Androgen Receptor Pathway-Independent Prostate Cancer Is Sustained through FGF Signaling

Summary
Androgen receptor (AR) signaling is a distinctive feature of prostate carcinoma (PC) and represents the major therapeutic target for treating metastatic prostate cancer (mPC). Though highly effective, AR antagonism can produce tumors that bypass a functional requirement for AR, often through neuroendocrine (NE) transdifferentiation. Through the molecular assessment of mPCs over two decades, we find a phenotypic shift has occurred in mPC with the emergence of an AR-null NE-null phenotype. These “double-negative” PCs are notable for elevated FGF and MAPK pathway activity, which can bypass AR dependence. Pharmacological inhibitors of MAPK or FGFR repressed the growth of double-negative PCs in vitro and in vivo. Our results indicate that FGF/MAPK blockade may be particularly efficacious against mPCs with an AR-null phenotype.

Significance
Targeting AR signaling in metastatic PC commonly produces robust clinical responses. However, disease progression is nearly universal. Potent AR antagonists appear to be shifting the phenotypes of resistant PCs to tumors that are devoid of AR activity, but the drivers of these resistant carcinomas are not known. Here we report that autocrine and paracrine FGF signaling is capable of bypassing a requirement for AR, and find that FGF and MAPK pathways are active in metastatic AR-null PCs. Suppressing FGF and MAPK inhibits the growth of AR-null PC indicating that targeting the FGF axis may represent a therapeutic approach for those cancers resistant to AR-directed therapies and may circumvent treatment resistance if combined with initial AR pathway blockade.

Introduction
Androgen deprivation therapy (ADT), achieved through surgical or pharmacological approaches, exploits the exquisite dependence of prostate carcinoma (PC) on androgen receptor (AR) signaling. Although initially highly effective as a treatment for metastatic PC, ADT is characterized by the predictable emergence of resistance, a disease state termed castration-resistant prostate cancer (CRPC). An important feature of CRPC is the reactivation of AR signaling, an event reflected by progressive rises in serum prostate-specific antigen (PSA), a gene product transcriptionally regulated by the AR. A substantial body of evidence has documented that essentially the entire AR cistrome is re-expressed in most CRPCs, and several mechanisms capable of maintaining AR activity have been established (Carver et al., 2011, Montgomery et al., 2008, Nelson et al., 2002, Taylor et al., 2010).

The continued importance of AR signaling in most advanced PCs has prompted the development of therapeutics directed toward further suppressing AR ligands or the AR itself. Several drugs, including improved AR antagonists and inhibitors of androgen synthesis, extend survival (de Bono et al., 2011, Scher et al., 2012), although to date complete remissions have been rare. While the intensive effort focused on completely repressing AR activity may completely eradicate a subset of PCs, this selective pressure has the potential to generate PCs reliant on survival mechanisms distinct from those regulated by AR or that substitute for vital AR functions.

Assessments of metastatic CRPCs have determined that patients may harbor tumor deposits that do not express AR following conventional ADT (Roudier et al., 2003, Shah et al., 2004). While a subset of AR-null tumors express markers of neuroendocrine (NE) differentiation, these neuroendocrine prostate cancers (NEPC) exist within a more complex spectrum of phenotypes ranging from anaplastic carcinomas, mixed prostatic adenocarcinomas with NE features, to pure small-cell carcinomas (Aparicio et al., 2011, Beltran et al., 2011, Tzelepi et al., 2012). Importantly, there are metastatic CRPCs that do not express the AR or markers of NE differentiation (Roudier et al., 2003, Wang and Epstein, 2008). Although conclusive data are lacking, evidence suggests that the widespread application of more effective AR pathway antagonists such as enzalutamide (ENZ) and abiraterone (ABI) is shifting the pattern of metastasis in patients with CRPC accompanied by alterations in their molecular landscapes (Beltran et al., 2014, Doctor et al., 2014). Anticipating that effective AR repression will more routinely result in CRPCs devoid of AR signaling, we sought to identify molecular pathways operating in CRPC that function to promote survival and growth in the absence of AR activity. The emergent signaling programs that confer resistance to AR-directed therapeutics may represent treatment targets for men with progressive CRPC.

