Androgen deprivation promotes neuroendocrine differentiation and angiogenesis through CREB-EZH2-TSP1 pathway in prostate cancers

The incidence of aggressive neuroendocrine prostate cancers (NEPC) related to androgen-deprivation therapy (ADT) is rising. NEPC is still poorly understood, such as its neuroendocrine differentiation (NED) and angiogenic phenotypes. Here we reveal that NED and angiogenesis are molecularly connected through EZH2 (enhancer of zeste homolog 2). NED and angiogenesis are both regulated by ADT-activated CREB (cAMP response element-binding protein) that in turn enhances EZH2 activity. We also uncover anti-angiogenic factor TSP1 (thrombospondin-1, THBS1) as a direct target of EZH2 epigenetic repression. TSP1 is downregulated in advanced prostate cancer patient samples and negatively correlates with NE markers and EZH2. Furthermore, castration activates the CREB/EZH2 axis, concordantly affecting TSP1, angiogenesis and NE phenotypes in tumor xenografts. Notably, repressing CREB inhibits the CREB/EZH2 axis, tumor growth, NED, and angiogenesis in vivo. Taken together, we elucidate a new critical pathway, consisting of CREB/EZH2/TSP1, underlying ADT-enhanced NED and angiogenesis during prostate cancer progression.

A ndrogen-deprivation therapies (ADT) are the mainstay treatment for prostate cancers. ADT is effective initially but a majority of tumors relapse with castration-resistant prostate cancer (CRPC), from which most patients eventually die. CRPC is driven primarily by aberrant activation of AR in the milieu of castrate serum levels of androgen 1 . On the other hand, approximately 25% of the men who die of prostate cancer have tumors with a neuroendocrine phenotype associated with low AR signaling and poor prognosis 2,3 . With the recent introduction of new generation potent AR pathway inhibitors, such as abiraterone and enzalutamide, the incidence of NEPC has increased, which is associated with a poor outcome 4,5 . Our knowledge of NEPC biology is still very limited and currently there is no effective treatment for NEPC. The mechanisms of CRPC progression, particularly pathways involved in the development of neuroendocrine prostate cancer (NEPC), need to be better understood for the development of future effective treatments for NEPC 3,4,6 .
We and others have previously shown that ADT leads to activation of CREB, which in turn promotes neuroendocrine differentiation (NED) of prostate cancer cells 7,8 . In AR-positive prostate cancer cells, CREB-binding protein (CBP), a histone acetyltransferase, has been shown to act as an AR coactivator in transcriptional activation of AR target genes 9 . However, it is still largely unclear how CREB activation promotes AR-indifferent NEPC. Elucidation of this mechanism is crucial for our understanding and developing treatments of CRPC/NEPC.
Another mediator potentially important for NEPC is polycomb repressive complex 2 (PRC2), which establishes transcriptional repression by tri-methylating lysine 27 of histone H3 (H3K27me3) 10,11 . The major enzyme for catalyzing this histone mark is EZH2 (enhanced zeste homolog 2) 12 , which is overexpressed in several solid tumors 11,13 . EZH2 expression and its PRC2 activity are particularly high in NEPC 6,14,15 . Overexpression of EZH2 in prostate cancer cells is known to promote prostate cancer cell proliferation and migration (review 11 ). It remains incompletely understood whether and how EZH2 directly contributes to NED, and what biological processes are responsible for elevated PRC2 activity in NEPC cells 6,16 .
Angiogenesis plays a crucial role in prostate cancer survival, progression, and metastasis 17 . NEPC is known to be highly vascularized 18,19 . Angiogenesis is a complicated process that is dependent on switching the balance between activators and inhibitors of angiogenesis 20 . VEGF and several neurosecretory peptides, such as bombesin and gastrin, are known to promote angiogenesis in NEPC 21 . However, it is unknown what endogenous angiogenic inhibitors are involved in angiogenesis regulation in NEPC and whether EZH2 overexpression in NEPC cells contributes to angiogenesis. Thrombospondin 1 (TSP1 or THBS1) was the first identified endogenous inhibitor of angiogenesis. It potently inhibits angiogenesis directly by interfering with endothelial cell migration and survival, and its suppression results in increased angiogenesis 22 . Interestingly, TSP1 is among a list of potential EZH2-repressed targets in gene expression profiles of prostate cancer cells upon EZH2 modulation 23 . However, confirmation and characterization of an EZH2-TSP1 relationship was still lacking.
Molecular links between NED and angiogenesis in NEPC have been largely unclear. In this study, we have uncovered functional connections among ADT, CREB activation, EZH2-mediated epigenetic repression, NE phenotypes, TSP1 expression, and angiogenesis in prostate cancer cells. Our results indicated that ADT-activated CREB promotes angiogenesis and NED through enhancing PRC2 activity of EZH2 that in turn upregulates NE markers and downregulates TSP1.

Results
ADT-induced CREB activation is critical for neuroendocrine phenotype. To determine whether androgen deprivation therapy (ADT) activates CREB, we found that enzalutamide (MDV3100) treatment leads to enhanced CREB activation (as indicated by p-CREB-S133 level) in AR-positive LNCaP and VCaP cells, which is reversed by androgen DHT (dihydrotestosterone) (Fig. 1a). In line with the notion that ADT induces NEPC progression, CREB is upregulated and activated in NEPC NCI-H660, NE1.3, and 144-13 cells, as compared to androgen-dependent prostate cancer (ADPC) LNCaP cells (Fig. 1b). NCI-H660 was isolated from a small cell prostate cancer, a tumor that is composed of pure malignant NEPC cells (ATCC). NE1.3 was derived from LNCaP cells upon long term culturing in charcoal stripped serum (CSS) medium that deprives hormone and mimics ADT 24 . 144-13 cells were derived from NEPC patient-derived xenograft (PDX) MDA PCA-144-13 25 .
