Abstract
Purpose: Current clinical parameters do not stratify indolent from aggressive prostate cancer. Aggressive prostate cancer, defined by the progression from localized disease to metastasis, is responsible for the majority of prostate cancer–associated mortality. Recent gene expression profiling has proven successful in predicting the outcome of prostate cancer patients; however, they have yet to provide targeted therapy approaches that could inhibit a patient's progression to metastatic disease.
Experimental Design: We have interrogated a total of seven primary prostate cancer cohorts (n = 1,900), two metastatic castration-resistant prostate cancer datasets (n = 293), and one prospective cohort (n = 1,385) to assess the impact of TOP2A and EZH2 expression on prostate cancer cellular program and patient outcomes. We also performed IHC staining for TOP2A and EZH2 in a cohort of primary prostate cancer patients (n = 89) with known outcome. Finally, we explored the therapeutic potential of a combination therapy targeting both TOP2A and EZH2 using novel prostate cancer–derived murine cell lines.
Results: We demonstrate by genome-wide analysis of independent primary and metastatic prostate cancer datasets that concurrent TOP2A and EZH2 mRNA and protein upregulation selected for a subgroup of primary and metastatic patients with more aggressive disease and notable overlap of genes involved in mitotic regulation. Importantly, TOP2A and EZH2 in prostate cancer cells act as key driving oncogenes, a fact highlighted by sensitivity to combination-targeted therapy.
Conclusions: Overall, our data support further assessment of TOP2A and EZH2 as biomarkers for early identification of patients with increased metastatic potential that may benefit from adjuvant or neoadjuvant targeted therapy approaches. Clin Cancer Res; 23(22); 7072–83. ©2017 AACR.
This article is featured in Highlights of This Issue, p. 6757
Metastatic prostate cancer accounts for the majority of prostate cancer–specific mortality. However, the ability to distinguish primary prostate cancer with metastatic potential has not been achieved. Here, we show that primary human prostate cancer with concurrent increased expression of TOP2A and EZH2 have a faster time to biochemical recurrence, are more susceptible to progress to a metastatic disease, and thereby result in increased prostate cancer–specific death. Importantly, we demonstrate, using preclinical prostate cancer mouse cell lines models that concurrent increase of TOP2A and EZH2 result in a hypersensitivity to combination treatment with etoposide, a TOP2 poison, with inhibitors of EZH2. Together, our work provides substantial evidence for the utility of TOP2A and EZH2 as prognostic biomarkers and therapeutic targets to identify and intercept aggressive prostate cancer progressing to a metastatic disease.
Introduction
Aggressive prostate cancer, as defined by the progression to metastatic disease after therapy for localized disease accounts for almost two thirds of prostate cancer–related deaths (1). Therefore, the need to more accurately identify lethal prostate cancer, in an effort to personalize medicine for those in need, has led to a large-scale push for biomarker development in the field (2–4). With this, it is important to recognize, at the earliest time point, which patients are at a higher risk for developing aggressive disease and intercept their disease progression with appropriate therapies. We have previously demonstrated that combination therapy using an EZH2 inhibitor with the TOP2 poison etoposide resulted in significantly increased therapeutic efficacy in models of lethal prostate cancer associated with enhanced DNA damage (5). While each of these genes were independently associated with metastatic prostate cancer (6–9), they have never been studied together in the context of early detection of patients with prostate cancer that will relapse after localized therapy with curative intent.
The systematic study of TOP2A and EZH2 mRNA expression in primary prostate cancer samples from two publicly available datasets (10, 11) allowed us to confirm that overexpression of these two genes can select a subgroup of high-risk patients with a decreased time to biochemical recurrence (BCR) and enriched for a mitotic gene signatures. Interrogation of additional datasets also revealed that prostate cancer patients with high expression of both TOP2A and EZH2 mRNA and/or protein were more likely to progress to a metastatic disease and even die of their disease. In support, prostate cancer metastatic datasets (12, 13) also demonstrated that patients with increased TOP2A and EZH2 mRNA expression maintained enrichment of mitotic-related gene signatures, suggesting that indeed TOP2A and EZH2 could provide early detection of primary tumors with metastatic potential. Altogether, our study supports the usefulness of TOP2A and EZH2 as valuable prognostic biomarkers to identify patients that are more likely to develop an aggressive, metastatic disease. Finally, using a novel cell line derived from a highly aggressive genetically engineered mouse models (GEMM) of prostate cancer, we provide the rational that a targeted combination therapy against TOP2A and EZH2 in a subset of patients may have the potential to intercept progression to metastatic disease in the context of neoadjuvant or adjuvant clinical trials.
