Proteogenomic Characterization of Patient-Derived Xenografts Highlights the Role of REST in Neuroendocrine Differentiation of Castration-Resistant Prostate Cancer

Death in prostate cancer patients mostly occurs after the development of castration-resistant metastatic disease (CRPC). Castration-resistant tumors develop after androgen ablation therapy or androgen receptor (AR) blockage and are driven by complex mechanisms that involve genetic, epigenetic, and posttranscriptional changes (1, 2). Extensive studies have been carried out to identify tumor driving genetic alterations in CRPC, but a unified genetic model of disease progression is yet to emerge. Amplification of the AR gene locus or point mutations in the AR gene occurs in 60% of tumors that have been subjected to androgen ablation therapy (3). The appearance of ligand-independent AR splice variants or AR posttranslational modification may also contribute to continued AR activation following castration (4). Other common genetic alterations in AR-expressing CRPC include heterozygous deletions of PTEN and, to a lesser extent, p53 inactivation or loss of Rb1 expression (5–8).

One recurrent cellular feature of CRPC is the presence of carcinoma cells that exhibit a neuroendocrine phenotype. Neuroendocrine differentiation (NED) of prostate cells is frequently associated with highly proliferative potential of tumors and increased chemoresistance, and as a consequence understanding the development of NED in CRCP can lead to the development of new therapeutic strategies for CRPC. Neuroendocrine prostate cancer (NEPC) constitutes a distinct class of castration-resistant tumors characterized by the expression of neuronal markers such as Chromogranin A (CHGA), Synaptophysin (SYP), and neural cell adhesion molecule (CD56; ref. 9). NEPCs show similarities to small-cell carcinoma of the prostate (SCCP), which accounts for 0.5% to 2% of all primary prostatic malignancies (10). Patients diagnosed with SCCP have poor prognosis, with a median survival time ranging from 5 to 17.5 months and the 5-year survival rate of 14.3% (11). Moreover, in patients with locally advanced tumors treated with androgen ablation therapy (alone or in a coadjuvant setting), and in patients with CRPC, the extent of NED has been shown to be an independent prognostic factor (12).

The molecular mechanisms driving the development of NE tumors after castration therapy are yet to be fully understood. The loss of AR expression, the identification of recurrent mutations in tumor suppressor genes such as Rb1 and p53, and the amplification of genes such as Aurora Kinase A (AURKA) and MYCN are thought to play important roles in NE transdifferentiation and the aggressiveness of prostate cancer (13–17). Additionally, we recently demonstrated that inactivation of the transcriptional repressor REST in prostate cancer cells results in the enhanced expression of a cluster of neuronal-specific genes including the NE markers CHGA and SYP. AR signaling enhances REST stability through the inhibition of the E3 ubiquitin ligase β-TrcP, which might explain why androgen ablation therapy can, in some cases, promote NE transdifferentiation (18). Recently, REST has been suggested as regulator of the epithelial-to-mesenchymal transition of hormone-refractory prostate cancer (19). Whether REST contributes to other aspects of the NE phenotype in prostate cancer, other than regulating the expression of neuron-specific genes, has not been studied in depth.

In this study, we characterized the proteome changes occurring during the differentiation from adenocarcinoma to NEPC in patient-derived xenograft (PDX) models and investigated the role of REST in this process by matching in vivo expression changes to REST-occupied chromatin regions. Functional studies also revealed an interplay between REST and the tumor suppressor p53 in regulating cell-cycle progression of prostate cancer cells.

Detailed information about the tissues of origin and generation of patient-derived primary xenografts has been published elsewhere (20).

The study cohort is composed of a consecutive series of men (n = 336) with clinically localized prostate cancer who underwent radical prostatectomy (RP) with curative intent from January 1, 2002, until December 31, 2005, at the Department of Urology, Rigshospitalet, Copenhagen, Denmark. This study was approved by the Danish National Committee on Health Research Ethics for the Capital Region (Journal no.: H-6-2014-111). Biochemical failure (BF) was defined as PSA ≥ 0.2 μg/L. The collection of data was approved by The Danish Data Protection Agency (file#2006-1-6256). The validation cohort (Malmö) was previously described (18, 21). All studies were conducted following the principles of the Declaration of Helsinki. All participants gave written or verbal consent. Verbal consent was documented by the physician in the patient journal.

