| Objective Most of Head and Neck Squamous Cell Carcinoma(HNSC)patients are often diagnosed as advanced stage.Traditional radiotherapy and chemotherapy do not significantly improve the prognosis of patients.Immunotherapy plays an increasingly important role in the comprehensive treatment of HNSC due to its high efficiency and safety.Heterogeneity and complexity of tumor immune microenvironment leads to discrepant immunotherapy effects among HNSC patients.This study aims to evaluate the tumor immune microenvironment of HNSC,and to identify multiple omics features such as immune-related differential mRNA,somatic mutant genes,and DNA methylation sites.The construction and comparison of multi-omics prognostic prediction models can provide methodological reference for survival analysis of highdimensional data,and provide reliable clues for searching immune-related prognostic biomarkers.MethodsThe mRNA expression,somatic mutation and DNA methylation data of HNSC patients were downloaded from The Cancer Genome Atlas(TCGA)database,and 493 patients were included according to the criteria.Immune and stromal score were calculated by the ESTIMATE algorithm and divided into high immune group(n=121)and low immune group(n=372)according to the best cut-point.T test or analysis of variance was used to compare the differences of immune/stromal scores in different clinical characteristics.The Log-rank test was used to compare the difference of overall survival between the high and low immune/stromal groups.The mRNA expression,somatic mutation frequency and the bate value of DNA methylation sites in the low and low immunization groups were difference analyzed.Differential multi-omics variables were included in univariate Coxregression.Patients were divided into training set and validation set in a ratio of 2:1,and univariate significant variables were included in Least Absolute Shrinkage and Selection Operator(LASSO)-Cox,CoxBoost and Random Survival Forest(RSF)-Coxmodels in training set.Area Under the Receiver Operating Characteristic Curve(AUC),Concordance index(C-index)and integrated brier score(IBS)were used to evaluate the models effect of training set and verification set.The risk score was calculated based on the optimal model,and the training set,validation set and total population were divided into high risk group and low risk group according to the optimal cut-point,and survival analysis was conducted.The subgroup analysis in different demographic and clinical characteristics were carried out by using multi-omics model.Multivariate Coxregression was used to analyze the independent prognostic power of multi-omics risk score,demographic,and clinical features.SPSS 23.0 and R 4.0.3 were used for statistical analysis,which was a two-sided test,α=0.05.Results1.The distant metastasis,higher lymph node grade,larger tumor size,higher tumor stage,and worse prognosis were associated with lower immune score(all P<0.05).The distant metastasis the worse the prognosis was associated with lower stromal score(P<0.05).The survival curves of high and low stromal scores intersected,and the effect on clinical features and prognosis was insignificant.Immune score was positively correlated with the expression of five immune checkpoints(all P<0.05).2.A total of 1,212 differential mRNAs were involved in immune-related biological processes such as immune response,signal transduction,inflammatory response and cytokine-cytokine receptor interaction pathways.There were 77 differential somatic mutation genes and 1,474 differential DNA methylation sites in the high and low immunization groups.Among all the mutations,83.8% of the types had higher mutation frequency in the low immunization group than in the high immunization group.A total of 55 DNA methylation positively correlated genes and 200 negatively correlated genes intersected by high and low immune groups were enriched in immunerelated biological processes.3.Univariate Coxregression model identified 468 significant prognostic factors,including 386 mRNAs,7 individual cell mutation genes,and 75 DNA methylation sites.Fifteen,21 and 22 variables were included in the LASSO-Coxmodel,CoxBoost model and RSF-Coxmodel,respectively.In the training set and validation set,the C-index and AUC of RSF-Coxwere the highest and IBS of RSF-Coxwas the lowest.The Cindex and 2-year AUC of RSF-Coxmodel in the training set were significantly higher than those of LASSO-Coxmodel(C-index: 0.791 vs 0.729,2-year AUC: 0.884 vs0.774,all P<0.05).4.In the training set,validation set and total population,the overall survival rate in the low risk group was significantly higher than in the high risk group(all P<0.05).Besides HPV positive population,the overall survival rate in the low risk group was significantly higher than in the high risk group in other subgroups(all P<0.05).Age,distant metastasis,histological grade,and risk score were independent prognostic factors(all P<0.05).Among them,risk score has the highest predictive power(1-,3-,5-year AUC = 0.721,0.693,0.662).The prediction effect of the combination of multiomics model,demographic,and clinical feature model was better than that of the simple multi-omics model(1-,3-,5-year AUC = 0.796,0.741,0.719).RSF-Coxwas the relative optimal model.Conclusions1.The higher the immune score,the better the prognosis of HNSC patients.Seventeen mRNAs,two individual cell mutation genes and four DNA methylation sites are immune-related prognostic factors among HNSC patients.2.Compared with LASSO-Cox and CoxBoost models,RSF-Coxmodel has the best prognosis prediction effect for HNSC patients,and has application value for dimension-reduction of high-dimensional tumor data.3.Elderly people aged 65 and above,distant metastasis,histological grade of Moderate differentiation,poor differentiation,no differentiation,and multi-omics high risk score are independent risk factors for prognosis in HNSC patients.Compared with the simple multi-omics model,the combination of multi-omics model,demographic,and clinical factors has a better prediction effect. |
PDF Full Text Request