Autophagy-related long noncoding RNAs can predict prognosis in patients with bladder cancer

We investigated whether autophagy-related long noncoding RNAs (lncRNAs) can predict prognosis in bladder cancer. We obtained bladder cancer lncRNA data from The Cancer Genome Atlas and autophagy-related genes from the Human Autophagy Database. Fifteen autophagy-related lncRNAs with prognostic significance were identified. Multivariate Cox analysis was used to construct a risk score model, which divided bladder cancer patients into high-risk and low-risk groups. We found that patients in the low-risk group had better survival than those in the high-risk group. Subgroup analysis showed that patients in the high-risk group also had worse OS than that in the low-risk group in subgroups based on age, gender, clinical stage, and TNM stage. We next established a nomogram according to the results of multivariate Cox regression, which included age, gender, clinical stage, TNM stage, and risk score. The area under the curve for 3- and 5-year overall survival predicted by the nomogram were 0.711 and 0.719, respectively. Bioinformatics analysis demonstrated that the 15 identified lncRNAs are involved in the cell cycle, DNA replication, cell adhesion, cancer pathway, WNT signaling pathway, and oxidative stress. These findings confirm that autophagy-related lncRNAs are predictive of prognosis in bladder cancer patients and may affect tumor progression.


INTRODUCTION
Bladder cancer is the tenth most common malignant tumor and the second most common urological malignancy of the world with approximately 549,000 new cases and 200,000 deaths in 2018 [1]. Seventy percent of patients diagnosed with bladder cancer have non-muscle-invasive bladder cancer, whereas the rest have muscle-invasive bladder cancer (MIBC) [2]. MIBC has a high recurrence rate and poor prognosis, with a 5-year survival rate of < 50% [3]. In addition, most bladder cancer patients are diagnosed with mid-or late-stage disease due to the lack of symptoms or signs of disease. MIBC is difficult to treat, and patients often experience recurrence, causing substantial physical, mental, and financial burdens [4]. Accurate prognostic and prediction tools for bladder cancer exist [5]. However, few studies have validated the predictive performance of these prognostic models, and novel biomarkers are needed to improve the utility of prediction tools for bladder cancer.
Autophagy contributes to tumor cell homeostasis by degrading and recycling damaged or unnecessary cytoplasmic components [6]. However, autophagy overactivation may promote the autophagic death of tumor cells, similar to apoptosis [7]. Long noncoding RNAs (lncRNAs) are a group of RNAs that participate in human physiological and pathological processes by interacting with other biological molecules. During autophagy, lncRNAs act as a molecular sponge and adsorb miRNA, avoiding the inhibitory effect of miRNA AGING on mRNA translation and regulating the expression of autophagy-related genes (ATGs) [8]. Moreover, lncRNAs can directly target the promoter region of ATG [9] or recruit other molecules to promote gene transcription [10], thereby regulating cell autophagy. Recently, abundant abnormally expressed lncRNAs were found to be a biomarker for early diagnosis and prognosis of bladder cancer [11].
Therefore, autophagy-related lncRNAs may have value as prognostic biomarkers in bladder cancer. Wu et al. showed that the lncRNA UCA1 decreased miR-582-5p expression and promoted bladder cancer progression and drug resistance through ATG7-mediated autophagy inhibition [12]. In addition, autophagy-related lncRNAs contribute to the pathogenesis of bladder cancer and could serve as diagnostic molecular markers [13].
The value of lncRNAs as prognostic indicators has been validated in several cancers [14][15][16]. However, there have been no studies of autophagy-related lncRNAs as prognostic indicators in bladder cancer. In this study, we established a prognostic model for bladder cancer based on autophagy-related lncRNAs and explored the biological functions of autophagy-related lncRNAs in cancer.

Acquisition of autophagy-related lncRNAs
This study was designed to investigate the prognostic significance of autophagy-related lncRNAs in bladder cancer, as shown in the flowchart in Figure 1. LncRNA sequencing data and corresponding clinical data of   Table 1). We set | log 2 FC | > 0.5 and false discovery rate (FDR) < 0.05 as the thresholds to recognize the differentially expressed genes and lncRNAs in tumors compared to paracancerous tissues. We found 454 differentially expressed lncRNAs, of which 161 were upregulated and 293 were downregulated in tumors ( Figure 2A). We also identified 66 differentially expressed autophagy-related genes, 34 of which were upregulated and 32 downregulated in tumors ( Figure   2B). Seventy-seven autophagy-related lncRNAs were determined using Pearson correlation analysis ( Table 1).

