m) Flow cytometry plots of CD44 and NCAM1 cell surface expression (top) and ALDH activity (bottom) in the indicated cell lines (mean ± SD). Diethylaminobenzaldehyde (DEAB) is as a control for background fluorescence.

Extended Data Fig. 2 |. ENZ-resistant lineage plastic tumours exhibit a distinct AR cistrome.

(a) Principal component analysis (PCA) reveals distinct AR binding patterns across prostate states. Each dot represents the genome-wide AR cistrome in an individual clinical specimen (6 normal prostate epithelial, 18 primary prostate cancer tumours, 15 PDX tumours derived from patient mCRPC; GEO:GSE130408) or indicated cell line. (b) Gene ontology (GO) pathways enriched surrounding AR binding sites in ENZ-resistant AR-driven (49F^ENZR) and lineage plastic (42D^ENZR) cells. The closest 2000 peaks in proximity to a transcriptional start site were used for pathway analysis. Statistical significance was determined using a hypergeometric test. Representative AR ChIP-seq tracks surrounding the KLK3/PSA locus are shown. (c) Heatmap indicating AR ChIP-seq signal intensity in 42D^ENZR cells and DHT-stimulated LNCaP cells from Jin et al and Zhang et al. The shade of green reflects binding intensity. Each horizontal line represents a 6-kb locus.

Extended Data Fig. 3 |. Lineage plastic NE-like tumours exhibit a unique chromatin accessibility profile.

(a) Heatmap of ATAC-seq signal intensity in a GEMM of prostate adenocarcinoma (SKO, Pten^f/f) evolution to a plastic, NE-like state (DKO, Pten^f/f/Rb1^f/f; TKO, Pten^f/f/Rb1^f/f/Tp53^f/f). Each horizontal line represents a 6-kb locus. (b) Motif analysis surrounding ATAC-seq peaks (250-bp) in NE-like DKO and TKO GEMMs compared to SKO. Transcription factor motifs identified by HOMER were plotted by ranks generated from their associated differential p values. (c) Significantly enriched pathways (gene set enrichment analysis) in accessible chromatin regions specific to NE-like DKO and TKO GEMMs compared to SKO.

Extended Data Fig. 4 |. The EZH2 cistrome is expanded in the lineage plastic state.

(a) Heatmap of EZH2 ChIP-seq signal intensity in CRPC 16D^CRPC and 42D^ENZR cell lines (left), with overlaid H3K27Ac and H3K27Me3 histone mark ChIP-seq (right). Each horizontal line represents a 6-kb locus. (b) Representative ChIP-seq tracks surrounding the WNT5A locus in 16D^CRPC and 42^ENZR cells. Regions of EZH2 co-occupancy with the active H3K27Ac histone mark are highlighted. (c) Relative expression of genes bound by EZH2 alone (EZH2-none) or co-operatively with H3K27Me3 (EZH2-me) and H3K27Ac (EZH2-ac) histone marks in 42D^ENZR and 42F^ENZR cell lines. Box plot shows mean and interquartile range. (d) Heatmap of H3K27Me3 and K3K27Ac ChIP-seq signal intensity surrounding AR:EZH2 co-occupied regions in 42D^ENZR cells. (e) Heatmap indicating AR and EZH2 ChIP-seq signal intensity at AR:EZH2 co-occupied sites (n = 2155) in 42D^ENZR cells, and EZH2 signal intensity at the corresponding sites in AR-negative cell lines: NCI-H660, DU145 (GEO: GSE135623), and PC-3 (GEO: GSE123204). The shade of green (AR) or blue (EZH2) reflects binding intensity. Each horizontal line represents a 6-kb locus.

Extended Data Fig. 5 |. Characterization of EZH2 phosphorylation.