Results
Emergence of an AR-Null and Neuroendocrine-Null Prostate Cancer Phenotype in Patients Following AR-Directed Therapy
To evaluate the shifting phenotypic and molecular landscapes of metastatic CRPC (mCRPC), we characterized metastatic tumors acquired from a long-standing tissue acquisition necropsy program spanning two decades. We classified tumors from 84 consecutive patients as androgen receptor pathway active prostate cancer (ARPC) if they expressed AR and the AR-regulated gene PSA, or NEPC if they expressed the NE gene synaptophysin (SYP). In a small minority of patients both ARPC and NEPC tumors were evident. In the era prior to the approval of the AR pathway antagonists ENZ and ABI (1997–2011), most CRPCs were ARPCs (85%) with rare NEPCs (10%) and rarer AR−/NE− tumors (5%), hereafter classified as “double-negative” PCs (DNPC) (Figures 1A and 1B ). In the contemporary era (2012–2016), we observed a shift in tumor phenotypes with a higher representation of DNPCs (Figure 1A). Gene expression programs of the tumors classified by immunohistochemistry (IHC) supported these distinct subtypes using 10-gene signatures that were concordant with previously published gene sets indicative of NE and AR pathway activity (Figures 1C, S1A, and S1B) (Beltran et al., 2016, Hieronymus et al., 2006).

While molecular characteristics of CRPCs with active AR and NE programs are well described, those of DNPC are not established. We used RNA sequencing (RNA-seq) to quantitate gene expression differences between DNPCs and ARPCs and identified 417 and 107 mRNAs with substantially increased or decreased levels, respectively (5-fold; q < 0.0001) (Figure 1D). In comparison with NEPC, 162 and 594 genes were significantly increased or decreased, respectively in DNPCs (5-fold; q < 0.0001) (Figure S1C). Gene set enrichment analysis (GSEA) identified numerous biological processes that differed between ARPC and DNPC, which complicated efforts to identify a predominant driver event or signaling pathway (Figure S1D). To prioritize efforts defining causal mechanisms underlying DNPC, we evaluated tumors for genomic alterations and partitioned mCRPCs that we previously characterized for genome-wide copy-number and mutation status (Kumar et al., 2016) into categories of ARPC, NEPC, and DNPC based on their expression profiles (Figures 1E, 1F, S1E, and S1F). Common aberrations in CRPCs such as TP53 mutation and PTEN loss did not differ significantly across groups with the exception of AR amplification, which was more frequent in ARPC (66%) compared with NEPC (13%) (p = 5.6 × 10−5) and RB1 loss, a hallmark of NEPC, which differed between NEPC (88%) and ARPC (16%) (p = 2.4 × 10−8) (Figure 1E). Several genomic regions differed in copy number between ARPC and DNPC, but no genes in these regions varied in expression by more than 2-fold (Figure 1F). With the caveat of limited tumor numbers, these data indicate that recurrent genomic aberrations do not underlie the marked phenotypic differences between ARPC and DNPC. AR Ablation Results in CRPC without Neuroendocrine Differentiation To provide insights into causal mechanisms capable of promoting survival in an AR-null state, we developed a model system that recapitulated the transition from a tumor initially dependent on AR activity to one capable of AR-independent growth. We began with the LNCaP cell line, a widely studied androgen-sensitive in vitro model of PC. LNCaP derivatives capable