LSD1-Mediated Epigenetic Reprogramming Drives CENPE Expression and Prostate Cancer Progression

Androgen receptor (AR) is highly expressed in prostate cancer cells and transactivates target genes that play essential role in tumour development and progression (1, 2). Patients with advanced and metastatic prostate cancer are typically treated with androgen-deprivation therapy (ADT) to block AR activity. Although initially effective, resistance to ADT often arises and results in castration-resistant prostate cancer (CRPC). Concordantly, AR expression is generally increased in CRPC and most AR-stimulated genes are highly expressed, indicating that AR transcriptional activity has been substantially restored (2). The recent clinical success of abiraterone (a CYP17A1 inhibitor that further suppresses androgen synthesis) and enzalutamide (a second-generation AR antagonist) has confirmed that AR signaling is a key driver of tumor progression in CRPC (3–5). However, these agents can only retard disease progression, and patients undergoing ADT generally relapse within 1 to 2 years. Therefore, developing novel therapeutic options are in urgent need for the treatment of CRPC patients.

Epigenetic regulators that control key epigenetic alterations, including histone modification and DNA methylation, have been recognized to play key roles in the dysfunction of transcriptional regulation in cancer cells. Among the epigenetic regulators, lysine-specific demethylase 1A (KDM1A/LSD1) is of particular interest in AR signaling. LSD1 is a histone-modifying enzyme responsible for demethylation of histone H3 lysine 4 (H3K4). Best characterized as a transcriptional repressor, LSD1 interacts with the repressive CoREST complex and demethylates enhancer-associated H3K4me1/2 (6–10). In contrast to its well-established transcriptional suppressive function, LSD1 has been reported to interact with AR and function as an upstream active regulator of AR signaling in prostate cancer (11–16).

LSD1 has been characterized as a potential oncogene and a therapeutic target in several cancer types (17–19). In acute myeloid leukemia (AML), LSD1 overexpression blocks differentiation and confers a poor prognosis (20, 21). Targeting LSD1 using small-molecule inhibitors in leukemia and small cell lung cancers is efficient and the efficacy is associated with a DNA hypomethylation signature (22). LSD1 promotes proliferation of androgen-dependent prostate cancer cells (12), yet its function in CRPC remains unclear. To fill this gap, we integrated RNA-seq and ChIP-seq data to systematically assess the function of LSD1 in CRPC. We identified LSD1-mediated epigenetic reprogramming in CRPC, which activates a group of cell-cycle genes, including CENPE, to drive prostate cancer progression. Taken together, our data provide a strong rationale to target CENPE in patients with CRPC.

LNCaP, 22Rv1, and VCaP cell lines were obtained from the American Type Culture Collection (ATCC) in September 2013. Abl cell line was established by Dr. Helmut Klocker's laboratory (23) and obtained by the He laboratory in September 2013. Castration-resistant cell lines V16A, MR42D, and MR42F were established by Dr. Amina Zoubeidi's laboratory and obtained by the He laboratory in October 2014. All prostate cancer cell lines were cultured as recommended by ATCC or the laboratories that generated the cell lines. Regular authentication was done by short-tandem repeat profiling. Mycoplasma testing was routinely done with the MycoAlert Mycoplasma Detection Kit (LT07-118, Lonza). The most recent reauthentication and mycoplasma testing was completed in March 2017. All the cell lines were resuscitated every 3 to 4 months. LNCaP cells were androgen-starved for 48 hours followed by 4 or 24 hours DHT stimulation (10 nmol/L). CRPC cell lines, including abl and V16A cells, were cultured in androgen-deprived condition. Colony-formation assays were performed as described previously (24). Cells were synchronized by the treatment of thymidine (G₁–S phase) or nocodazole (G₂–M phase). Cell-cycle assays were performed using standard PI staining.

siRNAs targeting LSD1/KDM1A and control siRNAs were purchased from GE Dharmacon. siRNAs targeting CENPE and control siRNA were purchased from ThermoFisher. Lipofectamine RNAiMAX transfection reagent (13778150, ThermoFisher) was used for siRNA transfection following the manufacturer's instructions. All target sequences were listed in Supplementary Table S1.