(a) EZH2 was immunoprecipitated in 42^ENZR cells, trypsin digested, and analyzed by mass spectrometry. Peptides covering 36% of EZH2 were recovered and analyzed for post-translational modifications. (n = 4 independent replicates). (b) Expression of total and phosphorylated (T350, S21, and T311 residues) EZH2 in the indicated cell lines. Protein abundance was assessed by densitometry and is reported relative to total EZH2. (c) IHC staining of pEZH2-S21 and pEZH2-T350 in serial sections from representative CRPC (n = 39) and NEPC (n = 26) patient tumours (Scale bar, 100 μm). Staining area and intensity was quantified and reported (mean ± SD; two-tailed unpaired t-test). (d) Expression of genes positively regulated by EZH2 when phosphorylated at S21 [defined by Xu et al.] in the indicated cell lines and patient tumours from the Beltran 2016 cohort. Statistical analysis was performed using a two-tailed unpaired t-test. Box plots show mean and interquartile range. ns, not significant. (e) qRT-PCR of NE lineage markers in CRPC^crEZH2 cells expressing myc-tagged EZH2^S21A or EZH2^S21D mutants, reported relative to empty vector transfected cells. (mean ± SD; two-tailed unpaired t-test, n = 3). Immunoblotting confirmed transgene expression. (f) Proliferation of parental 16D^CRPC (control) and CRPC^crEZH2 cells stably expressing EZH2^T350A and EZH2^T350D phospho-mutants assessed by IncuCyte (mean ± SD, n = 3 replicates). Immunoblotting confirmed transgene expression. (g) qRT-PCR of plasticity and NE markers in VCaP and C4–2 cell lines co-transfected with EZH2 siRNA and siRNA-resistant myc-tagged EZH2^WT, EZH2^T350A, or EZH2^T350D plasmid following treatment with ENZ (10 μM) for 7 days (mean ± SD; two-tailed unpaired t-test, n = 3).

Extended Data Fig. 6 |. pEZH2-T350 is associated with lineage plasticity.

(a) Frequency of patients with low and high EMT and ASC signature scores in the SU2C and Labrecque et al clinical cohorts that exhibit a high pEZH2-T350 score (indicative of pEZH2-T350 phosphorylation). Patients were defined as “lo” or “hi” for each signature based on +/− 1 standard deviation from the mean signature score in the respective cohort. (b) Frequency of patients with high pEZH2-T350 score (defined as ≥1 standard deviation from cohort mean) in adenocarcinoma, adenocarcinoma with genomic RB1/TP53 loss, and NEPC patient tumours from the SU2C clinical cohort. (c) CDK1 transcript abundance in a GEMM model of prostate adenocarcinoma to NE-like tumour evolution (GEO: GSE90891). DKO and TKO tumours mimic NE-like tumours. SKO, PBCre4:Pten^f/f; DKO, PBCre4:Pten^f/f;Rb1^f/f; TKO, PBCre4:Pten^f/f;Rb1^f/f;Trp53^f/f. Statistical analysis was performed using a two-tailed unpaired t-test. Box plots show mean and interquartile range. (d) CDK1 transcript abundance in benign prostate, adenocarcinoma (AdPC), and NEPC patient specimens from the 2011 Beltran cohort and 2016 Beltran cohort. Statistical analysis was performed using a two-tailed unpaired t-test. Box plots show mean and interquartile range. (e) CDK1 transcript abundance in a patient-derived xenograft (PDX) model of adenocarcinoma (LTL331) to NEPC (LTL331R) lineage conversion following androgen deprivation (castration) (ENA: PRJEB9660). Statistical analysis was performed using a two-tailed unpaired t-test. Mean with min/max range is reported. (f) Expression (qRT-PCR) of genes in 42D^ENZR cells following treatment with CDK1 inhibitor (5 μM RO-3306) for 24 hours. Data are reported relative to vehicle treated cells (mean ± SD; two-tailed unpaired t-test, n = 2).

Extended Data Fig. 7 |. Assessment of neuroendocrine differentiation following AR silencing.

(a) Immunoblot AR, EZH2, and H3K27Me3 (a surrogate marker of EZH2 activity) in 42D^ENZR cells following CRISPR-mediated AR deletion (crAR) or EZH2 inhibition (10 μM GSK126, 96 hrs). (b) Relative expression (qRT-PCR) of neuroendocrine lineage markers in 16D^CRPC and C4–2 cell lines following siRNA-mediated AR silencing for 96 hours. Data are reported relative to cells transfected with a non-silencing scrambled control (mean ± SD, n = 3). A fold change >2 is considered significant. Immunoblotting confirmed AR knockdown.

