Large-Scale Genome-Wide mRNA Expression Profiling of 1003 Colorectal Cancers
Shuji Ogino, Levi Waldron, Yujin Hoshida, Giovanni Parmigiani, Todd Golub, Curtis Huttenhower, Charles Fuchs. Brigham and Women's Hosp., Boston; Harvard School of Public Health, Boston; Broad Institute, Cambridge
Background: Over 20 million archival tissue samples are stored annually in the U.S. alone, as formalin-fixed, paraffin-embedded (FFPE) tissue blocks. If we can unfold information on genome-wide expression signatures from FFPE tissue, such tissue samples can offer an enormous resource to link the molecular mechanisms of disease to its impact on patient outcome. In this study, our aim was to identify reproducible colorectal cancer subtype clusters and genome-wide mRNA expression signatures in 1003 colorectal cancers, using DASL (cDNA-mediated Annealing, Selection, Extension, and Ligation) microarray technology.
Design: We analyzed mRNA expression of over 24,000 genes in 1003 colorectal cancers using the DASL HumanRef8 v3 (24K probes) microarray. We also analyzed MSI, KRAS, BRAF, and PIK3CA mutations, and other tissue markers as well as clinical outcome of patients.
Results: We established extensive quality check pipeline. After thoroughly validating expression profiling data, we applied complete linkage hierarchical clustering to tumor samples using Pearson correlation and assessed associations with sample metadata. Each metadata variable was tested using an appropriate association test for its class (linear regression, F-test, or chi-square test), and p-values were adjusted for false discovery rate. Supervised clustering was conducted with a randomly held-out training set and a validation (test) set to confirm subtype clustering. We were able to identify colorectal cancer subtypes, which were associated with KRAS mutation, or with BRAF mutation, MSI and CpG island methylator phenotype (CIMP).
Each subtype cluster showed a unique signature of up- and down-regulation of functionally-relevant gene sets.
Conclusions: Genome-wide mRNA expression analysis on 1003 colorectal cancers has enabled an identification of colorectal cancer molecular subtypes and differentially expressed genes. We anticipate genome-wide expression studies being broadly applicable to future studies of molecular events in large cancer patient populations.
Monday, March 19, 2012 1:00 PM
Poster Session II # 114, Monday Afternoon