The Master Neural Transcription Factor BRN2 Is an Androgen Receptor–Suppressed Driver of Neuroendocrine Differentiation in Prostate Cancer

Progression from primary prostate cancer to advanced metastatic disease is heavily dependent on the androgen receptor (AR), which fuels tumor survival. In men whose treatments for localized prostate tumors have failed, or in those who present with metastatic disease, androgen deprivation therapies (ADT) are used to deplete circulating androgens to abrogate AR signaling and prevent disease progression. Eventually, however, prostate cancer recurs after first-line ADT as castration-resistant prostate cancer (CRPC). Despite low levels of serum androgens in men with CRPC, reactivation of the AR occurs; thus, it remains central to tumor cell survival, proliferation, and metastatic spread. Targeting the AR is a cornerstone therapeutic intervention in patients with CRPC, and AR pathway inhibitors (API) that further prevent AR activation, such as enzalutamide (ENZ), have become mainstays in the prostate cancer treatment landscape (1). Despite its being a potent API, the treatment benefits of ENZ are short-lived in patients with CRPC and resistance rapidly occurs (2).

ENZ-resistant (ENZ^R) CRPC represents a significant clinical challenge not only due to the lack of third-line treatment options to prevent AR-driven tumor progression but also because it can be a precursor to rapidly progressing and lethal neuroendocrine prostate cancer (NEPC). Although NEPC can rarely arise de novo, it is increasingly defined as a variant of highly API-resistant CRPC (3, 4). Aside from the unique small-cell morphology and positive staining for neuroendocrine (NE) markers that characterize NEPC, it is often distinguished from prostatic adenocarcinoma by reduced AR expression or activity (5). Clinical presentation of NEPC reflects this shift away from reliance on the AR, as patients typically present with low circulating levels of PSA despite high metastatic burden in soft tissues, and are refractory to APIs (3). Importantly, it has been reported that under the strong selective pressure of potent APIs like ENZ, these “non–AR-driven” prostate cancers, which include NEPC, may constitute up to 25% of advanced, drug-resistant CRPC cases (6). Not surprisingly, therefore, the incidence of NEPC has significantly increased in recent years (7), coinciding with the widespread clinical use of APIs.

A number of molecular mechanisms likely facilitate the progression of CRPC to NEPC. These include loss of tumor suppressors, such as RB1 (8, 9) and p53 (10), amplification of MYCN (11), mitotic deregulation through AURKA (11) and PEG10 (12), epigenetic controls such as REST (13–15) and EZH2 (11, 16), and splicing factors like SSRM4 (14, 17). Importantly, the AR plays a crucial, albeit still mechanistically unclear, role in NEPC. Reports over many years have highlighted how ADT (18, 19) or loss of AR promotes the NE differentiation of prostate cancer cells (reviewed in ref. 20); as such, many genes associated with an NE phenotype, including ARG2 (21), HASH1 (22), and REST (14, 15) are controlled by the AR. Although this evidence underscores an inverse correlation between AR expression and/or activity and molecular events leading to NEPC, the mechanisms by which the AR directly influences the induction of an NEPC phenotype from CRPC under the selective pressure of APIs such as ENZ remain elusive.

Answering such questions requires a model of API-resistant CRPC that recapitulates the transdifferentiation of adenocarcinoma to NEPC that occurs in patients. Herein, we present an in vivo–derived model of ENZ^R, which, different from others (23–25), underscores the emergence of tumors with heterogeneous mechanisms of resistance to ENZ over multiple transplanted generations. These include the natural acquisition of known AR mutations found in ENZ^R patients (25–28) and the transdifferentiation of NEPC-like tumors through an AR⁺ state, without the manipulation of oncogenes typically used to establish NEPC in murine prostate cancer models (29–31). Using this model and data from patients with prostate cancer, we show that a master regulator of neuronal differentiation, the POU-domain transcription factor BRN2 (encoded by POU3F2; ref. 32), is directly transcriptionally repressed by the AR, is required for the expression of terminal NE markers and aggressive growth of ENZ^R CRPC, and is highly expressed in human NEPC and metastatic CRPC with low circulating PSA. Beyond suppressing BRN2 expression and activity, we also show that the AR inhibits BRN2 regulation of SOX2, another transcription factor associated with NEPC. These results suggest that relief of AR-mediated suppression of BRN2 is a consequence of ENZ treatment in CRPC that may facilitate the progression of NEPC, especially in men with “non–AR-driven” disease.

To model ENZ^R disease, we developed cell lines from LNCaP-CRPC and ENZ^R LNCaP-CRPC xenograft tumors. LNCaP cells were used to establish subcutaneous tumors in intact male athymic nude mice and, upon tumor growth and rising PSA, mice were castrated. Once tumors recurred (CRPC), mice were treated with vehicle or 10 mg/kg ENZ daily and monitored for tumor growth (Supplementary Fig. S1A and Supplementary Methods). Although ENZ treatment did slow tumor growth compared with vehicle control, it did not prevent tumor recurrence (Fig. 1A) and the majority (9 of 10, 90%) of ENZ-treated CRPC tumors increased in tumor volume with concomitant rise in PSA (Fig. 1B). Importantly, however, PSA was not required for tumor recurrence, as observed in one mouse (Fig. 1B and C). PSA⁺ CRPC tumors that grew in the presence of ENZ were serially transplanted into castrated male mice treated with 10 mg/kg ENZ to establish ENZ^R tumors (Supplementary Fig. S1A). Similar to the primary CRPC parental xenografts that recurred in the presence of ENZ, the majority (26 of 35, 74.3%) of transplanted tumors showed increasing volume associated with rising PSA (Supplementary Fig. S1B). However, over the course of serial transplantation, primary PSA⁺ ENZ^R xenografts also gave rise to 9 serially transplanted tumors out of 35 (25.7%) that grew without rise in PSA (Fig. 1D–F; Supplementary Fig. S1C). Cell lines were derived from vehicle-treated CRPC (referred to as 16D^CRPC) and multiple transplanted ENZ^R tumors (referred to as 42D^ENZR, 42F^ENZR, 49C^ENZR, 49F^ENZR, etc.; Supplementary Fig. S1A–S1C) and were screened for protein expression of AR and PSA. Reflecting in vivo data, the established cell lines displayed heterogeneous expression of PSA, yet all retained expression of the AR (Fig. 1G). Importantly, 42D^ENZR and 42F^ENZR cell lines derived from PSA⁻ tumors (Fig. 1E and Supplementary Fig. S1C and S1D) remained PSA⁻ in vitro (Fig. 1G), whereas 49C^ENZR and 49F^ENZR cells derived from PSA⁺ tumors (Fig. 1F; Supplementary Fig. S1B and S1D) retained PSA expression (Fig. 1G). Accordingly, sequencing the ligand binding domain (LBD) of the AR revealed the presence of the F878L AR activating mutation in PSA⁺ 49C^ENZR and 49F^ENZR cells but not in PSA⁻ 42D^ENZR and 42F^ENZR cells (Fig. 1H). Emergence of this mutation in only PSA⁺ ENZ^R cells supports previous results showing this alteration mediates resistance to ENZ (25–28) and suggests that it may be one mechanism by which the AR is reactivated specifically in this subset of ENZ^R cells.

