Introduction

Prostate cancer (PCa) is the second leading cause of cancer-related death in men1. The critical role of androgen receptor (AR) in castration-resistant PCa (CRPC) has led to the development of potent second-generation antiandrogens including apalutamide2, darolutamide3, and enzalutamide (ENZA)4,5. ENZA treatment improves progression-free survival and overall survival in both CRPC5,6,7,8,9 and hormone-sensitive PCa10. Yet while initially effective, resistance to ENZA eventually emerges in almost all patients11. While a subset (15–25%) of these ENZA-resistant tumors de-differentiate into a neuroendocrine state, most ENZA-resistant tumors still rely on AR signaling12. Several mechanisms of resistance to ENZA have been shown including: AR gene amplification13, somatic mutations in AR gene and enhancer region13, AR splice variants14, or aberrant glucocorticoid receptor (GR) activation15. In addition, recent work has demonstrated that epigenetic alterations can play a critical role in ENZA resistance by affecting genome accessibility through changes in DNA methylation, histone modifications, and chromatin remodeling16,17,18,19,20. These distinct epigenetic alterations reactivate latent developmental programs in advanced PCa tumor samples21,22,23,24,25,26, that facilitate adenocarcinoma lineage plasticity associated with alterations in AR transcriptional activity and ENZA resistance27,28,29. Several epigenetic modifiers, such as LSD1 and EZH2, have been shown to activate oncogene transcription in castrate conditions by altering their histone substrates30,31,32,33. Demonstrating the critical role of epigenetic modifiers in ENZA resistance, inhibitors of LSD134, BET-CBP/p30035, and EZH228,33 have been shown to inhibit the growth of preclinical ENZA-resistant models. From this work, the EZH2 inhibitor36 CPI-1205 has progressed to clinical studies for advanced PCa in combination with ENZA or abiraterone37. Considering their role in PCa progression, epigenetic modifiers are likely drivers of ENZA-resistance PCa and offer promising pharmacological targets.

Despite their critical role, only a handful of epigenetic modifying enzymes have been studied in ENZA-resistant CRPC. To systematically identify novel epigenetic regulators involved in ENZA resistance, we utilized an epigenetic-focused sgRNA library38 and performed unbiased CRISPR-Cas9 drop-out screens on multiple CRPC and ENZA-resistant models. From this, we identified and validated SMARCC2, a core subunit of SWI/SNF complex (also known as BAF complex), and DPF2, a canonical BAF complex specific subunit, that was selectively essential in ENZA-resistant models. Combined treatment of ENZA and BAF inhibitor/degrader resulted in an additive effect in ENZA-resistant cells while having synergistic effect on ENZA-sensitive parental cells. Suggesting a potential redistribution of the BAF complex, we observed a significant expansion of SMARCC2 and BRG1 binding sites in the ENZA-resistant model as compared to the parental model. ENZA-specific BAF regions were potentially bound by CRPC-associated transcription factors, that are highly accessible in clinical samples and located close to differentially regulated genes in ENZA-exposed samples. Collectively, these findings suggest that the SMARCC2-dependent SWI/SNF complex gains an essential role in ENZA resistance by expanding its chromatin regulation, which provides a therapeutic window for enzalutamide-resistant PCa treatment.

Results

Epigenetic-focused CRISPR screen identifies context-specific epigenetic vulnerabilities

To systematically test the role of epigenetic modifiers in ENZA resistance, we utilized a focused sgRNA library targeting all known epigenetic modifying enzymes. The EPIgenetic Knock-Out Library (EPIKOL) contains 7870 sgRNAs that target 719 epigenetic modifiers and 60 common essential genes (10 sgRNA/gene) in addition to 80 non-targeting control sgRNAs38. Using this library, we performed an unbiased CRISPR-Cas9 dropout screen in five different PCa cell lines: androgen-sensitive (LNCaP), androgen-insensitive (LNCaP-abl), two ENZA-resistant cell lines (MR49F and ENZA-R), and the immortalized non-neoplastic prostate epithelial cell line (RWPE-1) (Fig. 1a, Supplementary Fig. 1a). We focused on adenocarcinoma PCa cell lines derived from the same parental LNCaP model to provide a relatively homogenous genetic background that would allow us to selectively identify epigenetic dependencies specific to ENZA-resistant PCa. Using the read count data from sequencing, we conducted an initial quality check and observed that sgRNAs targeting pan-essential genes showed a significant decrease in read counts while nontargeting sgRNAs had no change (Supplementary Fig. 1b). Supporting the library design, the area under the curve value was greater than 0.5 for sgRNAs targeting pan-essential genes, equal to 0.5 for that of epigenetic modifiers, and lower than 0.5 for non-targeting sgRNAs (Supplementary Fig. 1c). Following MAGeCK-Robust Rank Aggregation (RRA)39 analysis of read counts, we focused on the genes with RRA score less than 0.05 and depletion by at least 5 independent sgRNAs. With this threshold, we identified numerous essential AR co-regulators that have been shown to impair the proliferation of LNCaP and its derivatives including LSD1, EZH2, BRD4, and MLL (Fig. 1b, c and Supplementary Fig. 1d)30,33,40,41. As expected, AR was required for LNCaP, LNCaP-abl, ENZA-R, and MR49F but not for RWPE-1 (AR-low/null cell line)35,42 (Fig. 1b, c and Supplementary Fig. 1e). To focus on PCa-associated epigenetic dependencies, we excluded those genes that were essential to either the non-neoplastic prostate epithelial cell line RWPE-1 or those that broadly affect cell growth and are defined as “common essential” in DepMap43. From this, we found 76 epigenetic modifiers that were essential in at least one cell line. Given the common genetic background, we observed a marked overlap in epigenetic dependencies with many of these genes being essential in both LNCaP and at least one of the CRPC cell lines (29/76) (Fig. 1c). Overall, our epigenetic-focused screen revealed context-specific dependencies that are supported by published literature and the cellular background of the cell lines.

