Bladder cancer is the most common cancer in the urinary system, which ranks 11th among all diagnosed cancers and has clear male predominance (Siegel et al., 2017; Witjes et al., 2021). More than 90% of bladder cancer cases were urothelial carcinoma, and approximately 75% of patients with bladder cancer were non-muscle infiltration of bladder cancer (Babjuk et al., 2019). Although the clinical outcomes have improved with the application of minimally invasive surgery, radiotherapy, neoadjuvant chemotherapy, and immunotherapy, about 25% of cases presented with muscle invasiveness or metastasis, which was still a tough problem to be solved (Wu et al., 2020; Witjes et al., 2021). Furthermore, compared with the traditional pathological prognostic indicators, an increasing number of researchers have paid more attention to prognostic and predictive molecular biomarkers with the widespread use of the next-generation sequence. For instance, molecular subtypes of bladder cancer have been accepted widely and might be incorporated into clinical management in the future, once some prospective studies validate molecular subtypes’ efficacy (McConkey and Choi, 2018).

Hypoxia was a hallmark of the tumor microenvironment, which was caused by rapid proliferation of tumor cells and the intercapillary distance longer than that of oxygen diffusion (Gilkes et al., 2014; Petrova et al., 2018). Hypoxia and hypoxia-inducible factors could regulate the expression of multiple genes in tumor cells and influence functions of tumor cells, such as proliferation, angiogenesis, invasion, metastasis, and immune invasion (Gilkes et al., 2014; Rankin and Giaccia, 2016; Lin et al., 2020; Liu et al., 2020; Mo et al., 2020). Moreover, hypoxia in the tumor microenvironment still played significant roles in treatment resistance, including radioresistance, chemoresistance, and immunosuppression (Zou et al., 2019; Lin et al., 2020; Zhang et al., 2020). Therefore, it is helpful to search for hypoxia-related genes and to change the hypoxia microenvironment for survival prediction and treatment. Lin et al. (2018) once reported that O(2)-generating MnO(2) nanoparticles succeeded in increasing the oxygen concentration of the bladder cancer microenvironment in vitro and in vivo, which enhanced the therapeutic effect of photodynamic therapy on bladder cancer. Hypoxia markers GLUT-1 and CAIX were identified as independent prognostic factors and associated with vascularity and proliferation in bladder cancer (Hoskin et al., 2003; Boström et al., 2016). Additionally, some hypoxia-related non-coding RNAs were mentioned in bladder cancer, like miR-210, circELP3, circRNA_403658, and lncRNA-UCA1, which could influence functions of tumor cells and act as therapeutic targets or predictive factors (Irlam-Jones et al., 2016; Xue et al., 2017; Su et al., 2019; Wei et al., 2019).

Although some hypoxia genes have been identified and have robust predictive performance, there are still no hypoxia signatures in bladder cancer to date. Herein, we constructed a hypoxia-related signature in TCGA datasets and investigated its performance and relationships with other clinicopathological variables in bladder cancer. Then, we validated our hypoxia signature in GSE13507 and GSE32548.

As the most common malignancy in the urinary system, bladder cancer ranks 11th among all diagnosed cancers in the world, which caused more than 16,000 deaths in the United States (Siegel et al., 2017; Babjuk et al., 2019; Witjes et al., 2021). Although neoadjuvant chemotherapy, immunotherapy, intensity-modulated radiation therapy, and minimally invasive surgeries have been applied for the treatment of bladder cancer and have improved the patients’ survival outcomes, the invasion and metastasis are still tough problems and shorten the survival times of patients with bladder cancer (Wu et al., 2020). To date, an increasing number of researchers have focused on hypoxia, a hallmark of tumor microenvironment, for it is closely related to invasion and metastasis of tumor (Gilkes et al., 2014; Rankin and Giaccia, 2016; Petrova et al., 2018). Yang et al. (2008) once reported that hypoxia and HIF-1alpha could directly regulate TWIST, which could promote epithelial–mesenchymal transition (EMT) and metastatic phenotypes in patients with head and neck cancer. Cox (Cox et al., 2015) found that lysyl oxidase (LOX) in the hypoxic cancer secretome could disrupt normal bone homeostasis and lead to the formation of focal pre-metastatic lesions in patients with estrogen-receptor-negative breast cancer. Furthermore, hypoxia and HIF-1alpha also promote the growth of tumor at distant sites via VEGF-α, which could stimulate the angiogenesis (Joyce and Pollard, 2009).