Mounting evidence suggests that “non–AR-driven” phenotypes of prostate cancer can emerge under the selective pressure of potent APIs like ENZ. Importantly, “non–AR-driven” tumors are not necessarily negative for AR expression; in fact, many retain AR but show reduced AR activity (33). Accordingly, our model of ENZ^R showed that approximately 25% of serially transplanted tumors, which emerged under constant presence of ENZ, may be non–AR-driven (Fig. 1D), a distribution that is echoed in clinical reports (4, 6). Indeed, RNA sequencing (Fig. 1I) and microarray analysis (Supplementary Fig. S1E) of PSA⁻ ENZ^R cells indicated that these cells exhibited divergence in global RNA expression on principal coordinate analysis (PoCA) from either CRPC or PSA⁺ (“AR-driven”) cells and displayed reduction not only in PSA (encoded by KLK3) but in many classic AR-regulated genes compared with 16D^CRPC cells (Fig. 1J; Supplementary Fig. S1F and Supplementary Table S1). Underscoring the AR dependency of AR-driven versus non–AR-driven ENZ^R phentoypes, targeting AR using siRNA significantly reduced cell proliferation in 49F^ENZR cells, whereas no effect was observed in 42D^ENZR and 42F^ENZR cells (Supplementary Fig. S1F). These data suggest that emergence of ENZ resistance does not require AR reactivation and that our ENZ^R model may have utility in studying mechanisms of both classic AR-driven and non-driven disease phenotypes, especially transdifferentiation of AR⁺ ENZ^R CRPC to NEPC.

Our data analysis revealed that 42D^ENZR and 42F^ENZR cells exhibited reduced expression of classic AR target genes (Fig. 1J; Supplementary Fig. S1G). Additionally, 42D^ENZR and 42F^ENZR cells showed increased expression of canonical transcription factors and markers associated with neuronal development and NEPC (5), such as neuron-specific enolase (NSE), synaptophysin (SYP), chromagranin A (CGA, encoded by CHGA), and neural cell adhesion marker 1 (NCAM1; Fig. 2A; Supplementary Fig. S2A and Supplementary Table S1). The increased expression of each of these terminal NE markers was validated at the mRNA (Fig. 2B and C) and protein levels (Fig. 2D) in 42D^ENZR and 42F^ENZR cell lines and in tumors compared with CRPC controls. Significantly increased surface expression of NCAM1 on 42D^ENZR and 42F^ENZR cells compared with 16D^CRPC cells was confirmed by flow cytometry (Fig. 2E). Importantly, we found that BRN2 was strongly upregulated in our RNA sequencing data in 42D^ENZR versus 16D^CRPC cells (Fig. 2A) as well as in microarray data comparing 42F^ENZR versus 16D^CRPC cells (Supplementary Fig. S2A). Across multiple prostate cancer cell lines, BRN2 mRNA expression (Fig. 2F) was highest in the NEPC tumor–derived NCIH660 cell line and second highest in 42D^ENZR and 42F^ENZR cells. These data were reflected in serially transplanted xenografts that gave rise to 42D^ENZR and 42F^ENZR cells but not primary CRPC tumors (Supplementary Fig. S2B). These ENZ^R cells showed the highest BRN2 activity (Fig. 2G), and increased BRN2 expression was validated at the protein level by Western blot (Fig. 2H) and IHC (Supplementary Fig. S2C) in 42D^ENZR and 42F^ENZR cells and tumors compared with 16D^CRPC. In addition to NCIH660 and our ENZ^R cell lines, BRN2 expression was also increased in NE-like TRAMP⁺ transgenic prostate tumors (34, 35), where it positively correlated with the expression of NSE, SYP, and CGA (Fig. 2I and J). These data suggest that non–AR-driven models of ENZ^R display NE characteristics that may be supported by the expression of BRN2.