Plasmids of pCMV6-CENPE and empty vector were purchased from OriGene (Cat# RC224375 and PS100001). Lipofectamine 3000 transfection reagent (L3000015, ThermoFisher) was used for plasmid transfection following the manufacturer's instructions.

Cells treated with thymidine/nocodazole or transfected with siRNA were fixed with pre-cold 70% ethanol. After fixation at −20 degrees overnight, cells were treated with 2 mg/mL RNase A solution, and incubated with 0.1 mg/mL PI solution for 30 minutes at room temperature. Data were collected on cytometer using CELLFIT and analyzed using FlowJo.

sgRNAs targeting CENPE were inserted into pLentiCRISPR vector that was generated by Dr. Zhang's lab (25). Lentiviral particle production and infection were performed as described previously (26). In brief, lentiviral vectors were cotransfected with psPAX2 and pMD2G vectors into HEK293T cells. Supernatant was collected at 24 and 48 hours after transfection and stored in −80°C. For infection, 5 × 10⁴ cells per well were seeded in six-well plates and infected with lentiviral supernatant on the following day. The transfected cells were selected by Puromycin for 3 days. The sequences of sgRNAs targeting CENPE were listed in Supplementary Table S1.

Chromatin immunoprecipitation (ChIP) assay was performed using LNCaP cells treated with DHT for 24 hours or abl cells cultured under androgen-deprived condition. Protein A (88845, ThermoFisher) and G (88847, ThermoFisher) Dynabeads were mixed at a 1:1 ratio, and preincubated with antibodies 3 hours before immuneprecipitation. Cells were crosslinked by 1% formaldehyde for 10 minutes and then quenched with 125 mmol/L glycine. After cold PBS wash, nuclear fraction was extracted by 10 mL of LB1 buffer (50 mmol/L HEPES–KOH, pH 7.5; 140 mmol/L NaCl; 1 mmol/L EDTA; 10% glycerol; 0.5% IGEPAL CA-630; 0.25% Triton X-100) for 10 minutes at 4°C. Nuclear fraction was then pelleted and resuspended in 10 mL of LB2 buffer (10 mmol/L Tris–HCL, pH8.0; 200 mmol/L NaCl; 1 mmol/L EDTA; 0.5 mmol/L EGTA) at 4°C for 5 minutes. Nuclear fraction was pelleted again and resuspended in LB3 buffer (10 mmol/L Tris–HCl, pH 8; 100 mmol/L NaCl; 1 mmol/L EDTA; 0.5 mmol/L EGTA; 0.1% Na–Deoxycholate; 0.5% N-lauroylsarcosine; Protease inhibitor cocktail). Nuclear fraction was sonicated in a water bath sonicator (Diagenode bioruptor) to generate chromatin fragments at ∼300 bp. The 1/10 volume of 10% Triton X-100 was added to chromatin lysate. Chromatin lysate was cleared by centrifugation and 1/10 of supernatant was taken as input DNA. The left chromatin lysate was divided equally to antibody-conjugated beads tubes and rotated at 4°C overnight. Antibodies used for ChIP assays are LSD1 (ab17721, Abcam), AR (sc-13062, Santa Cruz Biotechnology), POLII (ab5408, Abcam), and RB1 (#9313, Cell Signaling Technology). The beads were washed by RIPA buffer and elution buffer (0.1 mol/L NaHCO3; 1% SDS; Proteinase K) was used to reverse crosslinking of DNA–protein complex at 65°C for 8 to 16 hours. DNA was purified by phenol–chloroform extraction and subjected to qPCR. All primers used in this study were listed in Supplementary Table S1.

Cells were lysed using RIPA buffer. Cell lysates were cleared by centrifugation (20,000× g × 10 minutes, 4°C), and the supernatant was measured by the Bio-Rad Protein Assay Kit (#5000002, Bio-Rad) according to the manufacturer's protocol. Proteins were mixed with 4× Bolt LDS Sample Buffer (B0007, ThermoFisher Scientific), and boiled for 10 minutes. Protein samples were then separated by 4% to 12% SDS-PAGE (NW04120BOX, ThermoFisher Scientific), and transferred onto PVDF membranes. The immunoreactive proteins were then visualized by ECL substrate (34096, ThermoFisher Scientific). The antibodies used for Western blot were AR (sc-13062, Santa Cruz Biotechnology), LSD1 (ab17721, Abcam), GAPDH (sc-25778, Santa Cruz Biotechnology), CENPE (ab133583, Abcam), and RB1 (#9313, Cell Signaling Technology).

