Repurposing FDA approved drugs as radiosensitizers for treating hypoxic prostate cancer

Genes significantly differentially expressed after 24 h exposure to 1% oxygen in more than two cell lines (DU145, PC3, LNCaP, PNT2) were identified previously as seed genes. For each cell line, genes differentially expressed between normoxia and hypoxia conditions across triplicates were selected using a rank product probability of false positive rate < 0.05 [19]. The 848 seed genes were used to build gene co-expression networks using the publicly available GSE21032 or TCGA cohorts [18, 20]. Gene co-expression networks were assembled and partitioned into gene modules using the Louvain method [19, 21]. Each module was tested for enrichment of the hypoxia seed genes using the Chi-square test. The upregulated genes from the gene module(s) significantly associated with a high percentage of hypoxia seed genes (FDR < 0.01) were used as input for QUADrATiC.

The STRING database is a free online tool for exploring protein–protein interactions [22]. The user inputs their genes of interest, in this case the 103 GSE21032 and 66 TCGA hypoxia-associated genes, and the STRING database assembles a network of protein–protein interactions based on experimental evidence and predicted function. The associations in STRING include direct physical interactions and predicted indirect functional interactions. The protein–protein interactions are presented as networks, in which nodes represent proteins and the lines associations between proteins. The protein–protein association strength takes the form of a p-value that evaluates multiple channels of evidence as well as the chance of a random interaction between the protein pair [23].

QUADrATiC uses the LINCS database to identify connections between gene expression profiles and FDA-approved drugs [5]. The data in the LINCS database are compiled from in vitro cell line experiments. Two separate gene lists were used in the QUADrATiC software generated from GSE21032 (103 genes; Additional file 1: Table S2) and TCGA (66 genes; Additional file 1: Table S1). Genes were mapped to their corresponding probe(s) in the Affymetrix HG-133UA-na36 annotation (Additional file 1: Tables S3 and S4) and entered into the QUADrATIC software according to the most concordant drug response [24]. In other words, up-regulated genes associated with a poor prognosis (hazard ratio [HR] > 1) comprise the up-regulated set of the signature while up-regulated genes associated with a good prognosis (HR < 1) form the down-regulated set.

Three prostate cancer cell lines were used to test drugs in vitro: DU145 (HTB-81), PC3 (CRL-1435) and LNCaP (CRL-1740). The cell lines were obtained from the American Type Culture Collection and authenticated by short tandem repeat profiling using the Promega Powerplex 21 system. Cell lines were cultured under normal conditions (37˚C, 5% CO₂ in air), screened for mycoplasma and authenticated using short tandem repeat profiling. DU145 and LNCaP cells were cultured in RPMI 1640 (Sigma-Aldrich, UK) with 10% fetal bovine serum (FBS) and 2 mM L-glutamine (Sigma-Aldrich, UK). PC3 cells were cultured in Ham's F12 (Gibco, ThermoFisher, UK) with 10% FBS and 2 mM L-glutamine (Sigma-Aldrich, UK).

Menadione (Selleck Chem, Texas, USA), gemcitabine (Selleck Chem, Texas, USA) and tirapazamine (APExBIO, Texas, USA) were purchased in lyophilized form. Drugs were reconstituted to a concentration of 50 mM in dimethyl sulfoxide (DMSO) as recommended by the manufacturer. Stock solutions were aliquoted and stored at -80 °C; aliquots were not repeatedly freeze thawed.

Cells were seeded into 96 well plates and allowed to adhere overnight in the incubator. The next day plates were treated with 0, 2, 5, 10, 50 or 100 µM of drug or DMSO vehicle control. Following application of the drug, cells were immediately moved into a 0.1% O₂ hypoxia chamber (Don Whitley Scientific, Bingley, UK) or kept in a normoxia incubator for 24 h. The assay endpoint was either 24 h or 4 days post-treatment. For the 24 h time point, the cells were fixed and stained after 24 h incubation with the drug. For the 4-day time point, the drug was removed after 24 h and fresh media added to the cells. Plates were incubated for a further 4 days in normoxia. Cells were fixed and then stained with sulforhodamine B according to the published method [25].

