| ObjectiveA novel immunoscore model was established based on gene expression profile of tumor immune microenvironment.Meanwhile,ICPS(immunoscore-clinical prognostic signature)was built by combing immunoscore and other clinical variables.Method1.Data acquisition: We retrieved LUAD dataset from TCGA(the cancer genome atlas),named as TCGA-LUAD.TCGA-LUAD was used as the training set for immunoscore model development.Another two datasets from GEO(gene expression omnibus)were used as the testing sets: GSE31210(n = 219)as testing set 1;GSE68465(n = 356)as the testing set 2.Moreover,we combined these three datasets,forming the meta-set(n = 975).2.Immunoscore model construction and assessment: Immune-related genes from Immport database in TCGA-LUAD dataset were extracted.Univariate Cox regression was used to screen out immune-related risk genes.We next performed ss GSEA(single sample gene set enrichment analysis),calculating the standardized enrichment score of immune risk genes in 11 immune-related gene sets,which was then Z-score transformed.The immunoscore model was established by ridge regression using Z-score transformed enrichment score of 11 immune-related gene sets as predictive factors.After immunoscore model construction,we evaluated it in training set,testing sets and metaset using Kaplan-Meier survival analysis,univariate and multivariate Cox regression analysis.3.ICPS construction and assessment: ICPS was developed by combining immunoscore and clinical variables,based on coefficients from multivariate Cox regression from the training set.In addition,prognostic predictive power of immunoscore,tumor stage and ICPS were compared.4.Association study: The relationship between immunoscore and genomic alteration,tumor purity,cell infiltration in TME(tumor microenvironment)and biological phenotypes were explored.Result1.Immunoscore was related to patient survival.Kaplan-Meier survival analysis exhibited that patients in high-immunoscore subgroup had poor outcomes in all sets(P <0.001).Multivariable Cox regression analysis indicated that immunoscore was an independent risk factor in training set(HR(hazard ratio)= 2.96(2.24-3.9),P < 0.001),testing set 1(HR = 1.99(1.21-3.26),P=0.006),testing set 2(HR = 1.48(1.13-1.93),P =0.005),and meta-set(HR = 2.01(1.69-2.39),P < 0.001).2.ICPS was related to patient survival and had better prognostic predictive power than immunoscore or tumor stage.Kaplan-Meier survival analysis showed that patients in high-ICPS subgroup had poor outcomes in all sets(P < 0.001).In addition,by comparing C index of ICPS,immunoscore and tumor stage,we found that C indexes of ICPS were significantly higher than immunoscore and tumor stage in training set(0.72(ICPS)vs 0.7(immunoscore)or 0.59(tumor stage),P < 0.001 when compared with tumor stage),testing set 1(0.75(ICPS)vs 0.72(immunoscore)or 0.7(tumor stage),P = 0.015 when compared with tumor stage),testing set 2(0.65(ICPS)vs 0.61(immunoscore)or0.62(tumor stage),P < 0.001 when compared with tumor stage)and meta-set(0.7(ICPS)vs 0.66(immunoscore)or 0.64(tumor stage),P < 0.001 when compared with both immunoscore and tumor stage).3.In association study,Genome analysis indicated that TP53 mutation frequency was significantly different between the high and low immunoscore subgroups(71 vs 116,P <0.001),and immunoscore was positively correlated with tumor mutation burden(R = 0.22,P < 0.001).In exploring the relationship between immunoscore and TME,we discovered that immunoscore was positively correlated with tumor-associated fibroblast infiltration(R = 0.32,P < 0.001)and inversely correlated with CD8+T cell infiltration(R =-0.28,P< 0.001).Gene set enrichment analysis showed that most immune-related pathways were enriched in the low-immunoscore subgroup.ConclusionImmunoscore and ICPS could be used for prognostic prediction in stage Ⅰ-Ⅱ LUAD patients.Further studies are needed to validate and improve the immunoscore and ICPS model. |
PDF Full Text Request