It is due to the significant roles of hypoxia that some hypoxia-related gene signatures were constructed to predict patients’ survival outcomes in the era of genomics and precision medicine. To date, the hypoxia signatures of hepatocellular carcinoma, colorectal cancer, lung adenocarcinoma, glioma, gastric cancer, and prostate cancer have been developed (Yang et al., 2018; Zou et al., 2019; Lin et al., 2020; Liu et al., 2020; Mo et al., 2020; Zhang et al., 2020). Although some hypoxia genes have been identified as biomarkers in bladder cancer, there are still no hypoxia signatures about bladder cancer. Herein, we used AKAP12, ALDOB, CASP6, DTNA, HS3ST1, JUN, KDELR3, and STC1 to construct a hypoxia signature with a robust performance. The ROC of our signature in 1, 3, and 5 years were 0.663, 0.659, and 0.690, respectively, which were much higher than the ROC of the AJCC stage (Figure 8). Then, we validated our signature’s performance in GSE13507 and GSE32548, whose outcome variables and independent variables were different compared with TCGA datasets (Table 1 and Figure 2). The ROC of 1, 3, and 5 years of the hypoxia signature were still comparable with the ROC of the T stage and grade in GSE13507 and GSE32548, which illustrated the signature’s stability and wide applicability. After that, we constructed a predictive model with risk score, age, and stage in the training dataset and presented it with a nomogram. One thousand times bootstraps were used to validate the nomogram internally, and GSE13507 was utilized as an external validation set to validate the nomogram. Whatever the training dataset, the internal validation dataset, or the external validation dataset, the performance of this model is still steady and robust (Figure 13).

To deepen the understanding of the signature, we analyzed each gene in the signature. AKAP12 protein belonged to the family of kinase scaffolding protein and participated in signal transduction of cancer (Wu et al., 2018). Finger et al. (2015) observed that hypoxia-regulated AKAP12v2 could promote the invasion, migration, and metastasis of tumor via protein kinase A in melanoma. Tao et al. (2015) once reported that ALDOB inhibited metastasis and could serve as a prognostic biomarker in hepatocellular carcinoma, which concurred with Wang’s finding in clear cell renal carcinoma (Wang J. et al., 2019). Moreover, He et al. (2016) observed that downregulation of ALDOB is associated with poor prognosis of patients with gastric cancer. CASP6 encoded Caspase-6 protein which was the executor caspase inducing apoptosis by cleaving lamin A and other substrates. Lee et al. (2006) once analyzed the entire coding region and calculated the somatic mutations in gastric carcinoma. After that, he found that loss of caspase-6 might contribute to the pathogenesis of gastric cancers. DTNA encoded a scaffold protein, which maintained the structural integrity of muscle cells. Liu et al. (2017) once screened key genes for early-stage colon carcinoma and found that DTNA had the potential diagnosis value for Colon adenocarcinoma. HS3ST1 was involved in the biosynthesis of heparan sulfate, and Liu found that its transcriptional activity decreased significantly in glioma tissue compared with the para-tumorous tissue. In addition, the transcriptional activity in high-grade glioma tissue was lower than that in low-grade glioma (Ushakov et al., 2017). As a proto-oncogene, Jun was a subunit of AP-1 and participated in tumor proliferation, invasion, and metastasis. Wang et al. (2017) demonstrated that tobacco smoke elevated AP-1 activation in bladder cancer cells, and tobacco smoke-mediated cell differentiation and EMT could be reversed by AP-1 suppression. Zhao et al. (2018) reported that benzidine enhances the proliferation of bladder cells via activating the MAPK/AP-1 pathway. The researches about KDELR3 were few. Marie et al. (2020) once conducted melanoblast transcriptome analysis and identified that the loss of KDELR3 could impair experimental metastasis. STC1, a hypoxia-induced molecular target, could promote cell proliferations, invasion, migration, and metastasis in hepatocellular cancer, gastric cancer, colorectal cancer, breast cancer, and glioblastoma (Chang et al., 2015; Rezapour et al., 2016; Chan et al., 2017; Wang Y. et al., 2019; Xiong and Wang, 2019). With the exception of Jun, the remaining seven genes have not been studied in bladder cancer to date, but their ability to promote or weaken malignant phenotypes was well presented in our signature. For example, if a gene could promote the malignant phenotype, its expression was more frequent in patients with higher risk scores and vice versa (Figures 4G–I).

