Gene expression signature-based prognostic risk score in gastric cancer

Abstract

Purpose: Despite continual efforts to develop a prognostic model of gastric cancer by using clinical and pathologic parameters, a clinical test that can discriminate patients with good outcomes from those with poor outcomes after gastric cancer surgery has not been established. We aim to develop practical biomarker-based risk score that can predict relapse of gastric cancer after surgical treatment.

Experimental design: Microarray technologies were used to generate and analyze gene expression profiling data from 65 gastric cancer patients to identify biomarker genes associated with relapse. The association of expression patterns of identified genes with relapse and overall survival was validated in independent gastric cancer patients.

Results: We uncovered two subgroups of gastric cancer that were strongly associated with the prognosis. For the easy translation of our findings into practice, we developed a scoring system based on the expression of six genes that predicted the likelihood of relapse after curative resection. In multivariate analysis, the risk score was an independent predictor of relapse in a cohort of 96 patients. We were able to validate the robustness of the six-gene signature in an additional independent cohort.

Conclusions: The risk score derived from the six-gene set successfully prognosticated the relapse of gastric cancer patients after gastrectomy.

Fig. 1. Hierarchical clustering analysis of gene expression data from the YGC cohort

(A) Hierarchical clustering of gene expression data from 65 gastric cancer and 6 GIST patients in the YGC cohort. Genes with expression levels that were at least 2-fold different in at least 15 tissues, relative to the median value across tissues, were selected for hierarchical clustering analysis (2,077 gene features). The data are presented in matrix format, in which rows represent individual genes and columns represent each tissue. Each cell in the matrix represents the expression level of a gene feature in an individual tissue. The color red or green in cells reflects relative high or low expression levels, respectively, as indicated in the scale bar (log2 transformed scale). (B) Kaplan-Meier plots of three gastric cancer clusters in the YGC cohort. The six patients with GIST were excluded from the plotting. (C) Kaplan-Meier plots of stage III patients in 2 clusters (C1 and C2) in the YGC cohort. (No stage III patients were identified in C3.)

Fig. 2. Gene Expression signature unique to Cluster C1

Measured gene expression values were log 2-transformed and median-centered across samples before generating heatmap.

Fig. 3. Risk score based on six-gene signature and RFS of patients in GSH1

(A) The relative risk score based on the six-gene signature of each patient. (Each bar represents the risk score of an individual patient.) The regression coefficients of each gene were calculated by Cox regression analysis (Supplementary Table 1). The risk score was used to dichotomize patients into high- or low-risk groups, with the 50th percentile as the cut-off. To avoid the ambiguity of a risk score near the median value, patients in the upper and lower 5th percentiles from the risk score median were removed from Kaplan-Meier plotting. Blank bars near the median indicate these patients. (B) Kaplan-Meier plots of two risk score risk groups in the GSH1 cohort. P-values were obtained from the log-rank test.

Fig. 4. Kaplan–Meier survival plots of overall survival and relapse-free survival in AJCC stage III gastric cancer patients in GSH2 cohort

Patients were stratified by risk score in all stage III (A and D), risk score in IIIA (B and E), in IIIB (C and F). P-values were obtained from the log-rank test.