TMA sections were deparaffinized in xylene and rehydrated through graded ethanol- Antigen retrieval performed at pH6. IHC staining was performed using anti SCGN antibody (HPA006641, Sigma-Aldrich, diluted 1:250). The SCGN staining was quantified by scoring the intensity and the percentage of stained tumor cells (0 = negative or positive only in scattered neuroendocrine prostate cells; 1 = weak staining in less than 30% for the tumor cells; 2 = intense staining in less than 20% of the cells on a localized focus; 3 = moderated staining in more than 50% of the cells or intense staining in less than 50%; 4 = intense staining in more than 50% of the cells). Of all cores obtained from the same specimen, the highest expression was selected. Scores ≥2 were considered as high SCGN expression. The association between SCGN expression and clinicopathologic variables was analyzed using χ² test or Fisher exact test for categorical variables and the Mann–Whitney U test for continuous variables. The median time of follow-up was calculated using the reverse Kaplan–Meier method (22). Follow-up for BF was calculated until the latest PSA measurement, whereas time to death was calculated until the latest follow-up date. Cumulative incidences of study endpoints were analyzed using the Aalen–Johansen method for competing risks. Death before BF was treated as competing event when analyzing risk of BF. Other causes of mortality were treated as competing events when analyzing risk of prostate cancer–specific death. Gray test was used to assess differences in the cumulative incidences between biomarker subgroups (23).

Univariate and multivariate cause-specific Cox proportional hazard regression models were performed for risk of BF and prostate cancer–specific death. All tests were two-sided, and P < 0.05 was considered to be statistically significant. Statistical analyses were performed using SPSS (software version 22; IBM), R (R Development Core Team, Vienna, Austria), or GraphPad PRISM.

For the evaluation of the public MCKCC cohort (24), high SCGN tumors were defined as those with expression above the mean expression level of SCGN in nonmalignant neighboring tissue plus 2 times the standard deviation.

LNCaP cells were grown in normal medium to 80% confluency. Cells were crosslinked in 1% cold formaldehyde for 10 minutes and the fixation was stopped with 0.125 M glycine for 5 minutes. Cells were harvested in SDS buffer and then pelleted further to be resuspended in a cold IP-buffer supplemented with phosphatase and protease inhibitors. The DNA complexes were further sonicated to an average size of 300 to 500 bp using a Bioruptor plus sonication device (Diagenode; with settings 30 seconds on, 30 seconds off 7 times). Anti-rabbit IgG (10 μg; Millipore) or 5 μg of anti-REST (Millipore) was added to 50 μg of DNA lysate. Magnetic Dynabeads (Invitrogen) were added into the mix and incubated overnight at 4°C. After washes, the DNA was decrosslinked at 65°C for 12 hours and purified using a Minelute PCR purification kit (Qiagen). The DNA was sequenced at the National High-Throughput DNA Sequencing Centre at Copenhagen University. Raw reads were mapped to the human genome (hg19 assembly) using bowtie (25) and converted to bam- and bed-files using Samtools and Bedtools (26, 27). Only one read per chromosomal position per strand was allowed for downstream visualization and analysis. For peak finding, measurements of distances to TSSs, as well as visualization of tracks, superimposed tracks, and heat maps, data were imported into EaSeq (28) using default settings and an extension of each read from its 5′-end to a total length of 250 bases. Raw ChIP files are available at the Gene Expression Omnibus database (GSE119385).

Whole-protein extracts were purified from frozen specimens as previously described (29). Twenty-five to 40 g of protein extracts was trypsin digested following the Filter-Aid Sample Preparation (FASP) methodology (30). The resulting peptides were fractionated by Strong Anion Exchange chromatography (SAX) into five fractions to reduce sample complexity and maximize depth of proteome coverage (30). Each fraction was then analyzed by LC-MS/MS using the Q-Exactive mass spectrometer (Thermo scientific). Peptides were separated using a 4-hour gradient of water:acetonitrile, on a 30 cm C18 column. MS spectra were acquired in the Orbitrap with 70,000 resolution. MS/MS spectra were acquired in data-dependent mode, after Higher Energy Collisional Dissociation fragmentation, at a resolution of 17,500 (31). The obtained mass spectrometric raw data were analyzed in the MaxQuant environment, version 1.3.7.1 (32), with the integrated Andromeda searching engine and false discovery rate (FDR) cutoff for peptide identification of 0.1 (ref. 33; Fig. 1B). Proteins were identified by searching MS/MS data against the human proteome sequences from UniProt (UniprotKB, 2012). To compare NEPC PDXs and androgen-sensitive (AS) adenocarcinoma PDXs, the differences between the mean protein expression on each group, defined as label-free quantification (LFQ) intensity (34), were evaluated by the Student t test followed by Benjamini–Hochberg correction for multiple testing. FDR values smaller than 0.05 were considered statistically significant. Additionally, proteins with valid LFQ values for each of the NEPC tumors but ≤ 25% values in the adenocarcinoma group were considered upregulated in NEPC. Proteins with no detected in NEPC tumors but with LFQ values in >60% of adenocarcinoma tumors were defined as downregulated in NEPC (Supplementary Table S2).