Extended Data Fig. 8 |. Silencing EZH2 expression and/or activity reverts the lineage-infidelity phenotype.

(a-b) qRT-PCR in 42D^ENZR (a) and 42F^ENZR (b) cells following siRNA-mediated EZH2 silencing (siEZH2) for the indicated time, reported relative to non-transfected control cells at day 0 (mean ± SD; two-tailed unpaired t-test, n = 3). NTC, non-targeting control. (c-d) Spheroid formation and ALDH activity in 42D^ENZR (c) and 42F^ENZR (d) cells following siRNA-mediated EZH2 silencing (siEZH2; left) or treatment with increasing dose of EZH2 inhibitor (GSK126; right) for 8 days (mean ± SD; two-tailed unpaired t-test, n = 2). (e) qRT-PCR in 42D^ENZR cells treated with EZH2 inhibitor (10 μM GSK126) for 7 days, followed by removal (washout) for 14 days. Expression is reported relative to cells at day 0 (mean ± SD; two-tailed unpaired t-test, n = 3). Immunoblotting confirmed on-target effect.

Extended Data Fig. 9 |. Flow Cytometry Gating.

Flow cytometry gating strategy used in Fig. 4k and Extended Data Fig. 1m.

Fig. 1 |. The ENZ-resistant high-plasticity state possesses a distinct AR cistrome and chromatin accessibility landscape.

a, Partial least-squares discriminant analysis (PLS-DA) of the global transcriptome separates AR+neuroendocrine (NE)+ and AR−NE+ patient tumours from ref. . RNA-seq data from prostate cancer cell lines were projected onto the PLS-DA plot. Probability ellipsoid indicates the 95% confidence interval to group the samples. b, AR ChIP-seq signal intensity. Cell lines were grown in 5% FBS, with 42D^ENZR cells supplemented with 10 μM ENZ. The shade of green reflects the binding intensity. AU, arbitrary units. c, Overlapping AR peaks in 16D^CRPC and 42D^ENZR cells (left) and annotations plotted as a percentage of all peaks (right). d, The most significant AR DNA-binding motif in 16D^CRPC and 42D^ENZR cells identified by MEME-ChIP. ARE, androgen response element; hARE, half-ARE; FXBS, FOXA binding site. e, Transcription factor binding motifs within a 250-bp window surrounding AR ChIP-seq peaks in 42D^ENZR and 16D^CRPC cells plotted by ranks generated from their associated differential P values. f, Gene set enrichment ananlysis (GSEA) pathways enriched in 42D^EZNR cells (versus 16D^CRPC) and treatment-induced neuroendocrine-like patient tumours (versus non-neuroendocrine tumours from ref. ). Pathways with genes associated with AR binding (Fisher’s exact test, P < 0.05, fasle discovery rate (FDR) < 0.25) based on ChIP-seq profiles in 16D^CRPC and 42D^ENZR cells are highlighted. NES, normalized enrichment score. g, Overlap of AR ChIP-seq peaks in 42D^ENZR cells and prostate tumours following three months of ENZ therapy from the DARANA trial, along with gene ontology of the shared AR-bound genes. The AR cistrome from patients represents the common AR binding sites shared between patients (n = 3). h, Scatter plot of ATAC-seq counts per peak in 16D^CRPC and 42D^ENZR cell lines. Each dot represents an accessible region. i, Transcription factor binding motifs surrounding AR binding sites (AR ChIP-seq) and accessible chromatin (ATAC-seq) in 42D^ENZR versus 16D^CRPC cells were ranked on the basis of differential P value. Each dot represents a motif identified by HOMER and annotated on the basis of association with plasticity (Benporath ES 1 and Wong adult tissue stem module) or neuronal (GO neurogenesis) transcription factors in MSigDB. j, Heat map indicating AR ChIP-seq and ATAC-seq signal intensity in 16D^CRPC and 42D^ENZR cell lines. The shade of green (AR) and purple (ATACseq) reflects peak intensity. Each horizontal line represents a 6-kb locus.