As expected from our previous study 7 , CREB overexpression and activation induces the expression of NE markers in prostate cancer cells (Fig. 1c). Here, we further demonstrated that inhibitors of beta adrenergic and PKA/CREB signaling, such as ICI-118,551 (ICI), propranolol (PRO), and PKA inhibitor (PKI), downregulate NE markers CHGA and CHGB in NEPC NE1.3 and 144-13 cells (Fig. 1d). We next determined whether CREB itself is essential for ADT-induced NE marker expression. LNCaP cells expressing doxycycline (Dox)-inducible CREB shRNA (shCREB) or a dominate negative CREB polypeptide (ACREB) 26 were treated with CSS medium or MDV3100, respectively, without or with Dox induction. As shown in Fig. 1e, Dox-induced shCREB reduces CREB and p-CREB levels in LNCaP cells, which reverses NE marker induction by CSS. Similar effects were seen in Dox-induced ACREB with 24 h or 72 h of MDV3100 treatment in LNCaP cells (Fig. 1f). Together, these results indicate that CREB activation is induced by ADT, which is critical for ADT-induced NED of prostate cancer cells.
EZH2 is activated in NEPC. Consistent with the literature 27,28 , treating LNCaP cells with MDV3100 or culturing another ARpositive prostate cancer cell line 22Rv1 in CSS medium has resulted in induction of H3K27me3 and NE markers (Fig. 2a), which suggests that ADT activates EZH2's PRC2 activity. EZH2 activity is known to be elevated in NEPC patient samples and genetically engineered mouse (GEM) models 6,14 . As expected, we found that H3K27me3 levels are higher in NEPC NE1.3, NCI-H660, and 144-13 cells than in ADPC LNCaP cells, as well as higher in NEPC patient-derived xenograft (PDX) MDA-PCA-144-13 tumor than in adenocarcinoma PDX MDA PCA-133 tumor 29,30 (Fig. 2b). To confirm an increase in EZH2's function in NEPC cells, we performed RT-qPCR analysis for several known EZH2-repressed targets, including SLIT2, ADRB2, and DAB2IP 23,31-33 . Indeed, their expression is lower in NEPC NE1.3 and NCI-H660 cells than in LNCaP cells (Fig. 2c). Their expression is also lower in NEPC PDX MDA PCA-144-13 tumor than in adenocarcinoma PDX tumor MDA PCA-133 tumor (Fig. 2d). These results confirm that EZH2 activity is higher in NEPC cells and elevated by ADT.
ADT activates EZH2 through PKA/CREB signaling. Our preceding results are consistent with the literature that has separately indicated an elevation of CREB and EZH2 activity in NEPC cells. However, these two proteins have not been directly linked in NEPC and prostate cancer progression. We hypothesized that ADT and CREB activation enhances EZH2 epigenetic activity. To test this hypothesis, we employed pharmacological and genetic perturbations for CREB activity, followed by examination of EZH2 activity, as assessed by H3K27me3 level and EZH2 target gene expression. As showed in Fig. 3a, MDV3100 increases p-CREB and H3K27me3 levels in LNCaP cells, which is reversed by Dox-induced shCREB expression and CREB silencing. Conversely, overexpressing either wild type or constitutively active CREB cDNA induces H3K27me3 level in PC3 cells (Fig. 3b). Similarly, CREB activation by treatment with forskolin (FSK) or isoproterenol (ISO) also leads to an increase in H3K27me3 level ( Fig. 3c left panel, and Supplementary Fig. 1a). On the other hand, inhibition of CREB signaling through inhibitor treatments reduces H3K27me3 level in NEPC NE1.3 and 144-13 cells (Fig. 3c right panel, and Supplementary Fig. 1b) Fig. 2a). Similar results were obtained when using CSS medium as an approach for ADT in LNCaP cells ( Supplementary Fig. 2b). These results indicate that ADT activates EZH2 through PKA/CREB signaling. To further characterize the link between CREB and EZH2 activation in patient samples, we measured the levels of H3K27me3 and CREB activation (by pS133-CREB) in a tissue microarray (TM) with 78 cases of human prostate cancer and normal samples. The p-CREB level was found to positively correlate with the level of H3K27me3 (X 2 = 16.4, P = 0.003) ( Fig. 3g and Supplementary Table 1). These results indicate that the CREB/EZH2 axis may be active in human prostate tissues. Overall, these results support our hypothesis that CREB activation promotes EZH2's PRC2 functions.