Fresh-frozen tissues utilized in this study were from radical prostatectomy series and from the Rapid Autopsy Program at University Health Network (UHN). For all specimens, informed consent was obtained from all subjects in accordance with the requirements of the Institutional Review Board of UHN. All patient studies were conducted in accordance with the International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS). Tumor specimens were obtained from the UHN Pathology BioBank and evaluated by genitourinary pathologist (T.v.d.K) on scanned, hematoxylin and eosin (H&E)-stained slides. Samples with more than 70% tumor cellularity were used for RNA extraction using the PureLink RNA Mini Kit (12183018A, ThermoFisher Scientific).

Mice were anaesthetized using 2% isoflurane (inhalation) and 1 × 10⁶ V16A prostate cancer cells suspended in 100 ml of PBS with 50% Matrigel (BD Biosciences) were implanted subcutaneously into the dorsal flank on the right side of the mice. Once the tumors reached a palpable stage (∼200 mm³), the animals were randomized and treated with vehicle (DMSO), 50 or 100 mg/kg body weight GSK923295 intraperitonially, respectively, every other day for the first 2 weeks. Growth in tumor volume was recorded using digital calipers, and tumor volumes were estimated using the formula (π/6) (L × W²), where L is the length of tumor and W is the width. Loss of body weight during the course of the study was also monitored. Animal survival was determined based on the tumor sizes reaching maximal volumes allowable (1,500 mm³) under the University Health Network Institutional Animal Care and Use Committee (IACUC). At the end of the studies, mice were killed and tumors extracted and weighed. The tumor RNAs were extracted using PureLink RNA Mini Kit (12183018A, ThermoFisher Scientific). All cells planned for inoculation into mice require the authentication of being free of all mycoplasma contamination. Typically, qPCR profiling of mycoplasma contamination was performed prior to inoculation. All experimental procedures were approved by the University Health Network IACUC under the protocol AUP 4714.4.

LNCaP and abl cells were transfected by siRNAs targeting LSD1 or CENPE for 48 hours, and RNAs were extracted using the RNeasy Mini Kit (#74106, Qiagen) according to the manufacturer's instruction. The RNA-seq libraries were prepared using TruSeq Stranded mRNA Library Preparation Kit (RS-122-2101, Illumina) according to the manufacturer's Low Sample (LS) Protocol. Sequencing was performed at the Princess Margaret Genomic Centre. The sequencing output was aligned to Hg19 using TopHat2 version 2.1.0, and the differential expression was calculated using cuffdiff command of the Cufflinks version 2.1.1 software. The gene set enrichment analysis was made using DAVID 6.7 (27, 28). The geneset/pathway enrichment analysis has been performed using Metacore (Thomson Reuters).

The RNA-seq data for 333 primary samples and 51 benign samples were downloaded from The Cancer Genome Atlas (TCGA). The raw SRA files were downloaded for all samples from respective sources, and aligned to reference human genome version Hg19 using RSEM. The quantification of reads across the genes annotated in RefSeq was computed using DESeq2. The differential expression was calculated using negative binomial test using DESeq2. The read counts for genes were normalized for sequencing depth using DESeq2. Any gene with a fold change of 1.5 and q-value ≤ 0.1 is considered differentially expressed throughout the study unless otherwise specified.

The ChIP-seq libraries were prepared from ChIP DNAs using the ThruPLEX DNA-seq Kit (R400428, Rubicon Genomics) according to the manufacturer's protocol. Sequencing was performed at the Princess Margaret Genomic Centre. All ChIP-seq data were aligned against Hg19 using Bowtie2 version 2.0.5 and the peaks were called using MACS2 version 2.0.10 in its default setting. The total reads for each ChIP-seq data were subsampled to identical number when necessary to avoid bias in peak-calling due to variable sequencing depth.