Construction of the risk score model
Using autophagy-related lncRNA and clinical data, we performed univariate Cox regression and identified 23 prognosis-associated lncRNAs ( Figure 3A). We used least absolute shrinkage and selection operator (LASSO) regression analysis to further screen 15 pivotal lncRNAs ( Figure 3B and 3C). Next, we used multivariate Cox regression to calculate their respective coefficients (βi) to establish a risk score model. We set the median risk score as the cutoff value and divided 411 patients into AGING high-risk and low-risk groups. The overall survival (OS) in the low-risk group was significantly better than that in the high-risk group (P <0.001, Figure 4A). Subgroup analysis showed that patients in the high-risk group had worse OS than that in the low-risk group in subgroups based on age, gender, clinical stage, and TNM stage ( Figure 5). Furthermore, our results indicated that three lncRNAs (AC099850.3, MAFG-DT, and AL450326.1) were adverse prognostic factors for bladder cancer, whereas the other lncRNAs (LINC01589, AC010331.1, AL139089.1, and AF131215.5) were favorable prognostic factors.

Establishment and evaluation of the prognostic model
The results of univariate and multivariate Cox analysis indicated that the 15-lncRNA signature was a reliable predictor of OS in bladder cancer patients. Univariate Cox regression demonstrated that the clinical characteristics of age, gender, clinical stage, TMN stage, and risk score were associated with OS ( Figure 6A).

AGING
We conducted a multivariate Cox analysis of these clinical characteristics and found that the 15-lncRNA signature was an independent prognostic factor for bladder cancer (P <0.001; Figure 6B). Receiver operating characteristic (ROC) curve analysis also proved that the 15-lncRNA signature was an excellent predictive indicator of prognosis (area under the curve [AUC] = 0.731; Figure 6C).
Based on the results of multivariate Cox regression, we established a nomogram that included age, gender, clinical stage, TMN stage, and risk score ( Figure 7A). AUCs for 3-and 5-year OS predicted by the nomogram were 0.711 and 0.719, respectively ( Figure 7B), which confirmed the accuracy of the prediction. The calibration curve and decision curve analysis (DCA) of the prognostic model showed that the model had a good predictive ability ( Figure 7C-7F).

Gene set enrichment analysis
We performed Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis on the differentially expressed genes between the high-risk and low-risk groups. GO enrichment analysis showed that the genes were enriched in cell division, negative regulation of cell cycle, negative regulation of apoptosis, cell migration, oxidative stress, and the WNT pathway ( Figure 8). KEGG pathway analysis showed that these genes were involved in the cell cycle, DNA replication, cell adhesion, the cancer pathway, linoleic acid metabolism, and the WNT signaling pathway (Figure 9). This information may help researchers to conduct future studies on the mechanisms of autophagyrelated lncRNAs that affect bladder cancer pathogenesis.