We explore the relationship between the hypoxia signature with clinical parameters. Firstly, we stratified the patients with some clinical parameters in the TCGA dataset and found that the patients with lower risk scores still had better survival outcomes compared with the ones with higher risk scores in the subgroups of age ≥60, age <60, male, high AJCC-stage subgroup, low AJCC-stage subgroup, low T-stage, high T-stage subgroup, nodal metastasis-free, metastasis-free subgroup, and high pathological grade (Figure 5). In addition, we also validated the signature’s prognostic performance in subgroups stratified by clinical parameters in GSE13507 and GSE32548 (Figure 6). As for the insignificant differences of survival time in the subgroups of nodal metastasis, female subgroup, low pathological grade, and metastasis, we preliminarily thought that the small sample size in the subgroups might influence the predictive ability of our signature (Figures 5G,I and Supplementary Figures 2, 3). As regards some insignificant differences of the subgroups in GSE13507 and GSE32548, we suspected that the significant different outcome variables of three datasets and small sample size in the subgroups might influence the predictive performance (Figures 2, 6). Secondly, we compared the risk scores in different subgroups stratified by clinical parameters in the training dataset and two validation datasets. It was worth noting that the subgroups with higher pathological stages or grades usually had higher risk scores, which illustrated that our molecular signature closely related to pathological parameters and had the potential to be an important prognostic indicator (Figure 7).

Immune cell infiltration in tumor tissue played a significant role in promoting or preventing the proliferation, invasion, and migration of cancer cells, so that immunotherapy had become an important treatment for tumor (Petrova et al., 2018; Liu et al., 2020). To explore the relationships between the hypoxia signature and immune cell infiltration, we compared the immune cell content of different risk score groups and found that activated memory CD4+ T cells and M0 and M1 macrophages were all significantly different in the TCGA datasets and two validation datasets. The high-risk group had higher contents of M0 and M1 macrophages compared with the low-risk group. Li (Li et al., 2020) once reported that high expression of activated CD4 (+) memory T cells and low expression of M0 macrophage were associated with better clinical prognosis in bladder cancer. Tumor-associated macrophage (TAM) played a significant role in cancer progression, metastasis, and immunotherapy resistance, which were widely infiltrated in the tumor microenvironment (Cheng et al., 2021). Macrophage plasticity allows these innate immune cells to adopt their well-known M1–M2 polarization axis (Larionova et al., 2020). TAMs with M1 profile usually expressed MHC-II, CD68, CD80, and CD86 and had important antitumor and pro-inflammation function, while the M2 microphage in the tumor microenvironment could play a role in pro-tumor and anti-inflammation (Malfitano et al., 2020; Mantovani et al., 2021). TAMs are a polarized M2 subtype of macrophage in the microenvironment of some cancers, but the concrete mechanism is not well defined. Chai reviewed 99 urothelial carcinoma cases, immunostained pathological sections, and found that higher tumor-associated macrophage (TAM) infiltration was identified in high-expression HIF-1alpha cases rather than HIF-1alpha low-expression cases. Moreover, HIF-1alpha overexpression and high TAM count were associated with worse DFS (Chai et al., 2008). Koga et al. (2004) also conducted immunochemical analysis of TAM and Endothelial Per-Arnt-Sim domain protein 1 (EPAS1) induced by hypoxia and proved that EPAS1-expressing TAM counts were significantly associated with higher T stage and progression in bladder cancer. Theoretically, M1 macrophages with the roles of antitumor and pro-inflammation should be enriched into the low-risk group with better survival, while higher contents of M1 macrophages, not M2 macrophages, were enriched into the high-risk group in our study. Delprat et al. (2020) once reported that cycling hypoxia could induce unpolarized M0 macrophages into the pro-inflammation phenotype (M1) and amplify the pro-inflammatory phenotype of M1 macrophages via increase of C-jun. Therefore, we put forward a hypothesis that M1 macrophages enriched in the high-risk group had a certain association with hypoxia and JUN, which was highly expressed in the high-risk groups. Although activated memory CD4+ T cells were significantly different in three datasets, the difference of activated memory CD4+ T cell infiltration in both risk groups was inconsistent. In TCGA and GSE32548, the high-risk group had higher activated memory CD4+ T cells, while activated memory CD4+ T cells presented an opposite trend in GSE13507 (Figure 9). As for the inconsistence, perhaps more real-world experiments were needed to uncover the relationship between hypoxia and activated memory CD4+ T cells.