The normalized log₂ mRNA expression data for MSKCC were obtained from cBioPortal (29). To compute the Kaplan–Meier plot using the 12 CENPE-induced cell-cycle genes for a particular dataset, we adapted a previously described strategy to assign a “risk score” to each patient based on the expression of the 12 genes (30–32). A risk score for any patient is the sum of the product of the expression value of each gene and its Cox regression coefficient. The Cox regression coefficient for each gene is determined using the Glinsky dataset (33). These coefficient values were used to compute the risk score for each patient in the TCGA dataset to perform the survival analysis. The patient group is divided into high-risk and low-risk group at the median of the risk scores. A similar analysis for metastasis and prostate cancer–specific mortality endpoints was conducted using 235 patients treated with RP at the Mayo Clinic as previously described (34). The Cox regression coefficient and the KM plots are generated using the R package “survival” version 2.40 in R version 3.2.3. The significance of the difference between two survival curves is estimated using the “survdiff” function of the aforementioned package followed by χ² distribution function “pchisq.”

To explore the role of LSD1 in the progression to CRPC, we investigated the function of LSD1 in LNCaP, the canonical androgen-dependent (AD) prostate cancer cell line, and abl, an androgen-independent derivative of LNCaP cells. LSD1 mRNA and protein expression levels in LNCaP and abl cells are similar (Fig. 1A), consistent with the observation in clinical datasets that there is no significant difference in LSD1 mRNA level between primary and metastatic CRPC tumors (Fig. 1B; Supplementary Fig. S1A). Silencing LSD1 using two independent short interfering RNAs (siRNA) targeting LSD1 reduced cell proliferation and colony growth of both LNCaP and abl cells (Fig. 1C; Supplementary Fig. S1B). However, when we investigated the LSD1 target genes in both LNCaP and abl cells via RNA-seq with and without silencing LSD1, we observed a large shift in LSD1 target genes (Fig. 1D; Supplementary Fig. S1C). Only 10% of genes upregulated by LSD1 in LNCaP are also upregulated by LSD1 in abl cells. Similarly, only 29% of repressed by LSD1 in LNCaP are also regulated by LSD1 in abl cells. Further, there are many more LSD1 target genes in abl compared with LNCaP cells (Fig. 1D; Supplementary Fig. S1C). When we compared the LSD1-induced genes with gene expression data in patients with metastatic CRPC or primary prostate cancer (35), we observed that these genes are generally upregulated in metastatic CRPC compared with primary tumors (Fig. 1E); however, a larger fraction of the 142 abl-specific LSD1 induced genes (80%) are upregulated in metastatic CRPC tumor than the 57 LNCaP-specific LSD1 induced genes (67%; Fig. 1E). Upon further investigation, we identified that the LSD1-induced genes in abl cells are significantly overrepresented by cell-cycle genes functioning in metaphase checkpoint, whereas the LSD1-induced genes in LNCaP cells are not (Fig. 1F; Supplementary Tables S2–S5).

These data indicate that LSD1 targets different genes in LNCaP and abl cells, although mRNA and protein levels remain largely unchanged. We therefore analyzed the genome-wide binding (cistrome) of LSD1 by performing chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) in both LNCaP and abl cells. Only 27% of the LSD1 binding sites in LNCaP cells are also observed in abl cells (Fig. 2A; Supplementary Fig. S2A). Furthermore, there are 8,322 genomic regions where LSD1 binds only in abl cells, but not in LNCaP cells. When we computed the association between abl-specific LSD1 regulated genes and LSD1 binding sites in both LNCaP and abl cells, we observed that the association of abl-specific LSD1-regulated genes with abl-specific LSD1 binding sites is dramatically higher than with LNCaP-specific LSD1 binding sites (Supplementary Fig. S2B and S2C). This strongly suggests that the shift in LSD1 target genes in abl is due to reprogramming of LSD1 binding.

