Identification of an oncogenic network with prognostic and therapeutic value in prostate cancer

Abstract

Identifying critical pathways governing disease progression is essential for accurate prognosis and effective therapy. We developed a broadly applicable and novel systems-level gene discovery strategy. This approach focused on constitutively active androgen receptor (AR) splice variant-driven pathways as representative of an intractable mechanism of prostate cancer (PC) therapeutic resistance. We performed a meta-analysis of human prostate samples using weighted gene co-expression network analysis combined with experimental AR variant transcriptome analyses. An AR variant-driven gene module that is upregulated during human PC progression was identified. We filtered this module by identifying genes that functionally interacted with AR variants using a high-throughput synthetic genetic array screen in Schizosaccharomyces pombe This strategy identified seven AR variant-regulated genes that also enhance AR activity and drive cancer progression. Expression of the seven genes predicted poor disease-free survival in large independent PC patient cohorts. Pharmacologic inhibition of interacting members of the gene set potently and synergistically decreased PC cell proliferation. This unbiased and novel gene discovery strategy identified a clinically relevant, oncogenic, interacting gene hub with strong prognostic and therapeutic potential in PC.

Keywords: androgen receptor splice variants; castration resistance; mitotic gene signature; weighted gene co‐expression network analysis; yeast synthetic genetic array.

Figure 1. Multifaceted system‐level analyses identify seven prostate cancer hub genes

1. Module–trait relationships were established by WGCNA using eight independent microarray analyses comprising 375 human prostate samples. Gene modules (y‐axis) are denoted by an arbitrary color name. Bins show the Pearson correlation value between gene expression levels of each module within the noted phenotype/disease stage (x‐axis) and P‐values. A value of 1 (red) quantifies the strongest positive correlation (genes are upregulated), −1 (blue) the strongest negative correlation (genes are downregulated), and 0 (white) no correlation. Arrows indicate those modules whose genes were positively associated with PC.

2. Microarray analysis was performed following doxycycline‐regulated specific AR‐V7 depletion (using a tet‐pLKO backbone) in 22Rv1 PC cells compared to doxycycline‐treated shGFP controls. The genes that were significantly regulated by shAR‐V7 (in either direction, P‐value < 0.05) were distributed among the gene modules defined by WGCNA in panel (A). Upregulated genes (red) are those in which expression increased following AR‐V7 depletion, and conversely, downregulated genes (blue) are those that decreased following AR‐V7 depletion. Arrows indicate those modules whose genes were positively associated with PC.

3. AR‐V7 human functional interactome was generated using SGA screening in the yeast Schizosaccharomyces pombe, combined with STRING data to map protein–protein interactions, followed by the identification of the human orthologs. The colors denote the different types of genetic interactions: Red are genes that when deleted in yeast and crossed with AR‐V7‐expressing yeast suppressed growth, while green denotes genes that when deleted enhanced growth. White designates yeast essential genes (i.e., genes that are critical for yeast survival and thus could not be present in the yeast deletion library), but were incorporated into the network based on the criteria that they are known (based on literature) to physically interact with at least two of the red or green genes. Pink‐colored genes are a combination of essential and non‐essential genes identifying the same human protein.

4. Table summarizing the seven PC hub genes identified by the system‐level analyses.

5. Network interactions of the seven genes with the 50 most frequently altered neighbor genes were mapped using cbioportal.org. The types of gene‐to‐gene interactions are as follows: controls state change (blue), controls expression (green), and in complex with (brown).

Source data are available online for this figure.

Figure 2. Elevated expression of the seven‐gene set is associated with higher levels of AR‐V7 and serves as a prognostic biomarker for disease‐free survival (DFS) and chances of death in PC patients

1. Hörnberg et al (2011) gene expression profiling array data were analyzed to determine the expression levels of the seven genes in human CRPC bone metastases, grouped by their relative levels of AR variants, mainly AR‐V7 [high levels of AR‐V7 (top quartile) or low levels of AR‐V7 (quartiles 1–3)]. Data are plotted as the mean ± s.e.m. Non‐parametric Mann–Whitney test was performed (two‐tailed). Note that BUB1B expression was not measured in these microarrays. **P‐value < 0.05. N (AR‐V7 low) = 20; N (AR‐V7 high) = 10.

