Prognostic gene expression signature associated with two molecularly distinct subtypes of colorectal cancer

Abstract

Aims: Despite continual efforts to develop prognostic and predictive models of colorectal cancer by using clinicopathological and genetic parameters, a clinical test that can discriminate between patients with good or poor outcome after treatment has not been established. Thus, the authors aim to uncover subtypes of colorectal cancer that have distinct biological characteristics associated with prognosis and identify potential biomarkers that best reflect the biological and clinical characteristics of subtypes.

Methods: Unsupervised hierarchical clustering analysis was applied to gene expression data from 177 patients with colorectal cancer to determine a prognostic gene expression signature. Validation of the signature was sought in two independent patient groups. The association between the signature and prognosis of patients was assessed by Kaplan-Meier plots, log-rank tests and the Cox model.

Results: The authors identified a gene signature that was associated with overall survival and disease-free survival in 177 patients and validated in two independent cohorts of 213 patients. In multivariate analysis, the signature was an independent risk factor (HR 3.08; 95% CI 1.33 to 7.14; p=0.008 for overall survival). Subset analysis of patients with AJCC (American Joint Committee on Cancer) stage III cancer revealed that the signature can also identify the patients who have better outcome with adjuvant chemotherapy (CTX). Adjuvant chemotherapy significantly affected disease-free survival in patients in subtype B (3-year rate, 71.2% (CTX) vs 41.9% (no CTX); p=0.004). However, such benefit of adjuvant chemotherapy was not significant for patients in subtype A.

Conclusion: The gene signature is an independent predictor of response to chemotherapy and clinical outcome in patients with colorectal cancer.

Figures

Figure 1

Kaplan–Meier plots of the prognosis of patients with colon cancer in the Moffit cohort. Patients were stratified according to American Joint Committee on Cancer (AJCC) stage or gene expression patterns (two clusters). Disease free survival data from 32 patients are not available.

Figure 2

Construction of prediction model in the test cohort according to gene expression signatures from the Moffit cohort. (A) Schematic overview of the strategy used for the construction of prediction models and evaluation of predicted outcomes based on gene expression signatures. (B) Kaplan–Meier plots of OS. Patients were stratified according to AJCC stage or two subgroups predicted by CCP. p Values were obtained from the log-rank test. The ‘+’ symbols in the panels indicate censored data. AJCC, American Joint Committee on Cancer; CCP, compound covariate predictor; OS, overall survival.

Figure 3

Significant association of two subtypes with adjuvant chemotherapy. (A) Kaplan–Meier plots of disease-free survival (DFS) of patients with colorectal cancer in the combined cohort. Patients were plotted according to the prognostic expression signature of 80 genes (two subtypes). Patients in stage I, those in stage II and those in stage III with available adjuvant chemotherapy data were included for analysis (n=266). (B) Kaplan–Meier plots of DFS of patients with colorectal cancer in the combined cohort (patients in stage III, n=109). (C) Kaplan–Meier plots of patients in subtype B with stage III disease (n=59). Patients were plotted according to presence and absence of adjuvant chemotherapy (CTX). (D) Kaplan–Meier plots of patients in subtype A with stage III disease (n=50). Patients were plotted according to presence and absence of adjuvant chemotherapy (CTX).

Figure 4

Subtype-specific gene expression patterns conserved in all three cohorts of patients with colorectal cancer. (A) Venn diagram of genes with expression that differed significantly between patients with colorectal cancer in subtypes A and B in the three different cohorts. Univariate test (two-sample t test) with multivariate permutation test (10 000 random permutations) was applied. In each comparison, we applied a cut-off p value of less than 0.001 to retain genes with expression that differed significantly between the two groups of tissues examined. (B) Expression patterns of selected genes shared in the three colon cancer cohorts. The expressions of only 882 genes were commonly upregulated or downregulated in all three cohorts. Colored bars at the top of the heat map represent samples as indicated.