DISCUSSION
Bladder cancer is the second most common malignant tumor of the urinary system [1]. Approximately 30% of bladder cancer patients progress to MIBC, which is characterized by high rates of metastasis and recurrence and a 5-year OS < 50% [17]. Therefore, predicting bladder cancer prognosis is important to correctly stratify and treat bladder cancer patients. In this study, we developed a prognostic model based on autophagyrelated lncRNAs, which had excellent prediction performance in bladder cancer.
In tumors, autophagy can maintain cell homeostasis and survival by removing nonessential and dysfunctional substances; however, autophagy can also eradicate tumor cells through activation of the apoptosis pathway [18]. lncRNAs may affect autophagy in a variety of ways [19], and studies have shown that autophagyrelated lncRNAs are associated with tumor prognosis. Huang et al. showed that the lncRNA PVT1 triggered cytoprotective autophagy and promoted tumor AGING development in pancreatic ductal adenocarcinoma and that high PVT1 expression predicted poor prognosis of patients [20]. Luan et al. identified 10 autophagy-related lncRNAs in glioma and confirmed that these lncRNAs have prognostic value in glioma patients [15]. However, autophagy-related lncRNAs in bladder cancer had not been previously studied.
In this study, we identified a 15-lncRNA signature that predicted prognosis in bladder cancer patients and constructed a prognostic model based on the signature. We found that patients in the low-risk group according to the 15-lncRNA signature had longer OS than patients in the high-risk group. According to the results of multivariate Cox regression, a nomogram was     positively associated with OC. LncRNAs have received much attention as potential prognostic markers for tumors. Zhou et al. found that high expression of AC099850.3 predicted worse survival outcomes in patients with squamous cell carcinoma of the tongue [21]. In addition, in pancreatic ductal adenocarcinoma cells, LINC01589 expression is significantly increased, and LINC01589 expression is closely associated with OS in patients with this malignancy [22]. The latest research from our center has discovered that, in bladder cancer, the exosomal lncRNA LNMAT2 stimulates the formation and migration of lymphatic endothelial cells in vitro and intensifies cancer lymphangiogenesis and lymphatic metastasis in vivo [23]. Our laboratory has planned in vivo and in vitro experiments to further investigate the functions of these lncRNAs.
To determine the functions of the lncRNAs in the 15-lncRNA signature in bladder cancer, we performed GO enrichment and KEGG pathway analyses on the genes differentially expressed in the high-and low-risk groups. Gene set enrichment analysis showed that these lncRNAs were involved in the cell cycle, DNA replication, cell adhesion, cancer pathway, linoleic acid metabolism, WNT signaling pathway, and oxidative stress. Cell proliferation and cancer pathways have long been known to participate in autophagy regulation and tumor pathogenesis [24,25]. Furthermore, inhibition of the WNT signaling pathway in glioblastoma was reported to induce autophagic flux and consequently promote apoptosis of tumor cells [26]. Our findings demonstrate that autophagy-associated lncRNAs in bladder cancer might regulate tumor growth and progression through modulation of the cell cycle, DNA replication, cell adhesion, cancer pathway, and WNT signaling pathway.
Our study has some limitations. First, it is a retrospective study using data from the TCGA database, which lacks information on smoking history and treatment. Thus, the prognostic model was constructed using limited clinical data, restricting its predictive performance. Second, we included 15 lncRNAs in the prognostic algorithm, making it difficult to apply in clinical setting. However, the development of highthroughput sequencing may make it more feasible for a prognostic model based on multiple indicators to be applied in the clinic.
In summary, we developed a 15-lncRNA signature that can predict prognosis in patients with bladder cancer. Bioinformatics analysis suggested that autophagyrelated lncRNAs may regulate tumor pathogenesis through modulation of the cell cycle, DNA replication, cell adhesion, cancer pathway, and WNT signaling pathway. Our results indicate that autophagy-related lncRNAs can predict the prognosis of bladder cancer patients and play a key role in bladder cancer biology.

Sample sources and processing
We obtained information on lncRNAs and corresponding clinical data of patients with bladder cancer from the TCGA (https://cancergenome.nih.gov/). Autophagy-related genes were extracted from HADb (http://www.autophagy.lu/). We set | log 2 FC | > 0.5 and FDR < 0.05 as thresholds to recognize the differentially expressed genes and lncRNAs based on the edgeR package. We performed Pearson correlation analysis to identify autophagy-related lncRNAs with criteria of coefficients | R 2 | > 0.5 and P < 0.05. Univariate Cox regression was performed on autophagy-related lncRNAs and clinical data to identify prognosis-related lncRNAs. We used LASSO regression analysis to identify lncRNAs closely associated with OS.

Construction of the risk score model
LncRNAs selected by LASSO analysis were included in the multivariate Cox regression model to calculate their βi. Then, a risk score model consisting of βi and lncRNA expression levels (Expi) was established as follows: The risk score for each patient was calculated according to the equation. In addition, we divided patients into high-risk and low-risk groups based on the median risk score. The Kaplan-Meier survival curve showed a prognostic difference between high-risk and low-risk patients. We conducted a subgroup analysis to further validate the model.

Establishment and evaluation of the prognostic model
The risk score and clinical characteristics such as age, gender, clinical stage, and TNM stage were used in the prognostic model. A nomogram was established based on the results of multivariate Cox regression to predict each patient's 3and 5-year OS. We used calibration plots generated by the rms package to evaluate the properties of the nomogram. We further assessed the accuracy of the nomogram by performing ROC curve analysis to obtain AUCs. Then, the calibration curve and DCA were conducted to evaluate the model [27].

Gene set enrichment analysis
We performed GO enrichment and KEGG pathway analyses on the genes differentially expressed between AGING the high-risk and low-risk groups. The functions were derived by analyzing the gene set between two biological states. In addition, we explored whether the differentially expressed genes were enriched between the two groups during autophagy.

Statistical analysis
All statistical analysis was conducted using R version 3.6.2 (Institute for Statistics and Mathematics, Vienna, Austria; https://www.r-project.org) (Package: limma, pheatmap, survival, glmnet, survminer, survivalROC, rms, foreign, timeROC). The correlation was assessed using Pearson correlation analysis. The log-rank test compared the survival curves created using the Kaplan-Meier method. The categorical variables were compared using the χ 2 test. Univariate and multivariate Cox regression were used to analyze the correlation between clinicopathological features and risk scores and the OS of patients. ROC curves and AUCs were generated to assess the predictive power of the constructed model. The calibration curve and DCA were conducted to assess the model. Two-tailed P <0.05 was considered statistically significant.