The reprogramming of LSD1 binding offers a strong premise to identify genes that are critical in progression of androgen-dependent prostate cancer to CRPC. To pinpoint such targets, we focused on the metaphase checkpoint genes that are induced by LSD1 only in abl, but not in LNCaP. Genes from the metaphase checkpoint pathway are most significantly overrepresented among the LSD1-induced genes in abl (Fig. 1F; Supplementary Table S4). There are five metaphase checkpoint genes that are induced by LSD1 specifically in abl: INCENP, BIRC5, CENPE, CENPH, and CENPA. All of these genes except CENPH are also upregulated in CRPC compared with primary prostate cancer in at least one clinical cohort (Fig. 2B). CENPE has much stronger LSD1 binding in its promoters in abl cells compared with LNCaP cells, indicating that it is upregulated directly by the LSD1 reprogramming in CRPC (Fig. 2C; Supplementary Fig. S2D–S2F). When we analyzed the binding of AR, a key driver of prostate cancer and a known interactor of LSD1 (11, 12), we observed a dramatic increase of AR binding at the CENPE promoter specifically in the abl cells (Fig. 2C). The same region also has markedly stronger enrichment of RNA Polymerase II (POLII) and H3K4me2, in abl cells compared with LNCaP cells. CENPE is significantly upregulated in CRPC compared with primary tumor in prostate cancer patients (Fig. 2B; Supplementary Fig. S3A and S3B). We extended our analysis by measuring the expression of CENPE in freshly frozen tissues from 14 and 6 prostate cancer patients with primary and CRPC tumor, respectively. The CENPE expression is more than 13 times higher in patients with CRPC tumor compared with those with primary tumor (Fig. 2D). High expression of CENPE in the primary tumor is significantly predictive of reduced disease-free survival times (Fig. 2E; Supplementary Fig. S3C and S3D). The role of CENPE in prostate cancer progression is further reflected in the clinical samples as its expression increases in tumors with increasing Gleason score (Supplementary Fig. S3E). To complement the higher expression of CENPE gene in CRPC, we measured the protein level of CENPE across seven AR-positive prostate cancer cell lines: two androgen-dependent and five androgen-independent. Consistent with the expression data, we observed that CENPE protein level is much higher in the five androgen-independent compared with the two androgen-dependent cell lines (Supplementary Fig. S3F).

Upon silencing LSD1 by short interfering RNA (siRNA) targeting LSD1 mRNA, we validated the efficiency of the silencing by quantifying LSD1 mRNA using RT-PCR (Fig. 3A). We then measured the expression of CENPE in both LNCaP and abl cells and observed a significant decrease of CENPE expression upon LSD1 silencing only in abl cells, but not in LNCaP cells (Fig. 3A). Consistent with the previous finding that UBE2C is selectively upregulated by AR in CRPC cells (36), we also observed that LSD1 depletes UBE2C mRNA expression selectively in abl. We validated the binding of LSD1 and AR in CENPE promoter by chromatin immunoprecipitation followed by qPCR and observed that the binding of LSD1 and AR is significantly stronger in the CENPE promoter region in abl cells compared with LNCaP cells (Fig. 3B), consistent with the LSD1 and AR ChIP-seq data (Fig. 2C). In addition, silencing of LSD1 significantly decreases AR and POLII binding, as well as H3K4me2 and H3K27ac signals at the CENPE promoter region in abl cells (Supplementary Fig. S4A and S4B). These results collectively suggest that CENPE is a direct target of LSD1 in CRPC.

As an important regulator of metaphase checkpoint, CENPE is mainly expressed at G₂–M phase during the cell cycle progression (37). To determine whether the higher level of CENPE is caused by higher percentage of cells at G₂–M phase in abl cells, we synchronized LNCaP and abl cells to G₁–S and G₂–M phases (Supplementary Fig. S5), and tested the expression of CENPE as well as the binding of LSD1/AR to the CENPE promoter region. We found that the binding of both LSD1 and AR to CENPE promoter region is significantly higher in abl compared with LNCaP, regardless of whether the cells are in G₁–S or G₂–M phase (Fig. 3C). The enrichment of POLII shows a similar trend. Consistently, the expression level of CENPE is much higher in abl cells compared with the expression in LNCaP cells regardless of cell-cycle stages (Fig. 3D). Taken together, these data support that the higher level of CENPE in CRPC is mainly due to the higher level of transcription.

