Bladder cancer (BLCA), carrying a great social burden with about 170000 patients died annually 32, has been widely demonstrated to be able to benefit from immunotherapy 33, so exploring the underlying mechanism involving in the immune response of BLCA is necessary and meaningful. The increasing researches of ceRNA network suggested the upset of the equilibrium of regulatory ceRNA network played nonnegligible roles in the progress of diseases, especially cancers. A recent study 13 has constructed a ceRNA network, which was of high prognostic value, to achieve a deeper comprehension to the RNA regulatory network in BLCA. Besides, although several immune-relevant signature 34-36 have been constructed to predict the survival of BLCA, few studies concentrated on immune-relevant genes revealing the prognosis based on ceRNA hypothesis. In this study, we integrated the TCGA-BLCA database and GEO database to determine an immune-related risk signature and constructed an immune-relevant ceRNA network including 5 mRNAs, 24 miRNAs and 86 lncRNAs, leading to a better understanding of the oncological immunology of BLCA from the angle of RNA interactions.
Utilizing ssGSEA strategy, which has been proved to be effective for tumor grouping based on immune infiltrating status in multiple studies 17, 35, 37, 411 BLCA patients were labeled with high-immune-infiltrating and low-immune-infiltrating, and then genomic difference detection was performed to achieve the latent immune-relevant lncRNA, mRNA and miRNA. To identify the genes correlated with tumorigenesis of BLCA, we compared the genomic expression difference between the parcancerous tissues and malignant bladder tissues by “edgeR” package of R. WGCNA was implemented to obtain the lncRNAs mostly relevant to the process of immune response. Multiple databases, including MiRDB, miRcode, miRTarBase, TargetScan and starBase, were used to predict the possible binding profile of the lncRNAs, mRNAs and miRNAs in order to ensure the accuracy of the prediction of target molecules as possible. KM survival analysis, univariate Cox regression, LASSO algorithm and multivariate Cox regression were implemented to develop a risk model based on 182 overlapped mRNAs. Through the internal and external validation, 5 mRNAs including PCGF3, FASN, DPYSL2, TGFBI and NTF3 were identified as important prognostic biomarkers and the risk score was successfully constructed. ESTIMATE algorithm, CIBERSORT algorithm and the exploring of the correlation between risk score and crucial immune checkpoints were conducted, verifying our risk score was significantly correlated with BLCA immune process. At last, we established an immune-relevant ceRNA network based on these 5 critical mRNAs, providing novel insights of the tumor immunity of BLCA.
Among these five mRNAs, PCGF3, FASN, DPYSL2 and TGFBI were independent predictors and displayed significant prognostic value in the external validation dataset. Polycomb Group Ring Finger 3 (PCGF3), involving in polycomb group multiprotein PRC1-like complex, could inhibit the transcription of many genes in the developmental process 38. It has been reported that PCGF3 could serve as a biomarker for the efficacy of docetaxel, cisplatin and S-1 (DCS) on patients with advanced gastric cancer 39. Besides, some study reported the relatively high expression level of RING1 protein, such as PCGF3, could reveal a favorable prognosis in non-small cell lung cancer 40, implying PCGF3 might be an antioncogene during the tumorigenesis. In our study, we firstly found the expression of PCGF3 acted as a favorable predictor in BLCA and was negatively correlated with the expression of immune-checkpoint-relevant genes. Most of the current researches of PCGF3 are based on developmental biology, but the role PCGF3 played in cancer development, especially the process of antitumor immunity, is unclear.
Fatty Acid Synthase (FASN), Dihydropyrimidinase Like 2 (DPYSL2) and Transforming Growth Factor Beta Induced (TGFBI) were all associated with adverse clinical outcomes in BLCA patients. As an essential lipogenic enzyme synthesizing the fatty acids de novo, FASN was involved in energy metabolism of BLCA. Many studies have reported the metabolic reprogramming of tumor cells could inhibit the energy intake of immune cells 41. In addition, it has been found the fatty acids are able to induce the transformation of M1 macrophages to M2 macrophages, promoting tumor growth and metastasis 42. Recent study revealed blockade of FASN helped CD4 effector T cells effectively avoid restimulation-induced cell death (RICD) 43. All the proof above displayed the close relationship between FASN and tumor immune response, while few studies concentrated on this aspect in BLCA for the moment.