To investigate the clinical relevance of BRN2 in human NEPC, we assessed BRN2 expression in RNA sequencing data from prostatic hormone-naïve adenocarcinoma, CRPC, and NEPC patient tumors (Beltran cohorts; 2016, ref. 33, and 2011, ref. 11). In NEPC tumors, characterized by a high NEPC score and upregulation of canonical NE genes (Fig. 3A and B), BRN2 expression was significantly increased compared with CRPC or adenocarcinoma (Fig. 3C). Beyond this gene expression profile, NEPC is often associated with low AR activity; in these patient tumors, the AR activity score was significantly lower than adenocarcinoma or CRPC (Fig. 3D and E). Importantly, the AR score also decreased in CRPC compared with adenocarcinoma, whereas BRN2 was significantly increased (Fig. 3C and D), suggesting that BRN2 expression may be associated not only with NEPC but also with prostate cancer progression after ADT. Indeed, BRN2 expression and activity as well as NE marker expression were also increased in 16D^CRPC compared with parental LNCaP cells (Fig. 2F and G; Supplementary Fig. S2D), data that were in accordance with our observation in a human CRPC patient-derived xenograft (PDX) that transdifferentiated to NEPC in castrated mice (36). In this model, we found that BRN2 expression was highly upregulated in transdifferentiated NEPC versus adenocarcinoma (Fig. 3F). Increased BRN2 expression in NEPC was also observed by IHC staining (Fig. 3G and H). Importantly, IHC analysis of not only human NEPC but also CRPC and adenocarcinoma supported the inverse correlation between BRN2 and AR (Pearson R = −0.144, P = 0.0043) and positive correlations with CGA (Pearson R = 0.2686, P < 0.0001) and SYP (Pearson R = 0.2709, P < 0.0001). Furthermore, in patients, BRN2 was more highly expressed in metastatic CRPC than in localized adenocarcinoma (ref. 37; Supplementary Fig. S3A) and in metastatic than primary prostate cancer (ref. 38; Supplementary Fig. S3B). Finally, expression of BRN2 positively correlated with the NE-associated genes CGA, CGB, SYP, and MYCN (ref. 38; Supplementary Fig. S3C). These data show for the first time that BRN2 expression is strongly associated with severity of disease in prostate cancer, especially an NE phenotype, and that it is inversely correlated with AR activity.

Our observations in both NEPC patient tumors and our ENZ^R model led us to test the hypothesis that inhibition of the classical AR pathway increases the expression of BRN2. In accordance with our in vitro model where we observed PSA⁻ cell lines express BRN2, analysis of the CRPC samples in Grasso and colleagues (37) as well as the prostate adenocarcinoma data from The Cancer Genome Atlas showed an inverse correlation between BRN2 and serum PSA (Fig. 4A and Supplementary Fig. S3D). This trend was mirrored by immunohistochemistry analysis of a tissue microarray (TMA) of both CRPC specimens and treatment-naïve adenocarcinoma, where we found a significant inverse correlation between BRN2 staining intensity and circulating PSA levels in primary and CRPC patients. Moreover, BRN2 staining intensity significantly increased in progression from primary prostate cancer to CRPC only in patients with low levels of circulating PSA (Fig. 4B). In vitro studies further indicated that suppression of AR signaling regulates BRN2. RNA sequencing (Supplementary Fig. S4A), Western blot analysis (Supplementary Fig. S4B), and IHC (Supplementary Fig. S2C) showed that, compared with 16D^CRPC, the AR-driven, PSA⁺ ENZ^R cell lines did not upregulate BRN2 or markers of NE differentiation and showed significant reduction in BRN2 activity (Supplementary Fig. S4C). Serially transplanted tumors that gave rise to 49F cells (Supplementary Fig. S1B) were also negative for BRN2 and CGA (Supplementary Fig. S4D). Notably, in 16D^CRPC cells, BRN2 protein expression was increased after 2 days of ENZ treatment, which was followed by increased expression of the terminal NE markers CGA, NSE, and SYP over 7 days of ENZ treatment (Fig. 4C). Importantly, siRNA-mediated silencing of BRN2 over the course of ENZ treatment prevented the ENZ-induced upregulation of NE markers in 16D^CRPC cells (Fig. 4C) and in LAPC4 cells (Supplementary Fig. S4E). Similarly, deletion of BRN2 by CRISPR/Cas9 gene editing (Fig. 4D) or stable knockdown by shRNA (Supplementary Fig. S4F) in 16D^CRPC cells prevented ENZ-induced upregulation of NEPC markers, further confirming that BRN2 is a prerequisite for terminal NE marker expression. Reciprocally, transient overexpression of BRN2 in 16D^CRPC, PC3, LAPC4, and 49F^ENZR cells was sufficient to induce expression of NSE, CGA, SYP, and NCAM1 (Supplementary Fig. S5A–S5D) and enriched for an NEPC gene signature and neuronal associated pathways in BRN2-overexpressing 16D^CRPC cells (Supplementary Fig. S5E; Supplementary Tables S2 and S3). Moreover, the effect of BRN2 overexpression on NE markers was enhanced with ENZ treatment of 16D^CRPC cells (Fig. 4E). Importantly, overexpression of BRN2 in not only 16D^CRPC cells (Fig. 4F) but also AR-driven 49F^ENZR cells (Supplementary Fig. S5F) reciprocally downregulated AR target gene expression, which was associated with significantly reduced sensitivity to ENZ in both cell lines (Fig. 4G; Supplementary Fig. S5G). Together, these results suggest that the suppression of classic AR signaling with ENZ treatment leads to induction of an NE phenotype in CRPC that can be driven by the expression of the neural transcription factor BRN2, leading to ENZ^R.

To address AR regulation of BRN2 in CRPC, we assessed the effects of synthetic androgen (R1881) stimulation of AR on BRN2 activity and expression. We found that R1881 significantly reduced BRN2 reporter activity in 42F^ENZR (Fig. 5A), 42D^ENZR (Supplementary Fig. S6A), and in 16D^CRPC cells treated with ENZ (Supplementary Fig. S6B). The expression of BRN2 mRNA was also reduced by R1881 in multiple cell lines (Fig. 5B; Supplementary Fig. S6C–S6E) and was accompanied by a reduction in terminal NE markers (Fig. 5B; Supplementary Fig. S6C and S6D). Importantly, the R1881-dependent reduced expression of NE markers was rescued by BRN2 overexpression, suggesting that it is an important upstream androgen-regulated transcription factor responsible for NE marker expression (Fig. 5B; Supplementary Fig. S6C and S6D). Indeed, we identified an androgen response element (ARE) 8,733 bp upstream of the BRN2 transcriptional start site, and chromatin immunoprecipitation (ChIP) showed that stimulation with R1881 significantly increased AR occupancy at this ARE compared with androgen-deprived conditions (Fig. 5C). To address the effect of AR binding specifically to BRN2 at this ARE, a 20-bp DNA aptamer, which physically inhibits binding of other molecules to its complementary sequence (39), was designed to prevent AR binding at the ARE in the BRN2 enhancer region, and the effects of R1881 on BRN2 and NE marker expression were assessed. The reduction of BRN2 (Fig. 5D) and NE marker (Fig. 5E) expression by R1881 treatment in ENZ^R cells could be rescued in a dose-dependent fashion by introduction of the BRN2 ARE aptamer (ARE^Ap). No effects of aptamer treatment were seen on other AR-dependent genes such as PSA, FKBP5, and TMPRSS2, indicating the specific effects of aptamer treatment on AR binding to BRN2 (Supplementary Fig. S6F). In addition, treatment of 16D^CRPC cells with the BRN2 ARE^Ap increased expression of BRN2 and NE markers to similar levels as ENZ (Fig. 5F). Taken together, these results indicate that BRN2 is negatively regulated by AR activation in both ENZ^R cells and 16D^CRPC cells under the pressure of ENZ.