To compare expression of proteins between PDX on castrated and intact hosts, the difference between the mean LFQ of each group was evaluated by the Student t test (P < 0.05 was considered statistically significant). Additionally, proteins with no valid LFQ value on any of the tumors of one group but at least on 75% of the tumors of the other group were considered differentially expressed as well (Supplementary Table S3). Raw proteomic files are available at ProteomeXchange (PXD009636).

To compare gene-expression variation between mRNA and protein, tumors were ranked according to the expression levels for each gene and distributions were compared using Spearman correlation. The overall correlation to different ontological processes was performed using the 1D analysis tool implemented in the Perseus software as described elsewhere (35). Briefly, the score compares the mean RANK of the correlation values of given group (e.g., Ribosome) and the mean RANK of the correlation values of the rest of the proteins under the formula s = 2× (R1 − R2)/n, where R1 and R2 are the average ranks within the group under consideration and its complement (all remaining proteins in the experiment), respectively, and n is the total number of data points. Gene ontology classification and enrichment analysis were performed using DAVID (36), and results were graphically represented with Cytoscape (37).

C4-2B and PNT2 cells were transfected with two siRNAs targeting REST or a siRNA control, pulsed with BrdUrd (Life Technologies) for 30 minutes, collected by trypsinization, and fixed in methanol. Cells were permeabilized in 0.2 mol/L HCl containing 0.25% Triton X-100 for 30 minutes. Anti-BrdUrd antibody (Santa Cruz Biotechnology) was incubated at 1:200 dilution. As secondary antibody Alexa 488-anti-mouse (Life technologies) at 1:500 dilution was used. Cells were washed in PBS and resuspended in propidium iodide solution containing RNaseA for 30 minutes at 37C. Samples were analyzed on a FACSVerse instrument (BD Biosciences) using appropriate settings.

The Neon transfection system (Invitrogen) and Lipofectamine 2000 (Invitrogen) were used for transfection of siRNA into prostate cancer cells following the manufacturer's instructions. The preparation of cell lysates and Western blot has been described (38). Viability analyses were performed using the cell proliferation kit I (Roche), following the manufacture's guidelines. For evaluation of cell death, transfected cells were incubated with propidium iodide (Sigma-Aldrich) and Hoechst-33342 (Invitrogen) for 15 minutes at 37°C in the dark and of the number of dead and live cells estimated using the Celigo Imaging Cytometer (Nexcelom Bioscience). RT-PCR was performed as previously described (39). Primers used in this study: β-actin (ACTB), 5′-CTGGCTGCTGA-CCGAGG-3′ and 5′-GAAGGTCTCAAACATGATCTGGGT-3′; TP53, 5′-ACCTACCAGGGCAGCTACGGTTTC-3′ and 5′-GCCGCCCATGCAGGAACTGTTACA-3′; REST, 5′-GGAGGAGGAGGGCTGTTTAC-3′ and 5′-ACCGACCAGGTAATCACAGC-3′; p21 (CDKN1A), 5′-CCTGTCACTGTCTTGTACCCT-3′ and 5′-GCGTTTGGAGTGGTAGAAATCT-3′; MDM2, 5′-TGTTGTGAAAGAAGCAGTAGCA-3′ and 5′-CCTGATCCAACCAATCACCTG-3′; FAS, 5′-GGGCATCTGGACCCTCCTAC-3′ and 5′-GATAATCTAGCAACAGACGTAAGAACCA-3′; CHGA, 5′GCGGTGGAAGAGCCATCAT-3′ and 5′-TCTGTGGCTTCACCACTTTTCTC-3′; SYN1, 5′-GCACGTCCTGGCTGGGTTTCTGGG-3′ and 5′-AGGCTACCCGTCAGACATCCGTCTC-3′.