Tumor suppressor retinoblastoma-associated protein 1 (RB1) is often lost in prostate cancer (38–40) and has recently been reported to repress transcription in presence of androgen in prostate cancer (41). To investigate the underlying mechanism that leads to the absence of CENPE expression in LNCaP cells, we hypothesized that RB1 could play a repressive role in controlling the LSD1/AR binding to the CENPE promoter. To test the hypothesis, we first investigated the expression of RB1 in LNCaP and abl cells; both RB1 RNA and protein levels were significantly higher in LNCaP than in abl cells (Fig. 3E). More importantly, the binding of RB1 to the CENPE promoter region in LNCaP cells stimulated by DHT is significantly stronger than that in abl cells regardless of the presence or absence of DHT (Fig. 3F). Consistently, silencing of RB1 in LNCaP cells resulted in a modest increase of LSD1 binding and a significant increase of AR binding at the CENPE promoter region, as well as a significant increase in CENPE mRNA (Supplementary Fig. S6A and S6B). These observations suggest that RB1 binds to the CENPE promoter and represses the transcription of CENPE in LNCaP cells in the presence of androgen.

From the LSD1 silencing experiments in LNCaP and abl cells, we have observed that LSD1 drives the expression of CENPE only in abl but not in LNCaP cells. Because the higher expression of CENPE is association with poor prognosis, we hypothesized that LSD1 reprogramming activates CENPE and promotes transformation of androgen-dependent prostate tumor to CRPC. This would make the oncogenic characteristic of CENPE specific to abl, not LNCaP. To test this hypothesis, we transfected both LNCaP and abl cells with two independent siRNAs targeting CENPE. Upon observing the cell proliferation for up to four days, we found that silencing CENPE significantly reduces the growth of abl cells, but the growth of LNCaP cells remains unperturbed (Fig. 4A and B). Consistently, overexpression of CENPE significantly increases growth of LNCaP cells under hormone-deprived condition (Supplementary Fig. S7A and S7B). Because CENPE is a metaphase checkpoint gene and its primary function is to drive the cells to G₂–M phase during cell cycle, we next wanted to investigate its cell cycle function in both LNCaP and abl cells. We synchronized the LNCaP and abl cells into same cell-cycle stage and treated them with two independent siRNAs targeting CENPE. After the same period of time, the fraction of siCENPE-treated abl cells in G₂–M phase is significantly lower than the abl cells treated with siRNA controls; whereas siCENPE treatment did not have any obvious effect on LNCaP cells (Fig. 4C). This indicates that the progression of abl cells to G₂–M phase is heavily dependent on CENPE, whereas the LNCaP cells do not depend on CENPE to progress to G₂–M phase.

To further investigate the selective oncogenic function of CENPE in CRPC, we performed RNA-seq of both LNCaP and abl cells upon silencing of CENPE using siRNAs. The CENPE-dependent genes are vastly different between LNCaP and abl cells. Only 4 out of the 489 CENPE upregulated genes are also upregulated in LNCaP cells (Supplementary Fig. S8A). Similarly, only 4 out of the 84 CENPE downregulated genes are also downregulated in LNCaP cells (Supplementary Fig. S8B). Metacore gene set enrichment analysis (see Materials and Methods) reveals that CENPE-induced gene sets are also completely different between LNCaP and abl cells, while the four most enriched gene sets in abl cells all cell cycle associated (Fig. 4D; Supplementary Tables S6–S9). Cell-cycle genes that are induced by CENPE in abl but not in LNCaP cells are also generally upregulated in CRPC in clinical samples, and the basal levels of these genes are also generally higher in abl than in LNCaP cells (Fig. 4E). We further investigated the potential of using these 12 CENPE target cell-cycle genes as signature marker for CRPC. We computed a risk score for each patient in multiple cohorts using the expression values of these genes (see Materials and Methods). When we generated the Kaplan–Meier plot using the clinical data and the risk scores, we observed that the higher risk score is strongly associated with poor probability of disease-free survival of the patients (Fig. 4F).

To test whether CENPE could serve as a therapeutic target in CRPC, we further explored the effect of disrupting CENPE using CRISPR/Cas9 approach or CENPE-specific small molecular inhibitor, GSK923295 (42) in CRPC xenografts. As abl cells are difficult to engraft, we chose V16A, an LNCaP xenograft-derived CRPC cell line for the in vivo analysis. V16A also has high expression of CENPE, similar as the abl cells (Supplementary Fig. S3F). We established stable V16A cell lines with a scrambled small guide RNA (sgRNA) as well as two different sgRNAs targeting CENPE exons, following a protocol previously established (25). We found that the tumor growth of CENPE knocked-down xenografts was dramatically repressed, compared with scrambled negative control (Fig. 5A). After 7 weeks' observation, all eight mice in the scrambled sgRNA group developed tumors (100% incidence), while in the two CENPE–sgRNA groups, only one or two out of eight inoculated mice gave rise to tumors (Supplementary Fig. S9A). The mean volume of tumors in the scrambled sgRNA group was significantly higher than those of two CENPE-sgRNA groups (Supplementary Fig. S9B).

