Identification of differentially expressed RBPs in BLCA patients
We explored the role of RBPs in the prognosis of BLCA patients in our study (Figure 1). The transcriptome sequence gene expression data of 430 cases (tumor samples, 411; normal samples, 19 cases) in FPKM format and related clinical data in XML format were downloaded from the TCGA database, and the data was processed with R software packages, then DERPBs were discovered. Based on the previous studies, we examined the expression of 1542 RBPs in TCGA . As shown in Figure 2figure2.tif, 388 DERPBs were displayed (P<0.05, |log2FC| >1.0). Among 388 differently expressed RBPs, 219 were upregulated, and 169 were downregulated.
GO/KEGG enrichment analysis of DERBPs
To explore the potential roles which the DERBPs played in bladder urothelial cancer, GO enrichment and KEGG enrichment analysis were conducted. The results obtained from GO enrichment analysis displayed that the DERBPs were closely associated with RNA splicing, RNA catabolic process, ncRNA processing, catalytic activity acting on RNA (Figure 3Afigure3.tif). The KEGG enrichment analysis results showed that the DERBPs had a close connection with RNA transport, spliceosome, and mRNA surveillance pathway (Figure 3Bfigure3.tif).
Key module screening from PPI network
To further explore the function the DERBPs played in bladder urothelial cancer, the PPI network was built with online analysis tools STRING (Version 11) and visualized with Cytoscape (Version 3.7.2). All hub differently expressed RBPs with the number of nodes above ten were shown in Figure 4Afigure4.tif. To further investigate and acquire possible key modules, the co-expression network was constructed with the MODE tool. As shown in Figure 4Bfigure4.tif, 388 key DERBPs were obtained and participated in RNA splicing, RNA catabolic process, ncRNA processing, catalytic activity acting on RNA, RNA transport, spliceosome and mRNA surveillance pathway.
Identification of prognostic DERBPs.
In order to build RBP gene-associated prognostic model (RGPM), univariate Cox regression analysis was carried out, and then 11 hub RBPs which were associated with the prognosis of BLCA patients were discovered (P＜0.01). The results were shown in Figure 5Afigure5.tif. Six high-risk genes and five low-risk genes were screened.
Construction of RBP gene associated prognostic model (RPGM).
To constructing RBP gene associated prognostic model, multiple Cox regression analysis was performed, and coefficients of each hub RBP gene were obtained. As shown in Figure 5B, YARS, EFTUD2, TRIM71, and DARS2 were significantly associated with BLCA patients' survival. All of them were negative prognostic factors. OAS1 was positively related to the prognosis of BLCA patients. Subsequently, we constructed RGPM with five hub RBPs base on multiple univariate analyses. According to the following formula, the risk score of each bladder patients were calculated: Risk score = (0.6573* Exp YARS) + (0.6510* Exp EFTUD2) + (-0.1756* Exp OAS1) + (0.3804 * Exp DARS2) + (0.6038* Exp TRIM71).
Survival analysis of RBP scores.
In order to evaluate the predictive ability of RGPM, a survival analysis of RBP scores was carried out. The RBP gene risk score for each BLCA patient was calculated based on the RGPM. In the training group, BLCA patients were subdivided into low-risk groups and high-risk groups according to the median risk score. As shown in figure 6Afigure6.tif, patients in low-risk groups have a better survival probability. Then, a time-dependent ROC analysis was performed in order to assess the predictive ability further. Figure 6Bfigure6.tif showed that the area under the ROC curve (AUC) of our risk score model was 0.784. This indicated that the risk score model could be applied significantly. The expression heat map, survival status of patients, and risk score of the signature classified in the low-risk and high-risk groups are shown in Figure 6Cfigure6.tif. These results above illustrated that the prognosis-related genetic risk score model has moderate sensitivity and specificity. In general, survival analysis, time-dependent ROC analysis, and risk score analysis were performed in the training group (Figure 6figure6.tif).
Validation of the RGPM-based risk signature.
To validate the accuracy of the RGPM, we validated the RGPM in the test group at first. A similar analysis performed in the training group was executed in the test group (Figure 7figure7.tif). The results in Figure 7figure7.tif showed that patients in the low-risk group had a better prognosis than that in the high-risk group, which was consistent with results obtained from the training group. Subsequently, we validated the differential expression of five hub RBPs in normal tissue and BLCA tissue. We sought the results of immunohistochemistry of five hub RBPs from the HPA database. As shown in Figure 8Afigure8.tif, the expression of YARS, OAS1, EFTUD2, DARS2, and TRIM71 were higher in BLCA than normal tissues. The results of immunohistochemistry further confirmed the accuracy of our RGPM. In general, all of the results from the internal validation of our RGPM in the test group and the HPA database illustrated the RBP gene-associated prognostic model's accuracy.
Construction of a hub RBP-based prognostic nomogram
To better assess the impact of the five hub RBPs on the survival of BLCA patients, a nomogram based on the multivariate Cox analysis of five hub RBPs was constructed (Figure 8Bfigure8.tif). Moreover, we could clearly estimate the patient's 1-year survival rate, 2-year survival rate, and 3-year survival rate from the nomogram. Figure 8 figure8.tif showed that as the risk scores (YARS, EFTUD2, DARS2, TRIM71) increased, the 1-year, 2-year, and 3-year overall survival of BLCA patients declined OAS1 was positively related to the overall survival of BLCA patients. The risk nomogram was consistent with our above results, which confirmed the risk nomogram's prognostic value.