Predicting survival in bladder cancer with a novel apoptotic gene-related prognostic model Authors: Ding-ming Song, Kun Feng, Wen-fei Luo, Dong-shan Lv, Li-po Zhou, Yi-bo He, Yanyang Jin Published: 2024 ABSTRACT BackgroundApoptosis and apoptotic genes play a critical role in the carcinogenesis and progression of bladder cancer. However, there is no prognostic model established by apoptotic genes.MethodsMessenger RNA (mRNA), Expression data, and related clinical data were obtained from The Cancer Genome Atlas (TCGA) database and Gene Expression Omnibus (GEO) database. After extracting the apoptosis-related genes, the survival-related apoptosis genes were screened by univariate Cox regression analysis in the TCGA cohort. Following the Least Absolute Shrinkage and Selection Operator (LASSO) regression method, these genes were modeled by multivariate Cox analysis. The predictive abilities of the Apoptosis-Related Gene Model (ARGM) for overall survival (OS) rate, disease-specific survival (DSS) measures, and progression-free survival (PFS) were verified by the Kaplan–Meier(K-M)survival analysis and time-dependent Receiver Operating Characteristic (ROC) curve. Functional enrichment analyses were performed by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genome (KEGG). CIBERSORT and Single-Sample Gene Set Enrichment Analysis (ssGSEA) were used to calculate the type of immune cell infiltration and immune functions. The model’s predictive ability for immunotherapy were evaluated using Tumor Immune Dysfunction and Exclusion (TIDE) and the Imvigor210 study.The single-cell sequencing was used to display the expression level of the ARGM.Finally,qRT-PCR was executed to validate the expression level of ARGM.ResultsSeveral apoptosis genes were identified through the model, including ANXA1, CASP6, CD2, F2, PDGFRB, SATB1, and TSPO. The prognostic value of the model for OS, DSS, and PFS were verified using the TCGA and GEO cohort. The model can predict patient response to immunotherapy treatment as established through the model’s score which was linked to different types of immune cell infiltration and identified significant differences in the signal pathways between high-risk and low-risk groups. Nomogram variables, prompted from ARGM and clinical parameters, also generate a high predictive value for patient survival.ConclusionOurestablished apoptosis-related gene model (ARGM) has a substantial predictive value for prognosis and immunotherapy of bladder cancer. It may help with clinical consultation, clinical stratification, and treatment selection. The immune infiltration status and signal pathway of different risk groups also provide direction for further research.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-024-01575-z. Background New data estimates bladder cancer as the 11th most commonly diagnosed cancer in the world [1], 4th in new incidence diagnoses among males, 8th in mortality rate in the United States [2], and is confirmed as the most common urinary tumor in China [1–3]. By observing the invasion depth of the tumor, we can generally classify it as Non-Muscle-Invasive Bladder Cancer (NMIBC) and Muscle-Invasive Bladder Cancer (MIBC). NMIBC accounts for 70% of initial bladder cancer detection. After transurethral resection of bladder tumors (TUR-BT) Intravesical chemotherapy installation is the recommended treatment for patients with NMIBC, but 50% of patients relapse after surgery, and 20% make progress within 5 years [4]. After receiving neoadjuvant chemotherapy, radical cystectomy is the recommended standard treatment for MIBC, and postoperative radiotherapy, chemotherapy, immunotherapy, and other comprehensive treatments may be required according to pathological conditions [5, 6]. The pathology of bladder cancer is also an important prognostic factor. Patients with advanced grade tumors or carcinoma in situ may need more aggressive treatment. Nevertheless, the clinical prognosis of bladder cancer with the same clinical stage and pathological grade is significantly different [7]. Despite progress in robot-assisted surgery and immunotherapy based on immune checkpoint inhibitors (ICI), the survival time of bladder cancer patients was not fundamentally improved, and the overall recurrence-free survival of MIBC was around 60% in 5 years [6]. At present, more accurate stratification and markers are urgently needed to guide the treatment in clinical practice so that patients can get more appropriate treatment, avoid excessive or ineffective treatment, improve the life quality of patients, and prolong their survival. Apoptosis is a strictly regulated and evolutionarily conserved cellular process that plays a critical role in normal physiological processes, and resistance to apoptosis is an essential feature of cancer [8, 9]. The two main pathways of apoptosis are endogenous (also known as mitochondrial) and exogenous [10]. Endogenous pathways are mainly divided into caspase-dependent and non-caspase-dependent. Exogenous apoptosis mechanisms are mostly tumor necrosis factor-induced apoptosis and Fas–Fas Ligand (FasL) mediated apoptosis, both of which involve signal molecules of the TNF receptor family coupled with external signals. Studies has underscored the critical role of apoptosis-related genes in the development and progression of bladder cancer. Studies have shown that disruptions in apoptosis pathways, such as those involving key regulatory genes, can provide cancer cells with a survival advantage and contribute to tumor progression [8–10]. For instance, the activation or inhibition of apoptosis in urinary tract-associated diseases impacts disease severity and prognosis [8], while cancer cells often exploit the apoptosis paradox to evade cell death [10]. Understanding these mechanisms is essential to developing targeted therapies against bladder cancer. In our study, we systematically analyzed the expression and clinical data of TCGA and GEO databases, as well as establishing and verifying scores based on apoptotic genes using an apoptosis-related gene model (ARGM). Methods The data collection Gene expression and associated clinical data of 433 samples taken from bladder cancer patients were obtained from The Cancer Genome Atlas (TCGA) database. 