of proliferating in the absence of AR ligands typically continue to exhibit AR signaling (Sobel and Sadar, 2005). Furthermore, targeting the AR in these cells with antibodies, ribozymes, or RNAi induces apoptosis or growth arrest, indicating that the AR maintains vital functions (Cheng et al., 2006, Zegarra-Moro et al., 2002). To initiate the present studies, we used a LNCaP line stably transduced with a tetracycline (TET)-inducible anti-AR short hairpin RNA (shRNA) (Cheng et al., 2006), designated as LNCaPshAR. Repressing AR in the setting of castration-resistant LNCaPshAR growth results in tumor regression, but recurrent LNCaPshAR tumors re-express AR, due to the selective loss or silencing of the AR-directed shRNA (Snoek et al., 2009). To enforce AR ablation, we introduced an androgen response element (ARE)-driven thymidine kinase suicide gene designated pATK. In the resulting LNCaPshAR/pATK line, thymidine kinase is expressed in the setting of an active AR and induces cell death when treated with ganciclovir (Figures 2A and S2A–S2C). We subjected LNCaPshAR/pATK cells to increasingly severe AR pathway suppression (Figure 2A). After 2 weeks of androgen deprivation (ADT), medium was supplemented with 1 μg/mL doxycycline (Dox) to induce the anti-AR shRNA, which produced >99% cell death. After 5 months, a residual population of viable cells remained. This colony was treated with a 2-week course of ganciclovir to eliminate cells expressing functional AR. Surviving cells were designated LNCaP-AR Program-Independent Prostate Cancer (LNCaPAPIPC). AR and PSA were nearly undetectable in LNCaPAPIPC: AR expression was 45-fold lower and PSA expression was 30-fold lower than LNCaPshAR/pATK (Figures 2B and 2C). Transcripts comprising an AR activity signature were all substantially decreased in LNCaPAPIPC cells and showed no induction with androgen treatment (Figure 2D). We confirmed the absence of AR and PSA protein expression in LNCaPAPIPC grown in vivo as subcutaneous xenografts (Figure 2E).

Previous studies demonstrated that LNCaP cells grown in androgen-depleted medium or with AR antagonists display a transdifferentiated phenotype resembling NEPC (Mu et al., 2017, Zhang et al., 2003). NEPC is characterized by loss of AR expression and AR activity and increased expression of CHGA and SYP, and cells often exhibit small-cell morphology (Beltran et al., 2011). NE-associated genes were not upregulated in LNCaPAPIPC cells grown with or without androgen supplementation (Figure 2F). Furthermore, LNCaPshAR/pATK and LNCaPAPIPC grown as murine xenografts do not express CHGA or SYP protein (Figure 2E).

To further evaluate the characteristics of LNCaPAPIPC cells, we determined the effects of AR pathway-targeted therapies. In contrast to parental LNCaPshAR/pATK, LNCaPAPIPC grow robustly without androgen (Figure 2G). Furthermore, treatment of LNCaPshAR/pATK with ENZ completely inhibited growth, while LNCaPAPIPC was highly resistant to ENZ treatment (Figure 2G). PC cells with low AR transcriptional activity that accompanies advanced Gleason grade exhibit invasive and metastatic phenotypes (Aihara et al., 1994, Erbersdobler et al., 2009). LNCaPAPIPC cells displayed a slight but consistent increase in baseline migration (5%, p = 0.019) and invasion (12%, p = 0.006) when compared with LNCaPshAR/pATK, and also responded to a transwell serum gradient with a higher number of migratory and invasive cells (Figures 2H and 2I).