We analyzed the proteome of 17 prostate cancer PDX tumors, of which 4 tumors were obtained 3 weeks after castration of the host and in 2 cases tumors relapsed after castration and were collected (Fig. 1A; ref. 20). On average, 4,881 proteins were identified in each of the tumors (Fig. 1B). Unsupervised hierarchical cluster analysis revealed a high positive correlation between all samples (0.85 ± 0.05; Fig. 1C). Untreated tumors derived from the same patient such as PDXs in the LTL-313 series (0.93 ± 0.02) or PDXs LTL-352 and LTL-370 (0.96) are grouped together, reflecting a common genetic origin. On the other hand, the response to castration therapy shows a larger degree of variability. PDX LTL-311 shows little alteration of proteome expression after therapy (LTL-311-X), while PDX LTL-331 shows a more pronounced response. The proteome of PDX LTL-331 after 3 weeks of castration (LTL-331-X) shows higher similarity to tumors with NEPC PDXs (LTL-370 and LTL-352), than to the parental untreated PDX (LTL-331), suggesting that a castration-induced transdifferentiation process occurred in this case (Fig. 1C). Indeed, once castration resistant, the LTL-331R PDX shows a neuroendocrine phenotype characterized by low expression of AR-regulated genes and elevated expression of neuronal-specific genes (Supplementary Tables S1 and S2; ref. 20). In contrast, castration-resistant LTL-313BR PDX exhibits a proteome profile similar to the parental LTL-313B PDX, reflecting both the genetic similarity between LTL-313B and LTL-313BR and the AR-driven phenotype (Supplementary Table S2; ref. 20).

Interestingly, despite the efficacy of androgen ablation therapy in limiting tumor growth of advanced prostate cancer, limited proteome alterations were observed only in tumors collected 3 weeks after castration of the host compared with tumors propagated in intact mice (Supplementary Table S3). Tumors on castrated hosts showed signs of reduced androgen signaling, as suggested by the lower expression of androgen-regulated proteins such as SARG, DHCR24, and FASN and of proteins involved in fatty acid biosynthesis (FASN, PCCB, and ACSM3). Castration also resulted in reduced expression of proteins involved in rRNA processing and DNA metabolism (DDX56, HEATR1, NOB1, RPS27A, PCNA, and ATR), while leading to elevated expression of proapoptotic proteins such as BCL2L1. Moreover, proteins related to axon guidance (DPYSL2, MATN2, SPTAN1, and SPTBN1) increased their expression, although the canonical NE markers did not consistently show upregulation at this point after castration (Supplementary Table S3).

These findings demonstrate that label-free quantitative proteomic profiling of PDXs can precisely segregate tumor samples based on patient/genetic origin. In addition, it can also provide unbiased information on effects of treatment and tumor progression that cannot be obtained through genetic profiling.

Transcriptomic profiling is widely used to estimate protein expression in biological samples under the presumption that changes in mRNA levels would translate into changes in protein abundance. To test this hypothesis, specifically in prostate tissue, and to identify concordant and discordant mRNA/protein pairs, we compared how changes in mRNA levels correlate to protein expression changes for gene product across all the samples analyzed. We found a positive but limited correlation (average of 0.28) between mRNA and protein variation across samples (Fig. 1D). The correlation between protein/mRNA pairs variation measured across the different experiments does not seem to be a function of abundance of the mRNA or the protein, indicating that the calculated correlation coefficients are not likely to be an artifact of the measurement methodologies (Supplementary Fig. S1A). Overall, genes involved in intermediary metabolism of lipids and carbohydrates show positive mRNA/protein correlation, while genes related to ribosomal biogenesis, translational control, as well as mitochondrial proteins show close to random correlation (Fig. 1D). Therefore, transcriptomic studies involving the latest functional categories must by cautiously interpreted.

Androgen deprivation therapy (ADT) can induce regression of most advanced prostate cancer, at least temporarily, before CRPC arises. Primary or therapy-induced NEPCs are highly aggressive, but their proteome is poorly characterized, beyond the fact that they express high levels of neurosecretory proteins and have reduced expression of AR-regulated proteins (14). Therefore, we compared the protein expression profiles of NEPCs to those from AR-dependent adenocarcinomas. We identified 861 proteins (FDR < 0.05; Supplementary Table S2) differentially expressed between these tumor types. Among the proteins with differential expression, we found some previously validated by IHC (e.g., AR, PSA), as well as some of bona fide markers of NED (e.g., SYN1, SYP, and CHGA), thereby supporting the validity of our proteomic data (20). Gene ontology enrichment analysis of the biological processes connected to differentially expressed proteins showed that NEPCs have increased expression of proteins that are normally associated with neuronal differentiation and of proteins involved in mitotic cell-cycle and cell division associated with chromatin remodeling processes (Fig. 1E), in line with the highly proliferative phenotype often reported for NEPC (14). Additionally, we found a significant upregulation of proteins involved in DNA-repair mechanisms that have not been previously recognized to play a major role in NEPC (Fig. 1E). This includes proteins involved in homologous recombination (RAD51, PARP1, XRCC1, RECQL, and RPA2), nucleotide-excision repair (PCNA, POLD1, POLE, and RFC5) and mismatch repair (MSH2 and ERCC1) among others (Supplementary Table S2). Enzymes involved in nucleotide metabolism such as the thymidylate synthetase (TYMS) were also found upregulated in NEPCs, presumably to support DNA synthesis. On the other hand, NEPCs expressed overall lower levels of mitochondrial proteins, including those with functions in the catabolism of fatty acids, amino acids, and in the tricarboxylic acid cycle (Fig. 1F) and also of lysosomal proteins and of proteins involved in cellular response to oxidative stress. In good agreement with the protein/mRNA expression correlation, enrichment in GO terms associated with metabolic pathways involving mitochondrial and lysosomal proteins was only obvious in proteomic analysis and not through transcriptomics (Supplementary Figs. S1 and S2), suggesting a higher degree of posttranscriptional regulation on the expression of these genes.