165 samples were also downloaded from Gene Expression Omnibus (GEO), and the GEO data was used to verify the gene signature model established by TCGA. Disease-Specific Survival (DSS) and Progression-Free Survival (PFS) were retrieved from UCSC Xena TCGA Hub (https://tcga.xenahubs.net). Through the Molecular Signatures Database (MSigDB), 161 apoptosis-related genes were extracted from https://www.gsea-msigdb.org/gsea/msigdb/cards/HALLMARK_APOPTOSIS.html. The “SVA” package eliminates the batch effect between different databases. Gene expression and clinical data gathered on responses to PD-L1 therapy in metastatic urothelial cell carcinoma were obtained from the Imvigor210 study found at- http://research-pub.gene.com/IMvigor210CoreBiologies.Single-celltranscriptome data were sourced from the BioProject database. Establish an apoptotic related gene prognostic model Univariate Cox regression analyses was used to screen the 161 genes related to apoptosis in the TCGA cohort, together with the “survival” R package to carry out survival analysis. The TCGA cohorts were randomly divided into 70% train and 30% test groups. The apoptotic genes that can predict the overall survival in the training group were included in the LASSO-Cox regression analysis, these then determined the genes and coefficient that entered the model. The risk score for the ARGM model was shown as follows, Risk score = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sum }_{i=1}^{n}coef(i)*expr(i)$$\end{document}∑i=1ncoef(i)∗expr(i). The coefficient of each gene in the formula is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$coef(i)$$\end{document}coef(i), and the expression quantity of each gene is represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$expr(i)$$\end{document}expr(i). After establishing the model, we first used 30% TCGA cohort data for internal verification, and then used GSE13507 as the test group for external affirmation. Validation of model After the model is established, the TCGA and the GEO cohort are divided into high-risk and low-risk groups based on their respective median values through ARGM-score. Kaplan–Meier (K-M) log-rank test analysis calculated survival differences. “Survminer” R package was used to analyze survival differences. Univariate and multivariate regression analyses evaluated whether ARGM-score and other clinicopathological tumor parameters have independent prognostic values. Time-dependent receiver operating characteristics curve (ROC) and area under the curve (AUC) were used to evaluate the accuracy and specificity of the model and other parameters for survival prediction. After establishing the model, the clinical characteristics of the high-risk and low-risk groups were tested by chi-square or Fisher's precise test. Model comparison This study will consider several research findings from Min Zhang et al. who have reported that the gene signature based on ferroptosis predicts urothelial carcinoma's overall survival, including TP63, SLC2A12, CAPG, TXNIP, MAFG, SCD, and ZNF419 [11]. Rui Cao et al. have established an autophagy-related bladder cancer gene signature, including APOL1, ATF6, ATP6V0A1, EGFR, MYC, P4HB, SPNS1, TP53INP1, ZC3H12A [12], and Chaojie Xu et al. have discovered a tumor microenvironment (TME) related gene signature for bladder cancer, which enables the capacity of predicting prognosis and immunotherapy, including ANO9, BLOC1S3, CABP4, CXCL11, CEMIP, FBN2, ZNF432, and OAS1 [13]. We compared these gene signatures with our Apoptosis-Related Gene (ARG) model; The TCGA cohort was used when models were compared. The Concordance Index (C-index) is a method that calculates the proportion of the predicted results consistent with the actual results [14], and we used this method to evaluate the consistency of the models. The Restricted Mean Survival (RMS) curve was used to assess the prediction effectiveness of different models at different phases in time [15]. The “Survcomp” package was utilized to compare C-index, RMS with other models. Establish a nomogram for the prognosis of bladder cancer A nomogram is a very intuitive, concise, and easy-to-understand graphical interface often used to calculate and evaluate prognosis [16]. Based on the risk score of ARGM and characteristic clinical information with independent prognostic value, nomograms were established by R-package “RMS”, and the accuracy of nomograms was verified by the calibration curve. Differentially expressed genes The differentially expressed genes (DEGs) of low-risk and high-risk groups of the TCGA cohort were screened using |log2fold change(fc)|≥ 0.585 and False Discovery Rate (FDR) < 0.05) criteria. The LIMMA package was used to examine the different. Functional enrichment analysis We performed a functional enrichment analysis of DEGs in the TCGA cohort by Gene Ontology (GO) [17], and the Kyoto Encyclopedia of Genes and Genomes (KEGG) [18] was employed to analyze the functional enrichment differences. The R-package “clusterprofiler” was used for this analysis. Immune microenvironment “Cibersort” is an analytical tool we used to identify and evaluate 22 immune cells in the TCGA cohort through gene expression data [19]. The Spearman method was used to measure the correlation between model score, model genes, and the relative amount of immune cells. Single-Sample Gene Set Enrichment Analysis (ssGSEA) [20] analyzed the estimated different expression of 16 immune cells and 13 immune-related