Materials and Methods
Differentially expressed gene analysis
Gene expression data were analyzed using Bioconductor 3.1 (http://bioconductor.org), running on R 3.1.3. To identify significant differences in gene expression in TOP2A+/EZH2+ and other patients, moderated Student t tests were performed using empirical Bayes statistics in the “Limma” package, and the resulting P values were adjusted for multiple testing using the false discovery rate (FDR) Benjamini and Hochberg method; probe sets with adjusted P value FDR q < 0.05, and logFC > 1.5, were called differentially expressed (DEG).
Primary prostate cancer analysis
Data from The Cancer Genome Atlas (10) was collected from multiple institutes from primary prostatectomies of patients with prostate cancer. Samples then underwent RNA-sequencing on the Illumina HiSeq 2500 platform. Patient data (497 samples) were acquired as RSEM “counts” gene expression data from Firehose, a Broad Institute Software, with 52 matched normal samples. Data from The Memorial Sloan Kettering Cancer Center (11) set was collected from a single institute from primary prostatectomies of patients with prostate cancer. Samples then underwent microarray analysis on the Affymetrix Human Exon 1.0 ST Array platform. Patient data (131 samples) was acquired as log2-transformed gene expression data from the Gene Set Omnibus (GEO) with 29 matched normal samples. Expression data for TOP2A and EZH2 was isolated for each patient separately, then split into high or low expression based on z-scores over normal expression, with the designations high as z > 1.5 and other z < 1.5. Distributions for TOP2A and EZH2 were then overlapped, with patients in the high for both TOP2A and EZH2 being designated as TOP2A+/EZH2+ and all other samples being relegated to the “other” groups. Samples were then preprocessed, by removing all genes consisting of read outs of 0 for more than 80% of samples. The data underwent scale normalization using the “Limma” package in R statistical software, and were then Voom transformed. DEGs were isolated as described above. Those genes were then utilized to generate a scaled heatmap, constructed using the “Gplots” and “Heatmap.2” packages in R. Euclidian distances and hierarchal clustering was applied using “h.clust”, and further principal component analysis (PCA) was later conducted utilizing the “PrComp” package in R, to determine how similar samples were to one another.
Localized prostate cancer from decipher GRID analysis
We used a total of 3,565 genome-wide expression profiles from prostate cancer radical prostatectomy (RP) tissues from the Decipher Genomic Resource Information Database (GRID; Supplementary Table S1). Expression profiling, specimen selection, RNA extraction, and microarray hybridization were done in a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory facility (GenomeDx Biosciences) as described previously (14). Briefly, total RNA was extracted and purified using the RNeasy FFPE kit (Qiagen). RNA was amplified and labeled using the Ovation WTA FFPE system (NuGen) and hybridized to Human Exon 1.0 ST GeneChips (Affymetrix). Quality control was performed using Affymetrix Power Tools and normalization was performed using the Single Channel Array Normalization (SCAN) algorithm (15). For these analyses, patients with expression of TOP2A and/or EZH2 above the 75th percentile in the cohort were categorized as TOP2A+ and/or EZH2+. To test associations between TOP2A expression, EZH2 expression, and outcomes, we grouped patients into TOP2A+/EZH2+ (expression of both genes is greater than the 75th percentile), TOP2A−/EZH2− (expression of both genes is less than the 75th percentile) or patients with the expression of either gene greater than the 75th percentile (TOP2A+ or EZH2+). Then, we conducted survival analysis using the Kaplan–Meier method, with case-cohort reweighting when appropriate, across 4 case–cohort/cohort studies (Mayo Clinic Validation, n = 232; Thomas Jefferson University, n = 133; Johns Hopkins University post-RP, n = 262, Johns Hopkins University post-BCR, N = 213) to assess the prognostic impact of the four groups to predict metastasis outcome after RP or prostate cancer–specific mortality (PCSM) if the data were available. P values