Fig. 1: Epigenetic-focused CRISPR screen identifies cell line-specific epigenetic dependencies.
figure 1

a The experimental design of epigenome-wide CRISPR screen to identify epigenetic vulnerabilities in different PCa models. b Distribution of read count fold change based on gene rankings for LNCaP (androgen-sensitive), LNCaP-Abl (androgen independent LNCaP derivative) and MR49F + ENZA-R (ENZA-resistant LNCaP derivatives). Those top-scoring genes linked to prostate cancer are highlighted. c Heatmap of RRA scores across multiple cell lines to compare essential genes in different disease states. Arrows indicate the selected genes selected for validation.

Full size image

Hit epigenetic genes are selectively essential for enzalutamide-resistant cells

Focusing on epigenetic modifying enzymes that can potentially drive ENZA resistance, we stratified the genes that were selectively essential in CRPC models. From this, we identified several epigenetic modifying enzymes (SMARCC2, PAXIP1, SIN3B, CSNK2A1, SETDB1 and PCGF2) that showed a greater impact on the proliferation of resistant cells (LNCaP-abl, MR49F, or ENZA-R) than parental cells (LNCaP)(Fig. 1c). To validate the results from our CRISPR screen, we tested the effect of each gene on cell proliferation using an eGFP competition assay (Fig. 2a). In this, Cas9-expressing LNCaP and MR49F cells were transduced with sgRNA-eGFP constructs targeting the selected genes at a ~50% efficiency. We then quantified the relative proliferation rate of wildtype (non-fluorescent) and knockout cells (eGFP + ) by flow cytometry. Consistent with our pooled CRISPR screen, MR49F cells treated with sgRNAs targeting these genes showed a significant decrease in proliferation while LNCaP cells showed no significant change upon gene knockout except for PCGF2 (Fig. 2b, Supplementary Fig. 2a, Supplementary Fig. 6). Next, we asked whether the loss of any of these genes could drive epigenetic-mediated resensitization to ENZA. To test this, we repeated the eGFP competition assay with and without ENZA in MR49F cells. For all genes, ENZA treatment did not impact the proliferation of resistant cells suggesting that these epigenetic modifying enzymes do not re-sensitize cells to ENZA, but instead, provide a critical function that arose during the development of resistance (Supplementary Fig. 2b, Supplementary Fig. 6). Among the validated genes, we observed the greatest difference in proliferation following the knockout of SMARCC2, which reduced the eGFP population by 40% in MR49F cells but had no significant effect on LNCaP cell proliferation.

Fig. 2: Validated genes from CRISPR screen are selectively essential for enzalutamide-resistant cells.
figure 2

a Experimental design of eGFP competition assay by flow cytometry. Cells are transduced at 50% efficiency with sgRNA expressing eGFP vector and the fluorescence signal is monitored for 20 days. b Normalized eGFP signals of non-targeting (negative control) and SMARCC2 sgRNA-treated cells in multiple time points in Cas9-expressing LNCaP and MR49F cells. Data presented as mean +/- standard error (SE) from three biological replicates. c Western blot image of MR49F cells treated with SMARCC2 sgRNAs and non-targeting (NT) sgRNA. d Crystal violet staining of cell proliferation following SMARCC2 knockout in LNCaP, MR49F and ENZA-R cells. e The effect of SMARCC2 shRNA knockdown on viability in LNCaP, MR49F, and ENZA-R. Cell viability was assessed by CellTiter Glo. Data presented as mean +/- SE from 3 biological replicates. f The RRA scores of other SWI/SNF complex subunits in LNCaP and MR49F cells. cBAF: canonical BRG1/BRM-associated factor (BAF), PBAF: Polybromo-associated BAF complex, ncBAF: non-canonical BAF complex. ATPase module and core indicate shared modules in BAF complexes. g The effect of DPF2 knockdown on viability in LNCaP, MR49F, and ENZA-R. Cell viability was assessed by either CellTiter Glo or MTS. Data presented as mean +/- SE from 3 biological replicates. h Dose-inhibitory response matrix for ENZA and BRM014 combination treatment in LNCaP and MR49F. e+g Student’s t-test, p-value < 0.05 *, <0.01 **, <0.001 ***, <0.0001 ****.

Full size image

Loss of cBAF components SMARCC2 and DPF2 decreases proliferation in enzalutamide-resistant cells