Fig. 2 |. The AR functions in a non-canonical polycomb complex with EZH2.

a, Abundance of AR, FOXA1, SUZ12 and EED peptides detected using RIME with AR antibodies as bait. Each dot represents an independent replicate, with a solid line denoting the mean. b, SUZ12 immunoprecipitation (IP) followed by immunoblotting for AR and PRC2 subunits. The relative abundance of AR was normalized to SUZ12 pulldown. c, AR–EZH2 PLA and quantification of nuclear PLA signals (red dots) from a single plane (mean ±s.d.; P < 0.0001, two-tailed unpaired t-test; n = 3). Each dot represents the number of PLA signals in a single nucleus. Scale bar, 10 μm. d, Frequency of AR-bound genes with EZH2, SUZ12 and/or EED co-occupancy based on ChIP-seq peak annotation (±50 kb from the nearest TSS) in 42D^ENZR cells. e, Overlap of genomic regions co-occupied by AR and EZH2 ChIP-seq peaks (AR–EZH2 complex) with ChIP-seq peaks for the H3K27Me3 and H3K27Ac in 42D^ENZR cells. f, Heat map of AR and EZH2 ChIP-seq signal intensity in 16D^CRPC and 42D^ENZR cells, with corresponding ATAC-seq peak intensity. g, Overlap of AR and EZH2 ChIP-seq peaks in 16D^CRPC and 42D^ENZR cell lines. h, Overlap of AR and EZH2 ChIP-seq peaks in the Pten^f/f;Rb1^f/f (DKO) GEMM. i, Enriched reactome pathways with genes co-occupied by AR–EZH2 in 42D^ENZR cells and the Pten^f/f/Rb1^f/f GEMM. The size of each circular data point reflects the degree to which genes in the pathway are enriched based on RNA-seq from 42D^ENZR compared with 16D^CRPC cells. NS, not significant. j, Expression of AR–EZH2 co-bound genes in matched prostate tumours (P1–P3) pre- and post-ENZ therapy (n = 3) from the DARANA trial. Box plot shows mean and interquartile range. Statistical analysis was performed using a paired t-test. k, Venn diagram of overlap in genes downregulated (log₂FC < 1) in 42D^ENZR cells following depletion of AR using CRISPR (crAR) or EZH2 inhibition (10 μm GSK126; 96 h). The heat map depicts relative expression of select AR–EZH2 co-bound genes, reported relative to parental cells. l, Sequential ChIP (Re-ChIP) for selected binding sites in 42D^ENZR cells treated with vehicle or EZH2 inhibitor (10 μm GSK126, 96 h). Cells were first analysed by chromatin immunoprecipitation with AR antibody and then immunoprecipitated again with an AR or EZH2 antibody, as indicated. Results are reported relative to IgG control (mean ± s.d., n = 2).

Fig. 3 |. EZH2 is required to establish the lineage-infidelity state.

a, Expression of plasticity and neuroendocrine markers by real-time PCR (rtPCR) and Western blot in 16D^CRPC cells with CRISPR-mediated EZH2 knockout (16D^CRPC crEZH2) following 7 d ENZ treatment. Cells transfected with a non-silencing scrambled guide RNA (crSCR) served as a control. Data are reported relative to non-transfected cells (mean ± s.d., n = 3). Two-tailed unpaired t-test. b, Tumour growth velocity of CRPC cells with CRISPR-mediated EZH2 knockout transplanted subcutaneously into nude mice, followed by treatment with vehicle (veh) or ENZ (n = 5 mice per group). Box plots show mean and interquartile range. Mann–Whitney test. c, Gene expression analysis (by rtPCR) in 16D^CRPC control and crEZH2 xenograft tumours at the experimental end point. Data are reported relative to vehicle-treated mice (mean ± s.d.; *P = 0.05, two-tailed unpaired t-test; n = 3 mice per treatment group). d, Strategy used to establish the 16D^reporter cell line carrying GFP and mCherry fluorescent reporters in the endogenous OCT4 and ASCL1 loci, respectively. Fluorescence-activated cell sorting (FACS) plot shows gating used to isolate the individual cell populations. HL, left homology arm; HR, right homology arm. e, Immunofluorescence images for OCT4-GFP (green) and ASCL1-mCherry (red) in CRPC^reporter cells at the indicated time points after ENZ treatment. Single cells were tracked and are denoted with arrows. Scale bar, 100 μm. f, Fold change in transcript abundance of genes unique and common to the OCT4+, ASCL1+ and hybrid (OCT4+ASCL1+) FACS-isolated CRPC^reporter cell populations relative to the negative population (log₂FC cut-off of 1.5), by RNA-seq. g, MSigDB pathways enriched for common genes (n = 468) upregulated (defined as log₂FC > 1.5) across OCT4+, ASCL1+ and hybrid (OCT4+ASCL1+) CRPC^reporter populations relative to the negative population. Statistical analysis was performed using a hypergeometric test. h, EZH2 activity score, calculated on the basis of z-score-transformed expression of genes in the ‘Kondo EZH2 targets’ signature from MSigDB, in negative, OCT4+, ASCL1+ and hybrid (OCT4+ASCL1+) CRPC^reporter FACS-isolated cell populations. i, Quantification of GFP+ and ASCL1+ fluorescent CRPC^reporter cells following treatment with ENZ (10 μM) alone or in combination with EZH2 inhibitor (10 μM GSK126) using the IncuCyte fluorescent object counting algorithm (mean ± s.d., n = 2). Representative images at 8 d after treatment are shown. Scale bar, 50 μm.