were calculated with the log-rank test. We also conducted univariate and multivariate analyses to associate subtypes with clinical outcome after adjusting for other clinicopathologic variables including Gleason score, lymph node invasion, surgical margins, extracapsular extension, seminal vesicle invasion, and pre-operative prostate-specific antigen (PSA) levels in the Mayo Clinic Discovery (n = 545) case control cohort and a “Meta Cohort” from which we utilized genome-wide expression profiles of 751 patients with metastatic outcome follow-up from the Decipher GRID. These patients were pooled from four studies of either case–cohort or cohort design. Patients for these studies came from four institutes: John Hopkins University post-RP (n = 260; ref. 16), Mayo II (n = 235; ref. 17), Thomas Jefferson University (n = 139; ref. 18), and Durham VA (n = 117; ref. 19). A total of 120 nonrandomly selected patients from case–cohort studies were removed before pooling the studies to avoid bias in estimating the HR. A total of 631 patients were thus eligible for analysis, 70 of which developed metastasis. Median follow-up time for censored patients was 8 years and the median age at RP was 61 years. Finally, we used 2,293 prospective samples to associate the expression of TOP2A and EZH2 with metastatic outcome. But since we do not have metastasis outcome, we used the Decipher test risk stratification as a surrogate for metastasis given the well validated evidence that Decipher is the strongest predictor of metastasis currently available (20).
Metastatic prostate cancer analysis
Data from the Robinson and colleagues' metastatic castration-resistant prostate cancer (mCRPC) dataset was collected from 6 different institutes from metastatic lesions in patients diagnosed with prostate cancer (13). Samples then underwent RNA-sequencing on the Illumina HiSeq 2500 platform and patient data (n = 229) were acquired as transcripts per million (TPM). Data from Kumar and colleagues' mCRPC dataset was collected from a single institute from metastatic lesions in patients diagnosed with prostate cancer (12). Samples then underwent microarray analysis on the Agilent-016162 PEDB Whole Human Genome Microarray 4 × 44K platform. Patient data (154 samples from 64 patients) were acquired from the Gene Set Omnibus (GEO) using “GEO2R” as log2-transformed values. Expression data for TOP2A and EZH2 was isolated for each patient separately, then split into quartiles with the designations high, intermediate high, intermediate low, and low. Quartile distributions for TOP2A and EZH2 were then overlapped, with patients in the highest quartile for both TOP2A and EZH2 being designated as TOP2A+/EZH2+ and all other samples being relegated to the “other” group. Samples were then preprocessed, by removing all genes consisting of read outs of 0 for more than 80% of samples. The data underwent scale normalization using the “Limma” package in R statistical software, and were then Voom transformed. DEGs were isolated as described above. Those genes were then utilized to generate a scaled heatmap, constructed using the “Gplots” and “Heatmap.2” packages in R. Euclidian distances and hierarchal clustering was applied using “h.clust”, and further PCA was later conducted utilizing the “PrComp” package in R, to determine how similar samples were to one another.
In silico gene set enrichment analysis
A rank list for each dataset was constructed by taking the gene name and the log2 fold change value for each of the DEGs in each set. The Gene Set Enrichment Analysis (GSEA) tool (http://software.broadinstitute.org/gsea/index.jsp) was then used to analyze relationship of existing gene expression profiles (Hallmarks and Oncogenic Signatures) in the Molecular Signature Database (MSigDB) with the rank list generated. The GseaPreranked tool, with 1,000 permutations, and a failure to collapse dataset to gene symbols, was used for GSEA analysis. All statistically significant data (P < 0.05 and FDR < 0.15) are provided (Supplementary Tables S2 and S3). We acknowledge our use of the GSEA, GSEA software, and Molecular Signature Database (MSigDB; http://www.broad.mit.edu/gsea/; ref. 21).
Tissue microarray analysis of TOP2A and EZH2 (DFCI cohort)
Patients and samples.