Overall, these data suggest that rapidly growing NEPCs restrict their cellular activities to cell division. Reduced expression of mitochondria proteins suggests a greater reliance on glycolysis to fulfill their energetic needs (Fig. 1E and F).

Expression of neuron-specific proteins is a hallmark of NEPCs (Fig. 1). In previous studies, we identified the transcriptional repressor REST as an important regulator of NE transdifferentiation in response to castration and other stimuli in prostate cancer cells (18). In order to investigate whether REST function extends to the regulation of proliferation and/or metabolism-related genes, we performed an unbiased analysis of REST location on chromatin of prostate cancer cells using chromatin immunoprecipitation followed by DNA sequencing (ChIP-Seq), as an attempt to identify gene directly repressed by REST.

REST-bound chromatin regions were isolated from AR- and REST-expressing LNCaP cells and identified by next-generation sequencing in two independent experiments. The analysis identified 2,615 DNA regions significantly increased in REST-immunoprecipitated chromatin, common to both biological replicates, as compared with nonspecific IgG (Fig. 2A). The validity of these data was confirmed by the significant enrichment of the RE-1 DNA motif within REST-occupied regions (P < 10⁻³⁰⁷), the bona fide recognition site for REST binding on DNA. REST binding to the promoter regions of known target genes such as CHGA and SYP (Fig. 2B) as well as to BDNF, SYN1, SCGN, and CHGB (Supplementary Fig. S3) was also observed, further validating our experimental approach.

We then identified which of the genes with increased expression in the NE compared with adenocarcinoma PDX tumors (20) had REST binding regions in their vicinity (within 100 Kb), which occurred in 26.82% of the cases (Fig. 2C; Supplementary Table S4). Gene ontology enrichment analysis of the biological process of these groups of genes revealed that only genes involved in neuronal development related categories show an increased proportion in REST binding compared with the general distribution (P< 0.007, Fig. 3D).

In order to study the functional significance of the REST-bound regions, we analyzed how the distance between REST-occupied regions was related to the expression of genes in their vicinity. All genes significantly regulated between NEPC and AS adenocarcinoma tumors were sorted according to their distance to the nearest REST peak. We found that having a REST-occupied region within 10 kb of a gene significantly (Fisher P < 0.0001) increases the chances of that gene being upregulated in NEPC tumors compared with adenocarcinomas (Fig. 3A). Thus, of the genes within the vicinity of the REST-occupied region, there were 4 times more genes upregulated than were downregulated (217 vs. 51) in the NEPC tumors relative to AS tumors (Fig. 3B). Gene ontology enrichment analysis for the biological functions of these genes showed that genes overexpressed in NE located close to a REST-occupied region relate to neuronal differentiation and function, while genes related to proliferation, cell-cycle progression, and metabolism were mostly absent (Fig. 3C and D). These findings suggest that loss of REST during NED in prostate cancer is a key driver of the expression of neuron-specific genes but seems to have limited direct impact on other processes relevant to NEPC tumorigenesis, such as cell-cycle progression, DNA repair, cell motion, or changes in intermediary metabolism.

