Our study demonstrates how an integrative computational approach can take advantage of increasingly comprehensive genomic profiling data sets to understand the transcriptional regulation of miRNAs. Although miRNA transcriptional regulation mechanisms have been investigated for specific miRNAs in the past, genome-scale studies have been limited, partially due to the lack of well-annotated miRNA primary transcripts and their corresponding transcriptional start sites. MicroRNAs are excised through a series of processing steps from primary transcripts, which are highly variable in length and can be several kilobases long; identifying the TSS for miRNAs at a genome scale is therefore not trivial. Although experimental methods using chromatin signatures to identify miRNA promoters have been useful in identifying approximate locations of miRNA TSS in some cases (53
), such approaches have their limitations because miRNA expression is highly tissue specific and therefore experimental data applies to a relatively limited subset of miRNAs in specific cell line or tissue contexts. Our genome-wide approach relies heavily on transcript information derived from ESTs, both from EST clusters (AceView) and from ESTs derived from a common clone, to provide transcript-based evidence for defining the TSS of intergenic miRNAs. AceView is one of the most comprehensive assemblies of human transcriptome data, drawing cDNA sequences from dbEST, Genbank and RefSeq and representing a broad panel of tissue and cell type contexts. There are currently more than 8 million ESTs in dbEST and notably approximately 2 million 5′ capped ESTs in Genbank submitted by Kimura et al.
An important aspect of our study, therefore, is the genome-wide definition of presumptive miRNA TSS and promoters using such data, which provides a resource for further investigations of miRNA transcriptional regulation.
The key biological finding of this work is the identification of the p63 and p73 TFs as significant activators of miRNA overexpression in ovarian carcinoma. We found conserved TF binding sites for the p53/p63/p73 family upstream of 17 miRNA loci overexpressed in serous ovarian carcinoma, suggesting p63 and/or p73 may coordinately regulate an ensemble of miRNAs that could significantly modify the cellular gene expression landscape via repression of a cohort of mRNA targets. We also observed an overrepresentation of conserved EGR2
binding sites upstream of miRNA loci overexpressed in ovarian carcinoma, with 10 overexpressed miRNA loci containing EGR2
binding sites. In the TCGA ovarian carcinoma data set, EGR2
expression was significantly negatively correlated with the expression of miRNAs containing EGR2
binding sites. EGR2
has been suggested to be a tumor suppressor, based on the fact that it is frequently underexpressed in human cancers (including ovarian) and cancer cell lines (56
). We hypothesize that the 10 miRNA loci mentioned above are normally repressed by EGR2
, but that decreased EGR2
expression in ovarian carcinoma leads to their overexpression.
It is important to note that both p63 and p73 can be expressed as multiple isoforms that were not distinguishable on the expression platforms involved in this study. The isoforms of p63 and p73 can be divided into two classes: transactivating TAp63/TAp73 isoforms and the inhibitory ΔNp63/ΔNp73 isoforms. In ovarian carcinoma, multiple inhibitory isoforms of p73 and p63 are overexpressed and can be associated with poor prognosis (58
). In contrast, transactivating TAp73 increases the response rate to chemotherapy in BRCA-1-associated ovarian carcinoma (60
). We hypothesize that p73 and p63 may drive the expression of multiple miRNAs that contribute to tumor aggressiveness and/or treatment response. For instance, low levels of miR-200 family miRNA expression are associated with poor treatment response in ovarian carcinoma (61
); thus, TAp73-mediated activation of miR-200 could serve as a mechanism for increased sensitivity to chemotherapy. Future studies will be required to determine the role of specific isoforms of p73 and/or p63 in determining miRNA transcription in ovarian carcinoma.
In experimental validation of the pipeline predictions, we focused on the miR-200 family because of its important roles in tumor development and progression. Members of the miR-200 family inhibit EMT and suppress tumor invasion by directly repressing the TFs Zeb1 and Zeb2 (18–20
). Additionally, the miR-200 family has been shown to repress Bmi1 and Suz12, two essential components of polycomb repressor complexes that are responsible for the maintenance of tumor-initiating cells (21
). Previous work on miR-200 regulation has largely focused on mechanisms of repression, including transcriptional inhibition and epigenetic modifications (18
); mechanisms of transcriptional activation of this miRNA family are not well-understood. Our results with p63 and p73 shed light onto positive transcriptional regulation of the miR-200 family miRNAs, which has potential therapeutic value in the development of approaches aimed at modifying miR-200 family miRNA expression for the treatment of diverse forms of human cancer.
Although our work focused on miR-200 transcriptional regulation by p73 and p63, recent studies have identified different mechanisms by which the p53 family can modulate miR-200 expression. p63 can serve as a transcriptional regulator of Dicer
to modulate miRNA maturation, including processing of the miR-200 family (64
). Given our observation of p63 binding to the miR-200b/a/429 promoter, these studies suggest that p63 can act both directly and indirectly to alter miRNA expression, although the mechanism may be cellular context-dependent. Additionally, a recent report found that in mammary epithelial cells, p53 serves as a transcriptional activator of miR-200c, but not of the miR-200b/a/429 cluster (17
). In ovarian carcinomas, TP53
is commonly mutated and thus unlikely to drive the overexpression of miR-200 that we detected in our samples; however, TP53
mutations can frequently yield dominant negative proteins that interfere with the actions of p63 and p73 through promoter competition and through the formation of inactive heterotetrameric complexes (65
). Although we did not find a relationship between the TP53
mutational status of ovarian carcinomas and their expression of miRNAs with p53 family binding sites, we cannot exclude the role of p53 in the transcriptional regulation of the miR-200 family. In contrast, our results indicate that p63 and p73 are involved in overexpression of this miRNA family in ovarian carcinoma. Taken together with the report of p53-driven miR-200c expression in mammary epithelial cells, our study suggests that cellular context determines the relative importance of different p53 family members on the transcriptional activation of the miR-200 family.
Through our integrative analysis, we identified transcriptional mechanisms that govern miRNA expression in ovarian carcinoma and also found novel activators of miRNAs implicated in tumorigenesis. Our results indicate that p73 and potentially also p63 contribute to broad dysregulation of miRNA expression in ovarian carcinoma. Although we have directed our analysis here toward ovarian cancer, our approach can be broadly applied to other cancer types (e.g., TCGA data sets are expected to expand to include nearly 25 cancer types), or any biological state(s) in which mRNA and differential miRNA expression data are available (e.g., expression profiles across stages of development or across different cell differentiation states). Because different cancer and tissue types show unique mRNA and miRNA expression patterns, utilizing our computational pipeline with other systems may reveal distinct TFs that regulate miRNA activity in varied biological contexts.