We also used GSK923295, a CENPE-specific small molecular inhibitor (42, 43), to assess the effect of small-molecule mediated CENPE inhibition on the tumor growth of V16A xenografts. First, we tested the in vitro cytotoxic effects in LNCaP, abl, and V16A cells and found that abl was more sensitive to CENPE inhibition than LNCaP (Fig. 5B), which is consistent with the phenotypic impact of siCENPE transfection (Fig. 4A and B). Among these three cell lines, V16A was the most sensitive one to GSK923295 (Fig. 5B). We treated V16A xenograft-bearing mice with vehicle (DMSO) or two doses of GSK923295, and observed that GSK923295 decreased tumor volume in a dose-dependent manner (Fig. 5C; Supplementary Fig. S9C). The mean tumor weight of vehicle group was dramatically higher than those of GSK923295-treated groups, and the mean tumor weight of the lower dose (50 mg/kg) group was also significantly higher compared with the 100 mg/kg group (Fig. 5D). However, GSK923295 did not significantly affect the body weight of treated mice (Supplementary Fig. S9D). Furthermore, we tested the mRNA level of three CENPE-regulated cell-cycle genes that were identified in Fig. 4E (i.e., AURKB, PLK1, and E2F1) and observed that the expression of these genes was repressed by GSK923295 in a dose-dependent manner (Fig. 5E). All these results support the notion that CENPE is essential to drive CRPC cell proliferation and tumor growth and could be targeted by therapeutic interventions.

LSD1 has been previously known to be broadly associated with AR-regulated enhancers and functions as a coactivator for AR at these sites (11). Epigenetic reprogramming of LSD1 in CRPC, as revealed in this study, is a novel functional role of LSD1 that can activate a set of genes functionally specific to CRPC. As a lysine-specific histone demethylase, LSD1 suppresses or activates gene expression through demethylation of H3K4me1/2 or H3K9me1/2, respectively. Recent study reported that the demethylation of H3K9me1/2 is dependent on an alternatively spliced exon of LSD1 (44). However, this isoform is not or lowly expressed in prostate cancer. How LSD1 activates gene expression in prostate cancer, especially in CRPC, is still largely unknown and warrants further study.

Our analysis showed that CENPE, a cell-cycle gene, is one of the key LSD1-induced genes in CRPC and drives the progression of prostate cancer. We observed that LSD1 and AR bind to the CENPE promoter only in CRPC, which resulted in induction of CENPE expression. CENPE is a crucial kinetochore-associated kinesin motor protein with an essential role in metaphase chromosome alignment and satisfaction of the mitotic checkpoint (37, 45, 46). CENPE is highly expressed in CRPC cell lines including abl and V16A compared with androgen-dependent prostate cancer LNCaP cells. Consistently, the level of CENPE is much higher in CRPC tumors than in primary ones. There is no significant difference of LSD1 mRNA and protein levels between androgen-dependent and CRPC cells, how LSD1 reprograms its binding to CENPE promoter and activates CENPE expression is intriguing. Interestingly, RB1, a well-characterized tumor suppressor protein, is found to bind at the CENPE promoter region in androgen-dependent prostate cancer cells but the binding is lost in CRPC cells. Consistently, silencing of RB1 in LNCaP cells resulted in increased binding of AR and LSD1 at CENPE promoter region and consequently upregulation of CENPE. This indicates that the detachment of RB1 interaction caused by loss of RB1 expression may be important for LSD1/AR transactivation activity. Further investigation is required to investigate the mechanism of how RB1 regulates the binding of AR and LSD1, i.e., how RB1 prevents the binding of AR and LSD1 at the CENPE promoter in AR-dependent prostate cancer and what triggers the loss of RB1 binding in CRPC.