In order to functionally evaluate the effect of REST in cell proliferation, castration-resistant, AR-expressing C4-2B cells were transfected with two independent siRNAs targeting REST (Fig. 4A), and cell viability was measured by MTT assays and compared with cells transfected with control siRNA. Surprisingly, REST knockdown resulted in reduced C4-2B cell viability (Fig. 4B) and increased the number of dead cells over time (Fig. 4C). Cell-cycle profile analysis revealed that cell cultures with reduced levels of REST showed an increased number of cells in G₁ phase and fewer cells in S phase (Fig. 4E). These results suggest a function of REST in sustaining cell-cycle progression, which contrasts with the highly proliferative phenotype of low-REST-expressing NEPC tumors. We then asked whether REST depletion would alter the expression of checkpoint regulators such as p53 and Rb1, thereby resulting in cell-cycle arrest. REST knockdown in C4-2B cells did not affect total levels of Rb1 and had limited direct influence in p53 total levels (Supplementary Fig. S3C). However, the expression of the p53 downstream effectors such as p21 (CDKN1A), FAS, and MDM2 was clearly induced (Fig. 4A, D, and E), suggesting the p53 pathway becomes activated upon REST depletion. Importantly, simultaneous downregulation of REST and p53 was able to rescue the cell-cycle inhibition caused by REST knockdown alone (Fig. 4C, D and F), suggesting that checkpoint inactivation is essential for cell division in the context of reduced REST expression. Indeed, low levels of REST and mutation of TP53 are commonly observed in NEPC tumors (7, 40, 41). Further evidence of the interplay between REST and p53 was obtained by knocking down REST in the PNT2 prostate cell line. PNT2 cells were immortalized by transfecting normal prostatic epithelial cells with the SV40 large T and small t antigens (42). These viral proteins inhibit, among others, p53 function (43). Thus, REST knockdown in PNT2 cells resulted in increased expression of NE markers (Fig. 4G and H) but had no inhibitory effects in cell-cycle progression (Fig. 4I).

Overall, these results indicate that during NED in prostate, loss of REST induces the expression of neuronal genes, while negatively regulating cell-cycle progression through activation of p53. Thus, only tumors that have overcome p53 activation associated with REST (and AR; refs. 9 and 13) loss would be able to grow as NEPCs.

Prostate cancer tumors are heterogeneous and often exhibit mixed phenotypes, including the expression of neuronal proteins (9). Expression of neuroendocrine proteins has been associated with mutation-driven cell lineage switching (44) as well as reduced AR activity. Because both the processes are associated with increased tumor aggressiveness, we hypothesized that expression of REST-regulated proteins in primary tumors could have prognostic value. SCGN was the most highly upregulated protein in NEPC tumors (fold change > 150; P < 0.003) in our proteomic profile among the ones with REST binding in the promoter region (Supplementary Fig. S3A). We first tested the prognostic value of SCGN gene expression using data from a publicly available cohort of radical prostatectomies (n = 131; ref. 24). As expected, only a handful of tumors exhibited high SCGN expression, but most of them were associated with BF after RP (HR log-rank: 4.82; Supplementary Fig. S3D). This prompted us to validate these results using a Copenhagen Prostate Cancer cohort composed of 336 consecutive prostatectomies operated at Righshospitalet (Copenhagen, Denmark) between 2002 and 2005 (Fig. 5B, Supplementary Table S5). We analyzed the expression of SCGN by IHC (Fig. 5A). Tumors were classified according to the expression of SCGN as “high” and “low” (see Materials and Methods). High SCGN protein expression was detected in 46% of the cases (n = 147), and its expression correlated with RP Gleason score (GS; P = 0.02; Supplementary Table S5). Analysis of the 10-year cumulative incidence of BF was 36.2% (95% CI, 28.6–43.7) in the SCGN-low group compared with 50.1% (95% CI, 41.8–58.4) in the SCGN-high group (P = 0.0007; Fig. 5C). Moreover, univariate Cox regression analysis found a statistical association of SCGN expression with the risk of suffering BF after RP (HR, 1.7; P = 0.002; Table 1A). A trend toward this association being true was maintained after multivariate analysis (HR: 1.4; P = 0.07; Table 1A). No association was found, however, with other clinical endpoints such as development of CRPC or prostate cancer–specific death (Supplementary Table S6). We further validated these results using an independent cohort of RP collected at the Malmö University Hospital (18, 21). The incidence of tumors with high SCGN expression was lower in this cohort (19%). Given the foci pattern of SCGN expression in primary prostate tumors, this lower incidence is most likely related to the fewer cores per RP sampled when building the TMA: two in the Malmö cohort compared with 4 to 12 cores in the Copenhagen cohort. In this cohort, SCGN expression also tended to correlate with increased risk of suffering from BF after RP (HR = 2.8; P = 0.06; Supplementary Fig. S3D).

Clinical trials over the years have shown that prostate cancer patients bearing tumors with GS ≤ 3+4 have reduced risk of suffering biochemical relapse and of dying with prostate cancer after RP than those with higher histologic grades (45). Consequently, more conservative regimens such as “active surveillance” are being implemented to manage these patients. In order to test whether SCGN could serve as a prognostic biomarker for these patients diagnosed with “low-risk” tumors, we studied the association of SCGN expression to biochemical relapse in univariate and multivariate Cox regression analyses on patients with GS ≤ 3+4 tumors. Strikingly, both analyses retrieved statistically significant association of SCGN expression with BF (Table 1B). Moreover, the 10-year cumulative incidence of BF was 27.8% (95% CI, 19.9–35.7) in the SCGN-low group compared with 45.2% (95% CI, 35.2–55.2) in the SCGN-high group (P = 0.0007; Fig. 5D).