SMARCC2 is a core component of the ATP-dependent mammalian SWI/SNF chromatin remodeling complex. This multiprotein chromatin remodeling complex regulates chromatin accessibility and has recently been found to have an increasingly important role in PCa20,44,45,46,47,48,49,50. Focusing on this gene due to its profound impact on proliferation, we performed validation with orthogonal assays and methodologies. First, we knocked out SMARCC2 in both ENZA-sensitive and -resistant cell lines and performed both crystal violet staining (Fig. 2c, d) and proliferation assays (Supplementary Fig. 3a, b and Supplementary Fig. 7). We observed a significant decrease in the growth of MR49F and ENZA-R cells, while ENZA-sensitive LNCaP cells remained unaffected. To ensure this was not due to the sgRNAs used, SMARCC2 was silenced with RNA interference. Similar to our earlier experiments, MR49F and ENZA-R cells had significantly decreased proliferation following loss of SMARCC2 while there were minimal changes in LNCaP growth (Fig. 2e, Supplementary Fig. 3c). Next, we compared the expression of the paralog genes SMARCC1 and SMARCC2. The transcriptomic analysis revealed an overall higher expression of SMARCC2 than SMARCC1 in MR49F and LNCaP (Supplementary Fig. 3d). In agreement with the transcription levels, SMARCC2 exhibits higher overall protein expression than SMARCC1 in the cell lines tested (Supplementary Fig. 3e, Supplementary Fig. 8). This suggests that the selective dependency was likely due to changes in SWI/SNF activity rather than total protein amount or paralog protein expression as we observed no correlation between SMARCC1/SMARCC2 protein expression and dependency in these models. Supporting this, we observed no significant correlation between SMARCC2 protein expression and clinical disease stages (Supplementary Fig. 3f). Depending on the subunit composition, the SWI/SNF complex can form canonical (cBAF), polybromo-associated (PBAF), or non-canonical (ncBAF) complexes with distinct functional activities51,52,53,54. To determine if there were distinct SWI/SNF dependencies in ENZA resistant models, we re-examined our CRISPR screen data (Fig. 2f). While many common subunits such as SMARCB1, ARID1A and BRG1 were essential in both LNCaP and MR49F cells, we observed that DPF2 was depleted exclusively in MR49F cells. To explore this further, we characterized BAF dependencies from publically available DepMaP data of both the AR-variant expressing PCa cell line (22Rv1), a mechanism of ENZA resistance, and the AR-negative/NEPC-negative CRPC cell line (PC3)55 (Supplementary Fig. 3g). Common subunits SMARCB1, SMARCE1 and ACTL6A were found to be essential for both PC3 and 22Rv1. Interestingly, PBAF components ARID2 and BRD7 were found to be essential only for 22Rv1 and no complex-specific subunits were found to be essential in PC3. Although SMARCC2 is shared in both cBAF and PBAF complexes, DPF2 is selectively found in the cBAF complex56. When we tested the essentiality in ENZA-sensitive and -resistant cell lines, we observed that DPF2 loss selectively reduced the viability of ENZA-resistant MR49F and ENZA-R cells following either shRNA knockdown or CRISPR knockout (Fig. 2g, Supplementary Fig. 3h, i). Next, we assessed the expression of the paralog genes DPF1, DPF2 and DPF3. While DPF2 had overall higher expression than DPF1 and DPF3, the protein levels of DPF1 and DPF2 were relatively similar in MR49F and LNCaP (Supplementary Fig. 3j, k and Supplementary Fig. 9), suggesting that this dependency is unlikely mediated by altered expression of paralog genes. Overall, these results suggest that ENZA-resistant PCa has cBAF-specific dependencies.

Pharmacological targeting of the SWI/SNF component in ENZA-resistant PCa

Given the importance of the cBAF components in our ENZA-resistant models, we wanted to test if we could selectively target these tumors with SWI/SNF inhibitors. We tested several inhibitors including the BRG1/BRM ATPase inhibitor BRM01457 and BRG1/BRM/PBRM1 PROTAC degrader ACBI158 in both ENZA-resistant and -sensitive cell lines. We observed that all three cell lines were sensitive to both drugs with the PROTAC degrader being more potent than the ATPase inhibitor, possibly due to their difference in the mechanism of action (Supplementary Fig. 4a). As previously reported59, we observed a shift in IC50 values for both drugs when combined with ENZA treatment in all cell lines. We noted that in the presence of ENZA, LNCaP cells were more responsive to ACBI1 (IC50 = 4.9 nM) than BRM014 (IC50 = 24.4 nM). In contrast, ENZA resistant MR49F (3.13 nM and 7.6 nM for ACBI1 and BRM014) and ENZA-R (6.2 nM and 8.5 nM for ACBI1 and BRM014) cells responded similarly to both drugs (Supplementary Fig. 4b–e). As we observed that eliminating SMARCC2 didn’t re-sensitize the cells to ENZA, we then tested if BAF inhibitor/degrader would act synergistically with ENZA. Using Bliss/Loewe consensus score, we found that LNCaP exhibited synergistic activity with the combined BAF inhibitor/degrader and ENZA. In contrast, MR49F cells had an additive effect between ENZA and BAF inhibitors/degraders with no shift at ENZA IC50 (Fig. 2h, Supplementary Fig. 4f–i). Overall, this suggests an AR-independent role for the SWI/SNF complex in ENZA resistance.