A retrospective cohort of prostate cancer patients from a treatment facility of Dana-Farber Cancer Institute (DFCI, Boston, MA) Prostate Clinical Research Information Systems (CRIS) database was identified to include patients who had tissues available from RP or transurethral resections of the prostate (TURP) with 3 predefined tumor microarray (TMA). The IHC staining was performed and a multiplexed tyramide signal amplification method was performed on 4-μm sections of the TMA for detection of TOP2A and EZH2 proteins. TSA-plus fluorescence IHC combined with spectral imaging was used to measure TOP2A and EZH2 expressions/cell positivity. The analysis cohort included 89 patient samples after quality control to ensure for proper assessment of marker values (Supplementary Table S4). The analysis endpoints included time to BCR defined as time from RP to biochemical recurrence and time to lethal disease defined as time from RP to the development of metastases. The distribution of time to BCR or metastases according to TOP2A and EZH2 marker status (low vs. high) in the combination 4 groups used Kaplan–Meier method. Cox proportional hazards model was used to assess the associations of time to events and marker status with estimate HR, 95% confidence interval. The multivariate model adjusted for clinicopathologic factors of age, Gleason score, and pathologic grade was also used. Additional details are provided in the Supplementary Material and Methods.
Definition of TOP2A, EZH2 marker values.
The case-level positivity (= positive T-cell #/total T-cell # × 100 %) for both markers, used a cut-off value on cell intensity readout level to define TOP2A+ and EZH2+ cell. According to the cell data distribution, combined with representative images review, the top 90% cell intensity as chosen to be cut-off value for TOP2A and EZH2, respectively. The marker values were then dichotomized into low versus high categories using the prespecified lower quartile level as cutoff, that is, TOP2A− if ≤ lower quartile TOP2A cell positivity (%) or TOP2A+ otherwise; EZH2− if ≤lower quartile EZH2 cell positivity (%), or EZH2+ otherwise. Subsequently, four groups from the both markers were defined as: (1) TOP2A+/EZH2+ (reference group); (2) TOP2A−/EZH2−; (3) EZH2+; and (4) TOP2A+.
Analysis of gene and protein expression in sKO and dKO murine prostatic tissues
Total RNA extraction was performed using prostate dorsal and lateral lobes of of Pten−/− (sKO) or Pten−/−; Rb1−/− (dKO) prostate cancer mouse models (22). Reverse transcription and quantitative real-time PCR (qRT-PCR) were done as described previously (5) and data were normalized according to Rpl32 levels. The oligonucleotides used for the analysis of gene expression were Top2a (forward: AGGATTCCGCAGTTACGTGG; reverse: CATGTCTGCCGCCCTTAGAA), Ezh2 (forward: GCCAGACTGGGAAGAAATCTG; reverse: TGTGCTGGAAAATCCAAGTCA) and Rpl32 (forward: TTCCTGGTCCACAACGTCAAG; reverse: TGTGAGCGATCTCGGCAC). Detection of TOP2A and EZH2 by fluorescence IHC was done on formalin-fixed, paraffin-embedded prostate tissues from a sKO and a dKO mouse and performed as described for the TMA (Supplementary Materials and Methods).
sKO and dKO cell line establishment
A conditional reprogramming method was used to establish cell lines from sKO or dKO prostate cancer mouse models (22). Briefly, tumor samples from the prostate of a 51-week-old sKO mouse or a 38-week-old dKO mouse was chopped into 2-mm fragments followed by digestion in a mixture of Collagenase/Hyaluronidase (StemCell Technologies, #7912) at 37°C for three hours. Cell clumps were then incubated with 0.25% trypsin for 1 hour on ice followed by Dispase (StemCell Technologies, #7913) and DNaseI (Sigma, #DN25) for 1 minute. Cell suspensions were filtered through a 40-μm cell strainer, and then cocultured with irradiated Swiss-3T3 feeder layer in F medium containing 10 μmol/L Y-27632 (Enzo Life Sciences, #ALX-270-333) for two months. Epithelial colonies surrounded by irradiated fibroblasts were harvested and trypsinized into single cells and plated in a 96-well plate. sKO and dKO cell lines were established from the homogeneous population in one single well and then continuously maintained in F medium (described previously; ref. 24) with the supplement of 10 μmol/L Y-27632 and 1 nmol/L R1881. Both cell lines were finally cultured using DMEM containing 10% FBS.