DPYSL2, a member of dihydropyrimidinase-like protein family, which was acted as a regulator of the nervous system development, was firstly found to be an indicator for poor prognosis of BLCA. Some researchers 44, 45 have demonstrated the linkage between DPYSL2 and mTOR signaling pathway, and the activation of mTOR signaling pathway is crucial for T cell differentiation 46, suggesting the potential function of DPYSL2 in immunobiology of bladder cancer. TFGBI is an exocrine protein and could promote the malignant phenotypes of BLCA 47, 48. Recent study 49 has identified TGFBI as an immune-relevant biomarker for prognosis of clear cell renal cell carcinoma based on multiple datasets, verifying our interesting findings. In all, combining the literature and the outcomes of our study, we believe PCGF3, FASN, DPYSL2 and TGFBI could serve as important prognostic factors and involve in the process of immune response in BLCA.
Several ncRNAs of the 24 miRNAs and 86 lncRNAs in our ceRNA network have been reported to be correlated with immune response and tumorigenesis. For example, miR-216a, which served as a predictor for poor prognosis of BLCA (Figure. 6A), could promote the proliferation of renal cell carcinoma cell 50. We also found HCP5, a lncRNA obtained from the target prediction of miR-216a, was correlated with immune checkpoints, which was reported by a recent literature 51. Additionally, the previous studies offered us reliable experimental buttress to the prediction in our ceRNA network. The FASN-3’-UTR-luciferase assay was successfully performed to validate the ability of miR-150-5p to target FASN in the mice experiment 52. MiR-21 was able to induce the tumor growth by targeting TGFBI in non-small cell lung cancer cells 53 and pancreatic cancer cells 54, demonstrating the relative stability of our predictions though experimental investigation in BLCA is demanded.
Importantly, we constructed a gene-based risk scoring model, which showed high accuracy of prediction for overall survival both in the TCGA-BLCA cohort and GSE13507 cohort. By means of immune grouping and correlation test, we ensure the risk score is closely correlated with tumor immunity. Compared with other immune-related risk models 34-36, we firstly made use of ceRNA hypothesis to establish a prognostic signature in order to make sure the risk score was immune-related. The analysis results of functional annotation indicated that the immune-related pathways, such as complement and inflammatory response, were significantly enriched in the high-risk group, which once again verified the immunological correlation of risk score. We also explored the immune cell infiltration divergence between low- and high-risk group, and found the higher infiltration proportion of macrophages M0, M1 and M2 and lower infiltration ratio of memory B cells, activated dendritic cells and regulatory T cells in the high-risk cohort than that in the low-risk group. Multiple researches revealed that tumor-associated macrophages (TAMs) could positively regulate the tumor growth and affect the effectiveness of BCG instillation in BLCA 55, 56. Unexpectedly, M1 macrophages, which were regarded as anti-tumor cells, were found to be correlated with poor prognosis of BLCA in our study. It was reported that M1 macrophages could induce the expression of PD-L1 in hepatocellular carcinoma (HCC) cells, leading to the resistance to anti-tumor immunity 57. Our research also disclosed that the higher infiltrating level of M1 macrophages and higher expression level of PD-L1, which is also known as PDCD1, could be discovered in the high-risk group. Hence, there may be similar biological mechanisms in the tumorigenesis of BLCA, which needed to be further experimental validated. Furthermore, we revealed that the risk score is positively correlated with the immune score, stromal score and ESTIMATE score. Therefore, the cases in the high-risk group might have a significantly higher infiltration ratio of immune cells and stromal cells, along with the lower ratio of tumor cells, supporting the conclusion that the risk score could indicate the immune status of TME in patients with BLCA.
Meanwhile, the limitations of the present study are nonnegligible. First, PCGF3 and DPYSL2 were firstly reported to be prognostic biomarker and correlated with tumor immune response in BLCA patients, and further studies needed to explore the underlying mechanisms and stability of their predictive value though multicenter and large-scale researches. Second, we predict the possible targeted miRNAs and probable lncRNA competitors of DPYSL2, TGFBI, NTF3, PCGF3 and FASN, while few of the predictions have been verified in BLCA. Last, an immune-related scoring model has been constructed and validated primitively, and clinical traits ought to be performed to reconfirm the robustness.
To sum up, our study has successfully set up an immune-related risk signature, including DPYSL2, TGFBI, NTF3, PCGF3 and FASN, which showed high prognostic value in BLCA. Importantly, an immune-relevant ceRNA network, containing 5 mRNAs, 24 miRNAs and 86 lncRNAs, was established, leading to a deeper understanding of RNA interactions in the process of tumor immune response in BLCA.