SMARCC2 and BRG1 gain new binding sites in enzalutamide resistance

Given the sensitivity observed in ENZA-resistant models, we proposed that the cBAF complex caused a significant remodeling of chromatin accessibility during the progression to resistance. To explore this, we conducted chromatin immunoprecipitation sequencing (ChIP-seq) of both SMARCC2 and the ATPase subunit BRG1 in MR49F and LNCaP cells. Strikingly, we noted a marked expansion in the number of SMARCC2 binding sites in MR49F (n = 39,779) as compared to LNCaP (n = 17,811) (Fig. 3a). While the majority of LNCaP-binding sites were also found in MR49F cells, >20,000 additional SMARCC2 binding sites were gained in ENZA-resistant MR49F cells. A similar trend was observed for BRG1 binding sites (Supplementary Fig. 5a). SMARCC2 cistrome showed approximately 75% overlap with BRG1 cistrome in both cell lines, suggesting the recruitment of the whole SWI/SNF complex to these regions (BAF sites). To better understand the role of SMARCC2 in ENZA resistance, we categorized binding regions in MR49F as conserved adenocarcinoma-associated BAF sites (BAF-Adeno, n = 10,317), ENZA-resistant BAF sites (BAF-Enza, n = 18,923), and ENZA-resistant sites that only contain SMARCC2 but not BRG1 (SMA-Enza, n = 8866) (Fig. 3b, c and Supplementary Fig. 5b). BAF-Adeno sites include the binding regions that are co-occupied by SMARCC2 and BRG1 in both LNCaP and MR49F, whereas BAF-Enza sites are those co-occupied regions only found in MR49F. SMARCC2 binding sites were found to be primarily located in intronic and intergenic regions (>75%, Fig. 3d). Although BAF-Adeno, BAF-Enza and SMA-Enza showed very similar binding characteristics over different annotated regions, BAF-Adeno (11.6%) and BAF-Enza (13%) sites broadly tend to non-significantly occupy more promoter regions than SMA-Enza sites (5.1%) (Fisher test, p = 0.13 and p = 0.08 respectively) (Fig. 3d). To identify potential chromatin-binding proteins, we compared the different SMARCC2 binding sites to publicly available transcription factors experimental datasets with GIGGLE60. From this, we identified specific transcription factors that commonly bind to BAF-Adeno, BAF-Enza and SMA-Enza sites (Fig. 3e, Supplementary Fig. 5c). The similarity score was found to be low for SMA-Enza regions relative to the BAF-Enza and BAF-Adeno sites, suggesting that these regions may not be frequently bound by transcription factors. On the other hand, we observed that BAF-Adeno sites tend to be bound by AR and known cofactors including FOXA1, PIAS1, GATA2, GRHL2 and HOXB13. Additionally, SWI/SNF components BRG1 and ARID1A, and nuclear receptors GR/NR3C1 and ESR1 frequently overlap with these BAF-Adeno sites. BAF-Enza sites tend to be occupied by transcription-associated factors including RNA polymerase IIA, CDK9, BRD4 and NELFA. Interestingly, we observed the ETS transcription factors FLI1 and SPI1 preferably enriched in BAF-Enza regions, in addition to growth-promoting transcription factors MYC and MAPK1. Similar results were also obtained by motif analysis (Fig. 3f). Overall the chromatin binding profile supports an AR-independent role of SMARCC2 in ENZA resistance.

Fig. 3: SMARCC2 and BRG1 expand chromatin binding in enzalutamide resistance.
figure 3

a Density plot of SMARCC2 ChIP-seq peaks in LNCaP (blue) and MR49F (red). b ChIP-seq read density heatmap for SMARCC2 and BRG1 binding sites in LNCaP and MR49F cells. BAF regions indicates those sites co-occupied by SMARCC2 and BRG1. BAF-Adeno represents BAF regions found in both LNCaP and MR49F cells. BAF-Enza regions are only found in MR49F. SMA-Enza sites are SMARCC2 binding sites only in MR49F cells but not occupied by BRG1. c Venn diagram depicting different groups in SMARCC2 binding regions in MR49F cells. Overall there are 29240 sites (74% of total SMARCC2 binding regions) shared between SMARCC2 and BRG1 in MR49F cells (MR49F BAF), 26% of them are also found in LNCaP (BAF-Adeno). d Genomic annotation of SMARCC2 binding categories. UTR: untranslated region, TSS: transcription start site, TTS: transcription termination site. e Similarity scores of other transcription factors in BAF-Enza, BAF-Adeno and SMA-Enza regions obtained by GIGGLE analysis. f Motif ranking in BAF-Adeno and BAF-Enza regions. g Chromatin accessibility of BAF-Enza, BAF-Adeno and SMA-Enza regions in clinical samples and PDX models. RND: random sites. PPCA: primary prostate cancer, CRPC: castration-resistant prostate cancer. h Overview figure of proposed mechanism. SWI/SNF complex expanded binding regions and BAF units SMARCC2 and DPF2 during acquired ENZA-resistance. Green color indicates readily essential BAF complex components in parental cells. Red color indicates BAF subunits that has become essential upon acquired resistance.

Full size image

Impact of gained BAF sites on ENZA resistance

To expand these in vitro results to clinical studies, we compared the chromatin accessibility of these BAF-Adeno, BAF-Enza and SMA-Enza regions using published assay for transposase-accessible chromatin with sequencing (ATAC-seq) data obtained from primary PCa samples (n = 23)61 and patient-derived CRPC xenograft models (n = 6)62. Supporting the in vitro annotations, we observed that SMA-Enza sites generally have reduced chromatin accessibility (Fig. 3g), suggesting a low transcriptional activity from these regions. In contrast, both BAF-Adeno and BAF-Enza sites were highly accessible in CRPC samples (Fig. 3g) which corresponds with the previously observed global increase in accessibility at cis-regulatory elements in CRPC61,63,64. To determine how the expansion of binding sites impacts gene transcription, we assessed the functional significance of BAF region groups using Genomic Regions Enrichment of Annotations Tool (GREAT)65. From this, we observed that BAF-Enza regions were specifically enriched for those genes involved in the positive regulation of transcription and cell signaling as well as metabolic process (Supplementary Fig. 5d) in line with the findings from GIGGLE analysis. Additionally, we identified the differentially expressed genes in MR49F relative to LNCaP from publicly available datasets and searched for SWI/SNF binding regions in their proximities (Supplementary Fig. 5e). Supporting the proposed SMARCC2 mediated mechanism, we observed that those upregulated genes had a significant enrichment for BAF-Enza specific binding sites (+/-100kb) (Supplementary Fig. 5f). Similarly, BAF-Enza specific regions were enriched around differentially expressed genes identified in clinical samples that were exposed to the androgen pathway inhibitors abiraterone or ENZA (Supplementary Fig. 5g, h). Overall, these results support a potential regulatory role for SMARCC2-dependent BAF activity in ENZA resistance.

