EMT is the dominant program in human colon cancer

Abstract

Background: Colon cancer has been classically described by clinicopathologic features that permit the prediction of outcome only after surgical resection and staging.

Methods: We performed an unsupervised analysis of microarray data from 326 colon cancers to identify the first principal component (PC1) of the most variable set of genes. PC1 deciphered two primary, intrinsic molecular subtypes of colon cancer that predicted disease progression and recurrence.

Results: Here we report that the most dominant pattern of intrinsic gene expression in colon cancer (PC1) was tightly correlated (Pearson R = 0.92, P < 10(-135)) with the EMT signature-- both in gene identity and directionality. In a global micro-RNA screen, we further identified the most anti-correlated microRNA with PC1 as MiR200, known to regulate EMT.

Conclusions: These data demonstrate that the biology underpinning the native, molecular classification of human colon cancer--previously thought to be highly heterogeneous-- was clarified through the lens of comprehensive transcriptome analysis.

Figure 1

Intrinsic molecular stratification of human colorectal cancer. Unsupervised analysis and hierarchical clustering of global gene expression data derived from colorectal cancer cases identified 2 major "intrinsic" subclasses (cyan and magenta) distinguished by the first principal component (PC1) of the most variable genes. These two key native subtypes were clearly identified in both the (a) Moffitt Cancer Center (MCC) data set (n = 326) and the (b) EXPO dataset (n = 269). PC1 was later found to be tightly correlated with an EMT signature derived from cell lines, providing an explanation for the biology underpinning these two intrinsic classes in both datasets. PC1 clearly distinguishes two subclasses which were subsequently identified as epithelial vs. mesenchymal. On both panels (a) and (b), mean-centered probe intensities are shown, and probes are clustered using Pearson correlation based distance and Ward linkage. Also, rows represent samples, and columns represent array probes. Panel (c) shows scatter plot of EMT signature score and PC1 (First Principal Component Score) on Moffitt Cancer Center data set. Panel (D) shows the scatter plot between probe intensities for Vimentin (VIM) and E-cadherin probes in a panel of 93 Lung Cancer Cell Lines. Cell Lines exhibiting epithelial-like phenotype are shown in green; those exhibiting mesenchymal-like phenotype are shown in red.

Figure 2

Hierarchical cluster analysis of the top 100 genes assessed from a text mining approach were strongly associated with the EMT program as shown on 326 MCC colon tumors sorted by PC1. The 100 gene set contains individual genes (CDH1, CLDN9, FGFR1, FN1, TWIST 1 & 2, AXL, VIM) as well as signatures of genes (PC1, EMT, TGFbeta, Proliferation, MYC, and RAS) that are up-regulated in mesenchymal tumors (shown in magenta), and that are up-regulated in epithelial tumors (shown in cyan). Names for the relevant gene signatures are shown in black. Samples (rows) are sorted by PC1. Genes (columns) are clustered using Pearson correlation and Ward linkage. Heatmap shows mean-centered probe intensities.

Figure 3

Correlation of microRNA analysis (~700) with PC1/EMT across 49 colorectal cancers identified the MiR200 family as strongly, negatively correlated with PC1/EMT (upper plots). The Mir 200 family has been linked to inhibition of EMT (promotion of the epithelial phenotype) through inhibition of Zeb 1 & 2, known transcriptional repressors of CDH1. Waterfall plots show MiR 200 over-expression is correlated with more tumors classified as epithelial than mesenchymal and Mir 200 under-expression is correlated with fewer epithelial than mesenchymal tumors (lower plots).

Figure 4

Covariance matrix showing correlation of PC1 with disease recurrence. (a) PC1, despite being developed with unsupervised approaches, appeared to correlate well with EMT, disease recurrence, disease progression, and differentiation status, but not with gene signatures linked to adenoma vs. carcinoma, MSI status, or mucinous vs. nonmucinous cancers. Moreover, PC1 appeared to be anti-correlated with RAS, MYC, Proliferation, and colon laterality. PC1 distinguishes good and poor prognosis patients in the MCC data set (b) as well as in two independent test sets: (c) Netherlands Cancer Institute (NKI, #13) (L.V.V.) and (d) Lin et al data sets [21].