As in neural development, multiple transcription factors may enhance an NE phenotype in prostate cancer. One candidate that has been implicated in human NEPC (40) and cooperates with BRN2 in neural cells (41) is SOX2. Indeed, SOX2 was significantly upregulated in human NEPC compared with adenocarcinoma or CRPC (Fig. 6A) as well as in our NEPC PDX (Fig. 6B) and was positively correlated with BRN2 expression in metastatic prostate cancer (ref. 37; Supplementary Fig. S7A). Like BRN2, SOX2 expression was highly expressed in NCIH660 cells, as well as in 42D^ENZR and 42F^ENZR cells compared with other prostate cancer lines (Fig. 6C). In addition, genes co-bound and co-regulated by both BRN2 and SOX2 identified in neural progenitor cells (NPC; ref. 42) were also enriched in patients with NEPC (Fig. 6D, left) and in 42D^ENZR and 42F^ENZR compared with 16D^CRPC cells (Fig. 6D, right; Supplementary Fig. S7B). Coregulation of these genes may be mediated by BRN2–SOX2 protein–protein interaction, which we confirmed by coimmunoprecipitation of BRN2 and SOX2 in 42D^ENZR and 42F^ENZR cells (Fig. 6E), and Re-ChIP experiments showing BRN2 and SOX2 co-occupy enhancer regions of NES and RFX4 in 42D^ENZR (Fig. 6F). These results showed that BRN2 and SOX2 can physically interact in prostate cancer cells and supported further investigation into the hypothesis that these two transcription factors may work in concert to promote NEPC.

Similar to BRN2, SOX2 is also negatively regulated by the AR (43). Consistently, we found that SOX2 expression was induced by ENZ treatment in 16D^CRPC and LAPC4 cells (Supplementary Fig. S7C), whereas R1881 reduced SOX2 expression (Supplementary Fig. S7D). Beyond the AR, however, it remains unclear what regulates SOX2 expression in prostate cancer and whether BRN2 is involved. We found that SOX2 mRNA levels were reduced after BRN2 knockdown in ENZ^R (Fig. 6G), PC3, NCIH660, and ENZ-treated LAPC4 cells (Supplementary Fig. S7E–S7G), whereas overexpression of BRN2 in 16D^CRPC (Fig. 6H), PC3, LAPC4, or 49F^ENZR cells (Supplementary Fig. S7H) markedly increased SOX2, data that are in accordance with previous studies showing that BRN2 is required for SOX2 activity in neural development (44). Moreover, R1881-dependent suppression of SOX2 could be rescued in ENZ^R cells by the addition of our BRN2 ARE aptamer (Supplementary Fig. S7I), further supporting our data that BRN2 is AR-suppressed and regulates SOX2 expression. To investigate whether SOX2 may reciprocally regulate BRN2, we examined the enhancer region of BRN2 for potential SOX binding sites. Strikingly, we found a canonical SOX binding motif (42) that overlapped with the ARE in BRN2 (Fig. 6I, diagram). ChIP was used to validate binding of SOX2 to the same region as the AR upstream of BRN2 in ENZ^R cells in androgen-deprived versus stimulated conditions, and we found that SOX2 was able to bind this site in the absence, but not in the presence, of R1881 (Fig. 6I). Interestingly, however, we found that neither knockdown nor overexpression of SOX2 altered mRNA expression of BRN2 in ENZ-treated 16D^CRPC cells (Fig. 6J and K) or NCIH660-, PC3-, or ENZ-treated LAPC4 cells (Supplementary Fig. S7J–S7L). These results suggest a unidirectional regulation of SOX2 by BRN2 and not vice versa, and show that SOX2 binding to the BRN2 enhancer is regulated by AR.

These results strongly implicated BRN2 over SOX2 in driving an NE phenotype in CRPC cells. To investigate this hypothesis, we altered BRN2 and/or SOX2 expression with siRNA or by overexpression in multiple prostate cancer cell lines and found that although BRN2 knockdown reduced the expression of NE markers, SOX2 knockdown did not (Fig. 6J; Supplementary Fig. S7J). Reciprocally, forced expression of SOX2 marginally increased expression of terminal markers of NE differentiation in ENZ-treated 16D^CRPC and PC3 cells (Fig. 6K; Supplementary Fig. S7K), although not to the same extent as overexpression of BRN2 (Fig. 4E; Supplementary Fig. S5B), and did not increase NE markers in ENZ-treated LAPC4 cells (Supplementary Fig. S7L). Finally, in 16D^CRPC and LAPC4 cells where we simultaneously silenced BRN2 and overexpressed SOX2, we did not observe an increase in NE marker expression (Fig. 6L; Supplementary Fig. S7L), indicating that BRN2 is required for any SOX2-dependent induction of NE differentiation. Taken together, these data show, for the first time, that BRN2-dependent control of SOX2 in prostate cancer is inhibited by the AR and drives the expression of terminal NE markers. Moreover, they underscore the central role for BRN2 as a primary regulator of NEPC differentiation in prostate cancer.

