Expression Profiling and Molecular Classification of Colorectal Cancer: Meta-Analysis of 8 Independent mRNA Expression Datasets across the Globe
Levi Waldron, Giovanni Parmigiani, Charles Fuchs, Curtis Huttenhower, Shuji Ogino. Dana-Farber Cancer Institute, Boston, MA; Harvard School of Public Health, Boston, MA; Brigham and Women's Hospital, Harvard Medical School, Boston, MA
Background: Gene expression profiling has not reproducibly sub-classified colorectal cancers. In addition to the existing tumor biomarkers such as KRAS, BRAF, and MSI, gene expression profiling may provide clinically useful classification system, as well as biological insights.
Design: We meta-analyzed 8 genome-wide mRNA expression datasets (total N=2001). We performed gene expression profiling of 718 cases of archival colorectal cancer tumors, using the Illumina DASL microarray platform. We performed strict quality control for both samples and gene probes. We also obtained data from 7 independent studies on transcirptomic profiling from publicly available database; a total of 1,283 samples from fresh-frozen tissues. These 2,001 samples formed a basis for identifying and validating discrete or continuous subtypes present across research centers and study populations.
Results: We identified reproducible expression profiling clusters by principal component analysis (PCA). In addition, we were able to validate MSI-high-associated signatures across studies. Notably, gene set enrichment analysis (GSEA) showed that a number of immunity-related pathways represented reproducible signatures, which differentiated subtypes of colorectal cancers across the studies.
Conclusions: Genome-wide mRNA expression data can provide useful information in colorectal cancer molecular classification. In particular, MSI-high is associated with reproducible differentially up-regulated and down-regulated genes. Our data support that immunity-related pathways are important in colorectal tumor biology, reflecting the fact that tumor consists of both transformed epithelial cells and non-neoplastic stromal and immune cells, all of which interact in the tumor microenvironment.
Category: Special Category - Pan-genomic/Pan-proteomic approaches to Cancer
Tuesday, March 5, 2013 1:15 PM
Proffered Papers: Section H1, Tuesday Afternoon