Prostate cancer is the second leading cause of cancer death in American men, with approximately 26,000 deaths expected annually. While much research has focused on protein-coding genes, these account for only about 2% of all RNA transcripts in the human body. The vast majority of the genome is transcribed into non-coding RNAs -- including long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) -- whose roles in cancer are only beginning to be understood.
A major concept in RNA biology is the competing endogenous RNA (ceRNA) hypothesis, which proposes an intricate post-transcriptional regulatory network. In this framework, miRNAs act as silencers that bind to messenger RNAs (mRNAs) and block their translation into protein. Critically, lncRNAs and other non-coding RNAs can act as molecular sponges -- they contain the same miRNA binding sequences as mRNAs, so they compete for miRNA binding, effectively reducing the silencing of the mRNAs and allowing the genes to be expressed at higher levels.
This competitive sponging mechanism creates a complex regulatory triangle in which changes in lncRNA expression can indirectly influence which proteins a cancer cell makes. Understanding these networks in prostate cancer could reveal novel biomarkers of disease progression and identify new therapeutic targets beyond the conventional protein-coding genes that have traditionally been studied.
This study analyzed RNA expression data from the Cancer Genome Atlas (TCGA) database, comparing 499 prostate cancer tissue samples to 52 normal prostate tissue samples. Three types of RNA were profiled simultaneously using RNA sequencing: messenger RNAs (mRNAs) that encode proteins, long non-coding RNAs (lncRNAs), and microRNAs (miRNAs). Differentially expressed RNAs were identified using the edgeR statistical package, with a fold-change threshold of greater than 1.5 and a false discovery rate below 5%.
To construct the ceRNA regulatory network, the researchers used multiple validated databases: TargetScan, miRTarBase, and miRDB to predict which mRNAs are targeted by each differentially expressed miRNA, and miRanda Tools to predict which lncRNAs are targeted by those same miRNAs. Only miRNAs that were predicted to regulate both lncRNAs and mRNAs simultaneously were included, ensuring that the network nodes represented genuine competing interactions rather than coincidental sequence matches.
The resulting network was visualized using Cytoscape software, which generates interactive graphical representations of molecular interactions. Survival analysis was performed using Kaplan-Meier curves and Cox proportional hazards regression to identify which RNAs in the network were associated with overall patient survival. Gene ontology and KEGG pathway analyses were used to understand the biological functions of the genes involved.
Comparing prostate cancer to adjacent normal tissue revealed extensive RNA dysregulation. A total of 773 lncRNAs were differentially expressed -- 414 upregulated and 359 downregulated -- out of 14,254 lncRNAs examined. Among protein-coding mRNAs, 1,417 were differentially expressed (744 down, 673 up) out of nearly 20,000 tested. At the microRNA level, 58 miRNAs showed differential expression (42 upregulated, 16 downregulated).
From these thousands of candidates, only the RNA species that satisfied the strict ceRNA network criteria were retained. Thirteen of the 58 differentially expressed miRNAs were predicted to target 63 specific lncRNAs. Those same 13 miRNAs were also predicted to target 644 mRNAs, of which 18 overlapped with the differentially expressed mRNA list. The final ceRNA network therefore comprised 63 lncRNAs, 13 miRNAs, and 18 mRNAs -- a connected regulatory system specific to prostate cancer.
The most strongly downregulated mRNAs in cancer versus normal tissue included SERPINA5 (fold-change -6.8), EMX2 (fold-change -6.8), and CLDN2 (fold-change -7.9). Among lncRNAs, EMX2OS was the most strongly downregulated (fold-change -6.0), while several LINC (long intergenic non-coding) RNAs also showed large expression differences. These highly regulated molecules represent strong candidates for functional studies of prostate cancer biology.
Survival analysis of the 63 ceRNA network lncRNAs identified three with statistically significant associations with overall survival. LINC00355 and OSTN-AS1 were positively associated with survival -- meaning patients with higher expression of these lncRNAs lived longer. In contrast, LINC00308 was negatively associated with survival, meaning higher LINC00308 expression predicted worse outcomes (log-rank P less than 0.05 for all three).
