Integrating genomic, epigenomic, and transcriptomic features reveals modular signatures underlying poor prognosis in ovarian cancer

Abstract

Ovarian cancer has a poor prognosis, with different outcomes for different patients. The mechanism underlying this poor prognosis and heterogeneity is not well understood. We have developed an unbiased, adaptive clustering approach to integratively analyze ovarian cancer genome-wide gene expression, DNA methylation, microRNA expression, and copy number alteration profiles. We uncovered seven previously uncategorized subtypes of ovarian cancer that differ significantly in median survival time. We then developed an algorithm to uncover molecular signatures that distinguish cancer subtypes. Surprisingly, although the good-prognosis subtypes seem to have not been functionally selected, the poor-prognosis ones clearly have been. One subtype has an epithelial-mesenchymal transition signature and a cancer hallmark network, whereas the other two subtypes are enriched for a network centered on SRC and KRAS. Our results suggest molecular signatures that are highly predictive of clinical outcomes and spotlight "driver" genes that could be targeted by subtype-specific treatments.