Discussion

The introduction of ENZA has increased the overall survival in metastatic and non-metastatic CRPC5,6. However, in most tumors, this response is only temporary and the cancer will almost always progress to a lethal ENZA-resistant form of the disease. There is increasing evidence that resistance can be driven by dysregulation of epigenetic modifying enzymes. Yet only a handful of epigenetic modifying enzymes have been characterized. To expand the known epigenetic regulators in ENZA-resistant CRPC, we conducted a focused CRISPR screen of isogenic ENZA-sensitive and -resistant PCa cell lines. From this, we demonstrated that SMARCC2, a core component of the SWI/SNF complex, is selectively essential in ENZA-resistant cell lines.

The ATP-dependent SWI/SNF chromatin remodeling complex is a large evolutionarily conserved multiprotein complex that can have distinct protein combinations which impact their biological role51,53. Functional studies have revealed three main categories of mammalian SWI/SNF complexes: cBAF, PBAF, and ncBAF. Starting from the initial SMARCC1 or SMARCC2 nucleation dimer, every module in the complex has at least one specific subunit that determines its category56. Our EPIKOL screen data revealed SMARCB1, ARID1A and previously reported BRG146,49 to be commonly essential genes in parental and ENZA-resistant cell lines; however no PBAF-specific or ncBAF-specific components were essential in ENZA-resistant models, while the cBAF-specific component DPF2 was only essential for ENZA-resistant cells. DepMap analysis on AR-variant-driven ENZA-resistant CRPC model revealed PBAF-specific ARID2 and BRD7 dependencies. Further work in different models would shed light on whether SWI/SNF subunit-specific dependencies are model dependent. It should also be noted that MR49F have two different AR-driven mechanisms of ENZA resistance - an acquired AR mutation (F877L)66 and AR amplification19 which might cause a shift in cBAF dependency.  Of these mechanisms, AR amplification is the likely  driver of resistance  as clinical studies of ENZA-treated patients show that the AR-F877L mutation rarely emerges following resistance13,67. Based on published studies AR mutations are unlikely to alter SWI/SNF dependencies as these mutations primarily modify ligand agonism/antagonism66,68,69. Of note, the primary PCa patients included in this study have normal AR expression and do not contain AR mutations70, whereas CRPC clinical dataset used in this analysis were high AR expressing with varying AR mutations (Supplementary Table 4, Supplementary Data 1). Overall this study demonstrates that cBAF is strongly selective in ENZA-resistant models and may be a common adaptation mechanism.

The selective SMARCC2 dependency observed in ENZA-resistant cells can potentially be attributed to specific characteristics of the SWI/SNF complex. Mammalian SWI/SNF complexes have modular structures and combinatorial compositions, which allow residual complex formation even when a single BAF unit is eliminated56,71. Therefore, certain subunits compete for incorporation, and this behavior correlates to their compensation mechanism as well as the synthetic lethal relationship shown with SMARCC1-SMARCC2, BRG1-BRM, and ARID1A-ARID1B pairs71,72,73. Besides alternate SWI/SNF subunits, SWI/SNF loss can also be compensated by EP400 in cancer cells74, and become synthetically lethal with TMPRSS-ERG fusion and PTEN loss in PCa49,50. It has been shown that SMARCC2 depletion does not prevent either the formation or the activity of the BAF complex, suggesting that cells have a compensation mechanism71. To elucidate the effect of paralog gene expression on dependency shift, we analyzed the protein levels in both our cell lines and the public transcriptomic datasets for SMARCC1/SMARCC2 and DPF1/DPF2/DPF3. We observed that the ratio of paralog genes’ expression is maintained during acquired resistance, suggesting that the residual expression of the paralog gene(s) cannot compensate for SMARCC2 or DPF2 loss. In addition, SMARCC2 protein levels in cell line models and clinical samples showed no correlation with the disease status, suggesting that the SMARCC2 protein levels by itself is not sufficient to explain SMARCC2 dependency. Previous research has highlighted that the distinctive biological role of this multiprotein complex is governed by the redistribution of chromatin-binding regions along with its specific composition71,75. Our epigenetic-focused CRISPR screen didn’t reveal any other ENZA-specific SWI/SNF dependency that would suggest a change in subunit composition. Concordant to its redistribution of chromatin binding, the SWI/SNF activity is markedly different in ENZA-resistant models with a massive gain of binding sites. The shift in SMARCC2 dependency could be caused by the change in cBAF interactome in ENZA-resistant tumours. We observed that conserved adeno-specific BAF regions are more likely to be bound by AR and its co-factors76,77 and were shown to be broadly essential in both previous work44,78 and this CRISPR screen. This would support the synergistic behaviour observed with the AR antagonist and BAF inhibitor in ENZA-sensitive parental cells. In contrast, gained ENZA-specific BAF regions were enriched for proto-oncogenes such as MYC and ETS family of transcription factors. We also observed similar enrichment for transcription-related factors CDK9 and BRD4 that were negatively selected in our screen and shown to be essential for PCa in other studies41,79. We also observed that Adeno- and ENZA- BAF regions are highly accessible in clinical CRPC samples and found in close proximity to the genes upregulated upon acquired ENZA resistance in cell line and clinical models.  These results suggest an overall regulatory role for BAF regions in ENZA resistance that is independent of AR activity.

