The Proteogenomics of Prostate Cancer Radioresistance

Abstract

Prostate cancer is frequently treated with radiotherapy. Unfortunately, aggressive radioresistant relapses can arise, and the molecular underpinnings of radioresistance are unknown. Modern clinical radiotherapy is evolving to deliver higher doses of radiation in fewer fractions (hypofractionation). We therefore analyzed genomic, transcriptomic, and proteomic data to characterize prostate cancer radioresistance in cells treated with both conventionally fractionated and hypofractionated radiotherapy. Independent of fractionation schedule, resistance to radiotherapy involved massive genomic instability and abrogation of DNA mismatch repair. Specific prostate cancer driver genes were modulated at the RNA and protein levels, with distinct protein subcellular responses to radiotherapy. Conventional fractionation led to a far more aggressive biomolecular response than hypofractionation. Testing preclinical candidates identified in cell lines, we revealed POLQ (DNA Polymerase Theta) as a radiosensitizer. POLQ-modulated radioresistance in model systems and was predictive of it in large patient cohorts. The molecular response to radiation is highly multimodal and sheds light on prostate cancer lethality.

Significance: Radiation is standard of care in prostate cancer. Yet, we have little understanding of its failure. We demonstrate a new paradigm that radioresistance is fractionation specific and identified POLQ as a radioresistance modulator.

Figure 1

The genomic landscape of CF- and HF-resistant prostate cancer cells. A, Schematic of experimental design and workflow. B, Somatic SNV count for CF- vs. HF-resistant cells. CF-resistant cells gained twice more SNVs than HF (P = 0.1; Mann–Whitney U test). C, SNVs in cancer driver genes. Presented are all driver genes that are predicted to be strongly influenced by SNVs. Considered are SNVs that were identified in all three replicates for each cell type. Top, single-base substitution types. Bottom, the predicted annotation. D, Gained SNVs converged on partly similar cancer mutational signatures. Most signatures of known etiology, irrespective of the treatment schedule, are associated with defective DNA mismatch repair. Signature etiologies: SBS5, unknown; SBS26 and SBS15, defective DNA mismatch repair; SBS1, spontaneous deamination of 5-methylcytosine; SBS14, concurrent polymerase epsilon mutation and defective DNA mismatch repair; SBS20, concurrent POLD1 mutations and defective DNA mismatch repair; SBS44, defective DNA mismatch repair. E, Somatic SV counts for CF- and HF-resistant cells. The number of somatic SVs is similar between CF- and HF-resistant cells (P = 1; Mann–Whitney U test). F, Distinct SVs in CF-resistant cells compared to HF across the genome. Considered are SVs that were identified in all three replicates for each cell type. Chromosome numbers are presented on the x-axis. The colored lines represent types of SVs: DEL, deletion; DUP, duplication; INV, inversion; TRA, translocation. G, Fusion transcripts that were identified in either CF- and/or HF-resistant cells. Purple, the fusion transcript was identified; white, the fusion transcript was not identified. The results are presented for three replicates for each cell line.

Figure 2

Transcriptomic signatures of cancer-related genes in radioresistant cells. A, The differences in RNA abundances of driver cancer genes. The most abundant transcript out of all transcripts per gene was taken for visualization. B–C, The differences in RNA abundances of affected driver gene, compared to their abundance differences at the protein level. B, Differences between CF-resistant cells and the parental cells. C, Differences between HF-resistant cells and the parental cells. D, The RNA abundance profile of cancer hallmark genes in CF-resistant cells is distinct from HF-resistant cells and the parental profiles, which are relatively similar. The heatmap presents normalized counts of RNA isoforms that were significantly differentially abundant in CF-resistant cells compared to HF. Red, high abundance; blue, low abundance. For visualization, RNA abundances as a function of log₁₀ were converted to z-scores. E, Enrichment analysis of hallmark gene sets. The dot size represents the enrichment score and the dot color represents the directionality: for CF vs. PAR, HF vs. PAR, and HF vs. CF tests, orange represents upregulation toward CF-resistant cells, HF-resistant cells, and HF-resistant cells. F, The differences in RNA abundances of the lncRNA UCA1, the miRNAs 200c-3P and 200b-3P, and ZEB2 (the target of the miRNAs 200c-3P and 200b-3P). For UCA1 transcripts, the small letters represent different isoforms. In A and F, the dot size represents the log₂ (fold change) size and the dot color represents the directionality: for CF vs. PAR, HF vs. PAR, and HF vs. CF tests, red represents upregulation toward CF-resistant cells, HF-resistant cells, and HF-resistant cells. For all figures , three RNA-seq replicates were used for the radioresistant cell lines and two for the parental line. In B–C, three replicates were used for all cell types.

