Identification of a Novel Signature Predicting Overall Survival in Head and Neck Squamous Cell Carcinoma

Abstract

Background: Head and neck squamous cell carcinoma (HNSCC) is a highly heterogeneous tumor with a high incidence and poor prognosis. Therefore, effective predictive models are needed to evaluate patient outcomes and optimize treatment. Methods: Robust Rank Aggregation (RRA) method was used to identify highly robust differentially-expressed genes (DEGs) between HNSCC and normal tissue in 9 GEO and TCGA datasets. Univariate Cox regression analysis and Lasso Cox regression analysis were performed to identify DEGs related to the Overall survival (OS) and to construct a prognostic gene signature (HNSCCSig). External validation was performed using GSE65858 dataset. Moreover, comprehensive bioinformatics analyses were used to identify the association between HNSCCSig and tumor immune environment. Results: A total of 257 reliable DEGs were identified by differentially analysis result of TCGA and GSE65858 datasets. The HNSCCSig including 7 mRNAs (SLURP1, SCARA5, CLDN10, MYH11, CXCL13, HLF, and ITGA3) were developed and validated to identify high-risk group who had a worse OS than low-risk group in TCGA and GSE65858 datasets. Cox regression analysis showed that the HNSCCSig could independently predict OS in both the TCGA and the GSE65858 datasets. Further research demonstrated that the infiltration bundance of CD8 + T cells, B cells, neutrophils, and NK cells were significantly lower in the high-risk group. A nomogram was also constructed by combining the HNSCCSig and clinical characters. Conclusion: We established and validated the HNSCCSig consisting of SLURP1, SCARA5, CLDN10, MYH11, CXCL13, HLF, and ITGA3. A nomogram combining HNSCCSig and some clinical parameters was constructed to identify high-risk HNSCC-patients with poor prognosis.

Keywords: head and neck squamous cell carcinoma (HNSCC); nomogram; overall survival (OS); robust rank aggregation method; signature.

Figure 1

Overall flowchart of this study.

Figure 2

Integrated analysis of differentially expressed genes (DEGs). (A) Shows heat maps of the top 20 up-regulated and down-regulated differential genes screened from 9 GEO datasets based on “RobustRankAggre” method. (B) Venn diagram of the intersection of TCGA and GEO differential genes. (C) The heatmap of differential genes in the TCGA-HNSCC cohort.

Figure 3

Results of GO function (A) and KEGG pathway enrichment analyses (B).

Figure 4

Verification of the seven gene signatures in training and validation cohort. Kaplan–Meier analysis shows that patients with high risk scores have poorer OS, whether in the TCGA (A) or the GSE65858 data set (B). The ROC curve shows the accuracy of predicting the 3-year and 5-year OS of patients in the TCGA (C) and GSE65858 data sets (D).

Figure 5

Kaplan–Meier analysis of different subgroups in the TCGA test set. (A) T1-2 & T3-4 patients. (B) N0-1 & N2-3 patients. (C) G1-2 & G3-4 patients. (D) Stage III-IV patients.

Figure 6

Cox regression analysis to detect clinical independence of risk scores. Univariate (A) and Multivariate analysis (B) indicated that the risk score is an independent predictor variable in the TCGA dataset. (C, D) The Cox regression analysis also shows that the risk score can be independent of other clinical variables in the validation set GSE65858.

Figure 7

Verification of nomogram prediction performance. (A) A nomogram based on 7 genes and related clinical parameters. (B) The calibration curve reflects the accuracy of the nomogram to estimate the risk. (C) The Kaplan–Meier analysis of the nomogram. (D) The DCA curves evaluate the clinical benefit and the application range of the nomograms.

Figure 8

GSEA and immunity correlation analysis of seven prognostic genes. (A–C) The results showed that 1 KEGG pathway, 2 GO terms, and 3 oncological signatures were enriched in the high-risk group. (D) The immune score and estimated score of the high-risk group were significantly lower in the TCGA cohort. (E) The infiltration level of immune cells (including CD8+ T cells, B cells, neutrophils, and NK cells) in the low-risk group was significantly higher than that in the high-risk group in the TCGA cohort. *P < 0.05, **P < 0.01, and ***P < 0.001 respectively.