Overall, this work identified ENZA-specific epigenetic dependencies by performing CRISPR screens in multiple PCa cell line models. SMARCC2, a core SWI/SNF complex subunit, was identified and validated as selectively essential in ENZA-resistant models. Our data suggests a regulatory role for the SMARCC2-dependent SWI/SNF complex in ENZA-resistant phenotype (Fig. 3h) and  inhibiting the SWI/SNF complex can potentially provide benefits for ENZA-resistant PCa.

Methods

Cell Culture

LNCaP, RWPE-1 and HEK293T cell lines were purchased from ATCC. LNCaP-abl (Abl) cell line was a kind gift from Dr.Helmut Klocker and MR49F cell line was a kind gift from Dr. Amina Zoubeidi and Dr. Martin Gleave from Vancouver Prostate Center. ENZA-R cell line was generated by chronically treating LNCaP cells with increasing doses of enzalutamide in Koc University. LNCaP cells were cultured in RPMI 1640 (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco, USA) and 1% penicillin/streptomycin (Gibco, USA). LNCaP-abl cells were cultured in RPMI 1640 (Gibco, USA) supplemented with 5% charcoal-stripped fetal bovine serum (Gibco, USA) and 1% penicillin/streptomycin (Gibco, USA). MR49F and ENZA-R cells were cultured in RPMI 1640 (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco, USA), 1% penicillin/streptomycin (Gibco, USA) and 10 μM enzalutamide. RWPE-1 cells were cultured in Keratinocyte SFM media (Gibco, USA) supplemented with 0.05 mg mL-1 bovine pituitary extract and 5 ng mL-1 human recombinant epidermal growth factor (Gibco, USA). HEK293T cells were cultured in DMEM (Gibco, USA), supplemented with 10% FBS and 1% penicillin/streptomycin. All cell lines were maintained in a humidified incubator at 37 °C with 5% CO2. Cells were regularly tested for mycoplasma.

Viral packaging and titration

For viral packaging, 5×106 HEK293T cells were seeded onto 10 cm dishes 24 h before transfection. On the day of transfection, the media changed to DMEM with 2% fetal bovine serum at least 2 h before transfection. For transfection, lentiviral plasmid (6 μg) was diluted in 150 mM NaCl solution (0.8 mL) along with CMV-8.2dVPR (5.4 μg) and CMV-VSV-g (0.6 μg) plasmids. Transporter 5 (PolySciences Inc., USA) transfection reagent was added to a diluted plasmid mixture in a 1:3 (DNA:Reagent) ratio. The mixture was incubated for 15–20 mins and added dropwise. The media was refreshed 16 h post-transfection. Viral supernatant was collected at 48 h and 72 h. The combined viral supernatant was filtered by 0.45 μm Steriflips (Millipore) and precipitated by 50% PEG (Sigma) overnight. Next day, viral particles were spined down by centrifuge and diluted in 1 mL PBS. To determine viral titration, 0.2×106 cells were seeded onto 6 multiwell plates. Cells were transduced with different amounts of virus. After 48 h, transfected cells were selected with antibiotics for 3 days. At the end of selection, remaining cells were counted. Proportional to their volume, viral titration was back-calculated based on surviving cell numbers. All transduction steps were done in the presence of 8 μg mL-1 protamine sulfate (Sigma, USA).

CRISPR-Cas9 screens and high-throughput sequencing

For each cell line, Cas9-expressing cells were generated by lentiviral delivery of lentiCas9-blast (Addgene #52962) virus at multiplicity of infection (MOI) 5. Cells were selected with blasticidin for 5 days and maintained few passages before the CRISPR-Cas9 screen in selective media. For the screen, we used EPIKOL38 library in lentiGuide-puro (Addgene #52963) backbone that was generated in Koç University laboratories. The sgRNA library was transduced into Cas9-expressing stable PCa cell lines at low multiplicity of infection (MOI = 0.3) with a median coverage of 500x cells per sgRNA. 48 h after transduction, cells were selected with puromycin for 3 days. At the end of selection, 4×106 cells were collected for sequencing experiments (reference point) and 13.5×106 cells were maintained further up to 15 doubling time to minimize differences in growth rates between cell populations. At 15th doubling time, 4×106 cells were collected for sequencing experiments. From the collected cells, genomic DNA was isolated by PureLink Genomic DNA mini kit (Invitrogen, K1820) according to the manufacturer’s instructions. Integrated sgRNA sequences were amplified by PCR reaction. For this, 4 μg genomic DNA was amplified for 25 cycles by using Kapa HiFi HotStart ReadyMix (Roche, KK2602) mixed with 10 μM external forward and reverse primers. To introduce sequencing adapters, external PCR products were pooled and amplified for 15 cycles using internal primers. PCR product was gel-isolated. After quality check, PCR products were sequenced by Genewiz, USA. Each screen was repeated three times. Primers and PCR conditions are listed in Supplementary Table 1, 2.