FGFR and MAPK Signaling Pathways Are Activated in Androgen Receptor Pathway-Independent Prostate Cancer
The growth of LNCaPAPIPC cells in the absence of AR expression indicated that alternative survival pathways supplanted AR requirements and we next sought to identify them. We used RNA-seq to profile the gene expression program in LNCaPAPIPC and identified 548 differentially expressed transcripts relative to AR-intact LNCaPshAR/pATK cells (≥10-fold; q < 0.001) (Figure 3A). LNCaPAPIPC gained expression of basal cell genes such as TP63 and TRIM29, and retained expression genes expressed in luminal cells such as KRT8, KRT18, and HPN (Figure 3B). We used array CGH to identify copy-number aberrations harboring genes that could bypass a requirement for AR signaling. Overall, the genomes of LNCaPAPIPC and parental LNCaPshAR/pATK were nearly identical, with only seven regions differing in copy number between the two lines. Two genes, MAT2B and KIAA1328, exhibited concordant changes in copy number and expression, but transcript levels did not differ between ARPCs and DNPCs. Though located in the region of chromosome-3 copy gain, WNT7A transcripts were not measureable in LNCaPAPIPC cells (Figures S3A–S3C). Collectively, the few genomic aberrations identified do not explain the marked alterations in gene expression between LNCaPAPIPC and parental LNCaPshAR/pATK cells. To confirm lineage relationships, we compared the expression profiles of 15 PC cell lines with LNCaPAPIPC using unsupervised hierarchical clustering. LNCaPAPIPC grouped with other LNCaP derivatives, indicating that LNCaPAPIPC retains LNCaP characteristics even while lacking AR-regulated gene expression (Figure 3C). Notably, the removal of Dox from the culture medium of LNCaPAPIPC cells did not result in AR re-expression or a reversion of gene expression changes (Figure S4A). We also found no evidence of upregulation of the glucocorticoid receptor (GR/NR3C1), a nuclear hormone receptor previously shown to bypass AR requirements (Arora et al., 2013) (Figure 3D). Phosphatidylinositol 3-kinase (PI3K)/AKT signaling can influence the progression of CRPC and effectively compensate for reduced AR activity in PC models via reciprocal feedback activation (Carver et al., 2011, Mulholland et al., 2011). Therefore, we hypothesized that PI3K pathway upregulation was supporting LNCaPAPIPC growth. Consistent with previous studies, pAKT levels increased in AR-intact LNCaPshAR/pATK cells grown in androgen-depleted medium (Figure 3E). Surprisingly, pAKT was nearly undetectable in LNCaPAPIPC, suggesting that PI3K activity is not acting as a survival/growth pathway in these AR-null cells. Increased mitogen-activated protein kinase (MAPK) signaling is also postulated to support CRPC proliferation (Aytes et al., 2013, Mulholland et al., 2012, Ueda et al., 2002). MAPK signal transduction is activated through a variety of stimuli, and is closely associated with receptor tyrosine kinase (RTK) activity. Phosphorylated MEK and dually phosphorylated ERK1/2 (ppERK1/2) were elevated in LNCaPAPIPC compared with LNCaPshAR/pATK (Figure 3F). These data suggested that increased MAPK signaling may be sustaining AR-independent growth in LNCaPAPIPC. We evaluated RAS and RAF for alterations that could account for MAPK activation but found no evidence of altered expression or functional mutations (Figures S4B and S4C). We next evaluated the LNCaPAPIPC transcriptome for mechanisms plausibly contributing to MAPK activity and found that fibroblast growth factor 8 (FGF8) expression was substantially upregulated relative to AR-active LNCaPshAR/pATK (>100-fold, q < 0.001) (Figure 3G). FGF8 is transcribed as eight distinct isoforms (FGF8a–h), and of these FGF8b has the most oncogenic effects (MacArthur et al., 1995). LNCaPAPIPC expressed all active FGF8 isoforms at substantially higher levels than LNCaPshAR/pATK (FGF8a/g = 1,100-fold, p < 0.001; FGF8b = 600-fold, p < 0.001) (Figures 3H and 3I). FGF8 