Our data suggested that BRN2 contributes to ENZ resistance and is a master regulator of ENZ-induced NE differentiation in CRPC. Confirming the requirement for BRN2 in supporting an NE phenotype across multiple cell lines, we found that transient targeting of BRN2 yielded a marked reduction in mRNA levels of NSE, SYP, CGA, and NCAM1 in NCIH660 (Fig. 7A), 42D^ENZR, 42F^ENZR, PC3, and LAPC4 cells (Supplementary Fig. S8A–S8D). Similar results were observed in stable shBRN2 knockdown 42F^ENZR cells (Fig. 7B). In addition to the expression of terminal NE markers, we questioned whether BRN2 may be important in regulating the aggressive biology of NEPC. Therefore, we investigated the effects of BRN2 knockdown on cellular proliferation, migration, and invasion in vitro. BRN2 knockdown significantly reduced proliferation in both siBRN2 NCIH660 cells and shBRN2 42F^ENZR cells (Fig. 7C and D) and prevented wound closure in a one-dimensional scratch assay (Fig. 7E), as well as the capacity of shBRN2 42F^ENZR cells compared with sh-control cells to migrate through a Matrigel-coated Boyden chamber (Fig. 7F). Similar results were observed in shBRN2 16D^CRPC cells (Supplementary Fig. S8E–S8G). The reduced proliferative capacity of shBRN2 cells in vitro was translated in vivo; shBRN2 42F^ENZR subcutaneous tumors grown in castrated mice under the pressure of ENZ were smaller than sh-control tumors (Fig. 7G), and these tumors had reduced expression of BRN2 and terminal NE markers (Fig. 7H), indicating that this NEPC signature was associated with more aggressive growth. These in vitro and in vivo results indicate that ENZ-induced NE differentiation is mediated by BRN2, which can be targeted to reduce invasiveness and tumor proliferation in both ENZ^R and CRPC.

Clinical evidence suggests that the repercussion of potent AR suppression with APIs for a subset of patients with CRPC is the development of highly lethal NEPC tumors (20). Our work reveals several novel findings with implications for patients with CRPC and NEPC. First and foremost, we identify the master neural transcription factor BRN2 as a central and clinically relevant driver of NE marker expression in advanced prostate cancer. Utilizing a “non–AR-driven” in vivo–derived model of ENZ^R- and ENZ-treated CRPC, we identified BRN2 as a direct target suppressed by the AR that is both sufficient and required for NE differentiation in prostate cancer and mediates resistance to ENZ. Secondly, our data show for the first time the BRN2-dependent regulation of SOX2 in prostate cancer, and the importance of BRN2 over SOX2 in promoting NEPC. Lastly, our data reveal a striking overlap of AR and SOX binding motifs in the enhancer region of BRN2 that allow the AR to competitively inhibit the interaction between SOX2 and the BRN2 enhancer. These in vitro, in vivo, and human studies highlight BRN2 as a key driver of NE differentiation that may indicate progression toward a non–AR-driven or NE phenotype in patients with prostate cancer. Moreover, they suggest that relieving AR suppression of BRN2 could be a central mechanism driving NE differentiation, making BRN2 a strong potential therapeutic target for the treatment and/or prevention of NEPC.

BRN2 is a POU-domain transcription factor well described in developmental biology, where it plays an essential role in neural cell differentiation (32). In addition, BRN2 is highly expressed in NE small cell lung cancer (SCLC), where it acts upstream of other key regulators of neural programming (45) and is required for aggressive tumor growth (46). Complementing these reports, BRN2 was a highly expressed master neural transcription factor identified by RNA sequencing of our non–AR-driven ENZ^R cell lines, making it our top candidate for a potential driver of ENZ-induced NE differentiation in prostate cancer. Indeed, we found that BRN2 was not only sufficient to increase terminal markers of NE differentiation in CRPC but was also required for their expression in CRPC cells exposed to ENZ, in ENZ^R cell lines and in bona fide NEPC NCIH660 cells. Importantly, inhibiting BRN2 expression and concomitant reduction in NE markers in both ENZ^R and CRPC cells functionally led to significantly reduced proliferation, migration, and invasion, as well as decreased ENZ^R tumor growth in vivo and increased resistance to ENZ in vitro. These data mirror other reports showing the requirement for BRN2 in SCLC (46, 47) as well as in melanoma, where it is required for invasion and migration (48–50). Importantly, we have identified BRN2 as a major regulator of NE differentiation in a model of castration- and ENZ-resistant prostate cancer and in terminally differentiated NEPC cell lines.

Data from human specimens brought clinical relevance to BRN2 in aggressive prostate cancer tumors, including NEPC. RNA-sequencing data from patient cohorts showed that BRN2 was most highly expressed in clinically defined NEPC tumors compared with adenocarcinomas. Importantly, however, BRN2 was identified in CRPC specimens as well, indicating that BRN2 is not a specific marker of NEPC, but rather may indicate potential toward disease progression, especially in a setting of androgen deprivation. This was in accordance with our data showing that BRN2 is inducible in CRPC cells under the pressure of ENZ and supported our hypothesis that it is an androgen-suppressed gene. Indeed, we found BRN2 was most highly expressed in primary adenocarcinoma or CRPC tissue from patients with low levels of circulating PSA, and BRN2 expression significantly increased from progression to CRPC only in patients with low PSA levels. Publicly available data mirrored this inverse correlation between high BRN2 expression and low circulating PSA, and further underscored the association between BRN2 and the potential for NE-like disease, as it positively correlated with SYP and CGA expression.

