Gene expression profiling to predict recurrence of advanced squamous cell carcinoma of the tongue: discovery and external validation

Abstract

Objectives: To establish a prognostic signature for locally advanced tongue squamous cell carcinoma (TSCC) patients treated with surgery.

Results: In the discovery study, unsupervised hierarchical clustering analysis identified two clusters which differentiated the Kaplan-Meier curves of RFS [median RFS, 111 days vs. not reached; log-rank test, P = 0.023]. The 30 genes identified were combined into a dichotomous PI. In the validation cohort, classification according to the PI was associated with RFS [median RFS, 754 days vs. not reached; log-rank test, P = 0.026 in GSE31056] and DSS [median DSS, 540 days vs. not reached; log-rank test, P = 0.046 in GSE42743 and 443 days vs. not reached; P < 0.001 in GSE41613]. Among genes, positive immunohistochemical staining of cytokeratin 4 was associated with favorable prognostic values for RFS (hazard ratio (HR), 0.591, P = 0.045) and DSS (HR, 0.333, P = 0.004).

Materials and methods: We conducted gene expression profiling of 26 clinicopathologically homogeneous advanced TSCC tissue samples using cDNA microarray as a discovery study. Candidate genes were screened using clustering analysis and univariate Cox regression analysis for relapse-free survival (RFS). These were combined into a prognostic index (PI), which was validated using three public microarray datasets of tongue and oral cancer (123 patients). Some genes identified in discovery were immunohistochemically examined for protein expression in another 127 TSCC patients.

Conclusion: We identified robust molecular markers that showed significant associations with prognosis in TSCC patients. Gene expression profiling data were successfully converted to protein expression profiling data.

Keywords: cytokeratin 4; gene expression profile; keratin 4; oral squamous cell carcinoma; tongue squamous cell carcinoma.

Figure 1. Subgrouping through unsupervised hierarchical clustering in the discovery study

(A) Heatmap of the gene expression of 3755 probe sets used for USV clustering. (B) Kaplan-Meier curves for RFSs of two patient groups according to clustering analysis in the discovery cohort (Log-rank test, P value = 0.023). RFS, relapse-free survival.

Figure 2. Relationship of the expression of 30 genes between cluster A, cluster B and normal tissues

(A) Hierarchical cluster analysis of 30 samples (26 carcinoma tissues, 4 matched normal tissues) derived from 26 patients based on 30 gene expression patterns. (B) Normalized intensity values of predictive candidate genes of normal, cluster A and cluster B subgroups. Upper graph: five genes associated with a favorable prognosis. Lower graph: 25 genes associated with an unfavorable prognosis.

Figure 3. Kaplan-Meier curves of two subgroups divided based on a predictive formula in three independent cohorts

(A) GSE31056 for RFS (Log-rank test, P value = 0.026), (B) GSE42743 for DSS (Log-rank test, P value = 0.046), (C) GSE41613 for DSS (Log-rank test, P value <0.001).

Figure 4. Relationship between CK4 protein expression, RFS and DSS

(A) Micrograph of a representative TSCC expressing CK4 protein detected byimmunohistochemical staining. Left, hematoxylin & eosin; right, CK4. (B) Kaplan-Meier curves drawn according to CK4 protein expression by immunohistochemical staining. Left, RFS, Log-rank test, P value = 0.048. Right, DDS Log-rank test, P value = 0.002.

Figure 4. Relationship between CK4 protein expression, RFS and DSS

(A) Micrograph of a representative TSCC expressing CK4 protein detected byimmunohistochemical staining. Left, hematoxylin & eosin; right, CK4. (B) Kaplan-Meier curves drawn according to CK4 protein expression by immunohistochemical staining. Left, RFS, Log-rank test, P value = 0.048. Right, DDS Log-rank test, P value = 0.002.