2. The Kaplan–Meier curves for disease‐free survival (DFS) and overall survival were built using the TCGA Prostate Adenocarcinoma dataset (465 samples; upper graphs). Log rank tests were performed. The black curves denote cases with normal expression of the gene set, and red represents cases where the mRNA levels of the seven genes were upregulated (z‐score threshold ≤ 1.96). For DFS, P‐value = 0.0009; for death, P‐value = 0.0026. An independent dataset was analyzed (Prostate Adenocarcinoma MSKCC, Cancer Cell 2010, 123 samples; lower graphs). The black curves denote cases with normal expression of the gene set, and red represents cases where the mRNA levels of at least five genes of the gene set were upregulated (z‐score threshold ≤ 1.96). For DFS, P‐value = 0.0007; for death, P‐value = 0.00546.

Figure 3. Depletion of the expression of members of the seven‐gene set reduces CRPC cell proliferation and AR ligand‐independent transcriptional activity

1. A

   Cell proliferation was examined in the CRPC cell line 22Rv1 following individual depletion of mRNAs for the seven genes or shGFP controls, using shRNA against the coding region for each gene (shRNA #2). Cell number was measured using a non‐perturbing nuclear restricted dye and quantified after 72 h using Incucyte Zoom System. Data shown are mean ± s.e.m. of eight to 12 replicates normalized to their shGFP control. Kruskal–Wallis test (P‐value < 0.0001, two‐tailed) and Dunn's multiple comparisons test were performed. *P‐value < 0.05, **P‐value < 0.001.

2. B

   Representative images of 22Rv1 stably depleted of BUB1B or control (shGFP) are shown.

3. C

   22Rv1 stably depleted of each of the seven genes were transfected with a dual‐plasmid luciferase reporter system which quantifies AR activity and basal transcription. The assay was conducted in 2% CSS to measure AR ligand‐independent transcriptional activity. Data represent two independent experiments performed in triplicate, showing the mean ± s.e.m., and normalized to their shGFP controls. Kruskal–Wallis test (P‐value < 0.0001, two‐tailed) and Dunn's multiple comparisons test were performed. *P‐value < 0.05, **P‐value < 0.001.

4. D, E

   The expression of FKBP5 and UBE2C determined by RT–qPCR analysis and normalized to GAPDH mRNA levels was examined in 22Rv1 cells stably expressing shRNA for each of the seven genes. Cells were cultured in 2% CSS. Data represent two independent experiments performed in duplicate or triplicate, showing the mean ± s.e.m., and normalized to their shGFP controls. Kruskal–Wallis test (P‐value < 0.0001, two‐tailed) and Dunn's multiple comparisons test were performed. *P‐value < 0.05, **P‐value < 0.001.

Source data are available online for this figure.

Figure 4. Combined pharmacologic inhibition of TOP2A and CCNB1 synergistically inhibits CRPC cell proliferation

1. The CRPC cell line 22Rv1 was cultured in 2% CSS media and treated for 72 h with vehicle (DMSO), doxorubicin (DOX), N9‐isopropylolomoucine (N‐9), or the combination of DOX and N‐9 at different concentrations. Cell confluence was monitored using Incucyte Zoom System, and the experiments were done with eight replicates each. The data were analyzed using Compusyn software, and a normalized isobologram was built. The table shows the combination index (CI) for the different drug combinations. CI = 1 represents additivity, CI < 1 synergism, and CI > 1 antagonistic effects.

2. The non‐tumorigenic prostate epithelial cell line RWPE‐1, the AR‐null PC cell line PC‐3, the androgen‐dependent cell line LNCaP, and the CRPC cell lines C4‐2B and 22Rv1 were treated for 72 h with vehicle (DMSO), DOX [100 ng/ml (184 nM)], N‐9 [200 ng/ml (613 nM)], or the combination of DOX [100 ng/ml (184 nM)] and N‐9 [200 ng/ml (613 nM)]. C4‐2B and 22Rv1 cells were kept in 10% CSS media, and the other cell lines were kept in 10% FBS. Cell confluence was monitored using the Incucyte Zoom System. Data represent two independent experiments, with four to six replicates each, showing the mean ± s.e.m., and normalized to vehicle controls. Kruskal–Wallis test, (P‐value < 0.0001, two‐tailed) and Dunn's multiple comparisons test were performed. *P‐value < 0.05, **P‐value < 0.01, ***P‐value < 0.001.

3. The non‐tumorigenic prostate cell line RWPE‐1 and the CRPC cell line 22Rv1 were treated for 72 h with vehicle (DMSO) or the combination of DOX and N‐9 at 100 ng/ml and N‐9 200 ng/ml, respectively. Cell confluence was monitored using the Incucyte Zoom System and representative images are shown.

Source data are available online for this figure.

Figure 5

Schematic of pipeline developed, validated, and applied