While the expression of CENPE proves pivotal in driving CRPC, the gene is lowly expressed in LNCaP. Consistently, silencing CENPE has no obvious effect on the growth of LNCaP cells. The hormone-dependent prostate cancer may utilize other members of the centromere protein family to control metaphase checkpoint in the progression of the cell cycle. We observed CENPA and CENPH, two other centromere proteins, to be expressed in LNCaP cells. Interestingly, CENPH has been found to be overexpressed in many cancer types and facilitate cell-cycle progression and cell proliferation (47, 48). Nevertheless, the CRPC-biased expression of CENPE makes it an attractive therapeutic target for CRPC patients.

As an important regulator of AR transcriptional activity and as an epigenetic enzyme, LSD1 has been an attracting drug target for treatment of prostate cancer. Previous report as well as our preliminary study (data not shown) suggests that the two LSD1 inhibitors currently in clinical trials for AML and small cell lung cancer do not exhibit any cytotoxic effects on the cell proliferation of CRPC cells. However, targeting CRPC-specific LSD1-dependent genes resulting from the epigenetic reprogramming of the protein, such as CENPE, represents an alternative therapeutic measure for CRPC. Our data reveal that CENPE inhibition by small-molecule inhibitor GSK923295 significantly reduces the progression of CRPC both in vitro and in vivo in a dosage-dependent manner.

In summary, our study revealed LSD1 mediated epigenetic reprogramming as a novel mechanism in driving prostate cancer progression. LSD1-regulated CENPE may play an essential role in controlling the cell cycle and proliferation in CRPC cells specifically. Our data strongly support that CENPE can be a novel biomarker and therapeutic target for the treatment of CRPC.

E. Davicioni is Chief Scientific Officer at GenomeDx. N. Erho is R&D, Manager at GenomeDx Biosciences Inc. R.J. Karnes reports receiving other commercial research support from GenomeDx and is a consultant/advisory board member for Genomic Health. F.Y. Feng is founder of PFS Genomics and is a consultant/advisory board member for Medivation/Astellas, Janssen, Dendreon, Ferring, Sanofi, and EMD Serono. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Liang, M. Ahmed, H.H. He

Development of methodology: Y. Liang, M. Ahmed, H. Guo, N. Erho, H.H. He

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Liang, H. Guo, C. Lu, C. Poon, W. Han, J. Langstein, B. Li, E. Davicioni, R.J. Karnes, D. Chadwick, T. van der Kwast, A.M. Joshua, C. Cai, H.H. He

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Liang, M. Ahmed, H. Guo, S. Gao, M.B. Ekram, M. Takhar, N. Erho, P.C. Boutros, C.H. Arrowsmith, F.Y. Feng, H.H. He

Writing, review, and/or revision of the manuscript: Y. Liang, M. Ahmed, H. Guo, F. Soares, J.T. Hua, S. Gao, B. Li, M. Takhar, N. Erho, R.J. Karnes, T. van der Kwast, P.C. Boutros, F.Y. Feng, A.M. Joshua, A. Zoubeidi, C. Cai, H.H. He

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Guo, J.T. Hua, E. Davicioni, C.H. Arrowsmith, H.H. He

Study supervision: H.H. He

We thank the Princess Margaret Genomic Centre for high-throughput sequencing support. The results shown here are in whole or partly based upon data generated by the TCGA Research Network.

This work was supported by the Princess Margaret Cancer Foundation (886012001223 to H.H. He), the Canada Foundation for Innovation and Ontario Research Fund (CFI32372 to H.H. He), NSERC discovery grant (498706 to H.H. He), WICC Ontario 20th Anniversary Prostate Cancer Innovation Grant of the CCS (703800 to H.H. He) and CIHR operating grant (142246 to H.H. He), Movember Rising Star awards (RS2014-01 to P.C. Boutros and RS2016-1022 to H.H. He) and discovery grant (D2016-1115 to H.H. He) from Prostate Cancer Canada. H.H. He holds an OMIR Early Researcher Award (ER15-11-232), a Terry Fox New Investigator Award (1069), and a CIHR New Investigator Salary Award (147820). P.C. Boutros is supported by a Terry Fox New Investigator Award, a CIHR New Investigator Award, and by the Ontario Institute for Cancer Research through funding provided by the Government of Ontario. J. Hua is a CIHR Graduate Student Fellowship recipient (GSD146249).

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.