protein was detected in serum-free conditioned medium from LNCaPAPIPC but not from LNCaPshAR/pATK, concordant with transcript expression results (Figures 3J and S4D). To further assess FGF pathway activity, we measured a panel of transcripts shown to reflect the dynamic output of FGF receptor (FGFR) signaling (Delpuech et al., 2016). Several transcripts comprising this FGFR signature were increased more than 10-fold in LNCaPAPIPC cells including DUSP6, ETV4, and EGR1, and LNCaPAPIPC cells showed significant FGFR and MAPK pathway enrichment scores (Figure 3K). FGF pathway activation has been shown to occur in rare instances by FGFR genomic rearrangements in mCRPC (Wu et al., 2013), but we found no evidence of mutation, copy-number gain, or gene rearrangements involving FGF8 or FGFRs in LNCaPAPIPC (Figures S3B and S3C). Collectively, these data supported the hypothesis that an autocrine FGF signaling program is activated in LNCaPAPIPC in the absence of AR to maintain cell survival and growth via MAPK. FGFR and MAPK Signaling Are Active in DNPC and Are Inversely Associated with AR Activity We next sought to further evaluate FGF and MAPK signaling in DNPCs and confirm LNCaPAPIPC as a relevant model for this CRPC subtype. We determined that an LNCaPAPIPC gene signature is significantly enriched in DNPC metastases (false discovery rate [FDR] < 0.001) (Figure 4A), as are gene sets reflecting the activity of FGF signaling, MAPK activity, MEK/ERK, and EMT (Figure 4B). No single FGF ligand or receptor was universally increased across all DNPCs: individual tumors expressed high FGF1, FGF8, or FGF9, and different FGFRs. Each of these secreted FGF ligands has been shown to activate multiple FGFRs consistent with the finding that DNPCs exhibited consistently high MEK/ERK and FGF activity scores (Figures 4C and 4D). A small subset of ARPCs also expressed high MEK/ERK and FGFR pathway activity, and these tumors generally also had lower AR activity (Figure 4C). Across the full spectrum of CRPC metastases, AR activity was inversely associated with FGF8/9 expression, and FGFR activity (e.g., r = −0.48, p < 0.001 for FGF8) (Figure 4E). AR and FGF8/9 expression were inversely associated (r = −0.13) in an independent dataset of 150 metastatic CRPC tumors from the SU2C/PCF dataset (data not shown) (Robinson et al., 2015). Collectively, these results couple elevated FGF and MAPK signaling with a CRPC tumor phenotype, DNPC, which lacks AR activity and supports LNCaPAPIPC as a model that represents these attributes of DNPC. To address the challenge of deriving a generalized understanding of DNPC from a single model, we sought to develop additional systems with which to evaluate drivers of DNPC and identify effective therapeutics. As with LNCaPAPIPC, our objective was to begin with an AR-positive PC and then repress AR activity. We were unable to successfully eliminate AR in the commonly used VCaP or 22Rv1 PC lines by shRNA or CRISPR-based approaches (data not shown). However, using the PacMet-UT1 PC line that expresses a functional AR (Troyer et al., 2008), albeit with attenuated activity, we were able to excise AR using CRISPR/Cas9 editing and generate multiple PacMet AR-null sublines (Figures 5A and 5B ). AR loss was associated with 10-fold upregulation of FGF9 and enhancement of FGF and MAPK activity (Figures 5C and 5D). Notably, repressing AR activity in PacMet-UT1 cells did not result in an NEPC phenotype, and the expression of SOX2, a reprogramming factor associated with transdifferentiation to NEPC, was decreased (Figure 5C) (Mu et al., 2017). We were also successful in generating a patient-derived xenograft (PDX) model of DNPC, designated LuCaP173.2, initiated from a tumor acquired from a rapid autopsy procedure. Metastatic tumors from this individual had phenotypic variability, with one rib metastasis expressing AR and PSA and a second rib metastasis lacking AR or PSA staining (Figure 5E). We confirmed that the LuCaP173.2 