Large-scale association analysis of gene signatures links pathways, mechanisms, diseases, and drugs

AACR Annual Meeting-- Apr 14-18, 2007; Los Angeles, CA 5679 Several varieties of genome-scale analysis have been applied to the study of human biology and disease, including DNA microarrays used for global gene expression profiling. The Oncomine cancer profiling database represents a concerted effort to collect, standardize and analyze the global collection of cancer profiling data. In addition, researchers have profiled a wide variety of biological perturbations including drug treatments and the activated or repressed pathways. Here, we have attempted to collect and integrate all relevant gene signatures in the in the oncology knowledge space for large-scale association analysis. This work compares signatures representing thousands of cancer types and subtypes, biological perturbations, drug treatments, manually annotated pathways and protein interaction networks, predicted regulatory networks, gene ontologies and protein families. Our analysis integrates and builds upon Oncomine, the Connectivity Map and the work of the profiling and annotation communities to systematically connect pathways, mechanisms, diseases and drugs. We designated this project as the Molecular Concept Map, as the focus is on biological concepts that are represented by molecular signatures. First, we present an analysis of the Myc pathway, demonstrating that an in vitro Myc signature generated in mammary epithelial cells shares strong similarity with Myc signatures generated in alternative systems as well as disease subtypes with obvious Myc involvement, such as Ig-Myc positive lymphoma and n-Myc positive neuroblastoma. In addition, we identify a number of cancer types and subtypes displaying coordinate activation of the Myc pathway. Finally, we identified a class of compounds capable of indirectly repressing the Myc program in vitro, suggesting that such compounds could be used to treat the identified malignancies with Myc pathway activation. Second, we examined signatures of relapse in estrogen receptor (ER) positive and negative breast cancers. We found that distinct molecular concepts characterized ER+ and ER- tumors prone to relapse. The ER+ relapse-positive breast cancer signature was linked to cell cycle concepts, E2F and NF-Y promoter binding sites, 17q, 8q and 20q chromosome arms and inversely related to cell-cycle modulating compounds such as resveratrol. In contrast, the ER- relapse-positive breast cancer signature was linked to actin-binding genes, YY1 promoter binding sites, the Xp chromosome arm, the c-Src pathway and inversely related to valproic acid treatment, a histone deacetylase inhibitor. These analyses suggest that largely distinct pathways mediate relapse in ER- and ER+ tumors and furthermore suggest that different treatment strategies may be necessary.