EZH2 is critical for CREB induction of NE phenotypes. We next determined whether EZH2 is critical for NE phenotypes promoted by ADT and CREB activation. EZH2 has been shown to be overexpressed and/or activated in NEPC 6,14 . However, direct evidence of EZH2 in promoting NE phenotypes is scarce. We first showed that EZH2 indeed enhances NE marker expression. Overexpressing EZH2 increases H3K27me3 level and NE markers CHGA and CHGB in PC3 and LNCaP cells (Fig. 4a). On the other hand, inhibiting EZH2 by its inhibitor GSK126 or DZNEP decreases H3K27me3 level and NE marker expression in NEPC NE1.3 cells (Fig. 4b). Reduction of NE marker expression could also be achieved by expressing a validated EZH2 shRNA 34,35 in NE1.3 cells (Fig. 4c). Reduction of NE markers by shEZH2 was also observed in PC3 cells, which is rescued by overexpressing EZH2 cDNA (Fig. 4d). In determining whether EZH2 is critical for CREB-induced NE phenotypes, we found that CREB signaling activator FSK could no longer induce NE markers when the cells were co-treated with EZH2 inhibitor GSK126 or shEZH2 (Fig. 4e, f and Supplementary Fig. 2c). As expected, GSK126 alone reduces NE marker CHGA (Supplementary Fig. 2c).Together, these results indicate that EZH2 plays a critical role in promoting NE phenotypes and in mediating the process of CREB activation-mediated NE transition. TSP1 is a novel EZH2 target in prostate cancer cells. Previously we reported that CREB activation promote angiogenesis in part through reducing potent anti-angiogenic protein TSP1 expression in cancer cells 36 . NEPC is known to be highly angiogenic 21 . So far we have shown that the CREB/EZH2 axis is important for neuroendocrine phenotypes of prostate cancer cells. We next tested whether the effects of CREB activation in promoting NE phenotypes and angiogenesis are connected through EZH2. EZH2 is induced by VEGF in endothelial cells, which contributes to angiogenesis 37 . It was unclear whether EZH2 expression in cancer cells is critical for angiogenesis. Interestingly, TSP1 was among a list of genes identified as potential EZH2 repressed targets in prostate cancer cells, with no further validation 23 . We postulated that NE phenotype and angiogenesis are connected in NEPC cells through the CREB/EZH2/TSP1 pathway. We first examined whether TSP1 is indeed an EZH2 target in prostate cancer cells. In several prostate cancer cell lines, overexpressing EZH2 represses TSP1, while suppressing EZH2 with inhibitors or shRNA induces TSP1 (Fig. 5a-c). Chromatin Immunoprecipitation (ChIP) with H3K27me3 Ab, followed by  PCR of TSP1 promoter sequence revealed that TSP1 promoter is occupied with H3K27me3 mark, which is reduced by EZH2 inhibitor DZNEP (Fig. 5d). Using a TSP1 promoter luciferase construct 38 , we further demonstrated that transcriptional activity of TSP1 promoter is increased upon treatment of EZH2 inhibitor DZNEP (Fig. 5e). Moreover, upregulation of TSP1 by siEZH2 can be rescued by expressing a siRNA-resistant EZH2 cDNA (Fig. 5f). Collectively, these data establish that TSP1 is an EZH2 repressed target in prostate cancer cells.
TSP1 expression negatively correlates with EZH2 and NE markers. Further supporting the existing of an EZH2/TSP1 axis in human prostate cancers, we found that TSP1 and EZH2 are expressed significantly lower and higher, respectively, in metastatic CRPC than in localized prostate cancers (Grass-o_mCRPC 39 ) (Fig. 6a). Similarly, in the Beltran_NEPC dataset 6 , TSP1 expression is lower, while EZH2 expression is higher, in NEPC than in CRPC-adenocarcinoma (Fig. 6b). As expected, NE markers are upregulated in metastatic tumors vs localized tumors,  and are also expressed higher in NEPC than in CRPCadenocarcinoma ( Fig. 6a, b, and Supplementary Fig. 3a,3b). Consistently, mRNA expressions of EZH2 and TSP1 negatively correlate with each other in human prostate cancers, such as in Grasso_mCRPC (Spearman Correlation Rho = −0.55, P ≤ 1E-6) and in Beltran_NEPC (Rho = −0.53, P ≤ 8E-5) (Fig. 6c). Furthermore, expression of NE markers CHGA and ENO2 is negatively and positively correlated with that of TSP1 and EZH2, respectively ( Fig. 6d and Supplementary Fig. 3c). Similar correlations between NE markers, TSP1 and EZH2 are observed in other prostate cancer datasets, such as TCGA, SU2C 40 , FHCRC 41 prostate cancer datasets ( Fig. 6d and Supplementary Fig. 4). Interestingly, TSP1 expression is negatively correlated with that of EZH2 in the CCLE dataset for 1000 human cancer cell lines of many cancer types 42 (Fig. 6e, Spearman Correlation Rho = −0.51, P ≤ 9.96E-57), which suggests that the EZH2/TSP1 axis exist broadly in human cancer cells. Supporting this broader implication, EZH2 and TSP1 expression negatively correlates with each other in several large TCGA solid tumor datasets, such as breast, stomach, and colorectal cancers, as well as lung squamous carcinoma ( Fig. 6e and Supplementary Fig. 3d). Consistent with TSP1's downregulation in human mCRPCs (especially NEPCs), TSP1 protein and mRNA are downregulated by ADT (via MDV3100 treatment) in LNCaP cells, which is rescued by androgen DHT (Fig. 7a, b). As expected, TSP1 is expressed at lower levels in NEPC NE1.3 and 144-13 cells than in LNCaP cells (Fig. 7c). Mining gene expression profiles in an independent study, we also found that TSP1 is downregulated, while NE markers are upregulated, when LNCaP cells were cultured long term in CSS media in vitro 43 (Fig. 7d). Notably, we observed same expression trends in a LNCaP xenograft tumor dataset 44 , where LNCaP cell-derived xenografts relapsed from treatment with second generation androgen receptor targeted therapy, enzalutamide, and became NEPC (Fig. 7e). Furthermore, Changes in TSP1 expression and H3K27me3 levels upon EZH2 cDNA overexpression (a) and shRNA silencing (b) in prostate cancer cells. c Marked reduction of H3K27me3 and up-regulation of TSP1 in PC3, RWPE, and NE1.3 cells treated with 5 μM of EZH2 inhibitor DZNEP or GSK126 for 24 h. d PC3 cells were treated with 2.5 μM DZNEP or vehicle control (DMSO) for 24 h. A ChIP assay was conducted using IgG control or anti-H3K27me3 antibody. The level of H3K27me3 mark on TSP1 promoter was presented as % of the inputs (Y-axis), which was confirmed through DNA gel electrophoresis. e PC3 cells were transfected with pGL3 firefly luciferase vector with TSP1 promoter along with a pRL-TK renilla luciferase control vector, then treated with DZNEP 2.5 μM and 5 μM for 24 h, followed by Dual-luciferase reporter assay. Plotted on Y-axis are the ratios of firefly luciferase signals vs renilla luciferase control signals in each treatment. f PC3 cells carrying an empty vector or siRNA-resistant EZH2 cDNA were transfected with control siRNA (siCNL) or siEZH2. Total RNA was extracted 72 h after siRNA treatment, and RT-qPCR for EZH2 and TSP1 was carried out, using GAPDH as the internal control. Plotted on Y-axis are average fold changes with s.d.  45 , we found that TSP1 is downregulated, while EZH2 and NE markers are upregulated (Fig. 7f). Additional supporting evidences came from a published dataset for prostate cancer GEM models 46 , where TSP1 is expressed at significantly lower level in prostate tumors from TRAMP mice (a classic NEPC GEM model) than in prostate tumors from NP mice (a classic prostate adenocarcinoma GEM model) (GSE53202, P = 1.36E-07). Together, all these results indicate that TSP1 is repressed in CRPC, especially in NEPC, and its expression is negatively correlated with those of NE markers and EZH2. They also support that the EZH2/TSP1 axis may contribute to the progression of human prostate cancers.