MAGeCK and Area under the curve analysis

MAGeCK algorithm39 (version 0.5.9.2) was used through all steps of the analysis such as sgRNA quantification and identification of significantly depleted genes. For each sample, R1 fastq files were processed by MAGeCK’s count function to provide sgRNA level counts shown below. Biological replicates were presented as individual input files during sgRNA counting. All samples were combined and median normalized by ‘–norm-method “median”‘ argument of MAGeCK’s count. Identification of essential genes was calculated by MAGeCK’s test function. For each comparison, replicates were pooled as “-t Arep1,Arep2,Arep3” and “-c Brep1,Brep2,Brep3” for conditionA and conditionB respectively. In addition, negative control sgRNA were supplied with “--control-sgrna” function. Median normalized combined tables were used for visualization. R (version 2021.09.0) and ggplot2 package (version 3.3.5) were used for boxplot and density plots. For area under the curve (AUC) analysis, publicly available python code “AUC Calculation (https://github.com/mhegde)’ downloaded from the source80. AUC calculation was run following the instructions on GitHub (https://github.com/mhegde/auc-calculation). All calculated AUC data were plotted using R (version 2022.07.2).

eGFP competition assay

At least 2 sgRNAs per selected gene were cloned into pLKO5 (Addgene #57822) plasmid. Plasmids were packaged and infected into Cas9-expressing stable cell lines at MOI = 0.5. Cells were trypsinized and passaged as one third. The remaining cells were resuspended in PBS for flow cytometry. Initial eGFP expression was determined by flow cytometry using BD FACSCanto II and monitored at every 3–4 days following the previous measurement. Data is normalized to the initial eGFP percentage of the corresponding construct and non-targeting control. RPS11 (ribosomal gene) was used as a positive control for the experiment. Each experiment was repeated at least three times. Oligo sequences are provided in Supplementary Table 3.

Hit validations by cell viability assays upon knockout and knockdown experiments

For SMARCC2 and DFP2 genes, two sgRNAs per gene were cloned into LentiCRISPRv2 (Addgene #52961) for knockout experiments and two shRNAs per gene were cloned into pLKO1-puro (Addgene #8453) plasmid for knockdown experiments. Cells were transduced with constructs and selected with puromycin for three days. Selected cells were counted and seeded onto either 6 multiwell or 96 multiwell plates. Cell viability was measured with CellTiter Glo assay in 96 multiwell plates after 10 days for knockout experiments and after seven days for knockdown experiments. The readings were normalized to that of non-targeting control. 6 multiwell plates were stained with crystal violet solution (0.05% w/v) for 20 min, washed with tap water thoroughly and left to dry overnight. Gene knockout was confirmed by Western blot, and gene knockdowns were confirmed by RT-qPCR. GAPDH (Santa Cruz, sc-25778) and SMARCC2 (Cell Signaling, Cat.No.12760S) antibodies were used as primary antibodies. Goat anti-mouse IgG-HRP (Santa Cruz, sc-2055) and goat anti-rabbit IgG-HRP (Santa Cruz, sc-2054) were used as secondary antibodies. RT-qPCR primers are listed in Supplementary Table 3.

Western Blotting

Whole cell lysates were prepared using NP40 lysis buffer supplemented with ProBlock Gold protease inhibitor cocktail (GB-108-10, GoldBio). Whole cell lysates were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) using 8% or 15% gel. Proteins were then electrophoretically transferred to polyvinylidene difluoride (PVDF) membranes and probed with primary antibodies to SMARCC2 (Cell Signaling, 12760S), SMARCC1 (Invitrogen, PA5-30174), DPF2 (Santa Cruz, sc-514297), DPF1 (Life Span BioSciences, LS-C409145), B-actin (Santa Cruz, sc-47778) and Vinculin (SIGMA Millipore, V9131). Membranes were washed and probed with HRP-conjugated secondary antibodies, and visualized.

Immunohistochemistry analysis

Immunohistochemistry was performed using the Ventana DISCOVERY Ultra autostainer. Tissue microarray (TMA) section was deparaffinized and underwent antigen retrieval at 95 deg C for 64 min in Cell Conditioning 1 (Ventana). Anti-SMARCC2/BAF170 polyclonal antibody (A301-039A, Bethyl) was incubated at 1:200 for 1 h at room temperature. Next, antibodies were detected with the UltraMap DAB anti-Rb Detection Kit and counterstained with hematoxylin and bluing reagent (Ventana). SMARCC2-stained slides were scanned using the Aperio AT2 scanner (Leica) at magnification equivalent of 40x.

Pharmacological inhibition of SWI/SNF complex

Cells were seeded onto 96 multiwell plates in low density and treated with drugs serially diluted in corresponding media supplemented with DMSO. For enzalutamide treatment, a standard 10 μM drug concentration was used. Drug responses were measured after 5 days by CellTiter 96® AQueous One Solution Cell Proliferation Assay (MTS)(Promega, Cat.No. G3581). Data is normalized to the blank reading (0%) and lowest drug concentration (100%), then analyzed by GraphPad software Version 9.1.1. For drug synergy experiments, cells were seeded at low density and treated with both drugs in different concentrations in three independent biological replicates. The cells were allowed to grow either for a week or until untreated cells reached to confluency. The viability was measured by PrestoBlue assay, and the data was analyzed by SynergyFinder 3.081.