Figure 3

Fractionation-dependent protein profiles in radioresistant cells. A, The difference in relationships of consensus-module eigengenes and cellular fractions between CF- and HF-resistant cells. The colors represent the difference in correlations between consensus module eigengenes and a specific subcellular fraction of HF-resistant cells compared to CF: Blue, a higher correlation in CF-resistant cells; Red, a higher correlation in HF-resistant cells. At the top, each color represents a module, which is a detected group of positively correlated genes that are highly interconnected. B, Gene ontology enrichment of module genes for biological processes. The top two enrichments for each module are presented. C, Differences in protein abundances of driver, across different subcellular fractions and in whole cell lysates. Main, protein Cohen’s d effect sizes of significant proteins across cell fractions. Only significant changes at the level of FDR ≤ 0.025 were plotted. The dot color is the directionality: magenta and green represent upregulation and downregulation, respectively, toward CF- or HF-resistant cells. Right, the consensus modules that each gene was assigned to. For all figures, three replicates were used for CF- and HF- resistant cells. For the parental cells, at least two replicates were used.

Figure 4

Characteristics of radioresistance modulator candidates in primary prostate tumors. A, Significant associations (FDR ≤ 0.05) between CNAs in candidate genes and biochemical recurrence, as shown by fitting Cox proportional hazard models in ICGC. B,POLQ amplification is associated with biochemical relapse in ICGC (Cox proportional hazard model; FDR = 7.78 × 10⁻³). C, High POLQ RNA abundance is associated with BCR following an RT treatment in NCCS (Cox proportional hazard model; P = 0.06). D, Increased RNA abundance of POLQ is associated with a high mutation ratio in clinically relevant candidate genes based on linear regression in ICGC. P_adjusted levels after FDR correction are presented on the (right). E, Significant correlations between RNA abundances of clinically relevant candidates and POLQ RNA abundance. F and G, Strong correlation between POLQ and BRCA2 RNA abundance in ICGC (F) and NCCS (G). H, Increased RNA abundance of POLQ is associated with high Gleason grades. In F–H, ρ: Spearman correlation. P_ρ: the P_adjusted after FDR.

Figure 5

The proteome signature upon POLQ inhibition. A, Experimental design schematic. B, Genetic POLQ inhibition. The parental, CF-resistant, and HF-resistant cells were treated with either POLQ siRNA or scramble siRNA as the control. Significant radiosensitization was observed in all cell lines upon POLQ knockdown (P = 0.01, 0.002, and 0.02 for the parental, CF-resistant, and HF-resistant cells, respectively; paired t test). Three biological replicas of 4,000 cells per well and three of 6,000 cells per well were considered for each sample. C, Pharmacologic POLQ inhibition. CF-resistant, HF-resistant, and the parental cells were treated with 100 μmol/L novobiocin to achieve POLQ inhibition (Nvb 100 µmol/L). Significant radiosensitization was observed in all DU145 cell lines upon POLQ inhibition (P = 0.04, 0.02 and 0.01, for the parental, CF-resistant, and HF-resistant cells; paired t test). In B and C, both the control and the treated cells were irradiated with 0 or 4 Gy in two fractions. The surviving fraction of the treated cells (POLQ siRNA or Nvb 100 µmol/L) was normalized to the surviving fraction of the corresponding control of each cell line (black dots). D, The (top) 10 activated and suppressed biological processes at the protein level following POLQ depletion, in CF-resistant cells and the parental. The dot size represents the enrichment score, and the dot color is the directionality: orange shows upregulation toward POLQ-treated cells. E,POLQ genetic inhibition creates a proteomic signature, involving 12 affected genes in CF-resistant cells. Three replicates from each cell type were used for the analysis. Left: signature genes whose protein abundances were changed between POLQ-depleted cells and the control, in CF-resistant cells and the parental cells. The dot size represents the Cohen’s d effect size, and the dot color is the directionality: magenta, upregulation, and green, downregulation toward POLQ-treated cells. Middle: changes in protein abundances of signature genes, between CF-resistant cells and the parental. The dot size represents the Cohen’s d effect size, and the dot color is the directionality: magenta, upregulation, and green, downregulation toward CF-resistant cells. Right: changes in RNA abundances of signature genes between CF-resistant cells and the parental. The dot size represents the log₂ (fold change) values. The dot color represents the directionality: red shows upregulation toward CF-resistant cells. F, Investigating signature genes (that were affected by POLQ inhibition in cell lines) in primary patient data. Left: RNA-RNA Pearson correlations between POLQ and signature genes. Middle, Pearson correlation between the abundances of POLQ RNA and proteins of signature genes. Right: Abundance of signature genes in normal vs. tumor samples at the RNA and protein levels. For all panels, datasets refer to the cohort names used for analysis and the data type for the type of molecule tested for signature genes (i.e., RNA or protein).