ADT and CREB activation repress TSP1 through EZH2. We next asked whether TSP1 repression by ADT and CREB activation is mediated by EZH2. We first confirmed the downregulation   Supplementary Fig. 3 and 4 of TSP1 by MDV3100 in another AR-positive prostate cancer cell line 22Rv1 (Fig. 8a), similar to Fig. 7a for LNCaP cells. We next examined the role of CREB in repressing TSP1. Downregulation of TSP1 by CSS in LNCaP cells is reversed by PKI (Fig. 8b). Similarly, downregulation of TSP1 and induction of NE markers by MDV3100 is rescued by Dox-induced ACREB that represses CREB (Fig. 8c). On the other hand, overexpressing either wild type or constitutive active CREB cDNA represses TSP1 expression (Fig. 8d), which is rescued by simultaneous silence of EZH2 (Fig. 8e). Concordantly, CREB activation by PKA/CREB activators FSK can no longer effectively repress TSP1 expression when EZH2 is inhibited by GSK126 (Fig. 8f). Similar result was observed for another CREB signaling activator ISO when EZH2 was silenced (Fig. 8g).
To determine whether CREB activation has a direct impact on the H3K27me3 marks on the TSP1 promoter, PC3 and LNCaP cells were treated with beta adrenergic agonist ISO with or without beta adrenergic antagonist PRO, followed by ChIP with H3K27me3 Ab and PCR of TSP1 promoter sequence. As shown in Fig. 8h, the abundance of H3K27me3 marks on TSP1 promoter is increased by ISO treatment, which is reversed by additional treatment with PRO. The elevation of H3K27me3 mark on TSP1 promoter by ISO is more pronounced in LNCaP cells than in PC3 cells, which is consistent with the extent of TSP1 mRNA reduction in these two cell lines (Fig. 8i). Moreover, in line with enhanced EZH2 activity and elevated H3K27me3 level by ADT MDV3100 as shown above, MDV3100 treatment represses TSP1, which is rescued by EZH2 inhibitor GSK126 (Fig. 8j). These results indicate that ADT and CREB activation downregulates TSP1 expression through EZH2-mediated epigenetic repression.
CREB activation induces angiogenesis, depending on the EZH2/TSP1 axis. Given the well-established role of TSP1 in blocking angiogenesis, we next investigated whether activation of beta-adrenergic signaling increases tumor angiogenesis through the CREB/EZH2/TSP1 pathway. Indeed, ISO treatment in prostate cancer cells induces angiogenesis tube formation of SVEC4-10 endothelial cells, which is reversed either by treating prostate cancer cells with EZH2 inhibitor GSK126 or EPZ6438, or by adding TSP1 peptide to the conditioned medium of cancer cells (Fig. 9a). We next directly assessed the contribution of EZH2 expression in cancer cells to angiogenesis. Silencing EZH2 expression by shRNA in prostate cancer cells abrogates ISOinduced tube formation of SVEC4-10 endothelial cells (Fig. 9b). Treatment of NEPC NE1.3 cells with EZH2 inhibitor GSK126 or EPZ6438 also inhibits angiogenesis tube formation of SVEC-40 endothelial cells (Fig. 9c). Moreover, we evaluated the effects of prostate cancer cells on endothelial cell migration, another in vitro assay for angiogenesis, and found that conditioned media  ARTICLE from NEPC NE1.3 cells induced more migration of SVEC4-10 endothelial cells than the conditioned medium from ADPC LNCaP cells (Fig. 9d). We further determined whether TSP1 induction in prostate cancer cells is critical for angiogenesis inhibition upon blockade of CREB signaling. Interestingly, effective TSP1 silencing by two independent shRNAs induces NE markers in PC3 cells (Fig. 9e). As expected, conditioned medium from PC3 expressing an effective shTSP1 attracts more endothelial cell migration than that from PC3-scramble cells. Consistent with above results from angiogenesis tube formation assays, condition medium from PC3 cells treated with CREB signaling inhibitors PRO and ICI attracts fewer migrated SVEC4-10 cells than condition medium from untreated PC3 cells (Fig. 9f). However, PRO treatment in PC3-shTSP1 cells could no longer inhibit SVEC4-10 migration, which suggests that TSP1 is critical for angiogenesis inhibition by PRO treatment of PC3 cells h LNCaP and PC3 cells were treated with 10 μM PKA/CREB signaling activator ISO, with or without 10 μM PKA/CREB signaling inhibitor PRO, for 24 h, followed by ChIP assay using anti-H3K27me3 or IgG control antibody. The amount of H3K27me3 histone marks on TSP1 promoter was measured by quantitative PCR and presented as % of the input (Y-axis). DNA gel electrophoresis was performed to visualize the PCR results. i RT-PCR for TSP1mRNA changes in PC3 and LNCaP cells that were treated similarly as in the ChIP-qPCR above. j LNCaP cells were treated either with DMSO, 10 μM MDV3100 without or with 5 μM GSK126 for 72 h. RT-PCR was done for TSP1 and beta actin (Fig. 9f). Taken together, these results support our hypothesis that beta adrenergic signaling induces angiogenesis and NE phenotype through the CREB/EZH2/TSP1 pathway.   