These results suggest that in patients managed with conservative regimens, elevated SCGN expression might be considered as a sign of aggressiveness, suggesting the earlier application of curative treatments such as RP or radiation.

Patients with prostate carcinomas with NED have limited therapeutic options and subsequently poor prognosis. In order to better understand the processes involved in NED in prostate cancer, we have, for the first time, characterized the proteome differences between NEPC and prostate adenocarcinoma PDXs using quantitative mass spectrometry–based proteomics. NEPCs exhibit enhanced levels of proteins involved in the regulation of cell proliferation and response to DNA-damage stress and express lower levels of mitochondrial and lysosomal proteins, indicating the use of alternative pathways for energy production. Expression of neuron-specific proteins in NEPCs was directly linked to the reduced expression of the transcription repressor REST. Expression of one of these proteins, SCGN, correlated with increased risk of BF after RP.

We recently characterized the proteome of primary prostate tumors and detected increased mitochondrial activity and enhanced expression of proteins involved in oxidative phosphorylation and TCA cycle when compared with neighboring benign prostate tissue (29, 46). Here, we observe that NEPCs exhibit reduced levels of proteins involved in these metabolic processes compared with prostate adenocarcinomas (Fig. 1). These changes occur shortly after castration, before other NE features (increased expression of neuronal and proliferation related proteins) were revealed, suggesting this metabolic transition to be related to reduced AR activity. This hypothesis is further supported by the limited chromatin association of REST to metabolic gene promoters (Fig. 3) and the absence of changes in mitochondrial content observed in REST-depleted, AR-expressing cells (Fig. 4A). The opposite regulation of mitochondrial proteins in NEPC compared with prostate adenocarcinoma indicates the existence of fundamental differences on the pathways utilized for energy metabolism, and suggests that NEPCs would have relatively higher glycolytic activity. This conclusion is supported by studies using fluorodeoxyglucose (FDG) positron emission tomography (PET) in patients with suspected neuroendocrine tumors. Visceral metastases, the preferred site for NEPCs, show higher uptake of FDG as when compared to bone metastases, the preferred site for AR–expressing castration-resistant adenocarcinomas (47).

Interestingly, we also detect a significant downregulation of multiple proteins within several cytosolic compartments in NEPCs, including the lysosome, ER, and Golgi. The general reduction of these compartments would be consistent with the reduced cytosol volume that is characteristic of small-cell carcinomas (48). The reason for this phenotype is not clear. We detect reduction in enzymes involved in lipid biosynthesis, which could influence the size of these membrane-rich organelles. The normal prostate luminal epithelium has an important secretory function that is regulated by androgen signaling and, to some extent, maintained in primary tumors (49). Therefore, reduced levels of AR, resulting in reduced expression of AR-regulated genes in NEPCs (Supplementary Fig. S4A and S4B), may have widespread effects in secretory pathways, in addition to the regulation of mitochondria content, a hypothesis that deserves further exploration.

The transcriptional regulator REST plays a key role in repressing the expression of neuronal genes in nonneuron-related tissues. Loss of REST and the expression of REST splice variants unable to exert this repressor function have been implicated in the NED in prostate cancer cells (9, 18, 50). Thus, the expression of SRRM4, a splicing factor involved in the alternative splicing of REST, is commonly upregulated in NEPCs. Interestingly, we found that REST binds the chromatin regions in the vicinity of the SRRM4 gene (Supplementary Fig. S3B), suggesting a feed-forward regulatory loop that would result in promoting the expression of REST-repressed genes. To which extent the function of neuron-specific proteins is critical in the development of NEPCs aggressive phenotype remains to be defined.

Similar to REST (18), the expression of SCGN, a REST-repressed gene, in prostatectomy specimens correlated with BF after surgery, while no correlation with other clinical endpoints, such as disease-specific survival, was found (51). Strikingly, the expression of SCGN in primary adenocarcinomas also occurs in AR-positive cells (Supplementary Fig. S3E), suggesting that the expression of this neuroendocrine marker may occur independently or precede the loss of AR that is characteristic of NEPC. Whether or not SCGN and AR coexpressing tumors have increased chances to develop into full-blown NEPC with small-cell characteristics after RP or castration therapy remains to be elucidated.

