Fig 1.
Left box, Prostate cancer cell lines DU145 and LNCaP were purchased from the American Type Culture Collection and used to establish radioresistant cell lines. Gene copy number and expression profiles of radioresistant and corresponding age-matched non-irradiated radiosensitive parental cell lines were measured. Middle box, A prostate cancer-specific gene regulatory network was learned from gene expression and copy number data from 541 prostate cancer patients from The Cancer Genome Atlas (TCGA) and validated on 768 cancer cell lines of the Cancer Cell Line Encyclopedia (CCLE). This network was used to quantify putative impacts of genes with differential expression and directly underlying copy number alterations between radioresistant and radiosensitive cell lines (orange circle) on known marker genes of radioresistance (green circles) utilizing network propagation (red arrows). Right box, Identified potential radioresistance driver genes were evaluated for their potential to separate irradiated prostate cancer patients from TCGA into early and late relapse groups. In-depth literature analysis was done for all cell line-based candidate genes that were predictive for the relapse behavior of irradiated prostate cancer patients. Sophisticated experimental validations were done for the candidate gene VGF by analyzing the impact of siRNA-based VGF knockdowns on radiosensitivity. A detailed technical flow chart is shown in S1 Fig.
Fig 2.
Gene copy number alterations of DU145 and LNCaP.
Gene copy number profiles of DU145 (a) and LNCaP (b) comparing radioresistant to radiosensitive cell lines. Gene copy number alterations are quantified by log2-ratios of radioresistant versus radiosensitive and plotted in the chromosomal order of genes. Deviations of log2-ratios from zero (brown dashed line) indicate the presence of gene copy number alterations. Considered reduced (green dots below blue dashed line: log2-ratios < -0.1) or increased (red dots above blue dashed line: log2-ratios > 0.1) gene copy numbers in the corresponding radioresistant cell lines of DU145 and LNCaP are highlighted. Ends of chromosomes are marked by black dotted vertical lines. Unchanged genes on a chromosome are shown by alternating grey and black dots to further support the visual separation between chromosomes. An additional heatmap representation including comparisons of radioresistant and radiosensitive DU145 and LNCaP to normal reference DNA is shown in S2 Fig.
Fig 3.
Gene expression differences between DU145 and LNCaP.
Differentially expressed genes between radioresistant and radiosensitive cell lines were determined for DU145 and LNCaP. Identified under- (top panels) and overexpressed genes (bottom panels) in the radioresistant cell lines of DU145 and LNCaP were compared to each other at the single gene level (a) and at the level of cancer-relevant gene annotation categories (b; categories: oncogenes (OG), tumor suppressor genes (TS), cancer census genes (CC), phosphatases (PH), kinases (KI), metabolic pathway gene (MG), signaling pathway gene (SG), transcriptional regulator (TR)). Significant overlaps between categories are denoted by ‘*’ (b; grey columns, P < 0.001, Fisher’s exact test). Identified under- (c) and overexpressed genes (d) were further mapped to known cancer-relevant signaling pathways. Overrepresented pathways were highlighted by ‘*’ (P < 0.05, Fisher’s exact test) and ‘**’ (P < 0.01).
Fig 4.
Impacts of potential radioresistance driver genes on known radioresistance markers.
Impacts of differentially expressed genes with directly underlying copy number alterations in radioresistant DU145 (a) and radioresistant LNCaP (b) on known markers of radioresistance. The impact score represents the log10-ratio of the gene-specific impact on known radioresistance marker genes comparing the impact score reached for the prostate cancer specific network to the average impact score obtained under 10 random networks of same complexity (degree-preserving network permutations). Impact scores of genes with significantly greater impacts under the original network (q < 0.01) are shown by colored peaks (green: deleted and underexpressed; red: amplified and overexpressed for radioresistant vs. radiosensitive). The majority of gene names are shown. See S5 Table for names of all putative high impact genes. High impact genes that enabled a separation of TCGA prostate cancer patients into early and late relapse groups (Fig 5, S5 Fig) are highlighted in blue.
Table 1.
Summary of potential radioresistance drivers.
Fig 5.
Marker gene-based separation of irradiated prostate cancer patients into early and late relapse groups.
Potential radioresistance driver genes revealed from DU145 (top and middle row) and LNCaP (bottom row) were analyzed for their expression behavior in 32 irradiated prostate cancer patients from TCGA. Expression levels of each marker gene across the 32 patients were used to determine a marker gene-specific optimal cutoff for disease-free survival risk curves separating patients with low (blue curve) and high (red curve) marker gene expression with respect to the constraint that at least 8 patients must be assigned to each curve. Log-rank test p-values indicate that these selected marker genes enable a separation of irradiated prostate cancer patients into early and late relapse groups. Shown are standard approximate log-rank test p-values that only marginally deviated from exact log-rank p-values determined by exhaustive computations, except for FOXL1 that had a clearly less significant exact log-rank p-value of 0.076 (see Methods for details and S4 Fig). See S1 Text for a detailed discussion of the driver candidates in the context of the existing literature.
Fig 6.
Experimental validation of VGF as regulator of cell radioresistance.
(a) Increased VGF expression in radioresistant DU145 and LNCaP in comparison to their radiosensitive parental cell lines in our microarray data. Three biological replicates were considered for each condition. (b-d) RT-qPCR analysis of VGF expression under different conditions. (b) Increased VGF expression in four independent radioresistant DU145 and three independent radioresistant LNCaP clones relative to their radiosensitive parental cell lines. (c) Increased VGF expression in sphere relative to monolayer cultures of parental DU145 and LNCaP cells. (d) Reduced VGF expression in parental DU145 and LNCaP cells induced by siRNA-mediated gene silencing relative to negative controls. (e-f) Increased radiosensitivity of parental DU145 and LNCaP cells induced by siRNA mediated reduction of VGF expression. Shown are average fractions of surviving cells in log10-scale for increasing radiation dose. Error bars represent the standard error of the mean and ‘n’ specifies the number of biological replicates. Corresponding linear-quadratic (LQ) model curves are shown in S11 Fig.