36 . We next compared ADPC LNCaP and NEPC NE1.3 cell growth in vivo and also determined the effects of blocking PKA/CREB signaling in controlling NE1.3 cell growth and angiogenesis. NOD/SCID mice were implanted with two million LNCaP or NE1.3 cells that were previously labelled with luciferase, which were then treated twice daily with either saline for mice with LNCaP or NE1.3 CDXs, or 2 mg kg −1 of propranolol (PRO) for mice with NE1.3 CDXs. The tumor growth rate in these mice was monitored by bioluminescence imaging. At day 25 of CDX growth, as compared to day 1 after cell implantation, LNCaP CDX tumors grow by 4.6 folds in NOD/SCID mice, while NE1.3 CDX tumors grow by 44.3 folds (P = 0.020), which is consistent with the notion that NEPC cells represent more aggressive prostate cancer cells 24 . Notably, NE1.3 CDXs treated with PRO only grow by 10.6 folds ( Fig. 10a and Supplementary  Fig. 5a), indicating that PRO significantly blocks NEPC CDX growth (P = 0.024). Next, we examined the connections among CREB activation, H3K27 trimethylation, NE phenotype, TSP1 expression, and angiogenesis in mouse xenograft tumors. Firstly, IHC staining of xenograft tumors showed that the levels of p-CREB and H3K27me3 are both increased by ISO in LNCaP CDXs and both reduced by PRO in NE1.3 CDXs (Fig. 10b), which is consistent with their positive correlation in human patient samples (Fig. 3g). Secondly, western blotting results confirm the IHC results on p-CREB and H3K27me3, and further show the expected TSP1 changes in xenograft tumors (Fig. 10c). Thirdly, we found that microvessel density (MVD), a common readout of angiogenesis in vivo, is induced by ISO treatment in LNCaP CDXs and reduced by PRO treatment in NE1.3 CDXs (Fig. 10d,  Supplementary Fig. 5b and 5c). Lastly, castration of NOD/SCID mice carrying ADPC LNCaP xenografts activates the CREB/ EZH2 axis, increases H3K27me3 levels, represses TSP1 expression, as well as induces NE marker CHGA and angiogenesis marker CD31. Importantly, Doxycycline-induction of ACREB (a CREB inhibiting polypeptide) in vivo inhibits these effects of castration and xenograft growth (Fig. 10e, f). Together, these results indicate that the CREB/EZH2/TSP1 pathway is responsive to ADT-induced CREB activation in mouse xenografts of prostate cancer cells, and CREB signaling is critical for castration-induced EZH2 activation, angiogenesis, and NE phenotypes in vivo.
In summary, the results in our study illuminate that: (1) CREB activation enhances EZH2's PRC2 activity; (2) ADT activates the CREB/EZH2 axis to promote NED and angiogenesis; (3) NED links to angiogenesis and tumor progression in prostate cancer cells through EZH2-mediated epigenetic repression of TSP1; (4) This pathway is activated by castration in prostate tumor xenografts, which is reversed by repression of CREB signaling; (5) The components on this pathway are accordantly expressed in cancer patient samples. Taken together, our data propose a new model of prostate cancer progression, where ADT and beta adrenergic signaling concordantly regulate PKA/CREB activation, EZH2 activity, TSP1 expression, angiogenesis, NED, and NEPC progression (Fig. 10g).

Discussion
The transition from castration-resistant prostate adenocarcinoma (CRPC) to NEPC has emerged as an important mechanism of treatment resistance in prostate cancer. Through in vitro and in vivo studies, we investigated the still poorly understood mechanisms of NEPC progression, from which we identified a critical signaling axis, consisting of CREB/EZH2, for NED. We also studied the underlying mechanisms of heightened angiogenesis in NEPC, where we elucidated a role of the CREB/EZH2 axis in repressing anti-angiogenic factor TSP1. Our study connects a NE regulatory pathway to angiogenesis and provides critical new insights into the mechanisms of NEPC progression.
This work illuminates a previously unappreciated relationship between CREB activation and EZH2-mediated epigenetic regulation. Although these two biological processes are both critical for CRPC and NEPC development 8,26,47 and hence are expected to be coordinated, this has not been previously demonstrated. On the other hand, CREB is generally considered to function as an epigenetic activator because its activation recruits a histone acetyltransferase, CBP, to acetylate histones and activate transcription 48 . Therefore, it was previously unexpected that CREB activation would lead to an increase in epigenetic repression via EZH2. We have found that activation of CREB, through ADT or beta adrenergic-PKA signaling, robustly increases H3K27me3 levels and NE marker expression in vitro and in vivo, which can be reduced by CREB repression, EZH2 specific shRNA and inhibitors (Figs. 3, 4, and 10). Furthermore, CREB-induced NE phenotypes, TSP1 repression and angiogenesis are dependent on EZH2 (Figs. 4, 8, and 9). These results indicate that CREB and EZH2 are indeed coordinated in CPRC progression, and CREB activation enhances EZH2's epigenetic repression function. We are actively investigating the molecular mechanism of EZH2 activation by CREB signaling. One possible mechanism is through enhancing HDAC2 expression/activity by CREB activation, as we previously demonstrated in the context of chronic biobehavioral stress 36 . HDAC2 and HDAC1 are known to cooperate with EZH2 in epigenetic repression 49,50 . Supporting this potential mechanism, our preliminary study showed that TSA, a HDAC inhibitor, reverses ADT (MDV3100) induction of H3K27me3 level and NE markers, and rescues MDV3100-repressed TSP1 expression ( Supplementary Fig. 2d).