Analysis of protein expression
Analysis of protein expression in sKO and dKO cell lines was performed as described previously (5). Briefly, cells were harvested and whole-cell lysates was extracted using lysis buffer (PIPES 20 mmol/L, NaCl 150 mmol/L, EGTA 1 mmol/L, MgCl2 1.5 mmol/L, Triton X-100, pH 7.4) supplemented with both protease (Pierce, #88265) and phosphatase inhibitors (Pierce, #88667). Equal amount of denatured protein were resolved on a 4%–15% SDS-PAGE Mini-PROTEAN TGX gel (Bio-Rad, #456-1083) and transferred to polyvinylidene difluoride membrane. Membranes were blocked (5% BSA), washed and then probed with the following rabbit mAbs according to the manufacturer's instructions: anti-EZH2 (Cell Signaling Technology, #5246, 1:1,000), anti-TOP2A (Abcam, #52934, 1:500), or anti-GAPDH (Cell Signaling Technology, #2118, 1:10,000).
Adherent clonogenics
sKO and dKO were plated in a 6-well plate (5 × 104 cells/well). Following treatment as indicated with etoposide (Sigma-Aldrich), EPZ6438, GSK126 (Xcess Biosciences Inc.), or DMSO (vehicle) for 24 or 72 hours, cells were trypsinized and replated either 500 or 1,000 cells per well. Colony formation was assessed 10 days postplating by crystal violet staining.
Cell-cycle analysis
sKO and dKO cells (5 × 104 cells/well) were seeded into 6-well plates (BD Biosciences), left to adhere, and treated as indicated. Following treatments, adherent and nonadherent cells were collected and washed in 1× PBS, and fixed in 50% ethanol at 4°C overnight. Cells were stained with propidium iodide solution containing RNase A (Sigma) for 15 minutes at 37°C. DNA content analysis was performed by using a BD FACSCalibur cytometer.
Statistical analysis
Data are displayed as mean ± SEM. Differences were determined using two-tailed unpaired t tests, using GraphPad Prism software. P values <0.05 were considered statistically significant.
Results
We performed a cross-platform analysis to stratify high-risk prostate cancer patients (aggressive disease) from low-risk patients (indolent disease) within the primary disease setting. In The Cancer Genome Atlas (TCGA; ref. 10) and Taylor, and colleagues' (11) primary prostate cancer datasets, samples were divided into patients with increased levels of both TOP2A+/EZH2+ and other. Unsupervised hierarchical clustering and PCA demonstrated that TOP2A+/EZH2+ primary prostate cancer patients displayed a unique set of differentially expressed genes (DEG; ref. 21) compared with other patients (Fig. 1A and B; Supplementary Fig. S1A and S1B). In the TCGA dataset, high levels of TOP2A and EZH2 were associated with a shorter time to BCR (Fig. 1C), a feature that was also observed in the Taylor and colleagues dataset (Supplementary Fig. S1C). Comparing those patients' tumors with increased levels of TOP2A and EZH2 to patients without increased levels of both genes in the TCGA and Taylor and colleagues datasets generated a DEG list, consisting of 269 and 362 genes, respectively, with 86 of those genes shared between both primary tumor datasets (Fig. 1D; Supplementary Tables S5 and S6). GSEA using the Hallmark and Oncogenic Signatures gene sets revealed that DEG in patients with concurrent TOP2A+/EZH2+ expression in both datasets enriches for similar cell programs geared toward cellular proliferation (Fig. 1E; Supplementary Fig. S2; Supplementary Tables S2 and S3). Additional interrogation revealed that TCGA TOP2A+/EZH2+patients had statistically significant downregulation of CDKN1A, FGFR1, and PMP22, genes that were previously shown to be downregulated in lethal prostate cancer (ref. 23; Supplementary Fig. S3).