Fig. 4 |. EZH2 is reprogrammed by T350 phosphorylation to drive lineage infidelity and plasticity.

a, Immunoblot of total and phosphorylated EZH2 and CDK1 in the indicated prostate cancer cell lines. HPCS, high-plasticity cell state. b, Immunoblot of EZH2 and pEZH2-T350 following CDK1 inhibition (5 μM RO-3306, 6 h). c, Immunohistochemical staining for pEZH2-T350 and pCDK1-T161 in serial sections from treatment-naive (N, n = 30), CRPC (CR, n = 40) and NEPC (NE, n = 26) clinical samples. Scale bar, 100 μm. Staining intensity was quantified (mean ± s.d.; two-tailed unpaired t-test). d, SUZ12 and EED peptides detected by RIME using EZH2 and pEZH2-T350 antibodies as bait in 42^ENZR cells. Each dot represents an independent replicate, with a solid line denoting mean. Significance was defined as ≥4 peptides. e, Myc-tagged wild-type EZH2 (EZH2^WT) and T350 phospho-mimicking (EZH2^T350D) and phospho-dead (EZH2^T350A) mutants were transiently transfected into 16D^CRPC cells with endogenous EZH2 deletion for 72 h. Immunoprecipitation was performed using a Myc tag antibody. f, Distribution of pEZH2-T350, SUZ12 and EED ChIP-seq peaks in relation to the nearest TSS. The density of polycomb subunits and H3K27Ac are shown surrounding the WNT5A locus. g, Proportion of EZH2 and pEZH2-T350 ChIP-seq peaks overlapping with H3K27Me3 and H3K27Ac ChIP-seq peaks in 42D^ENZR cells. The distribution of H3K27Ac alone and co-occupied with pEZH2-T350 (pEZH2-ac) in relation to the TSS is shown. h, Single-sample GSEA (ssGSEA) score of MSigDB pathways in CRPC^crEZH2 cells expressing EZH2^T350A or EZH2^T350D mutant, and adenocarcinoma (CRPC-Adeno) and NEPC (CRPC-NE) patient specimens from the Beltran 2016 cohort. The ASC score is shown below each cell line or individual patient. i, rtPCR and immunoblot in 42D^ENZR cells with EZH2 knockdown, stably expressing siRNA-resistant Myc-tagged EZH2^WT or EZH2^T350A mutant for 72 h. Data are reported relative to cells transfected with empty vector (EV) (mean ± s.d.; two-tailed unpaired t-test, n = 2). j, Immunohistochemical staining for EZH2 and SYP in serial sections from CRPC^crEZH2 EZH2^T350A and EZH2^T350D mutant xenografts treated with vehicle or ENZ. Scale bar, 100 μm. SYP staining intensity was quantified; box plots show mean and interquartile range. k, Flow cytometry plots of CD44 and NCAM1 cell surface expression in dissociated tumour cells from EZH2^T350A and EZH2^T350D mutant xenografts.