Using several AR-positive ADPC cell lines, such as LNCaP, VCaP, and 22Rv1, we have demonstrated that ADT in vitro, by treatment with enzalutamide (MDV3100) or CSS medium, activates the CREB/EZH2 axis to induce NE makers and repress TSP1. These ADT effects are reversed by addition of androgen (DHT) or repression of the CREB/EZH2 axis. Notably, similar results were found in LNCaP xenograft tumors, using castration to mimic ADT and inducible CREB repression to inhibit the axis (Fig. 10e, f). These results suggest that the CREB/EZH2 axis may play an important role in mediating the effect of ADT to promote angiogenesis and NEPC progression. Several other proteins have been identified as important regulators of NEPC progression, such as RB1, TP53, EZH2, N-MYC, AURKA, SRRM4, REST, DEK, BRN2, PEG10, and SOX2 44,45,[51][52][53][54][55] . It would be interesting to determine to what extent our pathway is linked to these known NEPC regulators.
NEPC is highly angiogenic 18,19 . It was unknown how angiogenesis and neuroendocrine phenotype are connected and what endogenous angiogenic inhibitors are involved in angiogenesis regulation in NEPC. Our work elucidates that NE phenotype and angiogenesis are connected through CREB activation of EZH2 that in turn represses anti-angiogenic factor TSP1, induces neuroendocrine markers and angiogenesis. It is still unclear how EZH2 induces NE marker expression, which is under investigation in our lab.
EZH2 has been shown to play a role in angiogenesis in endothelial cells. Lu et al. 37 reported that VEGF induces EZH2 expression in endothelial cells that in turn represses VASH1. Kottakis et al. 56 identified another pathway in endothelial cells consisting FGF2-activated NDY1-miR-101-EZH2. It was still unclear whether and how EZH2 overexpression in cancer cells contributes to angiogenesis. We have demonstrated that silencing EZH2 or treatment of EZH2 inhibitors in cancer cells inhibit in vitro angiogenesis of endothelial cells, which provides direct evidences supporting a role of EZH2 in cancer cells contributing to angiogenesis. We have further found that anti-angiogenic protein TSP1 is downregulated in NEPC as a target of EZH2 epigenetic repression. Intriguingly, we also found that silencing TSP1 induces NE marker expression in PC3 cells (Fig. 9e), which further supports an intimate connection between angiogenesis and neuroendocrine phenotype. No effective treatment is currently available for NEPC. Consistent with others' reports 6,14 , our study indicates a critical role of EZH2 in NEPC, which suggests that EZH2 may be a valuable target for NEPC. EZH2 inhibitor EPZ6438 is currently under several clinical trials for multiple types of lymphoma, solid tumors and synovial sarcoma, such as clinical trials NCT03010982 and NCT01897571. It is conceivable that EPZ6438 or other EZH2 inhibitors may have some efficacy in treating NEPC, which should be formally investigated in additional preclinical and clinical studies.
Notably, we found that treatment with beta adrenergic antagonist propranolol (a beta blocker) significantly reduced the growth of NEPC cell-derived xenografts in NOD/SCID mice. This result is consistent with a critical role of CREB signaling in NE phenotype and angiogenesis, as demonstrated by us and others 7,8,36,47 . Epidemiology studies have demonstrated that usage of beta blockers, especially PRO, for cardiovascular diseases in some cancer patients are associated with better clinical outcomes in multiple cancer types, such as melanoma, prostate, lung, and breast cancers [57][58][59] . These retrospective studies in cancer patients are in line with mounting evidences supporting an antitumor effect of PRO in cancer cell culture and in mouse xenografts [60][61][62] . Given that beta blockers have been safely used for decades for hypertension and heart diseases, our result suggests that beta blockers may become new therapies for NEPC, likely through combinations with other cancer therapeutics. According to clinicaltrials.gov, beta blocker PRO is being tested in several clinical trials for solid tumors. A main obstacle is the lack of clear understandings of PRO's mechanism of action in vivo as well as a shortage of biomarkers for patient selection and efficacy monitoring 58,63 . Our study suggests that PRO may exert its activity, at least in part, through regulating the CREB/EZH2/TSP1 pathway. The activity of this pathway may also become biomarkers for patient selection and efficacy monitoring, which warrants further studies in the future.
Lineage plasticity promotes conversion of a cancer cell that is dependent on a therapeutic target to another cell that is indifferent to the function of that target. Similar to conversion from AR-dependent adenocarcinomas to AR-indifferent NEPCs, some EGFR-mutant and initially EGFR-dependent lung adenocarcinomas relapse with the appearance of histologically distinct variants that lack EGFR and RB1 expression, but express neuroendocrine lineage markers 64 . Future studies are warranted to investigate the contributions of the CREB/EZH2/TSP1 pathway shown here in prostate cancer cells to the lineage plasticity of lung cancer cells.

Methods
Cell culture. 22Rv1, RWPE, LNCaP, and NCI-H660 cells were purchased from ATCC. The PC3 prostate cancer cells used in this study represent a poorly metastatic PC3 variant that was kindly provided by Fidler 65 and was matched to PC3 cells from ATCC by DNA STR fingerprinting (Biosynthesis Inc). LNCaP, VCaP, 22Rv1, and PC3 cells were maintained in RPMI 1640 media (Mediatech), supplemented with 10% FBS (Gibco) and 1% penicillin-streptomycin. NEPC NE1.3 cells were derived from LNCaP cells after long term culturing in charcoal striped serum (CSS) medium 24 and cultured in phenol red-free RPMI 1640 medium supplemented with 5% CSS (Gibco) and 1% penicillin and streptomycin. The LNCaP and NE1.3 lines were matched to ATCC LNCaP profile by DNA STR fingerprinting (Biosynthesis Inc), confirming that the NE1.3 line was derived from LNCaP cells. Neuroendocrine small cell prostate carcinoma cells NCI-H660 and immortalized non-tumorigenic prostate epithelial cells RWPE-1 were obtained from ATCC and cultured according to ATCC guidance. NEPC/SCPC cell line 144-13 were kindly provided by Maity and cultured as described 25 . 293T cells and mouse endothelial cells SVEC4-10, originally ordered from ATCC, were cultured in DMEM media (Mediatech), supplemented with 10% FBS and 1% penicillinstreptomycin. LNCaP cells carrying Doxycycline-inducible shCREB and ACREB were kindly provided by Hu and cultured as described 26 . Cultures were grown in a 37°C incubator with 5% CO 2 . All cell lines were routinely confirmed to be mycoplasma-free using the Lonza MycoAlert Detection kit (LT07-218).