PDX lacks AR and PSA expression and does not express classic NE markers such as chromogranin or synaptophysin, thus fulfilling criteria for DNPC (Figure 5F). However, other genes associated with an NE phenotype such as EZH2 and MYCN are expressed in this PDX line, suggesting a continuum of tumor differentiation (Figure S5). In accord with findings in DNPC metastases, LuCaP173.2 expresses high FGF9 and FGFR1 levels with low AR and NEPC program scores and a high FGFR activity score (Figure 5G). FGF Activates MAPK Signaling and Bypasses a Requirement for Androgens and the AR in Promoting Prostate Cancer Growth We next sought to determine whether FGF signaling is necessary and sufficient for bypassing a requirement for AR activity. We hypothesized that the substantial upregulation of FGF8 in LNCaPAPIPC cells comprises an autocrine loop to sustain cell survival in the absence of AR. The introduction of FGF8-specific small interfering RNAs (siRNAs) reduced LNCaPAPIPC growth by 80% (p < 0.001) (Figure 6A). In contrast, siRNA knockdown of FGF9, which is not upregulated in LNCaPAPIPC, had no effect. Exogenous FGF8b increased the growth of parental LNCaPshAR/pATK in androgen-depleted conditions (p < 0.001) and the addition of concentrated LNCaPAPIPC conditioned medium (CM) showed a small but statistically significant increase in proliferation (11%, p = 0.01), whereas LNCaPshAR/pATK CM had no effect (Figure 6B). The addition of exogenous FGF8b increased ERK1/2 phosphorylation in both LNCaPshAR/pATK and LNCaPAPIPC. FGF8-induced growth in androgen-depleted conditions and ERK1/2 phosphorylation were blocked by treatment with the FGFR inhibitor PD173074 (Mohammadi et al., 1998) (Figures 6C and 6D). To demonstrate that FGF8 was sufficient to promote the growth of cells cultured under total AR pathway suppression, we treated parental LNCaPshAR/pATK grown in androgen-deprived conditions with Dox to suppress AR expression, and added FGF8b. FGF8b maintained cell proliferation during AR pathway ablation (30% increase in cell number compared with untreated LNCaPshAR/pATK; p = 0.019), albeit at a lower rate than AR-intact LNCaPshAR/pATK (75% increase in cell number compared with untreated LNCaPshAR/pATK; p = 0.018) (Figure 6E). In a parallel experiment, LNCaPshAR/pATK cells were cultured in androgen-depleted medium and AR expression was suppressed by pre-treatment with Dox for 72 hr. Addition of exogenous FGF8b rescued the growth inhibition by ADT and AR suppression (58% increase in growth compared with untreated LNCaPshAR/pATK; p = 0.003) (Figure S6A). The FGFR antagonist PD173074 is a nanomolar inhibitor of FGFR1 but is also a submicromolar inhibitor of vascular endothelial growth factor receptor 2/kinase domain receptor (VEGFR2/KDR) (Mohammadi et al., 1998). To confirm that FGFR antagonism is mediating the growth repression in DNPC, we treated LNCaPAPIPC with a second FGFR antagonist CH-5183284, which potently and selectively inhibits FGFR1–3 (IC50 of 8–22 nM) without significant biological effects toward VEGFR2/KDR or other kinases (Nakanishi et al., 2014). At concentrations of 0.1–1.0 μM, CH-5183281 substantially inhibited the viability and increased apoptosis rates in LNCaPAPIPC with effects far exceeding those observed in wild-type LNCaP cells (Figures 6F and 6G). CH-5183281 also reduced the viability of AR-null PacMet-UT1 cells relative to the AR-intact parental line (Figure 6H). Confirming that MAPK activity is required for FGF8-mediated castration-resistant proliferation, the MEK1/2 inhibitor U0126 blocked the growth of LNCaPshAR/pATK induced by FGF8 in androgen-depleted conditions (p < 0.001; Figure 6I) and repressed LNCaPAPIPC proliferation. Co-treatment of a second androgen-sensitive PC line, 22RV1, with U0126 and FGF8 led to a 46% decrease in cell number compared with cells treated with FGF8 alone (p < 0.001; Figure S6B). We next sought to determine whether suppressing FGF signaling would inhibit the growth of DNPC