Given that high levels of both TOPA2 and EZH2 expression in patients are associated to a poorly differentiated primary disease and a shorter time to BCR, we hypothesized that those patients might be more prone to progress to an advanced metastatic disease. To test this hypothesis, we used 5 retrospective RP cohorts with outcome data (total N = 1,272) and one prospective cohort (N = 2,293; Supplementary Table S1) from the GRID. In the Mayo Clinic Validation study (n = 232), a case–cohort consisting of high-risk prostate cancer patients (17) with a median follow-up of 7 years, we found that TOP2A+/EZH2+ patients had a faster progression to metastatic disease (Fig. 2A). This finding was also confirmed in the Thomas Jefferson University dataset (n = 133), a cohort of prostate cancer patients treated with radiotherapy (ref. 18; Fig. 2B) and in two Johns Hopkins University case–cohorts consisting of post-RP patients who received no therapy until the onset of metastasis (n = 262; ref. 16; Fig. 2C) or of patients who had developed BCR (post-BCR) and again received no therapy until the onset of metastasis (n = 213; ref. 24; Fig. 2D). However, concurrent TOP2A and EZH2 expression in those datasets was not associated to a shorter time to BCR (Supplementary Fig. S4). Importantly, TOP2A+/EZH2+ patients in both Johns Hopkins University cohorts were also more likely to die of their disease compared with patients with low levels of TOP2A and EZH2 expression (both low) as shown by a shorter time to prostate cancer-specific mortality (PCSM; Fig. 2E and F). Strikingly, univariate and multivariate analyses using the Mayo Clinic Discovery dataset (n = 545; ref. 25) demonstrate that concurrent TOP2A and EZH2 expression had a similar prognostic power than a high Gleason grade (>8) in predicting the development of an aggressive, metastatic disease and outperformed all other clinicopathologic parameters analyzed (Fig. 2G and H). Additional univariate and multivariate analyses using a “Meta Cohort” (see the Materials and Methods for details) confirmed the prognostic power of TOP2A and EZH2 expression in predicting prostate cancer progression (Supplementary Tables S7 and S8). Finally, using 2,293 prospective RP patients whose prostate cancer agressivness was stratified by the Decipher test, a validated surrogate of metastasis (20), we found that almost all patients with high expression of TOP2A and EZH2 were categorized as high risk of progressing to a metastatic disease (Fig. 2I). Moreover, multivariate survival analysis of Decipher adjusting for TOP2A+/EZH2+ and other clinicopathologic risk factors revealed that both Decipher and TOP2A+/EZH2+ are significantly associated with metastasis and contain unique prognostic information (Supplementary Table S9). Altogether, we found that high levels of TOP2A and EZH2 expression is consistently associated to the progression to a metastatic and lethal disease.
To evaluate whether TOP2A and EZH2 protein levels could also discriminate patients bearing an indolent from those bearing an aggressive tumor, we performed IHC staining utilizing a tissue microarray (TMA) of primary prostate cancer patients who had a prostatectomy and either did or did not relapse with metastatic disease (Dana-Farber Cancer Institute cohort; Supplementary Table S4). TOP2A and EZH2 protein expression was quantified and patients were then classified in one of four categories (TOP2A+/EZH2+, EZH2+, TOP2A+ or TOP2A−/EZH2−; Fig. 3A). Kaplan–Meier analysis showed that patients with high levels of both proteins had a higher rate and significantly shorter time to BCR and metastatic disease progression (Fig. 3B and C). Moving forward, this initial finding requires external validation with a larger sample size, but our data shows that in addition to TOP2A and EZH2 gene expression levels, protein expression levels predict for patients who will develop metastatic/lethal prostate cancer with high accuracy in small tissue samples (Fig. 3D).
