Prognostic characterization of immune molecular subtypes in non-small cell lung cancer to immunotherapy

Abstract

Background: Non-small cell lung cancer (NSCLC) was usually associated with poor prognosis and invalid therapeutical response to immunotherapy due to biological heterogeneity. It is urgent to screen reliable biomarkers, especially immunotherapy-associated biomarkers, that can predict outcomes of these patients.

Methods: Gene expression profiles of 1026 NSCLC patients were collected from The Cancer Genome Atlas (TCGA) datasets with their corresponding clinical and somatic mutation data. Based on immune infiltration scores, molecular clustering classification was performed to identify immune subtypes in NSCLC. After the functional enrichment analysis of subtypes, hub genes were further screened using univariate Cox, Lasso, and multivariate Cox regression analysis, and the risk score was defined to construct the prognostic model. Other microarray data and corresponding clinical information of 603 NSCLC patients from the GEO datasets were applied to conduct random forest models for the prognosis of NSCLC with 100 runs of cross-validation. Finally, external datasets with immunotherapy and chemotherapy were further applied to explore the significance of risk-scores in clinical immunotherapy response for NSCLC patients.

Results: Compared with Subtype-B, the Subtype-A, associated with better outcomes, was characterized by significantly higher stromal and immune scores, T lymphocytes infiltration scores and up-regulation of immunotherapy markers. In addition, we found and validated an eleven -gene signatures for better application of distinguishing high- and low-risk NSCLC patients and predict patients' prognosis and therapeutical response to immunotherapy. Furthermore, combined with other clinical characteristics based on multivariate Cox regression analysis, we successfully constructed and validated a nomogram to effectively predict the survival rate of NSCLC patients. External immunotherapy and chemotherapy cohorts validated the patients with higher risk-scores exhibited significant therapeutic response and clinical benefits.

Conclusion: These results demonstrated the immunological and prognostic heterogeneity within NSCLC and provided a new clinical application in predicting the prognosis and benefits of immunotherapy for the disease.

Keywords: Immune molecular subtype; Immunotherapy; Non-small cell lung cancer; Prognosis; Risk stratification.

Fig. 1

Identification of immune molecular subtypes and characteristics of subtypes in LUSC. A Consensus clustering matrix for k = 2 in LUSC patients. B Heatmap of immune cells infiltration and clinicopathologic features of the two subtypes. C The box plots showing the difference of immune cells infiltration between SubA and SubB. D Kaplan–Meier curves of overall survival (OS) for the NSCLC patients in two subtypes. E The expression of immune check points between SubA and SubB groups. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 2

Identification of immune molecular subtypes and characteristics of subtypes in LUAD. A Consensus clustering matrix for k = 2 in LUAD patients. B Heatmap of immune cells infiltration and clinicopathologic features of the two subtypes. C The box plots showing the difference of immune cells infiltration between SubA and SubB. D Kaplan–Meier curves of overall survival (OS) for the NSCLC patients in two subtypes. E The expression of immune check points between SubA and SubB groups. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 3

Identification of DEGs of subtypes and functional enrichment analysis. A Volcano plots displayed the up-regulated and down-regulated DEGs between two subgroups in LUSC and LUAD cohorts. B The bubble diagram showed the results of GO enrichment analysis of the subtypes. C The results of GSEA of SubA and SubB in LUAD and LUSC respectively. D Venn chart exhibited the common 252 DEGs among these subgroups in NSCLC

Fig. 4

Establishment and assessment of the risk prognosis signatures through LASSO and multivariate Cox regression analysis; Correlation between risk prognosis signatures with clinical and immune characteristics. A LASSO coefficient profiles of 16 prognostic immune-related genes. B 10-times cross-validation for tuning parameter selection in the LASSO model. C Heatmap of the expression of 11 risk genes after multivariate Cox regression analysis. D Kaplan–Meier curves of overall survival (OS) for the NSCLC patients in high- and low-risk groups. E Time-dependent receiver operating curves of 1/3/5-years survival for NSCLC patients using risk scores. F The distribution of risk scores and the relationship between risk scores and survival times. G The different levels of risk scores between different phenotypic terms. H The discriminative levels of immune cells infiltration between high- and low-risk groups. I The distinguishing expression levels of immune check points between high- and low-risk groups in NSCLC patients. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 5

Evaluation of the prognostic model for NSCLC patients. A The forest plot showing the multivariable Cox model results of risk scores and other clinical features. B A combined nomogram for predicting the probability of 1/3/5-year survival for NSCLC patients. C, D The calibration curve of the established nomogram with 3-year and 5-year survival respectively. F DCA curve of the established nomogram showing risk scores could bring more benefit to the prognosis of NSCLC. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 6

Validation of the prognostic model for NSCLC patients using external datasets. A–C Kaplan–Meier curves and ROC curves for the overall survival of NSCLC in three GEO datasets. D Validation of the correlation of risk scores and clinical characteristics in external datasets. E The distribution of risk scores and the relationship between risk scores and survival times in GEO datasets. F Receiver operating characteristic curve of the combined risk models for the prognosis of NSCLC with the mean AUC value 0.784. G Variable importance of risk scores and clinical variables of predicting the prognosis of NSCLC. Mean decrease accuracy represents the decrease of accuracy in the model when one variable is excluded, and mean decrease Gini represents the specific diagnostic capabilities of variables in the construction of the predicting model. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 7

Exploration of the significance of risk-scores in clinical chemotherapy and immunotherapy response. A The expression of these risk genes (ARHGDIA, CDKN1A, and CCDC85B) remarkably increased in tumor patients using immunohistochemistry from HPA database. B the effective response rate of immunotherapy was significantly higher in the high-risk score group than in low-risk cohorts; C Difference of IC50 value between high- and low-risk groups for common chemotherapeutics drugs including Cisplatin, Docetaxel, Etoposide, Gemcitabine, and Vinorelbine