Human data implicating AR control of BRN2 were complemented by mechanistic studies using ENZ^R cell lines, which showed that BRN2 is suppressed by the AR through ligand-dependent binding to an ARE in the enhancer region of BRN2. AR binding to this site resulted in reduced levels of not only NE markers but also SOX2, which we also found highly expressed in human NEPC. SOX2 is a well-defined transcription factor that supports the proliferation and invasiveness of prostate cancer (43, 51–53), is associated with NEPC (40), and is required for the function and maintenance of NPCs (54). Importantly, however, SOX2 can only drive a neural development program by cooperating with other master transcription factors, especially POU family members (41). In particular, BRN2 and SOX2 co-bind upstream of many genes with central roles in neural cell fate and function (42, 55). Moreover, BRN2 is a key regulator of SOX2 activity in NPCs, a function that is highly evolutionarily conserved (44, 55). Despite extensive research into this “pou-sox code” in neural development, how SOX2 is regulated in prostate cancer is largely unexplored. Our study shows that BRN2 is required for SOX2 expression in both CRPC cells treated with ENZ as well as ENZ^R cells. The regulation of SOX2 by AR may be compounded by direct enhancer binding (43) and via AR-dependent suppression of BRN2. Indeed, as we found for BRN2, SOX2 is more highly expressed in AR⁻ than AR+ prostate cancer cell lines (40) and is higher in NEPC and metastatic CRPC than in adenocarcinoma (40, 43, 56). Altogether, these data suggest that the link between SOX2 and progression to AR-independent CRPC or NEPC may be a result of increased BRN2 expression.

Our results showing BRN2-mediated regulation of SOX2 also shed light on the importance of BRN2 over SOX2 in driving NE differentiation in prostate cancer cells. Although SOX2 is present in NE tumors of not only the prostate (40) but also the lung (57, 58) and skin (59), these studies have not shown a direct requirement of SOX2 in supporting this phenotype. Our results suggest that whereas SOX2 overexpression alone in CRPC cells can marginally increase expression of CGA, NSE, SYP, and NCAM1, SOX2 requires the presence of BRN2 to significantly upregulate the expression of these NE markers. The hypothesis that BRN2 and SOX2 work together to drive a neural program in CRPC cells is further supported by our data showing direct BRN2–SOX2 protein–protein interaction at the enhancer region of neuronal genes, leading to their upregulation in ENZ^R cells, which is in accordance with previously reported ChIP-seq analysis (42). Importantly, however, although BRN2 and SOX2 can both bind upstream of each other in NPCs (42), the controlling signals as well as the consequences of these binding events remain unclear. Intriguingly, our observation that SOX2 bound to a canonical SOX motif that overlapped with the ARE in the enhancer region of BRN2 suggests that the AR may play an important role in controlling SOX protein DNA binding in prostate cancer cells. Although the consequence of this inhibition of SOX2 binding to the BRN2 enhancer in the presence of androgen remains unclear, it may be that this interaction supports the two factors coordinating to drive downstream neural gene expression.

Although multiple pathways likely converge to drive the emergence of NEPC, understanding the contribution of the AR to this disease is critical for better implementation of current API therapies and novel drug design. Together, data from our human cohorts and in vitro mechanistic analysis strongly implicate BRN2 as an androgen-suppressed transcription factor that plays a significant role in the progression of prostate cancer from adenocarcinoma to NEPC, making it a potentially attractive and novel therapeutic target.

The detailed procedure for generation of CRPC and ENZ^R tumors and cell lines is found in Supplementary Methods as well as our previously published report (60). A schematic depicting model generation and growth of individual xenografts from which cell lines are derived is shown in Supplementary Fig. S1.

PC3, NCIH660, and LAPC4 cells were obtained from the ATCC in 2013. LNCaP cells were kindly provided by Dr. Leland W.K. Chung (Emory University) and authenticated in January 2013. CRPC and ENZ^R cell lines were generated from LNCaP cells (60), tested, and authenticated by whole-genome and whole-transcriptome sequencing (Illumina Genome Analyzer IIx, 2012). Cells were maintained in RPMI-1640 (LNCaP-derived and PC3) or Iscove's Modified Dulbecco's Medium (LAPC4), containing 10% FBS, 100 U/mL penicillin-G, 100 mg/mL streptomycin (all Hyclone), ±10 μmol/L ENZ (Haoyuan Chemexpress), or DMSO (Sigma-Aldrich) vehicle (for ENZ^R vs. CRPC, PC3 did not receive ENZ). Where indicated, CRPC or LAPC4 cells induced to an NE phenotype were cultured in RPMI-1640, 10% FBS, 100 U/mL penicillin-G, 100 mg/mL streptomycin, +10 μmol/L ENZ for 7 days prior to downstream analysis. Cells were seeded at a density of 10⁶ cells/10 mL media and harvested after 72 hours unless otherwise noted.

Cells were transfected with 500-ng GeneArt Platinum Cas9 nuclease (Thermo) and 125-ng guide RNA (gRNA) using Lipofectamine CRISPRMAX (Thermo). The targeting gRNA sequence 5′-GCTGTAGTGGTTAGACGCTG-3′ was used to edit exon 1 of the POU3F1 locus. At 72 hours after transfection, cells were harvested for analysis of genome modification efficiency using the GeneArt Genomic Cleavage Detection Kit (Thermo) with the forward primer 5′-AAATCAAAGGGCGCGGCGCC-3′ and reverse primer 5′-GCCGCCGCCGTGGGACAG-3′. Ten individual clones were isolated and assessed for indels at the POU3F2 locus by Sanger sequencing.

Microarray gene expression was performed as previously described (36) using Agilent SurePrint G3 Human GE 8 × 60 K slides (Design ID 028004) and analyzed using Agilent GeneSpring 11.5.1 and Ingenuity Knowledge Base (Ingenuity Systems). Specimens were prepared for RNA sequencing using the TruSeq RNA Library Preparation Kit v2, and transcriptome sequencing (RNA-seq) was performed using Illumina HiSeq 2000 (Illumina Inc.) or HiSeq 2500 (human tumors) according to standard protocols. Sequence data mapping and processing was performed as previously described (61), except normalization was performed using reads per million. Quantification of gene expression was performed via RSEQtools using GENCODE v19 as reference gene annotation set. Expression levels (RPKM) were estimated by counting all nucleotides mapped to the gene and normalized by the total number of mapped nucleotides (per million) and the gene length (per kilobase). Sequencing of the AR ligand binding domain was performed exactly as previously described (26). To assess global differences in gene expression in microarray and RNA-sequencing data, multidimensional scaling to analyze differences between cell line gene expression data was performed using the PCoA tool in XLSTAT software (Addinsoft).