Western blotting analysis. Cells were washed in ice-cold PBS and lysed in lysis buffer (30 mM Tris, 200 mM NaCl, 1.5 mM MgCl 2 , 0.4 mM EDTA, 20% Glycerol, 1% NP-40, 1 mM DTT) with complete mini protease inhibitor cocktail and PhosSTOP phosphatase inhibitor cocktail (Roche Applied Science). Protein concentrations were determined using Pierce BCA protein assay kit (Thermo Scientific). The samples were then separated by SDS-PAGE and transferred to PVDF membrane (Bio-Rad). The membrane was blocked with 5% skimmed milk in TBST for 1 h at room temperature, followed by incubation of a primary antibody overnight at 4°C. The dilutions and catalog numbers of primary antibodies used are listed in Supplementary Table 2. After washes, the membrane was incubated with HRP-conjugated secondary antibodies for 2 h at room temperature. The blots were then detected by Pierce ECL Western Blotting Substrate (Thermo Scientific) on Blue Basic Autoradiography Films (Bioexpress). The uncropped scans of the most important Western blots and RT-PCR DNA gel pictures are presented in Supplementary Fig. 6 and Supplementary Fig. 7.
Chromatin Immunoprecipitation (ChIP). DNA binding proteins in cells were cross-linked to DNA by 1% formaldehyde for 10 min at room temperature, which was quenched with glycine. Cells were then lysed in SDS Lysis Buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl, pH 8.1 and freshly added protease/phosphatase inhibitors) and sonicated to shear DNA to 300-500 bp fragments using Branson Low Power Ultrasonic Systems 2000 LPt/LPe sonicator (Fisher Scientific). Fifty microliters of supernatant was diluted in 450 µl dilution buffer (1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.1, 150 mM NaCl supplemented with 0.1% NP40, protease and phosphatase inhibitors). Samples were pre-cleared with protein A/G agrose beads for 2 h. Twenty microliters of the post-cleared supernatant was kept as input. The remaining supernatants were incubated overnight with 2 µg anti-H3K27me3 (Millipore) or anti-IgG antibody, followed by 1 h incubation with protein A/G agrose beads at 4°C. The immunoprecipitates were subjected to multiple washes for 5 min each at 4°C in low salt buffer with 150 mM NaCl, high salt buffer with 500 mM NaCl, LiCl buffer with 250 mM LiCl and finally the TE buffer. DNA was recovered after reversion of the protein-DNA cross-links with 0.2 M NaCl and proteinase K. Subsequently, DNA was extracted with phenolchloroform and precipitated with ethanol. Five microliters of DNA was subjected to real time PCR. Primers used to measure the enrichment of TSP1 promoter DNA sequence containing H3K27me3 marks are: F (5′-tggctgtttgcagcagtcggg-3′), and R (5′-ggatctcagcacgggcaggg-3′). The enrichment of ChIP DNA was calculated as percentage of input. The PCR products were resolved electrophoretically on a 2% agarose gel and visualized by ethidium bromide staining.
Luciferase assay. TSP1 promoter firefly luciferase construct that contains TSP1 promoter sequence from −2033 to +750 off the transcription starting site was generously provided by Xiao 38 . Prostate cancer cells were co-transfected with the TSP1 promoter luciferase construct and a TK-renilla luciferase construct that is the internal control to normalize for transfection efficiency. Twenty-four hours later, the cells were treated with either DMSO or EZH2 inhibitor DZNEP at 2.5 μM or 5 μM for another 24 h. The luciferase activities (firefly and renilla luciferase) were determined using the Dual-Luciferase Reporter Assay System (Promega).
Endothelial cell tube formation assay. PC3 and NE1.3 cells were treated as indicated overnight in RPMI1640 with 0% FBS. SVEC4-10 mouse endothelial cells were grown till 70% confluence, serum starved overnight, trypsinized and resuspended in the conditioned media collected from PC3 and NE1.3 cells. Growth Factor Reduced Matrigel (Corning) was thawed at 4°C and 50 μl were quickly added to each well of a 96-well plate and allowed to solidify for 30 min at 37°C. The wells were then incubated at 37°C with SVEC4-10 cells in conditioned medium (20,000 cells per well) and monitored regularly to observe the tube formation. Pictures were taken under 4× and 10× magnification and the number of branches were quantified.
Endothelial cell migration assay. LNCaP and NE1.3 cells were grown till 70% confluence and starved in 0% FBS RPMI1640 medium overnight. PC3 cells expressing shScramble or shTSP1 were either untreated or treated with PRO for 48 h, followed by serum starvation overnight. SVEC4-10 cells were grown till 70% confluence, starved in 0% FBS DMEM medium overnight, trypsinized and resuspended in 0% FBS DMEM. 50,000 SVEC4-10 cells were seeded on top of each Boyden chamber insert (8 μm, BD Biosciences). Conditioned media collected from starved prostate cancer cells (treated or not) were added to the wells underneath the inserts as an attractant. SVEC4-10 endothelial cells were allowed to migrate for 4 h and the inserts were fixed and stained with crystal violet dye to observe migrated cells. Cell migration was analyzed qualitatively by counting the numbers of migrated cells within each high power fields.