Chromatin immunoprecipitation and analysis

BRG1 and SMARCC2 ChIP-seq were done using a double cross-linking strategy to stabilize the SWI/SNF complex. The cells were first fixed at room temperature with 2 mM DSG (Thermo Scientific, Cat.No. 20593) for 45 min prior to 1% formaldehyde fixation for 10 min. The cells sequentially lysed with LB1 (50 mM HEPES-KOH (pH 7.6), 140 mM NaCl, 1 mM EDTA, 10% (vol/vol) glycerol, 0.5% (vol/vol) NP-40/Igepal CA-630 and 0.25% (vol/vol) Triton X-100), LB2 (10 mM Tris-HCl (pH 8.0), 200 mM NaCl, 1 mM EDTA and 0.5 mM EGTA) to enrich for nucleus. The nucleus was further lysed in LB3 and sonicated for 10 mins (30 sec ‘On’ and 30 sec ‘Off’ cycles) using Bioruptor (Diagenode). The immunoprecipitation was performed overnight at 4 °C with 4 µg of BRG1 (Abcam, ab110641) or SMARCC2 (Bethyl Lab A301-039A) conjugated protein G magnetic beads (Thermo Scientific). The DNA was then purified using the Monarch DNA cleanup kit (NEB) after overnight de-crosslinking at 65 °C. The library for high throughput sequencing was prepared using the Ultra II DNA library kit (NEB) according to the manufacturer’s instructions. The ChIP library was further sequenced on an Illumina NovoSeq platform with a minimum of 30 M paired-end reads per sample. FASTQ files were checked with FASTq (v0.11.9) for quality. Paired-end reads were concatenated and aligned to hg38 human genome using BWA82. Unique paired-end reads with MAPQ > 30 are collected by samtools (-F3076 -f2). Peaks were called by MACS383 with default options (-g hs -q 0.01) and peaks with -logFDR>5 are considered for further analysis. Bigwig signals over the genome were generated by genomecov function in BEDtools84. For GIGGLE analysis, Toolkit for Cistrome data browser (http://dbtoolkit.cistrome.org) was used85,86. Gene annotation analysis was performed using annotatePeaks.pl script of HOMER (v4.11) on hg3887. Peaks were called from reference genome hg38 and scanned for known motifs using Hocomoco Human v11 with fimo utility of MEME suite88. The motif rank was calculated based on the differences of log10 transformed q-values for each transcription factor. The GREAT analysis was conducted using the R package rGREAT (v1.26.0)89, and the dot plots were created with ggplot2 (v3.4.2, https://cran.r-project.org/web/packages/ggplot2/index.html).

Analysis of patient chromatin accessibility

To provide a background, matching length random genomic loci were generated, and defined as RND. The normalized bigWig files for ATAC-seq samples were collected from hormone-sensitive PC (HSPC)61 and castration-resistant PC (CRPC) (GSE156291)62. The total signal on every region for each sample was calculated and the resulting data matrix was TMM normalized. Finally, the signal on each BAF binding site group were averaged per each individual and plotted as a boxplot.

Analysis of AR status in patient-derived xenograft models

Exome sequencing data from LuCaP cells were obtained from the SRA database under the accession number SRA03739590. Paired-end sequencing reads were aligned to the human reference genome (hg38) using BWA-MEM (v0.7.17)82. The resulting BAM files were sorted using samtools (v1.16.1)91. To remove PCR duplicates, Picard’s MarkDuplicates (v2.26.10, http://broadinstitute.github.io/picard/) was used, and read groups were added to the BAM file with picard AddOrReplaceReadGroups (v2.26.10) for proper sample identification. Base quality score recalibration (BQSR) was performed using GATK92’s BaseRecalibrator and ApplyBQSR, incorporating known variant sites to adjust base quality scores. Somatic mutations were called with GATK Mutect2 (v4.3), generating raw variant calls in VCF format, which were subsequently filtered using FilterMutectCalls to remove low-confidence variants. Finally, the filtered variants were annotated with ANNOVAR (v2020Jun08)93 using multiple databases (e.g., refGene, avsnp150, 1000g2015aug, clinvar), providing functional and clinical insights. The resulting annotated VCF files contained high-confidence somatic variants for downstream analysis. The AR status is provided in Supplementary Table 4 and Supplementary Data 1.

Analysis of the binding sites distance to differentially regulated genes in the public datasets

The raw RNA-seq data for LNCaP and MR49F cell lines are collected from GSE13846094 and GSE12337995. FASTQ files were checked for quality by FastQC (v0.11.9). Paired-end reads were aligned to hg38 human RefSeq reference set of transcripts using STAR aligner (PMID: 23104886) with following options:--outSAMtype BAM SortedByCoordinate --outSAMunmapped Within --outSAMattributes Standard. Differentially expressed genes were called by DESeq296 using a cut-off at +/-1 for log fold change (LFC) with FDR < 0.05. The gene body locations are collected from the RefSeq GTF files. For the cumulative density function (CDF), for each differentially expressed gene the closes binding type (BAF-Adeno, BAF-Enza, SMA-Enza) is measured and plotted as CDF. The SMARCC2 binding sites around DEG with maximum +/- 100 kb are matched with each gene using BEDtools84 window function (-w10000), and the number of binding sites was plotted. The clinical information and transcriptomic data of patient samples were accessed from SU2C/PCF Dream Team, PNAS 201997 cohort deposited in cBioportal. Differentially regulated genes based on abiraterone/ENZA exposure status were determined and the analysis of binding site distance were performed as described.

Statistics and Reproducibility

Data are presented as the mean ± standard error of mean. Individual data are shown in graphs whenever possible or provided in Supplementary Information. Experiments were repeated as three independent biological replicates. Unless stated otherwise, the two-tailed Student t-test was used to determine the mean differences between the two groups. P-value less than 0.05 is considered significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.