The tumor suppression function of REST suggested by its prognostic value for tumor aggressiveness (18) is challenged by our experimental results. Depletion of REST in C4-2B cells resulted in decreased viability, increased cell death and cell-cycle arrest. This phenotype can be rescued by simultaneous p53 depletion (Fig. 4). This suggests that a defective p53 pathway is required for the survival of tumors with low REST expression. Because combined p53 and REST genetic inactivation in mice results in more frequent brain neuroepithelial tumors than caused by the inactivation of p53 alone (52), we cannot entirely exclude that REST inactivation may also contribute to prostate neuroendocrine tumor growth in vivo, through mechanisms not yet fully explored here, such as the paracrine regulation of the tumor microenvironment. The significance of p53 for NE prostate malignancies is further supported by the upregulation of the PEG10 protein in NEPCs, as the inhibition of p53 is also required for the cell-cycle promoting actions of PEG10 in NEPC (53). The fact that tumor suppressor p53 is often mutated in NE prostate tumors (13) supports our hypothesis that depletion of REST expression (as a result of ADTs) would be deleterious for the prostate cancer cells if combined with an active p53 signaling pathway (Fig. 4C).

Expression of proteins that regulate cell-cycle progression is a hallmark of NEPC. Many of these proteins are controlled through the activation of the E2F1 transcription factor (54). We could validate this in the NEPC by analyzing the E2F1 chromatin binding regions in LNCaP cells (55) in relation to our profile of differentially regulated genes in NEPC compared with prostate adenocarcinoma (Supplementary Fig. S4C and S4D). Importantly, E2F1 also binds chromatin in the proximity of two suggested driver genes of NE prostate cancer: MYCN and AURKB (ref. 17; Supplementary Fig. S4E). The mechanisms leading to E2F1 activation during NED in prostate cancer remain unknown; however, these would likely be related to deletion of the Rb1 gene, an E2F1 inhibitor, commonly found mutated in NEPC (7).

After integrating proteomic, transcriptomic, and ChIP-seq profiling, our data suggest that relapse from ADT as NEPC requires a series of concurrent events including loss of AR and REST protein expression, likely in a context of p53 inactivation and inhibition of RB1 function leading to E2F activation and cell-cycle progression (Fig. 5E). Under these conditions, cells would adapt from a predominant oxidative phosphorylation-based metabolism, typical of prostate adenocarcinomas, to a more glycolytic one. Reduction in mitochondrial content would suggest enhanced sensitivity to drugs targeting glycolytic pathways that could be exploited as therapeutic alternative. Finally, expression of neuron-specific proteins, characteristic of NE tumors, is mostly a consequence of low REST expression, which, as judged by our data, has limited direct effect in promoting the highly proliferative phenotype of these tumors. Whether or not the expression of neuronal proteins contributes to survival of prostate cancer cells under conditions of low androgen availability requires further investigation.

No potential conflicts of interest were disclosed.

Conception and design: A. Flores-Morales, Y. Wang, D. Iglesias-Gato

Development of methodology: D. Lin, A. Bartels, J.M. Moreira

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Flores-Morales, T.B. Bergmann, C. Lavallee, T.S. Batth, D. Lin, S. Friis, A. Bartels, A. Krzyzanowska, H. Xue, J.M. Moreira, A. Bjartell, Y. Wang, J.V. Olsen, C.C. Collins, D. Iglesias-Gato

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Flores-Morales, T.B. Bergmann, C. Lavallee, T.S. Batth, M. Lerdrup, G. Kristensen, K.H. Hansen, M.A. Røder, K. Brasso, J.M. Moreira, D. Iglesias-Gato

Writing, review, and/or revision of the manuscript: A. Flores-Morales, S. Friis, G. Kristensen, M.A. Røder, K. Brasso, J.M. Moreira, A. Bjartell, Y. Wang, D. Iglesias-Gato

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Flores-Morales, G. Kristensen, A. Krzyzanowska, K. Brasso, Y. Wang

Study supervision: A. Flores-Morales

Other (pathology): L. Fazli

This work was supported by grants from the Danish Research Council (DFF 4004-00450, A. Flores-Morales), the Movember Foundation (A. Flores-Morales, C.C. Collins), and the Danish Cancer Society (R90-A6060-14-S2, A. Flores-Morales). This work was also supported by The Canadian Institutes of Health Research (Y. Wang), The Terry Fox Research Institute (Y. Wang, C.C. Collins), The Lundbeck Foundation, and The Danish Cancer Research Foundation (S. Friis). The Novo Nordisk Foundation Center for Protein Research, University of Copenhagen is financially supported by the Novo Nordisk Foundation (grant agreement NNF14CC0001).

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.