Cells were seeded into plug seal T25 flasks and allowed to adhere overnight in the incubator, with plug seal caps kept loose. The next day media were removed and replaced with fresh media containing the drug (10 µM menadione or 8 nM gemcitabine) or DMSO (vehicle control). Untreated control flasks had fresh media applied. Cells exposed to normoxia, with or without the drug, were incubated with plug seal caps left loose in the incubator for 24 h. Cells exposed to 0.1% oxygen for 24 h, with or without the drug, were incubated with plug seal caps left loose in the H35 hypoxystation (Don Whitley Scientific, Bingley, UK). After 24 h plug seal caps were tightened on the flasks in the hypoxystation to maintain hypoxic conditions during irradiation. Cells were irradiated with x-rays delivered at 0.95 Gy/min using a Faxitron X-ray machine (Tucson, Arizona, USA). After irradiation cells were immediately harvested, counted and seeded onto 6-well plates [26]. Plates were incubated for 7–21 days until colonies had formed, fixed and stained using crystal violet solution (70% methanol (v/v), 0.1% crystal violet). Plates were imaged on a GelCount machine (Oxford Optronix, Abingdon, UK) and colonies counted using an optimized CHARM algorithm in the GelCount software. The surviving fraction was calculated for each biological repeat (experiments ran on separate days with cells of a different passage) using the mean number of colonies across the six individual wells. The concentration of DMSO was 0.02% (v/v) for menadione and 0.000016% (v/v) for gemcitabine. Sensitizer enhancement ratio (SER) was calculated as the ratio between the doses needed for 1 log kill with or without the drug.

Co-expression networks in the GSE21032 and TCGA cohorts identified hypoxia-associated genes (Additional file 1: Tables S1 and S2). Proteins encoded by the genes in the two networks interacted significantly (p < 1 × 10^–16). Figure 1 illustrates the high level of connections between the proteins in the networks. The network plots show only proteins with at least one connection. Sixty-nine of the 103 genes (67%) in the GSE21032 list had at least one connection and there was an average of 3.6 connections per node. The TCGA gene list produced a more highly connected network of protein–protein interactions with 55 of the 66 genes (83%) having at least one connection and an average of 8.2 connections. The gene lists were applied independently to the QUADrATiC connectivity mapping software to identify FDA-approved drugs.

Drugs with negative connections were considered able to target the phenotype of interest, in this case hypoxia in prostate cancer. Using the GSE21032 hypoxia-associated genes there were 5,348 drugs with negative z-scores, of which 2,405 were nominally significant (p < 0.05). Using the TCGA hypoxia-associated genes there were 4,827 drugs with negative z-scores, of which 2,270 were nominally significant (p < 0.05). The QUADrATiC interface provided summary visualizations of the top connections as bubble charts and drug and cell line connections (Additional files 2, 3: Figure S1 and S2).

The top 10 drugs from each of the two gene lists were ranked based on Z-score (Table 1). There were two prostate cell lines, PC3 and VCaP, in the connectivity mapping analysis but drug selection was not restricted to the highest ranking drugs identified in these cell lines (Table 2). The strength of the connection between the genes and the drug was considered more important than the cell line in which it was identified because (i) there were only two prostate cancer cell lines in the program and (ii) the importance of hypoxia across solid tumours. Two candidate drugs were selected for in vitro validation, menadione and gemcitabine, because they ranked in the top 10 in the GSE21032 and TCGA datasets in all (Table 1) and the prostate cancer (Table 2) cell lines. Menadione appeared twice in the top 10 ranked drugs in the GSE21032 and TCGA (Table 1) and was the only drug with a strong connection to the VCaP cell line in both datasets (Table 2). Gemcitabine appeared twice in the top 10 ranked drugs in the GSE21032 and once in the TCGA (Table 1) and had a strong connection to the PC3 cell line in both datasets (Table 2). Menadione and gemcitabine also had negative Z-scores in the PC3 and VCaP cell lines (Additional file 1: Table S5). The approved application, mechanism of action and reported peak plasma concentrations for menadione and gemcitabine are summarized in Table 3.

Menadione and gemcitabine were tested alongside the known bioreductively activated tirapazamine. In the DU145, PC3 and LNCaP cell lines tirapazamine demonstrated hypoxia selective cytotoxicity 24 h (Additional file 4: Figure S3) and 4 days (Fig. 2) following drug exposure). There was no loss in cytotoxicity for menadione and gemcitabine 4 days post-treatment in any of the three cell lines. However, hypoxic PC3 cells were more sensitive to 5 µM menadione and 10 µM gemcitabine than the normoxic cells after 24 h drug exposure (Additional file 4: Figure S3).

The radiosensitizing ability of the two FDA approved drugs was studied in DU145 and PC3 cells. The rationale for selecting DU145 in addition to PC3 for the radiosensitization experiments is because this cell line was derived from human tissue, whereas the LNCaP cell line was initially derived from human tissue but cultured in a mouse model.Vehicle control experiments confirmed DMSO did not alter surviving fraction compared to the untreated controls and DMSO plus radiation did not alter surviving fraction compared to radiation alone (Additional file 5: Figure S4). Figure 3 shows survival curves for the cells irradiated in normoxia and hypoxia. Oxygen enhancement ratio (OERs) calculated at the 10% survival level were 1.34 for DU145 and 1.69 for PC3 cells. Figure 4 shows menadione was a weak radiosensitizer in DU145 cells only. The SER for menadione in DU145 cells was 1.02 in normoxia and 1.15 in hypoxia. Figure 5 shows gemcitabine was also a weak radiosensitizer in DU145 cells only. The SER for gemcitabine was 1.27 in normoxia and 1.09 in hypoxia.