MicroRNA expression and activity in high-grade serous ovarian cancer

miRNAs are critical modulators in the development and progression of cancer. Emerging evidence suggests that they are drivers of ovarian cancer. A better understanding of the molecular underpinnings of the development, progression and chemoresistance of the disease is critical for the development of new, more effective therapies. Here we explore the expression patterns of miRNAs as they relate to gene expression, as they differ across molecular subtypes of the disease. We examine the correlation structure of miRNA expression with mRNA expression in two distinct genomic datasets and report on patterns in correlation structure in several subsets of the data. We find that the datasets show consistency in their correlation structure, and in the specific miRNA-mRNA pairs that are either highly positively or negatively correlated. The data include a larger number of strong positive and strong negative correlations than would be expected by chance, indicating that biological relationships between the types of data are detectable in these datasets. We further find an enrichment for positively-correlated miRNA-mRNA pairs in which the miRNA is encoded in close proximity to the mRNA. The correlation of miRNA and mRNA is apparently unaffected by miRNA and mRNA expression level; similarly the two molecular subtypes do not contain differences in their correlation. We find that the recently described poorer prognosis, or angiogenic, subtype has a generally lower miRNA activity than the second, non-angiogenic, subtype. The subtypes are characterized by a consistent pattern of differential miRNA expression. We also report on a switch-like relationship between the expression levels of certain miRNAs and the genes that are anticorrelated with them. We propose these miRNAs drive many of the differences in the subtypes both directly, by RISC-mediated repression of target messages and indirectly, by repressing transcription factors that regulate expression in the cell. We build models of patient survival and time-to-relapse based on these miRNA expression data and inferred miRNA activity scores, using several types of univariate and variable selection models. We find essentially no survival-predictive information provided by the RE score data. While the direct miRNA expression measurements may contain some predictive power, we find that a larger dataset and the segretation of that dataset into distinct molecular phenotypes is likely to be necessary to produce a useful model of survival in ovarian cancer.EThOS - Electronic Theses Online ServiceGBUnited Kingdo