pathways between high-risk and low-risk groups in TCGA and GSE13507 cohorts. The correlation between ARGM-score and immune checkpoint gene expression was calculated using the Spearman method, and the “corrplot” package was used for analysis. Immunotherapy response Tumor Immune Dysfunction and Exclusion (TIDE) can be applied to predict the patient's immunotherapy response through published data and gene expression data. We obtained TIDE scores by uploading gene expression data calculated at the official website- (http://tide.dfci.harvard.edu/). The differences in TIDE scores between high and low-risk groups were compared in the TCGA cohort, and then we estimated the immunotherapy responses of patients in different risk groups. In the imvigor210 cohort [21], high and low-risk groups were determined by our ARGM-score, and the survival differences between risk groups were compared using K-M analysis. The scores between the complete response (CR) group, partial response (PR), stable disease (SD), and progression disease (PD) groups were also compared. Single-cell sequencing “Seurat”package and “scRNAtoolVis”package were used to identify the different cluster of single-cell sequencing data, and then we displayed the expression level of ARGM genes in the single-cell sequencing data to show the influence ARGM genes made to the Tumor immune microenvironment.11 bladder cancer including three single-cell samples of normal bladder tissue and eight single-cell samples of in situ bladder cancer were used. Cell culture Bladder cancer cell lines (T24, RT-112, and 5637), along with the human normal bladder epithelium cell line HCV-29, were sourced from the Cell Bank of the Chinese Academy of Sciences in Shanghai, China. All these cell lines were cultivated in RPMI-1640 medium (Gibco, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, USA). The cells were grown under controlled conditions at 37 °C in a. humidified incubator with 5% CO2. qRT-PCR Total RNA was isolated from the cultured cells using Trizol reagent (Beyotime, Shanghai, China), and cDNA synthesis was performed with NovoScript® Plus 1st Strand cDNA Synthesis SuperMix (Novoprotein Scientific Inc., Shanghai, China). Following this, quantitative reverse transcription PCR (qRT-PCR) was carried out using SYBR High-Sensitivity qPCR SuperMix (Novoprotein Scientific Inc., Shanghai, China). Transcriptional levels were normalized against the internal control gene, GAPDH. The sequences of the primers used are as follows: GAPDH forward 5′-ATCATCAGCAATGCCTCC-3′and reverse 5′-CATCACGCCACAGTTTCC-G-3′; ANXA1 forward 5′-TTTTGGCATCAAGAACTAAC-3′and reverse 5′-CCTCAGATCGGTCACCCT-3′ CASP6 forward 5′-ATTAACTGGCTTGTTCAAAG-3′and reverse 5′-GCAGCGTGTAAACGGAGG-3′ CD2 forward 5′-CACAGCAGGGAACAAAGT-3′and reverse 5′-CAGTGCCACAAAGACCAT-3′ F2 forward 5′-TTGTGAGACAGCGAGGACG-3′and reverse 5′-GGATGGGTAGTGGAGTTGATT-3′ PDGFRB forward 5′-CCCTCATGTGCATTGTGAT-3′and reverse 5′-ATGGAGCGGATGTGGTAA-3′ SATB1 forward 5′-AGGTGTCTTCCGAAATCTA-3′and reverse 5′-CAAGCCCTGAGTTCTGTT-3′ TSPO forward 5′-GGGCCTTGGTGGATCTCCT-3′and reverse 5-TGTCCCGCCATACGCAGT-3′ The relative expression levels of the target gene were determined using the 2-ΔΔCT method. Statistical methods R.4.1.2 is used for statistical analysis and graphic visualization. The Fisher test or chi-square test is used for the difference of count data composition ratio. Mann Whitney Wilcoxon test was used to test the inter-group differences of continuous variables. P < 0.05 was considered statistically significant. Results Model establishment based on apoptosis gene The overall flow of the article is shown in Fig. 1: through the MSigDB database, we obtained 161 apoptosis-related genes. The apoptotic gene expression data of 406 TCGA patients with clinical data and 165 GEO (GSE13507) patients were extracted. TCGA is used to establish the model, and the GSE13507 cohort is used as the test group to verify the model. In the establishment of the model, Univariate Cox regression was used to screen out 28 genes significantly associated with OS in the TCGA cohort firstly. (Additional file1: Table Figure S1). After filtering the significant genes by LASSO regression (Fig. 2, A, B), we ultimately built a model of 7 genes by multivariate Cox analysis (Additional file3: Table S1). The formula is as follows: ARGM-score = 2028.366416141768 * exp (0.173952736241888 * ANXA1-0.284261956865169 * CASP6-0.438724535364491 * CD2-1.31030529547207 * F2 + 0.238739174665763 * PDGFRb-0.279726650468274 * SATB1-0.17817126370784 * TSPO). The risk score was calculated using the above formula and divided into high and low-risk groups based on their respective median values (Additional file4: Table S2 for TCGA and Additional file5: Table S3 for GSE13507). In Fig. 2, we can see that in terms of overall survival (K-M log-rank test), there are statistical differences between the high-risk group and low-risk group in the TCGA (Fig. 2C, P < 0.001). Additionally, the GSE13507 cohorts (Fig. 2D, P = 0.002) and the OS of the low-risk group display significantly higher values than that of the high-risk group. The risk profile also indicates that the higher the ARGM-score, the more death cases transpire (Fig. 2E for TCGA; Fig. 2F for GSE13507). Principal component analysis (PCA) reveals the internal structure of data by dimensionality reduction, which helps us better understand data variability [22]. PCA analysis reveals that different risk groups could differentiate bladder cancer patients into two subgroups well after dimension reduction (Fig. 2G for TCGA; Fig. 2H for GSE13507).Fig. 1Overall design flow chart of this studyFig. 