Animal experiments. All animal studies followed protocols approved by the Animal Welfare Committee at the University of Texas Health Science Center at Houston. We have complied with all relevant ethical regulations. LNCaP and NE1.3 cells were transfected to express luciferase. Male 6-7 week old NOD/SCID mice (Charles River Laboratories) were implanted subcutaneously with two million of LNCaP or NE1.3 cells in 100 µl 1:1 of PBS and matrigel in both sides of each mouse. The mice were then randomly divided and treated twice daily with either saline for mice with LNCaP or NE1.3 xenograft tumors (8 mice, n = 16 tumors in either group), or 2 mg kg −1 propranolol (PRO) for mice with NE1.3 xenograft tumors (8 mice, n = 16 tumors). The tumor growth rate in these mice was monitored by bioluminescence imaging. Mice were anaesthetized using isoflurane and injected intraperitoneally with 150 mg kg −1 luciferin (Caliper Life Sciences), and tumors were imaged using an IVIS Lumina II platform (Caliper Life Sciences) and analyzed with Live Image software (Caliper Life Sciences).
For the castration and CREB in vivo repression experiment, two million of LNCaP-Dox-inducible-ACREB cells were implanted subcutaneously in both flanks of 15 male 6-7 week old NOD/SCID mice (Envigo). When the majority of LNCaP xenograft tumors were palpable, the 15 mice were randomly divided into three groups (5 mice in each group, n = 10): first group uncastrated and fed with regular chow, second group castrated and then fed with regular chow, third group castrated and then fed with chow containing 200 mg kg −1 Doxycycline (Bio-Serv). According to power calculation and common practice in the fields, xenograft tumor's sample size of 10 is sufficient to detect statistically significant differences among different groups. When the tumors in uncastrated group reached the end point, all mice in the three groups were sacrificed. Xenograft tumors were weighted, photographed and then fixed with formalin and/or snap-frozen. All xenograft tumors were included to compare the differences in mean tumor weights. No visible tumors were excluded.
IHC staining of human prostate tissue microarray. TMA containing 78 cases of formalin fixed and paraffin embedded normal or cancer prostate samples were obtained through clinical protocols approved by the Institutional Review Board of Baylor College of Medicine. We have complied with all relevant ethical regulations. The TMAs were dewaxed in 60°C oven for 2 h and deparaffinized, and rehydrated through incubating in xylene and alcohol series. Tissue sections were subjected to antigen retrieval in 10 mM sodium citrate buffer (pH 6.0) in a food steamer for 45 min. The Universal Elite ABC kit (Vector Laboratories) was used for immunohistochemistry (IHC) staining, according to the manufacturer's instruction. After suppressing endogenous peroxidase activity, the sections were incubated in normal horse serum to prevent nonspecific immunoglobulin binding, then incubated with primary antibody overnight at 4°C. Primary antibodies used were anti-p-CREB and anti-H3K27me3 from Cell Signaling Technology. A streptavidin-HRP detection system was used to reveal specific binding. The stained slides were scored by two investigators who reached consensus, as following: staining intensity −/+, <25% positive cells (weak, score 1); staining intensity ++, 25-50% positive cells (intermediate, score 2); and staining intensity +++, >50% positive cells (strong, score 3).
IHC staining of xenograft tumors. For immunohistochemical staining of xenograft tumors, 5-μm serial sections were deparaffinized, rehydrated and subjected to antigen retrieval with 10 mM sodium citrate (pH 6.0) for 45 min. Before staining, non-specific binding was blocked by incubation with hydrogen peroxide as peroxidase suppressor (Thermo Scientific) and normal horse serum (Vector Laboratories) as blocking buffers, followed by incubating with 1:100 anti-CD31 (Abcam) antibodies in antibody diluent (Biogenex) at 4°C overnight. All sections were briefly washed with phosphate-buffered saline and incubated at room temperature with horseradish peroxidase-conjugated secondary anti-rabbit antibody. Color was then developed by incubation with a DAB substrate kit (Vector Laboratories). Nuclei were counterstained blue with hematoxylin (Sigma Aldrich) and mounted in VectaMount Permanent mounting medium (Vector Laboratories). Isotype IgG controls were used as negative controls for the staining.
Microvessel density. MVD of tumor tissues was assessed through immunohistochemical analysis of the endothelial marker CD31 and determined according to the method previously described 69 . The immunostained sections were initially screened at low magnification (×50) to identify hot spots of the neovascularization. Any yellow-brown stained endothelial cell or endothelial cell cluster clearly separate from the adjacent microvessels, tumor cells and other connective tissue elements was considered a single, countable microvessel. Within the hot spot area, the average vessel count in three hot spots with a 200-fold magnification in each tumor section was considered the value of MVD. The stained slides were scored by two investigators whose consensus was reported.
Microarray data mining. All non-TCGA and TCGA datasets indicated genes were downloaded from cBioPortal 70 and the GEO database (http://www.ncbi.nlm.nih. gov/gds). The transformed and normalized gene expression values from these sources were used in our analysis and statistical calculation.
Statistical analyses. Statistical analyses were performed using GraphPad software and/or online statistics tools. P values were obtained through Student t-test with two tails and unequal variance, unless otherwise indicated. Spearman correlation coefficient and associated P values for gene expression were calculated using GraphPad or a statistics tool at http://vassarstats.net/corr_rank.html and confirmed by another online tool: http://www.socscistatistics.com/tests/spearman/default2. aspx. Chi-square test was used to assess the correlation of p-CREB and H3K27me3 IHC staining on human samples. P values < 0.05 are considered significant. All error bars are defined as s.d. All central values are defined as mean.

Data availability
The authors declare that the data supporting the findings of this study are available within the article and its supplementary information files, or are available upon reasonable requests to the authors.