With evidence that expression of TOP2A and EZH2 predict metastatic-free survival from primary prostate cancer samples, it was hypothesized that metastatic prostate cancer samples harboring concurrent upregulation of TOP2A and EZH2 would enrich for similar gene expression sets as identified in our primary prostate cancer analysis. Using the Robinson and colleagues' (13) and Kumar and colleagues' (12) mCRPC datasets, we observed that TOP2A+/EZH2+ patients expressed a unique gene signature as assessed by hierarchical clustering and PCA (Fig. 4A and B). A generated DEG list showed a total of 315 and 144 genes driven by high TOP2A/EZH2 expression in the Robinson and colleagues' and the Kumar and colleagues' datasets, respectively, with a 107 gene overlap (Fig. 4C; Supplementary Tables S5 and S6). Interestingly, as observed with primary prostate cancer, GSEA using DEG characteristic of TOP2A+/EZH2+ patients revealed a consistent enrichment for gene sets related to mitosis in metastatic tumors (Fig. 4D; Supplementary Tables S2 and S3), a hallmark that has been previously associated with aggressive disease (26, 27).
Next we compared primary and metastatic TOP2A+/EZH2+ patient gene signatures from RNA-sequencing and microarray platforms. The DEG lists provided a high level of overlap between the metastatic and primary prostate cancer samples (Fig. 4E and F; Supplementary Table S6) with a core of 45 DEGs conserved in all four dataset (Supplementary Table S6), as well as a high level of overlap in mitotic spindle deregulation and G2–M/mitotic hallmarks (Figs. 1F and 4D). This study and others demonstrate that biomarkers of aggressive prostate cancer are often associated with mitotic deregulation (27–29). Mitotic deregulation and mitotic gene conversion is associated with therapy resistance and increased genome instability (30, 31). This knowledge may give insight into underlying mechanisms as to why our identified patients with increased enrichment for mitotic-related DEG progress more rapidly with BCR, metastatic disease, and ultimately prostate cancer–specific mortality.
To test our biomarkers as targetable actions for therapeutic intervention, we took advantage of newly generated preclinical model of highly aggressive GEMM of prostate cancer (22). While prostate-specific deletion of Pten (sKO) is sufficient to initiate prostate cancer development, the codeletion of Pten and Retinoblastoma-1 (Rb1; dKO) results in a robust prostate cancer progression with spontaneous metastasis, a feature not observed in the sKO mice (22). Interestingly, dKO primary tumors show greater levels of TOP2A/EZH2 mRNA and protein compared with sKO GEMMs (Fig. 5A and B), a feature that prompted us to generate cell line models (32) from primary tumors resected from those GEMM of prostate cancer (Fig. 5C). Importantly, the dKO cell line retained high levels of TOP2A and EZH2 expression compared with the sKO cell line (Fig. 5D). Our previous work identified a significant increase in DNA damage and therapeutic efficacy by combining etoposide with EZH2 inhibitors (EZH2i) in preclinical models of lethal prostate cancer (5). With this, we proposed that dKO cells would be more sensitive to combination treatment with etoposide and clinically relevant EZH2i, namely GSK126 and EPZ6438 (33, 34). Both sKO and dKO cells were treated with either the vehicle control (DMSO) or combination therapy consisting of 1 μmol/L etoposide and 1 μmol/L or 5 μmol/L of EZH2i. Although intrinsically more proliferative than sKO cells, dKO cell lines lose their clonogenic ability upon combined inhibition of TOP2A and EZH2, a feature not observed in sKO cells (Fig. 5E; Supplementary Fig. S5A). This hypersensitivity to TOP2A and EZH2 inhibition is likely related to the significant increase in accumulation of a sub-G1 population (Fig. 5F; Supplementary Fig. S5B), indicative of apoptosis associated with DNA damage as we have demonstrated previously (5).
Discussion
While an improved understanding of the molecular basis of prostate cancer tumorigenesis has generated increased prognostic and predictive measures, the early identification of aggressive primary prostate cancer is still not resolved. Genetic pathways and/or gene expression panels to predict prostate cancer outcome and response to therapeutic interventions (35) are promising. Three recent gene expression panels have reported success in predicting prostate cancer patient outcome, in terms of BCR, mortality, and metastatic progression (36–40). While these panels have indicated patient outcomes, they have yet to provide and/or validate targeted therapy approaches that could intercept a patient's progression to metastatic disease. Provoked by this and our previous work (5, 41), this study presents data regarding two genes associated with metastatic prostate cancer, which have never been considered as potential cooperating partners in the identification and treatment of aggressive primary prostate cancer. Results within convincingly demonstrate the ability to predict metastatic potential of primary prostate cancer by expression analysis of TOP2A and EZH2. Furthermore, in vitro prostate cancer models demonstrate the ability to predict response to a combination therapy approach based on TOP2A and EZH2 gene or protein expression.