2Establishment and verification of the ARGM: Profile of a lasso coefficient (A) and partial likelihood deviation of apoptosis genes (B) in TCGA cohort. Risk curve profile of OS for ARGM score in TCGA cohort (C) and GSE13507 cohort (D) (With the increase of score, the number of overall deaths and disease-specific deaths increases). KM survival curve of OS based on ARGM score in TCGA cohort (E) (log-rank test, p < 0.001) and GSE13507 cohort (F) (log-rank test, p = 0.002); PCA diagram of OS based on ARGM score in TCGA cohort (G) and GSE13507 cohort (H) (Different risk groups can be well divided into two subgroups) Overall design flow chart of this study Establishment and verification of the ARGM: Profile of a lasso coefficient (A) and partial likelihood deviation of apoptosis genes (B) in TCGA cohort. Risk curve profile of OS for ARGM score in TCGA cohort (C) and GSE13507 cohort (D) (With the increase of score, the number of overall deaths and disease-specific deaths increases). KM survival curve of OS based on ARGM score in TCGA cohort (E) (log-rank test, p < 0.001) and GSE13507 cohort (F) (log-rank test, p = 0.002); PCA diagram of OS based on ARGM score in TCGA cohort (G) and GSE13507 cohort (H) (Different risk groups can be well divided into two subgroups) Prediction of other survival indicators by the model: Considering the critical value of DSS and PFS in cancer, we further evaluated our scoring model’s prediction ability on DSS and PFS (Fig. 3). The survival curve of DSS and PFS was significantly different between low and high-risk groups, as shown through the Kaplan Meier (log-rank test) in the TCGA cohort. Both DSS (Fig. 3A, P < 0.001) and PFS (Fig. 3B, P < 0.001) were significantly higher in the low-risk group than in the high-risk group. The low-risk group also had a substantially higher DSS than the high-risk group in the GSE13507 cohort (Fig. 3C, P < 0.001).Fig. 3Verify the score on other survival indicators: KM survival curve analysis of DSS (A) (log-rank test, p < 0.001) and PFS (B) for patients based on risk score in the TCGA cohort. C KM survival curve of DSS in GSE13507 cohort (log-rank test, p < 0.001) Verify the score on other survival indicators: KM survival curve analysis of DSS (A) (log-rank test, p < 0.001) and PFS (B) for patients based on risk score in the TCGA cohort. C KM survival curve of DSS in GSE13507 cohort (log-rank test, p < 0.001) Predictive value of the model and its comparison with other models The prediction ability of the ARGM is evaluated by the area under the ROC curve, as shown in Fig. 4. The values of AUC for OS prediction of one, three, and 5 years in the TCGA cohort, are 0.737, 0.711, and 0.713, respectively (Fig. 4A). The values of AUC for OS prediction of one, three, and 5 years in GSE13507 cohort are 0.701, 0.682, and 0.682, respectively (Fig. 4B). We found several typical gene signatures from the literature, including a gene signature composed of seven ferroptosis-related genes [11], a gene signature consisting of nine autophagy-related genes [12], and a gene signature composed of eight tumor immune microenvironment related genes [13]. We compared these gene signatures with our ARG model: In the ROC curve of predicting OS, except that Cao signature [12] has a slight advantage in one-year AUC, the AUC of our ARG-model is higher than that of other gene signatures. (Fig. 4C–G). The consistency of the model is essential, and we evaluated the consistency of several models by C-index, the results of which were 0.683, 0.669, 0.628, 0.686 in our model, Zhang signature [11], XU signature [13], Cao signature [12], respectively. These discoveries suggest that the model has good consistency. In addition, the RMS curve also proves the precise prediction ability of ARGM at different time points. Considering our ARG signature’s number of genes, consistency, and prediction ability, it is an excellent gene signature.Fig. 4The prediction ability of ARGM and comparison with other models. A–D ARGM, Zhang signature, Xu signature, Cao signature ROC curve of 1-year, 3-year, and 5-year for overall-survival prediction in TCGA cohort; E ROC curve of ARGM score predicting overall survival at 1, 3, and 5 years in GSE13507 cohort; Comparison chart of C-index (F) and diagram of RMS (G) curve for ARGM, Zhang signature, Xu signature, and Cao signature The prediction ability of ARGM and comparison with other models. A–D ARGM, Zhang signature, Xu signature, Cao signature ROC curve of 1-year, 3-year, and 5-year for overall-survival prediction in TCGA cohort; E ROC curve of ARGM score predicting overall survival at 1, 3, and 5 years in GSE13507 cohort; Comparison chart of C-index (F) and diagram of RMS (G) curve for ARGM, Zhang signature, Xu signature, and Cao signature Independent prognostic ability of the model The ability of ARGM and other clinical features to predict OS was evaluated by univariate Cox and multivariate Cox analysis (Fig. 5A–D). ARGM-score, Age, T stage, and lymph node stage were independent prognostic factors in the multivariate Cox analysis of the TCGA cohort (Fig. 5B). In the multivariate Cox analysis of GSE13507 cohort, Age, T stage, M stage, and ARGM-score were independent prognostic factors (Fig. 5D).Fig. 5The prognostic capacity of ARGM score and other clinical features on OS: Forest map of univariate (A) and multivariate Cox (B) regression analysis on the predictive ability of ARGM score and other clinical parameters to predict OS in TCGA cohort; Forest map of univariate Cox (C) and multivariate Cox (D) regression analysis on the ability of ARGM score and other clinical parameters to predict OS in GSE13507 cohort; The prognostic capacity of ARGM score and other clinical features on OS: Forest map of univariate (A) and multivariate Cox (B) regression analysis on the predictive ability of ARGM score and other clinical parameters to predict OS in TCGA cohort; Forest map of univariate Cox (C) and multivariate Cox (D) regression analysis on the ability of ARGM score and other clinical parameters to predict OS in GSE13507 cohort; Moreover, we also assessed the