Fig. 5 |. EZH2 T350 phosphorylation correlates with RB1 loss in neuroendocrine-like tumours.

a, Heat map depicting normalized z-score for genes in the pEZH2-T350 signature in NPp53 GEMMs (Gene Expression Omnibus (GEO) accession: GSE92721). ABI, abiraterone acetate; NR, exceptional non-responder. b, pEZH2-T350 signature score in patient tumours from Aggarwal (ref. ) (P = 0.0019) and Beltran (ref. ) (P = 0.012). Violin plots show mean and interquartile range, with significance assessed using a two-tailed unpaired t-test. AdPC, prostate adenocarcinoma; tNEPC, treatment-induced neuroendocrine prostate cancer. c, Correlation between pEZH2-T350 and ASC scores in NEPC tumours from ref. . Each dot represents a patient tumour showing RB1 loss-of-function score. d, Immunohistochemical staining in patient-derived NEPC organoids. Scale bar, 100 μm. e, Immunohistochemical staining in PB-Cre4:Pten^f/f and PB-Cre4:Pten^f/f;Rb1^f/f GEMMs. Scale bars: 200 μm (prostate), 1 mm (liver), 20 μm (insets). f, pEZH2-T350 score in PTEN-null GEMMs with RB1 deletion alone (P = 0.022) or concomitant with TP53 loss (P = 0074; GEO accession: GSE90891). Dots represent individual tumours. Box plot shows mean and interquartile range, with significance assessed using a two-tailed unpaired t-test. g, Volcano plot of RNA-seq from PB-Cre4:Pten^f/f and PB-Cre4:Pten^f/f;Rb1^f/f GEMMs (GEO accession: GSE90891). Each dot represents a gene, with those in the ‘Reactome neuronal system’ (neuronal) and ‘Wong embryonic stem cell core’ (plasticity) MSigDB pathways highlighted. h, Immunoblot in 16D^CRPC cells with stable RB1 knockdown treated with CDK1 inhibitor (5 μM RO-3306, 6 h). shNS, non-silencing control. i, Heat map of gene expression (by rtPCR) in CRPC^crEZH2 cells with shRB1 expressing EZH2^WT, EZH2^T350A or EZH2^T350D mutant constructs following treatment with DMSO or ENZ (10 μM) for 7 d. Expression is reported relative to parental cells (n = 3). j, Left: quantification of spheroids in CRPC^crEZH2 cells with shRB1 expressing EZH2^T350A or EZH2^T350D treated with DMSO or ENZ (10 μM). Data reported as mean ± s.d., with significance evaluated at the end point (P = 0.003, two-tailed unpaired t-test; n = 3). Right: images at 5 d after ENZ treatment. Scale bar, 100 μm. k, Left: neurite length measured using IncuCyte in CRPC^crEZH2 cells with shRB1 expressing EZH2^T350A or EZH2^T350D and treated with DMSO or ENZ (10 μM). Right: phase-contrast images at 96 h after ENZ treatment, with neurite-like extensions highlighted by NeuroTrack. Data reported as mean ± s.d. with significance evaluated at end-point (P = 0.0003, two-tailed unpaired t-test; n = 3). Scale bar, 100 μm.

Fig. 6 |. AR and pEZH2-T350 co-operate to activate lineage-plastic transcriptional programmes.

