Integrative Analysis Of Immune Molecular Subtypes And Microenvironment Characteristics Of Bladder Cancer Based On Machine Learning

Objective: The purpose of this study was to explore the immune molecular subtypes of bladder cancer and the potential molecular mechanism of immune escape using the relevant principles of machine learning,and to explore the relationship between the immune subtypes and immunotherapy.Methods: Transcriptome data and simple nucleotide variation(SNV)data of bladder cancer,which included 412 tumors and 19 paracancerous tissues samples,were downloaded from the Cancer Genomic Atlas(TCGA)database.R 4.0.0 software was used for data extraction and subsequent analysis.First,differential expressed genes(DEGs)of bladder cancer were analyzed via edge R package of R software.The DEGs were intersected with 1830 immune-related genes obtained from IMMport database to obtain the differentially expressed immune genes.Consensus Cluster Plus R package and unsupervised cluster analysis were used to identify the immune molecular subtypes of bladder cancer using the expression profiles of differentially expressed immune genes as characteristic variables.The clinical and immunological characteristics of different immune molecular subtypes were compared by comparing various immune cells infiltration checkpoint genes and human leukocyte antigen(HLA)genes expressing tumor mutation burden(TMB).Then Weighted Gene Co-Expression Network Analysis(WGCNA)was conducted for immune-related DEGs in each immune subtype,and the genes were divided into different modules via WGCNA methods.The Kyoto Encyclopedia of Genes and Genomes(KEGG)pathway enrichment analysis was conducted to explore the main biological pathways associated with tumor immune subtypes and mutual transformation of different subtypes.Finally,we constructed the decision tree model by the random forest principle and use the genes that are most related to the immune subtypes as variables.30% of the samples in the training set were randomly selected for internal validation,and GSE133624 of the Gene Expression Omnibus(GEO)database was used as the external validation set for external validation.Results: The differential expression analysis of cancer and paracancerous transcriptome data of bladder cancer obtained 3203 DEGs,and 303 differentially expressed immune genes were screened after intersection calculation with 1830 immune genes from IMMport database.Four molecular subtypes of bladder cancer were identified using the expression profiles of 303 differentially expressed immune genes as characteristic variables.Among the four subtypes,cluster 2 has a better clinical prognosis,with the lowest immune cells,immune checkpoint and HLA genes expression,which indicated that the cluster 2 owns the most serious degree of immune escape.Analysis results of SNV data showed that cluster 2 had higher mutation frequency and TMB was the highest than other subtypes.However,the expression of various immune features was high in cluster 4,which was similar to the immune microenvironment of normal bladder tissue,and the degree of immune escape was the least.Then,the differentially expressed immune genes in bladder cancer were divided into 6 modules by WGCNA analysis,among which the brown module was most correlated with the molecular subtypes.KEGG enrichment analysis showed that MAPK,Rap1,Ras,IL-17 Erb B,B cell receptor etc.were associated with the formation of subtypes,and these pathways were involved in immune escape and immune molecular subtype of bladder cancer.Finally,five genes,including ANXA6,NRP2,TGFB3,Grem1 and FGF7,in the brown module were selected as variables by the random forest principle to construct the decision tree model.Its accuracy was 90.5% in the training set and 91.8% in the internal verification set.Analysis of the external validation set GSE133624 showed that the immune characteristics of patients with the predicted subtypes were in high agreement with those in the training set,indicating that the model has good general applicability and can predict the immune molecular subtypes of patients with clinical bladder cancer sequencing accurately.Conclusions: The main conclusions of this study are as follows:(1)Bladder cancer patients can be divided into four immune molecular subtypes according to the differentially expressed immune genes,and patients in different subtypes have different clinical characteristics,among which cluster 2 has a better prognosis and cluster 4 has the worst prognosis.(2)The four immune subtypes had different degrees of immune escape,among which cluster 2 happened the most serious immune escape,followed by cluster 1,cluster 3 and cluster 4,and the tumor immune micro-environment of cluster4 subtype was almost similar to normal bladder tissue.(3)The immune characteristics of cluster 2 subtype patients were consistent with those of immunotherapy respondents,indicating that this subtype may have a good effect of immunotherapy.(4)KEGG analysis showed that the potential molecular pathways involved in immune molecular typing and immune escape of bladder cancer include signaling pathways such as MAPK,Rap1,Ras,IL-17,Erb B,B cell receptor etc.(5)By constructing a decision tree model,accurate subtype prediction of clinical sequenced patients with bladder cancer and the efficacy of immunotherapy can be realized.In summary,we have explored the immune molecular subtypes of bladder cancer and their potential molecular mechanisms,and these results may provide guidance for the formulation of bladder cancer immunotherapy strategies.