Lots of researches have demonstrated that tumor mutational burden was closely related to the immunotherapy effect (Hugo et al., 2017; June et al., 2018). The more mutational genes existed in tumor cells, the more mutation-associated RNAs and proteins might be generated, which could be recognized and targeted by the immune system (Rizvi et al., 2015). Therefore, a high mutation burden in tumors was associated with improved immunotherapy response, durable clinical benefit, and progression-free survival (Rizvi et al., 2015). In our study, we listed the top 20 mutational genes in the low- and high-risk groups and then compared risk scores between the wild types and the mutation types of top mutational genes. Finally, we found that the risk scores in the wild type of MUC16 and FGFR3 were significantly higher than those in the mutation types of MUC16 and FGFR3, while the risk scores in the mutational TP53 and RB1 were higher than those in the wild type (Figure 10). TP53 was a cancer-suppressor gene, which encoded p53 protein involved in regulating many target genes. Mutations in TP53 were frequently observed in cancers and were closely related with prognosis of patients with bladder cancer (Ciccarese et al., 2017; Wu et al., 2019). To date, it has become an attractive therapy that restoring functional p53 protein in cancer cells by small peptide molecules (Ciccarese et al., 2017). Similarly, RB1 is also a cancer-suppressive gene and its mutation is common in bladder cancer samples (Felsenstein and Theodorescu, 2018). Choi once reported that bladder cancer with mutation of RB1 and NFE2L2 can be enriched into basal subtype of muscle-invasive bladder cancer (Choi et al., 2017), which could be divided into epithelial–basal and more clinically aggressive mesenchymal subtypes (Guo et al., 2019). Cotton et al. (2017) once screened prognostic biomarkers and identified MUC16 as a poor prognostic biomarker in advanced-stage bladder tumors. FGFR3 was a carcinogenic driver, and the mutation, activation, and overexpression of FGFR3 are common in bladder cancer (Guancial et al., 2014; Pouessel et al., 2016). Ahmad once explored the frequency of FGFR3 mutation in Indian bladder cancer patients and found that FGFR3 mutations were more common in the low pathological stage and low-grade tumors (Ahmad et al., 2018). Taken together, the effect caused by mutation of TP53, RB1, MUC16, and FGFR3 were in accord with risks predicted by hypoxia risk scores in our study.

Finally, we enriched different genes in different risk groups and found that cancer- and immune-associated pathways were enriched in the high-risk group, while the metabolism-related pathway belonged to the low-risk group. The cancer-associated pathways were enriched in the high-risk group, which might imply the poor prognosis of patients with high risk scores. Meanwhile, the immune-related pathways in the high-risk groups included autoimmune disease pathways and inflammatory pathways, which was not related to immune response or immune-suppressive pathways. Bladder cancer patients with higher hypoxia risk scores had higher IC50 of chemotherapy drugs in the three independent cohorts, which seemed to be in accordance with the current opinion that hypoxia in the tumor microenvironment played significant roles in treatment resistance.

This is the first hypoxia-related signature in bladder cancer, which could accurately predict the survival outcomes in patients with bladder cancer compared with the traditional pathological parameters. Moreover, the molecular signature has close relationships with clinical–pathological parameters, some infiltrating immune cells, and mutational genes in tumors. As for the stability and practicability of our signature, it still needs to be validated in future clinical practices.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

Ethical review and approval was not required for the study on human participants in accordance with the Local Legislation and Institutional Requirements. Written informed consent for participation was not required for this study in accordance with the National Legislation and the Institutional Requirements.

FZ and XW designed the study. HH, DL, and YB collected the transcriptome and clinical data. YB, YY, HH, JW, YT, HM, and DF took part in the data collection. FZ drafted the manuscript. PH revised the final manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.