a, Heat map of AR, EZH2 and pEZH2-T350 ChIP-seq binding intensity in 42D^ENZR cells. Each horizontal line represents a 6-kb locus. b, Frequency of AR ChIP-seq peaks overlapping with EZH2 and pEZH2-T350 ChIP-seq peaks in 42D^ENZR cells. c, Distribution of AR–EZH2 and AR–pEZH2 co-bound peaks in relation to the TSS. Peaks were mapped into 5-kb bins. d, PLA analysis of the interaction between AR and pEZH2-T350, and quantification of nuclear PLA signals (red dots) from a single plane (mean ± s.d.; P = 3.8 × 10⁻¹⁰, two-tailed unpaired t-test; n = 3). Each dot represents the number of PLA signals in a single nucleus. Scale bar, 10 μm. e, Overlap of genes co-bound to AR–EZH2 occupied by SUZ12- and/or EED, based on ChIP-seq peak annotation in 42D^ENZR cells. Gene annotation was restricted to ±50 kb from TSS. f, Expression of genes with promoter-bound (defined as ±3 kb from TSS) AR alone or co-occupancy with EZH2 or pEZH2-T350 in 42D^ENZR and 42F^ENZR cell lines. Data are mean expression ± s.d., with significance assessed using a two-tailed unpaired t-test. g, Expression of AR–pEZH2 co-bound genes in matched individual patient tumours pre- and post-ENZ therapy from the DARANA trial (n = 3). Box plots show mean and interquartile range. Statistical analysis was performed using a paired t-test. h, Gene ontology signatures from MSigDB enriched for AR–EZH2 and AR–pEZH2 co-bound genes in 42D^ENZR cells. Statistical analysis was performed using a hypergeometric test. i, Immunohistochemical staining for AR, pEZH2-T350 and SYP (neuroendocrine marker) in serial sections from non-treated (naive) and neoadjuvant ADT/TAX-treated (4.5 months) prostate tumours from the CALGB 90203 clinical trial. Treated tumours were binned on the basis of pEZH2-T350 staining intensity, and matched NanoString-based sequencing was used to assess the expression of plasticity factors in pEZH2-low (n = 8) and pEZH2-high (n = 4) tumours. Box plots show mean and interquartile range of z-score-transformed expression values with significance assessed using a two-tailed unpaired t-test. Scale bar, 100 μm.

Fig. 7 |. The lineage-infidelity state exhibits dynamic plasticity.

a, PCA of global transcriptome in the indicated cell lines. 42D^ENZR cells with AR knockout (AR KO) and inhibited EZH2 activity (10 μM GSK126, 96 h) are shown. b, ASC and NEPC scores in patient tumours from ref. (adenocarcinoma cluster 5, n = 28; AR+NE+, n = 10; AR−NE+, n = 3) and the indicated cell lines. c, GSEA signatures enriched (Fisher’s exact test, P < 0.05) in 42D^ENZR cells following AR knockout or EZH2 inhibition (10 μM GSK126, 96 h). d, Volcano plot of peptides detected by RIME using EZH2 antibodies as bait in 42D^ENZR cells treated with DMSO or EZH2 inhibitor (10 μM GSK126, 96 h). Statistical analysis was performed using a two-tailed unpaired t-test (n = 3). e, Immunoprecipitation of EZH2 in 42D^ENZR cells treated with 10 μM GSK126 for 96 h followed by immunoblotting. f, Immunoblot of SUZ12 in nuclear soluble and chromatin-bound fractions in 42D^ENZR cells treated with 10 μM GSK126 for 96 h. g, PLA analysis of AR–EZH2 in 42D^ENZR cells following EZH2 inhibition (10 μM GSK126, 96 h). Nuclear PLA signals from a single plane were quantified (mean ± s.d.; P = 3.1 × 10⁻¹⁶, two-tailed unpaired t-test; n = 3). Scale bar, 10 μm. h, Chromatin immunoprecipitation–PCR (ChIP–PCR) for AR at the AREs within the KLK3 enhancer in 42D^ENZR cells following treatment with EZH2 inhibitor (10 μM GSK126, 96 h). Results reported relative to IgG control (mean ± s.d.; P = 0.018, two-tailed unpaired t-test; n = 4). F, forward; R, reverse. i, rtPCR in 42D^ENZR cells treated with EZH2 inhibitor (10 μM GSK126 or GSK343, 96 h) or EED inhibitor (1 μM A-395, 96 h). Data reported relative to vehicle-treated cells (mean ± s.d., two-tailed unpaired t-test; n = 3). Western blot confirmed PRC2 inhibition. j, Confluency measured using IncuCyte (mean ± s.d., n = 2). At 48 h after seeding, cells were treated with EZH2 inhibitor (10 mM GSK126). k, Proliferation of 42D^ENZR cells treated with ENZ (10 μM) and EZH2 inhibitor (2 μM GSK126) alone or in combination, measured using IncuCyte. EZH2 inhibitor was removed (washout) at 96 h. Data plotted are mean ± s.d. (n = 3), with significance evaluated using a two-tailed unpaired t-test at the end point.