in vivo. PD173074 significantly reduced LNCaPAPIPC xenograft growth rates: the study was terminated at 40 days due to large tumors in the control group at which time tumor volumes were 1,147 mm3 in the vehicle and 571 mm3 in PD173074 arms (p < 0.001) (Figure 6J). To confirm these findings, we treated LuCaP173.2 DNPC PDX tumors with CH-5183284. At the study endpoint of 24 days, tumor volumes were 814 mm3 and 170 mm3 in the vehicle and CH-5183284 arms, respectively (p < 0.001) (Figure 6K). The expression of FGFR pathway genes as well as composite FGFR and MEK/ERK pathway activity were significantly reduced in LuCaP173.2 tumors resected 3 days and 24 days on CH-5183284 treatment (Figure 6L). FGF- and MAPK-Induced ID1 Contributes to AR-Null Prostate Cancer Growth We next evaluated LNCaPAPIPC for downstream mediators of FGF/MAPK signaling that could promote the dedifferentiated phenotype of DNPC and support survival in the absence of AR activity. We identified a strong candidate for this role, inhibitor of differentiation 1 (ID1), which was upregulated in LNCaPAPIPC compared with LNCaPshAR/pATK (∼10-fold by RNA-seq; q < 0.001; 5-fold by qRT-PCR) (Figure 7A). ID1 expression is induced by exogenous FGF and bone morphogenetic protein via MAPK pathway activation (Langenfeld and Langenfeld, 2004, Passiatore et al., 2011), prevents differentiation by binding cell lineage-specific transcription factors (Perk et al., 2005), and has been associated with poorly differentiated PC (Coppe et al., 2004, Sharma et al., 2012). Notably, other ID family members were also increased in LNCaPAPIPC and the LuCaP173.2 DNPC PDX (Figure 7B). ID1 levels are significantly higher in DNPC metastases relative to ARPCs (p < 0.005) (Figure 7C), and ID1 and AR expression are inversely associated in mCRPC (Pearson correlation = −0.39) (Figure 7D). Stimulation of LNCaPshAR/pATK cells with FGF8 resulted in a 4-fold (p = 0.006) increase in ID1 mRNA and protein (Figures 7E and 7F). MEK inhibition reduced FGF8-mediated ID1 induction by approximately 30% (p = 0.005) (Figure 7E). Although ID1 levels were already elevated, stimulation of LNCaPAPIPC with exogenous FGF8 resulted in a further 1.6-fold increase (p < 0.001), and treatment with U0126 alone decreased baseline ID1 expression by approximately 40% (p = 0.006) (Figure 7E). We also observed a significant upregulation of ID1 in response to FGF8 treatment in androgen-sensitive 22Rv1 cells (Figures S7A and S7B). The enhanced activity of specific RTKs is associated with ligand-independent activation of AR transcription in some models (Gregory et al., 2005, Yang et al., 2003); however, we did not observe a change in AR, PSA, or TMPRSS2 expression in response to FGF8b stimulation in androgen-deprived LNCaPshAR/pATK, LNCaPAPIPC, or 22Rv1 cells (Figures 7G and S7C). ID1 has been shown to influence PC differentiation and proliferation (Ling et al., 2011, Ouyang et al., 2002), and we hypothesized that ID1 could mediate a component of the growth-promoting effects of FGF/MAPK activity. In support of this hypothesis, levels of ID1–3 transcripts were diminished in the LuCaP173.2 DNPC PDX tumors treated with the FGFR inhibitor CH-5183284 (Figure 7H). ID1 knockdown did not significantly affect LNCaPshAR/pATK growth compared with a scrambled control siRNA (siUNI). In contrast, two independent ID1-targeting siRNAs decreased LNCaPAPIPC growth by 32% (p = 0.003) and 43% (p < 0.001) (Figure 7I). When LNCaPshAR/pATK were treated with FGF8, ID1 knockdown significantly attenuated FGF8-induced proliferation by ∼35% (p < 0.001). The effect of ID1 knockdown was enhanced in LNCaPAPIPC with ID1 siRNAs suppressing FGF8-induced growth by 39%–50% (p < 0.001) (Figure 7I). These effects were replicated in 22RV1 cells grown in androgen-deprived conditions (Figure S7D). Discussion Therapeutic approaches designed to impair AR activity remain first-line therapy for men with metastatic PC. While resistance to AR-directed therapeutics is