Previous studies have already reported both coding mRNAs and non-coding RNAs to predict the prognosis of patients with malignant tumors, but almost all these prognosis models were established based on the expression levels of coding genes[20, 21]. In the present study, we established a prognostic risk model by pairing immune-related genes for the first time and constructed a risk model with two lncRNA pairs, without adopting their exact expression value. First, we downloaded raw data of lncRNAs from TCGA-BLCA, acquired DE-lncRNAs, conducted a coexpression analysis to detect DE-ir-lncRNAs, and validated the obtained DE-ir-lncRNA-pairs using a 0-or-1 matrix. Second, we applied a modified Lasso penalized regression model, including procedures of the cross, multiple repetitions of validation, and random stimulation to determine DE-ir-lncRNA pairs. Third, we delineated ROC curves and calculated the AUC values to acquire the optimized model. In addition, we calculated the AIC value of each point on the AUC to detect the top cutoff to differentiate the high-or low-risk groups among patients with bladder cancer. Finally, we verified this novel risk model in different situations, including clinical characteristics, various immune cells, chemotherapeutics, and immunotherapeutic biomarkers.
Various studies have established a signature based on the expression of several lncRNAs for predicting the survival of patients with bladder cancer. It is well known that high expression of lncRNAs generally has crucial biological functions. Lian et al. established an 8-lncRNA signature containing APCDD1L-AS1, FAM225B, LINC00626, LINC00958, LOC100996694, LOC441601, LOC101928111, and ZSWIM8-AS1 as candidate prognostic markers for bladder cancer. Based on the signature, they constructed a risk model and validated its prognostic value. However, all these studies were based on the exact expression of lncRNAs. In the study, we detected the DE-ir-lncRNAs based on the immure-related genes and constructed conspicuous ir-lncRNA pairs. Then, we detected pairs using sorting parameters instead of examining the specific expression value of each lncRNA. Some of the DE-ir-lncRNAs detected in this study, such as TRPM2-AS, LINC01605, AC104041.1, and UCA1, have been confirmed to play significant roles in the progression of bladder cancer. Avgeris et al. detected that downregulation of UCA1 was correlated with a higher risk of short-term relapse in bladder cancer. Tian et al. reported that TRPM2-AS promoted bladder cancer by targeting miR-22-3p and regulating the expression of GINS2 mRNA. Qin et al. found that high LINC01605 expression promoted the progression of bladder cancer by upregulating MMP9. In this study, we established a Cox regression model that was confirmed using the Lasso regression model to enhance the efficacy of the predicted value of the prognostic model. Thus, we calculated every AUC value to detect the maximum value for a risk model and to obtain the cutoff value for the risk model with the AIC point. In addition, we analyzed the therapeutic effect of chemotherapeutics for BLCA, tumor environment immune infiltration, and immunotherapy-related markers.
Previous preclinical studies have confirmed that increased infiltration of CD4 + or CD8 + immune cells[26–28] leads to a better response to immune checkpoint inhibitors. In the present study, we further used various methods, including CIBERSORT, XCELL, CIBERSORT-ABS, QUANTISEQ, MCPcounter, EPIC, and TIMER to estimate the tumor-infiltrating cells, as well as their association with risk scores. Our results showed that CD4 + T-cells, monocytes, macrophages, cancer-associated fibroblasts, and myeloid dendritic cells were enriched in the high-risk group. Previous studies have demonstrated that LINC00665 enhanced the efficacy of immunotherapy in bladder cancer[29–31]. In this study, as signatures were not shown to be positively correlated with checkpoint-related biomarkers, novel biomarkers, other than signatures, should be detected and validated. Our risk model implied that the high-risk group was more sensitive to methotrexate and metformin, whereas the low-risk group was more sensitive to cisplatin, docetaxel, and paclitaxel.
There were some limitations to our study. First, the raw data downloaded from the TCGA database were relatively insufficient for initial analysis. Second, external validation was necessary to verify the risk model established in this study. To remedy this difference, we screened lncRNA pairs initially using the 0-or-1 matrix, which was optimal in this study. We aim to gather more clinical samples for further verification; however, this collection will take time. In summary, we detected a novel model comprising ir-lncRNAs, which did not require the exact expression level of lncRNAs and would help clinicians identify patients who might benefit from immunotherapy.