predictive capacity of the ARGM-score for OS at different clinical stratification levels and discovered statistical differences in OS between the low-risk and high-risk groups in T1 and T2 (P = 0.021), T3 and T4 (P < 0.001), N0 (P < 0.001), N1, N2 and N3 (P = 0.025), Men (P < 0.001) and Women (P < 0.001) in TCGA cohort. There are statistical differences in OS between low-risk and high-risk groups aged less than 65 years (P = 0.035), N0 (P = 0.004) and M0 (P = 0.002) in GSE13507 cohort (Fig. 6A–I).Fig. 6Survival analysis of clinical stratification ARGM score in TCGA cohort in stratified clinical analysis, The KM survival curve of OS in TCGA cohort. A T1 and T2 groups, B T3 and T4 groups, C N0 groups, D N1, N2, and N3 groups, E Men (P < 0.001) group, F women groups; The KM survival curve of OS in GSE13507 cohort, G age less than 65 years group, H N0 group, I M0 group Survival analysis of clinical stratification ARGM score in TCGA cohort in stratified clinical analysis, The KM survival curve of OS in TCGA cohort. A T1 and T2 groups, B T3 and T4 groups, C N0 groups, D N1, N2, and N3 groups, E Men (P < 0.001) group, F women groups; The KM survival curve of OS in GSE13507 cohort, G age less than 65 years group, H N0 group, I M0 group We also considered the relationship between ARGM-score and clinical features in the TCGA cohort. The boxplot uncovers that stages N, M, and T have significant differences in high-risk and low-risk groups (Chi-squared test, P < 0.05). The gene expression in high-risk and low-risk groups also had different expression trends, as shown in Fig. 4G. Among the comparison of risk scores between different groups: T1-T2 versus T3-T4 group, N0 versus N1 group, M0 versus M1 group, stage I-II versus stage III-IV group, the risk scores were shown to have statistically different results (P < 0.05). If the grade and stages are higher, so is the risk score (Additional file2.Figure S2). Establish a nomogram base on an apoptosis-related genes model We created a nomogram to predict OS by adding the clinical parameters, which have the independent prognostic ability in multivariate Cox analysis, including age, T stage, and N stage. We included the clinical parameters here: (Fig. 7A). By verifying the curve through calibration, we can see the nomogram has a solid prediction value in 1 year, 3 years, and 5 years (Fig. 7B–D).Fig. 7Establish and verify nomogram (A) Nomogram was established by ARGM score and age, T, and N; B, C, D Calibration curve for 1-year, 3-year, and 5-year OS prediction Establish and verify nomogram (A) Nomogram was established by ARGM score and age, T, and N; B, C, D Calibration curve for 1-year, 3-year, and 5-year OS prediction Functional analysis of differentially expressed genes Between the high-risk and low-risk groups of ARGM-score in the TCGA cohort, we screened out 467 DEGs in the TCGA cohort (Additional file6: table S4). Through the functional enrichment analysis of these DEGs, the signal pathways in GO analysis are mainly enriched in external encapsulating structure organization, extracellular matrix organization, extracellular structure organization, collagen-containing extracellular matrix structural, wound healing, skin development, endoplasmic reticulum lumen, extracellular matrix structural constituent (Fig. 8A). We found that the DEGs were mainly enriched in focal adhesion, PI3K-Akt signaling pathway, Human papillomavirus infection, Proteoglycans in cancer, ECM − receptor interaction, Protein digestion and absorption, Amoebiasis, AGE − RAGE signaling pathway in diabetic complications, Bladder cancer (Fig. 8B) by employing KEGG enrichment analysis. These results suggest that ARGM-related genes might enhance extracellular matrix formation and influence cancer phenotypes by promoting cell fibrosis through the PI3K-Akt signaling pathway. Bladder tumor cells may evade apoptosis or foster tumor growth by modifying their microenvironment, including changes to extracellular matrix structure and signaling. Additionally, they might regulate immune-related pathways, such as IL-17 and TGF-beta, to influence apoptosis and drive tumor progression. Alterations in lipid metabolism pathways may further help these cells evade apoptosis or adapt to harsh conditions. Furthermore, the activation of signaling pathways like PI3K-Akt and AGE-RAGE could inhibit apoptosis and support tumor growth. These hypotheses will require further experimental validation.Fig. 8Go (A) and KEGG (B) were used for functional enrichment analysis of DEGs in the TCGA cohort Go (A) and KEGG (B) were used for functional enrichment analysis of DEGs in the TCGA cohort Correlation between ARGM-score and immune cell infiltration With the aid of the Cibersort algorithm, we calculated the relative amount of immune cell infiltration expressed in the TCGA cohort and determined a correlation between the abundance of immune cells and ARGM-score (Fig. 9A-J). The finding established the scores were negatively correlated with B cells naive (P = 0.0064, R = − 0.21), T cell regulatory (P = 2.9e-08, R = − 0.4), T cell memory active (P = -0.00081, R = − 0.25), monocytes (P = 8.8e-07, R = − 0.36), T cells CD8 (P = 1.2e-15, R = − 0.56), plasma cells (P = 0.027, R = − 0.17). ARGM-score were positively correlated with mast cell activated (P = 0.0002, R = 0.23), macrophages M0 (P = 0.0002, R = 0.45,), neutrophils (P = 0.043, R = 0.31), T cells CD4 memory resetting (P = 0.043, R = 0.15,). The genes involved in the model significantly correlated with the expression of most of the immune cells (Fig. 9k). From the ssGSEA analysis of immune cell infiltration in the high-risk and low-risk groups (Fig. 10A for TCGA Fig. 10B for GSE13507), as shown in Fig. 10B, it was found that except for the tendency of macrophages and mast cells to be higher in the high-risk group, all the remaining cells were relatively