Among the 18 ceRNA network mRNAs, RRM2 (ribonucleotide reductase regulatory subunit M2) was significantly associated with worse overall survival. RRM2 had previously been linked to poor outcomes in colorectal, lung, and pancreatic cancers, but its role in prostate cancer had not been established. The finding that higher RRM2 expression predicts shorter survival in prostate cancer is consistent with its known function in promoting cell proliferation by supplying building blocks for DNA synthesis in rapidly dividing cells.
To understand what biological processes might be affected by LINC00308, the researchers constructed the miRNA-mRNA interactions downstream of this lncRNA. Two miRNAs -- hsa-mir-137 and hsa-mir-93-5p -- were identified as its regulatory intermediaries, with downstream target genes including RORA, GIGYF1, and NCOA3. RORA has been linked to risk in breast, prostate, and lung cancer, while NCOA3 is overexpressed in prostate cancer and correlates with Gleason score, clinical stage, and PSA level.
Going beyond the ceRNA network, the researchers performed a genome-wide survival analysis of all differentially expressed mRNAs. Univariate Cox regression identified 21 mRNAs significantly associated with overall survival (P less than 0.01). These were then entered into multivariate Cox regression, and a model selection algorithm identified the combination of four mRNAs that best predicted survival: HOXB5, GPC2, PGA5, and AMBN.
The 4-mRNA survival risk score was calculated as a weighted sum of the expression levels: (0.420 x HOXB5) + (0.794 x GPC2) + (0.947 x PGA5) + (0.473 x AMBN). All four mRNAs had positive coefficients, meaning higher expression of any of them was associated with shorter survival. Using the median risk score as a cutoff, 499 patients were divided into equal high-risk and low-risk groups, and Kaplan-Meier curves showed a significant survival difference between the two groups.
The predictive performance was evaluated by receiver operating characteristic (ROC) curve analysis, yielding an area under the curve (AUC) of 0.904 -- indicating high sensitivity and specificity for predicting survival outcomes. In the high-risk group, 10-year overall survival was only 46.3%, compared to much better outcomes in the low-risk group, demonstrating the clinical significance of the score for long-term prognosis in prostate cancer patients.
Gene ontology analysis of the differentially expressed mRNAs in the ceRNA network identified enriched functions including sequence-specific DNA binding, negative regulation of endopeptidase activity, and anterior-posterior pattern specification. The involvement of DNA-binding regulation suggests these genes participate in controlling which other genes are switched on or off in cancer cells -- consistent with a broad epigenetic or transcriptional reprogramming in prostate cancer.
KEGG pathway analysis identified 19 enriched pathways, including chemical carcinogenesis, drug metabolism via cytochrome P450, complement and coagulation cascades, and salivary and gastric secretion pathways. The enrichment of cytochrome P450 pathways is notable because these enzymes metabolize steroid hormones including androgens -- the primary driver of prostate cancer growth -- suggesting the differentially expressed mRNAs may interact with the androgen signaling axis.
The presence of complement and coagulation pathway enrichment aligns with growing evidence that these systems are dysregulated in cancer and contribute to tumor immune evasion and metastasis. While some pathways -- such as salivary and digestive secretion -- appear at first to be unrelated to prostate cancer, they likely reflect shared molecular machinery (such as secretory proteins and protease inhibitors) that plays different roles in different tissue contexts.
This study provides the first mRNA-based prognostic model built from TCGA RNA-seq data specifically for prostate cancer that incorporates ceRNA network context. The 4-mRNA signature identifies a molecular subgroup of prostate cancer patients at substantially elevated risk of death, which could guide more intensive surveillance or earlier escalation of treatment for high-risk patients who might otherwise appear to have manageable disease by conventional staging criteria.
The study has important limitations that the authors acknowledged openly. The TCGA dataset, while large and comprehensive, consists of heterogeneous samples that were not collected under standardized conditions. Most importantly, the prognostic model was derived and tested within the same dataset without a fully independent validation cohort -- a critical requirement before any biomarker can be considered clinically ready. The statistical performance (AUC 0.904) may be optimistic due to overfitting.
A practical challenge also exists: some of the identified mRNAs may not produce detectable proteins in immunohistochemistry assays, complicating translation to standard laboratory tests. Despite these limitations, the ceRNA framework represents a valuable conceptual advance -- it reveals how lncRNAs such as LINC00308, OSTN-AS1, and LINC00355 may coordinate gene expression programs in prostate cancer through miRNA intermediaries, establishing a roadmap for functional experiments to validate causality and explore therapeutic targeting.