CRPC or ENZ^R cells were seeded at a density of 10⁶ cells/10 mL complete media in 10-cm tissue culture dishes (Corning Life Sciences) 18 to 24 hours prior to transfection with siRNA, shRNA, plasmid overexpression, or DNA aptamer.

Cells were transfected with 10 nmol/L BRN2#1 or control siRNA (Santa Cruz Biotechnology), 10 nmol/L of BRN2#2 and siSOX2 (Life Technology) using Oligofectamine (Invitrogen), and OPTI-MEM media (Gibco). After 18 hours, cells were retransfected. After 4 hours, OPTI-MEM media were replaced with complete media, and cells were harvested after 48 hours.

The same protocol as siRNA was used for shBRN2 transfections using shBRN2 or control shRNA (Santa Cruz Biotechnology) and successfully transfected clones were selected for and expanded in complete media containing 10 μg/mL puromycin.

SOX2 plasmid (1 μg; Addgene, #16353) or 8 μg BRN2 plasmid (Addgene, #19711) was transfected using Mirus T20/20 and OPTI-MEM media (Invitrogen) according to the manufacturer's instructions. After 18 to 24 hours, OPTI-MEM media were replaced with complete media ± 10 μmol/L ENZ, and cells were harvested after 7 days or at indicated time points. For 7-day experiments, CRPC cells were retransfected on day 4.

Cells were plasmid transfected with Mirus T20/20 on day 1 and the following day transfected with siRNA with Oligofectamine. After 18 to 24 hours cells were retransfected with siRNA, and OPTI-MEM was replaced with complete media ±10 μmol/L ENZ and harvested 48 hours later (total 5 days).

Indicated doses of BRN2–ARE aptamer were transfected into cells using Oligofectamine. After 18 hours, cells were transfected for a second time for 4 hours. This was repeated on day 4, and cells were either in OPTI-MEM or in complete media + 10 μmol/L ENZ. 42D^ENZR cells were transfected with the aptamer, and after 18 hours transfected a second time. Following this, they were exposed to R1881 for 48 hours, and samples were harvested. For aptamer sequence please see Supplementary Table S4.

Total RNA was extracted from cells using TRIzol reagent (Life Technology) and 2 μg was reversed transcribed using MMLV reverse transcriptase and random hexamers (Invitrogen). Real-time PCR was performed using SyberGreen ROX Master Mix (Roche Applied Science). Target gene expression was normalized to GAPDH levels in three experimental replicates per sample. For primer sequences, please see Supplementary Table S4.

Immunoprecipitation was performed using the ImmunoCruz IP/WB Optima B System (Santa Cruz Biotechnology) based on the manufacturer's guideline. Dilution (1/50) of primary antibody was used for immunoprecipitation. Total proteins were extracted from adherent cells grown in vitro using RIPA lysis buffer. Forty micrograms of protein was resolved by SDS-PAGE, and the following antibodies were used for Western blot: AR, PSA, BRN2, SYP (Cell Signaling Technology), NSE (Dako), SOX2 (Millipore), and CGA and vinculin (Sigma-Aldrich). Blots were incubated overnight at 4°C with designated primary antibodies at 1:1,000 dilution, unless noted otherwise. Proteins were visualized using the Odyssey system (Li-Cor Biosciences).

CRPC and ENZ^R cells were plated in 6-well plates (2 × 10⁴ cells/cm²) and transfected with a BRN2 luciferase reporter (courtesy of Dr. Goding, Ludwig Institute for Cancer Research, Oxford, UK; ref. 62) or Renilla (Promega) using Lipofectin (Invitrogen). The total plasmid DNA used was normalized to 0.5 μg per well by the addition of Renilla. At 24 hours after transfection, cells were incubated with or without 10 nmol/L R1881 for 24 hours, and luciferase activities were measured using a Dual-Luciferase Reporter Assay System (Promega) and a Tecan Infinite 200 PRO microplate luminometer (Tecan). BRN2 luciferase activities were corrected by the corresponding Renilla luciferase activities. Results are expressed in arbitrary light units.

Cells were harvested using citric saline for 10 minutes at room temperature and washed in RPMI+10% FBS. Before antibody addition, cells were incubated with human Fc receptor Binding Inhibitor (eBioscience) for 20 minutes on ice. Flow cytometry staining was performed using anti-human NCAM1 followed by staining with viability marker 7-aminoactinomycin D (7-AAD; both eBioscience, per instructions) and fixation in 2% paraformaldehyde (PFA). Data were acquired (minimum 30K events) on a Canto II (BD Biosciences) and analyzed with FlowJo (TreeStar).

The WST-1 (Promega) assay was used to assess cell growth according to the manufacturer's protocol. One thousand cells were seeded in 96-well plates in complete media, and absorbance at 450 nm was measured after 72 hours.

Cell migration was assessed in a wound-healing assay. Cells were plated on Essen ImageLock 96-well plates (Essen Instruments) and incubated for 2 hours with mitomycin (5 μg/mL) prior to wound scratching with a wound scratcher (Essen Instruments) 24 hours after plating. Wound confluence was monitored with the IncuCyte Zoom Live-Cell Imaging System and software (Essen Instruments). Wound closure was measured every 6 hours for 24 hours by comparing the mean relative wound density of three replicates.

Invasion was assessed by the invasion of 2.5 × 10⁴ cells through BioCoat Matrigel-coated Transwell inserts with 8-μm pore size (BD Biosciences). After 24 hours, the Transwell insert was removed and fixed for 10 minutes in 100% methanol (Sigma-Aldrich) at −20°C and mounted on glass cover slips with Vectashield Mounting Media with DAPI (Vector Laboratories). Filters were imaged using Zeiss Axioplan II microscope (Zeiss) and cells invaded in membrane were quantified.