low expressed in the high-risk group. This result is similar to the Cibersort algorithm. However, the differences between different risk groups regarding immune cell function were mostly insignificant (Fig. 10C for TCGA; Fig. 10D for GSE13507), mainly in APC co-inhibition, HLA, and T cell co-stimulation. The ARGM-score was significantly negatively correlated with the gene expression of most immune checkpoints (Fig. 10E).Fig. 9Correlation analysis of ARGM score, score gene and immune cell infiltration; A–J Relationship between ARGM score and related amount of immune cell infiltration, B cells naive (R = − 0.21, P = 0.0064), T cell regulatory (R = − 0.4, P = 2.9e-08), T cell memory active (R = − 0.25, P = − 0.00081), Monocytes (R = − 0.36, P = 8.8e-07), T cells CD8 (R = − 0.56, P = 1.2e-15), Plasma cells (R = − 0.17, P = 0.027). K Correlation between scoring genes and relative quantities of various immune cell infiltration(Spearman’s Correlation test, *P < 0.05; **P < 0.01; ***P < 0.001)Fig. 10Analysis of relative quantity and function of ssGSEA immune cells(Wilcoxon test, *P < 0.05; **P < 0.01; ***P < 0.001): Analysis of the difference of relative quantity of immune cell between high and low-risk groups in TCGA cohort (A) and GSE13507 cohort (B); The difference of immune cell function between high and low-risk groups in TCGA cohort (C) and GSE13507 cohort (D); Correlation analysis between ARGM score and common immune checkpoints (E) (Spearman’s Correlation test, *P < 0.05; **P < 0.01; ***P < 0.001) Correlation analysis of ARGM score, score gene and immune cell infiltration; A–J Relationship between ARGM score and related amount of immune cell infiltration, B cells naive (R = − 0.21, P = 0.0064), T cell regulatory (R = − 0.4, P = 2.9e-08), T cell memory active (R = − 0.25, P = − 0.00081), Monocytes (R = − 0.36, P = 8.8e-07), T cells CD8 (R = − 0.56, P = 1.2e-15), Plasma cells (R = − 0.17, P = 0.027). K Correlation between scoring genes and relative quantities of various immune cell infiltration(Spearman’s Correlation test, *P < 0.05; **P < 0.01; ***P < 0.001) Analysis of relative quantity and function of ssGSEA immune cells(Wilcoxon test, *P < 0.05; **P < 0.01; ***P < 0.001): Analysis of the difference of relative quantity of immune cell between high and low-risk groups in TCGA cohort (A) and GSE13507 cohort (B); The difference of immune cell function between high and low-risk groups in TCGA cohort (C) and GSE13507 cohort (D); Correlation analysis between ARGM score and common immune checkpoints (E) (Spearman’s Correlation test, *P < 0.05; **P < 0.01; ***P < 0.001) The model score predicts the immunotherapy effect By evaluating our model score in the Imvigor210 cohort, we validated a significant difference in OS between high and low-risk groups in patients with urothelial carcinoma receiving immunotherapy (K-M log-rank test, P = 0.018) (Fig. 11A). The ARGM-score of the CR group was significantly lower than that of the PR, SD, and PD groups (Fig. 11B). Through the TIDE algorithm, we can see the high-risk group's TIDE score and exclusion score are higher than those of the low-risk group, and the difference is significant (Fig. 11C, D). These findings suggest that the effect of immunotherapy in a high-risk group may be relatively poor, which may be related to immune rejection.Fig. 11Prediction of ARGM score on immunotherapy: A The KM survival curve of OS based on risk score in Imvigor210 cohort (log-rank test, p = 0.018). B Analysis of the difference of ARGM score between CR group and PR/SD/PD group in imvigor210. Analysis of the difference of TIDE score (C) and exclusion score (D) between high-risk and low-risk groups in TCGA database; (B, C, D, Wilcoxon test, *P < 0.05; **P < 0.01; ***P < 0.001) Prediction of ARGM score on immunotherapy: A The KM survival curve of OS based on risk score in Imvigor210 cohort (log-rank test, p = 0.018). B Analysis of the difference of ARGM score between CR group and PR/SD/PD group in imvigor210. Analysis of the difference of TIDE score (C) and exclusion score (D) between high-risk and low-risk groups in TCGA database; (B, C, D, Wilcoxon test, *P < 0.05; **P < 0.01; ***P < 0.001) ARGM expression in different clusters In Fig. 12, we used the single-cell transcriptome and identified different cell clusters, including smooth muscle cells, B cells, endothelial cells, fibroblasts, epithelial cells, myeloid cells, T cells, mast cells (Fig. 12A). At the same time, we explored the expression levels of ARGM-related genes in each cluster, and used umap to show the specific expression levels of each gene (Fig. 12B–H). The differential expression of these genes in different clusters revealed that ARGM-related genes achieve relevant biological effects by regulating different types of cells in the bladder cancer immune microenvironment. CD2 and CASP6 are mainly distributed in T cells, PDGFRB is mainly expressed in fibroblasts, SATB1 is mainly distributed in myeloid, endothelial cells and fibroblasts, and TSPO and ANXA1 are expressed in all cells.Fig. 12Single cell sequencing validation of BLCA: A Different cluster of scRNA sequencing. B ARGM genes expression in different clusters. C–H ARGM genes expression in Single cell sequencing Single cell sequencing validation of BLCA: A Different cluster of scRNA sequencing. B ARGM genes expression in different clusters. C–H ARGM genes expression in Single cell sequencing qRT-PCR validation We performed qRT-PCR detection for each ARGM-related gene, and each experiment was repeated three times, and then verified the change of expression trend of ARGM-related genes, which further verified the reliability of the model prediction (Fig. 13A–G).Fig. 13PCR validation of model: A–G The qRT-PCR results show the mRNA expression level of ARGM in bladder cancer cells (T24, 5637, HT-1376) and normal bladder epithelial cells (HCV-29) PCR validation of model: A–G The qRT-PCR results show the mRNA expression level of ARGM in bladder cancer cells (T24, 5637, HT-1376) and normal bladder epithelial cells (HCV-29) Discussion Due to the development of bioinformatics technology, the research on cancer has reached a new stage [23]. The tumor immune microenvironment is very complex, but thanks to the development of bioinformatics technology and the updating of the understanding of the tumor immune microenvironment, we now have more means to explore the tumor immune microenvironment [24–27]. Bladder cancer ranks 13th in mortality rate worldwide [1]. In developing countries, the mortality rate of bladder cancer in men is still increasing [28]. Significant progress has been made in bladder cancer prognosis and treatment, both in surgery and immunotherapy [27, 29, 30]. However, the recurrence rate of bladder cancer is high, the life expectancy of MIBC is relatively short, and the heterogeneity of bladder cancer patients is strong [4–6, 31]. There is a lack of an accurate and efficient scoring system for evaluating prognosis in medical and health clinics, which can be used for patient consultation and guiding treatment. Apoptosis is one of the most critical components in cell physiological processes. Normal apoptosis is closely regulated and maintained in a balanced state. Bladder tumors may resist apoptosis through mechanisms such as upregulation of BCL2 and Survivin [8, 32], autophagy and upregulation of beclin-1 and Autophagy Related Protein 7(Atg7), cell surface Fas downregulation, and release of solid FAS, downregulation of caspase-3 [33]. Although bladder cancer is the most common urinary tract cancer, we still have little understanding of the mechanism of apoptosis regulation in urinary tract tumors. In recent years, research has made great strides in understanding the heterogeneity of bladder cancer [7, 34]. The information in many gene signature models have helped to further the efforts in prognosis and immunophenotype of bladder cancer, which supports the precise treatment of bladder cancer [3]. However, many prognostic gene signatures are based on various gene sets or immune infiltration types [4, 35–41]. Although an apoptosis-related gene signature for bladder cancer has been previously reported [42], our study introduces a more streamlined model with a reduced number of apoptosis-related genes. This reduction in gene number could enhance model efficiency without compromising accuracy. Additionally, we have experimentally validated our gene signature, further verifying its reliability and potential for clinical application. Establishing a prognostic model of bladder cancer apoptotic related gene signature may help to enhance understanding and advance the study of the mechanism of bladder cancer apoptosis in addition to guiding prognosis. With the help of our gene modeling process, we screened survival-related apoptotic genes through univariate Cox analysis. Additionally, seven genes, including ANXA1, CASP6, CD2, F2, PDGFRB, SATB1, and TSPO, were screened out using LASSO-Cox analysis Subsequent verification confirmed that our model has accurate capacity for a wide range of prognoses. In the TCGA cohort, it has predictive ability not only for the OS but also for DSS and PFS. It can also predict OS and DSS in GSE13507 cohort. Our model has also been verified in Imvigor210 [21], and there are significant differences in OS between different ARGM-score groups among patients receiving immunotherapy. While ensuring the prediction ability, the number of genes in the signature should be reduced as much as possible to reduce the difficulty of its use. In addition, the model should be as convenient to use as possible with reliable consistency. There is considerable evidence which suggests our model is excellent in these aspects. In the genes that we were involved in building the model: The study of Chien Feng Li et al., found the expression of ANXA1 correlates to T stage and tumor grade, suggesting it is related to tumor progression [43]. Other bladder cancer gene signature models also confirmed the important prognostic effect of ANXA1 [44]. CASP6 is crucial in alternatively activated macrophages, which may promote tissue remodeling, angiogenesis, and tumor progression [45]. In breast cancer, CD2 is essential in treating breast cancer by regulating the tumor microenvironment [46]. A tumor can lead to abnormal coagulation, and the factors can also promote tumor progression through various mechanisms [47]. F2 codes thrombin, and its signal affects the tumor microenvironment of different human tissue types through the protease-activated receptor (PAR) family and can promote tumor progression [48]. PDGFRB is an essential target for target drugs such as Sorafenib. In Jiayu Feng’s study, PDGFRB can be used as a biomarker for NMIBC [49]. SATB1 plays a critical role in tumor progression and EMT [50]. In the study of Samia Hussein et al., SATB1 has been used as a marker of bladder tumor progression and poor prognosis [51]. Translocation protein is a high-affinity mitochondrial protein that binds to cholesterol and drugs. It is involved in a variety of cellular functions, including steroidogenesis, apoptosis, and proliferation, and its gene expression alters a variety of cancers [52]. Yi Fu et al. confirmed in the study of glioma cells that the loss of TPSO will lead to mitochondrial dysfunction, can regulate the balance between mitochondrial oxidative phosphorylation and glycolysis, and promote glioma growth and angiogenesis [53]. In the follow-up study of immune infiltration, we described how ARGM scores correlate with immune cells. To our surprise, mast cell activated, macrophages M0, neutrophils, and T cells CD4 memory resetting were positively correlated with the model score. Previous studies have confirmed