PMC

Unsupervised clustering of miRNA expression data had identified three subtypes (C1–C3) of high-grade serous ovarian tumors. For miRNAs and genes differing significantly between the groups (t-test P<0.01, fold change>1.5, any subtype compared to the other tumors), predicted miRNA:mRNA functional pairs were defined, based on both anti-correlation in expression and predicted miRNA targeting interaction (both miRanda and TargetScan). For each miRNA:mRNA group (e.g. miRNA high/gene low in C1 versus other tumors), expression patterns are represented as heat maps (rows, miRNAs or gene transcripts; columns, profiled samples).

Analogous to , scatter plot showing mean correlation versus significance of enrichment for predicted target interactions (enrichment expressed as a Fisher's exact z-score), when separately considering the following potential interactions: 7mer seed sequence in 3′-UTR (black dotted line), 7mer seed sequence in 5′-UTR (black dashed line), 7mer seed sequence in coding sequence region (“cds,” gray dotted line), miRanda prediction (black solid line), TargetScan prediction (gray solid line). Plot uses bins of 10000 miRNA:mRNA pairs (total number of pairs represented: 191 miRNAs X 8547 genes). Fisher's exact z-score of +/−2.57 corresponds to significant enrichment (nominal P<0.01) for predicted targets within miRNA:mRNA pairs.

Using a training dataset of TCGA miRNA expression profiles, a prognostic miRNA signature was defined (left panel) and then applied to a test dataset (right panel), each tumor being assigned a score measuring how well the tumor's expression patterns reflected those of the signature pattern. Kaplan-Meier analysis (log-rank tests) compares time to death for ovarian cancer patients showing higher risk (prognostic score>0) versus lower risk (prognostic score<0). Univariate Cox test treats the prognostic score as a continuous variable. (miR-923 was recently determined to be an rRNA fragment, according to miRBase.).

(A) Correlation of gene expression with miR-29a expression (genes from , same ordering). Predicted miR-29a targets are indicated. (B) Top eight words (of all 5, 6 and 7mers) enriched in 3′-UTRs of mRNAs anti-correlated with miR-29a expression (FDR<1e-6). (C) QPCR analysis showing relative quantity of selected miR-29a anti-correlated gene targets after miR-29a overexpression in HEYA8 ovarian cancer cells (SCR, scrambled control; WT, untreated; two-sided t-test P<0.05, mir-29a vs SCR and miR-29a vs WT, each comparison, except for SAE1). (D) MTS assays demonstrating the effect of miR-29a overexpression on proliferation of HEYA8 and OVCAR-8 cells (Lipo, lipofectamine-treated alone, no miRNA; two-sided t-test P<0.001, miR-29a vs each of three control groups at both 48 h and 72 h). (E) Effect of miR-29a (72 h) on proliferation, under a range of concentrations of cisplatin treatment (at 48 h). Error bars represent standard error.

(A) MiRNAs and their predicted gene targets tend to be anti-correlated within ovarian tumors. Scatter plot showing mean correlation and fraction of predicted target interactions (miRanda-mirSVR score>0.1), using bins of 10000 miRNA:mRNA pairs (total number of pairs represented: 191 miRNAs X 8547 genes). Dashed line corresponds to chance expected baseline fraction (13.2%) of predicted target interactions. Correlations were computed using both Pearson's (gray line) and a simple linear regression model to account for ‘noise’ due to DNA copy alterations (black line). (B) Hierarchical clustering matrix (with Pearson's correlation coefficient as distance metric, Ward's Linkage) of correlation coefficients for all miRNA:mRNA pairs having a strong negative correlation (regression coefficient smaller than −7.0, only negative correlations represented). For each gene cluster, enriched gene classes are indicated. *, significant anti-enrichment (P<0.001, one-sided Spearman's rank, TargetScan or miRanda) for predicted targets within miRNA:mRNA correlations.

(A) MiRNAs correlated with CNA. Left plot shows hemizygous loss (N< = 1 copies) versus homozygous deletion (N = 0) for miRNA precursors. Correlation (Pearson's) between precursor DNA copy number and mature miRNA expression is indicated by symbol size, and only miRNAs with a minimum correlation (R>0.3) are included here; cytoband regions in bold represent focal CNA events. Similarly, right plot shows gain (N> = 3) versus amplification (N> = 4) for miRNAs having R>0.3 for precursor copy versus mature expression. (B) miRNAs are frequently coexpressed with neighboring miRNAs. Plot shows relationship between the distance separating miRNA loci and their coordinate expression. Each miRNA was paired with each of the others lying in the same orientation on the same chromosome. For each pair, the distance between the two loci (right axis) was ranked, and the correlation coefficient for their expression was plotted according this rank (left axis). (C) miRNAs are frequently coexpressed with host genes. For each of 188 miRNA-host gene pairs (same orientation), the correlation was computed both between miRNA and host gene expression; pair orderings are the same (“X”, no corresponding mRNA or gene copy data; p-values by two-sided t-statistic).