In addition, we found that high TOP2A and EZH2 gene expression in TCGA kidney cancer datasets (42–44) also selected for a majority of patients with worse progression-free survival involving progression to metastatic disease (Supplementary Fig. S6A and S6B). However, stratification of primary TCGA bladder cancer patients (45) by our methods did not predict metastatic potential or progression-free survival (data not shown). Overall, these data indicate that the potential of TOP2A and EZH2 expression to identify patients with metastatic potential can be extended to other genitourinary cancers and should be further explored in additional models.
While this study focused on the use of etoposide combination with EZH2 inhibitors, other targeted therapies maybe considered. Recently, exciting clinical results showed the dramatic impact of PARP inhibitors for treatment of patients with metastatic prostate cancer with defects in DNA repair genes (46). Because our data highlight a subgroup of patients with increased mitotic DEGs and potential of increased genomic instability, PARP inhibitors may offer an alternative therapeutic opportunity. Overall these data could provide a novel direction for prostate cancer risk stratification and the clinical management of primary disease (Fig. 6).
Disclosure of Potential Conflicts of Interest
E. Davicioni holds ownership interest (including patents) in GenomeDx Biosciences. R.J. Karnes reports receiving other commercial research support from GenomeDx Biosciences. A.E. Ross holds ownership interest (including patents) in and is a consultant/advisory board member for GenomeDx Biosciences. F.Y. Feng is a consultant/advisory board member for Celgene, Dendreon, EMD Serono, Ferring, Medivation/Astellas, and Sanofi. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: D.P. Labbé, S. Rosario, K.H. Eng, L. Ellis
Development of methodology: D.P. Labbé, P. Galbo, S. Rosario, S.-Y. Ku, R.B. Jenkins, M. Bowden, Y. Huang, D.E. Spratt, K.H. Eng
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.J. Sweeney, P. Galbo, S. Rosario, K.M. Wadosky, E. Davicioni, E.M. Schaeffer, R.B. Jenkins, R.B. Den, A.E. Ross, F.Y. Feng, D.E. Spratt, D.W. Goodrich, L. Ellis
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.P. Labbé, C.J. Sweeney, P. Galbo, S. Rosario, M. Sjöström, M. Alshalalfa, N. Erho, E. Davicioni, R.J. Karnes, E.M. Schaeffer, R.B. Jenkins, M. Bowden, Y. Huang, K.P. Gray, D.E. Spratt, K.H. Eng, L. Ellis
Writing, review, and/or revision of the manuscript: D.P. Labbé, C.J. Sweeney, M. Brown, S. Rosario, M. Sjöström, N. Erho, E. Davicioni, R.J. Karnes, E.M. Schaeffer, R.B. Jenkins, R.B. Den, A.E. Ross, Y. Huang, K.P. Gray, D.E. Spratt, D.W. Goodrich, K.H. Eng, L. Ellis
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.P. Labbé, P. Galbo, S. Rosario, S.-Y. Ku, E. Davicioni, R.J. Karnes, L. Ellis
Study supervision: D.P. Labbé, M. Bowden, D.E. Spratt, D.W. Goodrich, L. Ellis
Acknowledgments
The authors thank Clyde Bango for technical assistance and Noriko Uetani for figure design and drawing.
Grant Support
D.P. Labbé is supported by a Prostate Cancer Foundation Young Investigator Award, a Cancer Research Society Next Generation of Scientist Scholarship, and a Canadian Institute of Health Research Fellowship. L. Ellis is supported by a Roswell Park Cancer Institute (RPCI) and a Dana-Farber Cancer Institute faculty start-up funds, and a Prostate Cancer Foundation Young Investigator Award. This study used shared resources supported by the RPCI Cancer Center Support Grant from the NCI (P30CA016056) and was supported by grants from the NIH (5P01CA163227 to M. Brown; R21CA179907, to D.W. Goodrich; R01CA207757, to D.W. Goodrich and L. Ellis).
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.