that mast cells may promote tumor growth by releasing mediators, including growth and angiogenesis factors [54], and may penetrate tumor cells to enhance tumor invasion and cell proliferation. In the study at Shanghai General Hospital, the ratio of neutrophils to lymphocytes was calculated by collecting clinical samples. It was verified that patients with a high neutrophil ratio had a worse prognosis; This experimental result was also confirmed in the TCGA database [55]. Similarly, accumulation of Macrophages M0 has been shown to be associated with bladder cancer and predicts a worse prognosis [56]. Moreover, activated memory T cells and T cells CD8 involved in cellular immunity were negatively correlated with an increased ARGM score, suggesting that the cellular immunity of the high-risk group may be inhibited. At the same time, the score was significantly negatively correlated with immune checkpoints such as CTLA4 and PDCD1. In previous studies, low PD1 expression in colon cancer has been associated with poor prognosis [57]. Further research may be required when considering the complex relationship between immune cells and immune checkpoints [58]. In addition, our studies show that KEGG function is enriched in focal adhesion and PI3K/Akt signaling pathway. Previous studies, have shown that high expression of focal adhesion was positively correlated with the pathological grade and lymph node metastasis of bladder cancer [59]. A PI3K/Akt signaling pathway is a key regulator of apoptosis [8–10]. In GO and KEGG functional analysis, the extracellular matrix and structure of ECM are also different. ECM components may have a close relationship with the invasion and metastasis of bladder cancer [60–62]. This scovery needs further research. The widespread use of immune checkpoint inhibitors is significant in treating bladder cancer [63, 64]. At present, PD-1, CTLA4, and other immune checkpoint inhibitors have made some progress in immunotherapy. However, due to the complexity of tumors evading immune surveillance subsequent rejection, its therapeutic effect is not satisfactory [65]. At present, immunotherapy is still a second-line treatment for bladder cancer [5, 6]. Through TIDE calculation our study found that the TIDE score and exclusion score in the high-risk group were significantly higher. This suggests the possibility of substantial immune rejection in the high-risk group. Reliable sources reveal that about 1/3 of patients with bladder cancer receiving immunotherapy are in an immune rejection state [66]. In addition, in terms of immunotherapy, the CR rate of the high-risk group was significantly lower, which was consistent with the previous results. The combined use of the ARGM-score may comprehensively guide us to adopt more active treatment schemes in high-risk groups, such as immune checkpoint inhibitor (ICI) combined with chemotherapy [65]. In conclusion, we developed ARGM-scores for bladder cancer based on apoptotic genes. This model has reliable prognostic value for OS, DSS, and PFS. After adding clinical parameters with independent prognostic ability, the nomogram forecasts compelling predictions for OS at one, three, and 5 years. The revelation of relevant signal pathways, such as ECM and PI3K/Akt signaling pathway, needs further research. Our model also has a significant correlation between the abundance of immune cells and immune cell infiltration. Our study also has some limitations. To start with, the results are based on biological information analysis and lack experimental verification. We are also hindered by a lack further studies on the related mechanisms. The absence of practical use and validation of gene signatures in a health setting causes a limited benefit to clinical guidance. However, with the continuous improvement and establishment of the model, combined with clinical practice, in the future, our findings may contribute to prognosis guidance, immunotherapy typing, prediction of immunotherapy effect, and even the selection of treatment options. Conclusion In conclusion, we constructed a prediction model based on apoptotic genes. The model and nomogram based on this model have a strong ability to predict the survival of bladder tumors. The immune infiltration status and signal pathway of different risk groups also provide direction for further research. Supplementary Information Additional file 1: Figure S1. Forest map of 28 genes related to overall prognosis in the training group of the TCGA cohortAdditional file 2: FigureS2. There were significant differences in ARGM scores between different clinical groups(Wilcoxon test,*P < 0.05; **P < 0.01; ***P < 0.001), A-D, stage I–II vs stage III–IV group, T1-T2 group vs T3-T4 group, N0 vs N1 group, M0 vs M1 groupAdditional file 3: TableS1. Genes involved in model construction and their coefficientAdditional file 4: TableS2. High-risk and low-risk groups of ARGM score in TCGA cohortAdditional file 5: TableS3. High and low-risk groups of ARGM score in GSE13507 cohortAdditional file 6: TableS4. 467 differentially expressed genes grouped by ARGM score in TCGA cohort Additional file 1: Figure S1. Forest map of 28 genes related to overall prognosis in the training group of the TCGA cohort Additional file 2: FigureS2. There were significant differences in ARGM scores between different clinical groups(Wilcoxon test,*P < 0.05; **P < 0.01; ***P < 0.001), A-D, stage I–II vs stage III–IV group, T1-T2 group vs T3-T4 group, N0 vs N1 group, M0 vs M1 group Additional file 3: TableS1. Genes involved in model construction and their coefficient Additional file 4: TableS2. High-risk and low-risk groups of ARGM score in TCGA cohort Additional file 5: TableS3. High and low-risk groups of ARGM score in GSE13507 cohort Additional file 6: TableS4. 467 differentially expressed genes grouped by ARGM score in TCGA cohort