Twenty‑four signature genes predict the prognosis of oral squamous cell carcinoma with high accuracy and repeatability

Abstract

Oral squamous cell carcinoma (OSCC) is the sixth most common type cancer worldwide, with poor prognosis. The present study aimed to identify gene signatures that could classify OSCC and predict prognosis in different stages. A training data set (GSE41613) and two validation data sets (GSE42743 and GSE26549) were acquired from the online Gene Expression Omnibus database. In the training data set, patients were classified based on the tumor‑node‑metastasis staging system, and subsequently grouped into low stage (L) or high stage (H). Signature genes between L and H stages were selected by disparity index analysis, and classification was performed by the expression of these signature genes. The established classification was compared with the L and H classification, and fivefold cross validation was used to evaluate the stability. Enrichment analysis for the signature genes was implemented by the Database for Annotation, Visualization and Integration Discovery. Two validation data sets were used to determine the precise of classification. Survival analysis was conducted followed each classification using the package 'survival' in R software. A set of 24 signature genes was identified based on the classification model with the Fi value of 0.47, which was used to distinguish OSCC samples in two different stages. Overall survival of patients in the H stage was higher than those in the L stage. Signature genes were primarily enriched in 'ether lipid metabolism' pathway and biological processes such as 'positive regulation of adaptive immune response' and 'apoptotic cell clearance'. The results provided a novel 24‑gene set that may be used as biomarkers to predict OSCC prognosis with high accuracy, which may be used to determine an appropriate treatment program for patients with OSCC in addition to the traditional evaluation index.

Figure 1.

Kaplan-Meier curve analysis indicated a significant difference in survival between H (n=41) and L (n=35) stages samples based on tumor-node-metastasis classification. P=2.00×10⁻⁰⁵. H, high; L, low.

Figure 2.

Optimal threshold selection in the classification model. (A) Overlap scale of classifications using the model with the High and Low classifications. Gene sets with |F_i|>k were selected, and k from 0–1 was set with a step size of 0.01; F_i=0.47 (vertical line) was used as the cut-off value to classify the samples. (B) 5-fold cross validation results for 10 iterations, which are indicated by the different colored lines. F_i, disparity index; k, iteration step.

Figure 3.

Clustering analysis of signature gene expressions and corresponding survival analyses. (A) Signature genes were used to mark samples in H and L classifications under the score-classification model. (B) Kaplan-Meier curve indicating a significant difference in survival between Cluster 1 and Cluster2 classified with the boundary of ‘0’. (C) Heat map of signature gene expressions in two cluster samples. (D) Kaplan-Meier curve indicating a significant difference in survival between Cluster1 and Cluster2 that were identified by expression profiling of the signature genes. H, high; L, low.

Figure 4.

Multivariate survival analysis of 24 signature genes. (A) ROC curve; AUC=0.97. (B) Kaplan-Meier curve indicating a significant difference in survival between high-risk and low-risk samples identified by the multivariate prognosis of 24 genes; P=2.48×10⁻¹⁹. AUC, area under the ROC curve; ROC, receiver operating characteristic.

Figure 5.

Validation of survival analysis using two additional data sets. (A) ROC curve of the classification in the GSE42743 data set; AUC=0.994. (B) Kaplan-Meier curve indicating a significant difference in survival between high and low risk samples in GSE42743; P=4.55×10⁻¹⁵. (C) ROC curve of the recurrent classification in the GSE26549 data set; AUC=0.984. (D) Kaplan-Meier curve indicating a significant difference between high and low risk samples in GSE26549; P=1.41×10⁻¹⁴. AUC, area under the ROC curve; ROC, receiver operating characteristic.