Cells treated with or without 1 nmol/L R1881 for 24 hours were cross-linked with PFA (Sigma-Aldrich) and sonicated to shear DNA. ChIP assay was performed using the ChIP Assay Kit (Agarose Beads) according to the manufacturer's protocol (Millipore) and antibodies against AR (N20; Santa Cruz Biotechnology) and SOX2 (Millipore). Negative control primers were designed for the regions approximately 1600 bp upstream and 1800 bp downstream of the ARE 8733 bp upstream of BRN2 TSS using Primer Express 3 (Thermo Fisher). Re-ChIP was performed via a Re-ChIP-IT kit (Active Motif Inc.), based on the manufacturer's protocol using antibodies against BRN2 (Cell Signaling Technology) and SOX2 (Millipore). IgG was used as a negative control for antibodies, and negative control primers for each binding site were designed as listed in Supplemental Experimental Procedures. Using Primer Express 3 (Thermo Fisher), primers were designed around BRN2/SOX2 consensus binding elements approximately 7,500 bp upstream of NES start codon and 35,000 bp downstream of RFX4 start codon. Negative control primers were approximately 5,400 bp upstream of NES start codon and 38,000 bp downstream of RFX4 start codon. For primer sequences please see Supplementary Table S4.

shBRN2 and shCTR ENZ^R tumors were grown and monitored in vivo in castrated male athymic mice (Harlan Sprague-Dawley) under pressure of daily oral 10 mg/kg ENZ. Tumors were monitored for growth, and blood was drawn for PSA screening weekly as previously described (60). When tumors reached 1,000 mm³ or greater than 10% animal body weight, tumors were harvested and processed for RNA analysis by qRT-PCR. All animal procedures were conducted according to the guidelines of the Canadian Council on Animal Care.

RNA-seq was performed as above on samples from the Weill Cornell College of Medicine: Beltran 2016 (phs000909.v.p1, cBioportal; ref. 33) = 68 adenocarcinoma, 34 CRPC-Adeno, and 15 CRPC-NE; Beltran 2011 (phs000310 .v.p1, cBioportal) = 30 Adeno and 7 NEPC (11). Tumors were classified by the following criteria based on histomorphology (11, 33): Adeno, usual prostate adenocarcinoma without neuroendocrine differentiation (from radical prostatectomy); CRPC, tumor obtained from CRPC adenocarcinoma metastasis without neuroendocrine differentiation; and NEPC, either of the following categories, adenocarcinoma with >20%, neuroendocrine differentiation, small cell carcinoma, large cell neuroendocrine carcinoma, or mixed small cell carcinoma—adenocarcinoma. For IHC, prostate cancer specimens were obtained from the Vancouver Prostate Centre Tissue Bank and were classified as above (Adeno, n = 93; CRPC, n = 30; NEPC, n = 11). Tissue microarrays of duplicate 1-mm cores were constructed manually (Beecher Instruments). Samples were from radical prostatectomy or transurethral resection of prostate. Immunohistochemical staining was conducted as previously described (36) using the Ventana DiscoverXT Autostainer (Ventana Medical System) with enzyme labeled biotin streptavidin system and a solvent-resistant DAB Map Kit by using 1 of 150 concentration of BRN2 (Abcam) and 1 of 25 concentrations of CGA and AR (Sigma) antibodies. Specimens were graded from 0 to +3 intensity by visual scoring, representing negative-heavy staining. Automated quantitative image analysis was conducted using pro-plus image software. Scoring was conducted at 200× magnification.

Pearson correlations (95% confidence interval) were performed using GraphPad Prism (GraphPad Software) for IHC scoring and BRN2 versus serum PSA. In bar graphs, unpaired, two-tailed, Student t tests were performed to analyze statistical significance between groups using GraphPad Prism (GraphPad Software). Significance is indicated as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Graphs show pooled data with error bars representing SEM obtained from at least three independent experiments.

C.C. Collins is a consultant/advisory board member for Accuragen. No potential conflicts of interest were disclosed by the authors.

Conception and design: J.L. Bishop, D. Thaper, S. Vahid, M.E. Gleave, A. Zoubeidi

Development of methodology: J.L. Bishop, D. Thaper, S. Vahid, A. Davies, K.M. Nip, A. Zoubeidi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.L. Bishop, D. Thaper, S. Vahid, H. Kuruma, A. Angeles, F. Johnson, M.E. Gleave, M.A. Rubin, Y. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.L. Bishop, D. Thaper, S. Vahid, K. Ketola, H. Kuruma, R. Jama, K.M. Nip, A.W. Wyatt, C.C. Collins, Y. Wang, H. Beltran, A. Zoubeidi

Writing, review, and/or revision of the manuscript: J.L. Bishop, D. Thaper, S. Vahid, A. Davies, K. Ketola, R. Jama, C.C. Collins, H. Beltran, A. Zoubeidi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.L. Bishop, R. Jama, K.M. Nip, A. Angeles, D. Lin, C.C. Collins

Study supervision: J.L. Bishop, A. Zoubeidi

Other (pathology): L. Fazli

Tissue procurement and tissue microarray construction at the Vancouver Prostate Centre were supported by the Terry Fox New Frontiers Program Project Grant. We thank Dr. Goding (Ludwig Institute for Cancer Research, Oxford, UK) for providing the BRN2 luciferase plasmid and Mary Bowden and Darrell Trendall for technical support in animal studies.

This work is supported by the Prostate Cancer Foundation USA and Prostate Cancer Canada, and proudly funded by the Movember Foundation (T2013-01). J.L. Bishop was supported by the Prostate Cancer Foundation Young Investigator Award and the Urology Care Foundation. K. Ketola was supported by the U.S. Department of Defense (PC121341). D. Thaper and A. Davies were supported by a Canadian Institute of Health Research scholarship.

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