usually accompanied by reactivation of AR signaling, newer drugs with potent AR pathway antagonism appear to be shifting the phenotypes of resistant PC to anaplastic and NE carcinomas that are devoid of AR activity (Figure 7J). The AR-null/NE-null tumors evaluated in the present study were acquired from men after initial responses to AR antagonists, indicating that these agents effectively eliminated tumor clones dependent on the AR, but failed to eradicate cell populations that no longer required AR signaling. Defining the drivers of these resistant carcinomas is critical for the development of effective treatment strategies. We determined that complete AR pathway independence was associated with elevated autocrine FGF signaling in vitro and elevated FGFR and MAPK pathway activity in mCRPC. FGF ligands and receptors have previously been shown to influence the development and progression of PC (Acevedo et al., 2007, Feng et al., 2015). Of relevance to the present study, a PDX model of PC devoid of AR signaling was shown to express high levels of FGF9, which promoted tumor growth, induced an osteoblastic tumor microenvironment, and responded to FGF-directed therapy (Li et al., 2008). MAPK signaling promotes poorly differentiated tumor growth in models of PC (Mulholland et al., 2012), and constitutive ERK1/2 activity is associated with castration resistance (Gioeli et al., 1999, Oka et al., 2005, Rodriguez-Berriguete et al., 2012). While there is evidence suggesting that MAPK can stimulate ligand-independent AR activity (Feldman and Feldman, 2001), FGF/MAPK signaling did not promote the re-expression of AR-regulated genes in our models and FGFR activity was inversely associated with the expression and activity of AR in CRPCs. At this time, the mechanism(s) influencing FGF expression in LNCaPAPIPC or other DNPCs is not known. As we found no recurrent genomic events involving FGFs/FGFRs, other processes capable of influencing FGF transcription, including epigenetic regulation, are likely operative. Notably, a small subset of CRPCs exhibiting FGFR/MAPK activity did not express high levels of FGF ligands, suggesting that in some circumstances paracrine FGF derived from microenvironment constituents may promote pathway activity and drive treatment resistance (Lawson et al., 2010). While AR repression can allow for cellular reprogramming and transdifferentiation to NE carcinoma (Ku et al., 2017, Zou et al., 2017), our results indicate that the acquisition of NE characteristics may represent a continuum of differentiation from ARPC to DNPC to NEPC, although the acquisition of NE characteristics does not appear to be a certainty following AR ablation (Figure 7J). Importantly, alternative cell fates may associate with unique therapeutic vulnerabilities. Given that NE and anaplastic tumors are more common following sustained AR pathway suppression and appear to arise from adenocarcinomas in vivo based on shared genomic aberrations (Beltran et al., 2011, Beltran et al., 2016), it is quite likely that the incidence of AR pathway-independent PCs will increase with the deployment of increasingly potent AR inhibition. Whether acute and more complete AR repression will eliminate PCs or consistently generate AR-null variants remains to be determined. Early results from an ongoing clinical trial (NCT00831792) of the FGFR antagonist dovitinib in men with metastatic CRPC unselected for loss of AR activity reported a 26% response rate in bone and soft tissue lesions (Wan et al., 2014). Our results suggest that FGFR inhibition may have modest effects in AR-active CRPCs, but be particularly active in the subset of CRPCs with absent or limited AR function. A clinical trial of FGFR or MAPK antagonists may be fruitful in patients stratified by AR activity status. Furthermore, co-targeting of predictable AR bypass pathways capable of providing robust cell survival and proliferation signals may prolong responses to initial AR antagonism.