Detection of a microRNA signal in an in vivo expression set of mRNAs

Abstract

Background: microRNAs (miRNAs) are approximately 21 nucleotide non-coding transcripts capable of regulating gene expression. The most widely studied mechanism of regulation involves binding of a miRNA to the target mRNA. As a result, translation of the target mRNA is inhibited and the mRNA may be destabilized. The inhibitory effects of miRNAs have been linked to diverse cellular processes including malignant proliferation, apoptosis, development, differentiation, and metabolic processes. We asked whether endogenous fluctuations in a set of mRNA and miRNA profiles contain correlated changes that are statistically distinguishable from the many other fluctuations in the data set.

Methodology/principal findings: RNA was extracted from 12 human primary brain tumor biopsies. These samples were used to determine genome-wide mRNA expression levels by microarray analysis and a miRNA profile by real-time reverse transcription PCR. Correlation coefficients were determined for all possible mRNA-miRNA pairs and the distribution of these correlations compared to the random distribution. An excess of high positive and negative correlation pairs were observed at the tails of these distributions. Most of these highest correlation pairs do not contain sufficiently complementary sequences to predict a target relationship; nor do they lie in physical proximity to each other. However, by examining pairs in which the significance of the correlation coefficients is modestly relaxed, negative correlations do tend to predict targets and positive correlations tend to predict physically proximate pairs. A subset of high correlation pairs were experimentally validated by over-expressing or suppressing a miRNA and measuring the correlated mRNAs.

Conclusions/significance: Sufficient information exists within a set of tumor samples to detect endogenous correlations between miRNA and mRNA levels. Based on the validations the causal arrow for these correlations is likely to be directed from the miRNAs to the mRNAs. From these data sets, we inferred and validated a tumor suppression pathway linked to miR-181c.

Figure 1. Distribution of correlation coefficients.

(A) The histogram consists of 40 bins ranging from −1 to 1. (B) The ratio of number of correlation coefficients in the experimental data to the random case for bin in Figureô 1A. At high correlations, the number of microRNA-mRNA pairs is larger than expected in the random case. This result suggests information exists in the correlation coefficients between miRNAs and mRNAs.

Figure 2. Percentage of predicted targets across a range of correlation coefficients.

The percentage of targets over the range of correlation defined in Figureô 1 decreases with the increasing correlation between miRNAs and mRNAs. This implies that the correlation indeed captures information about target prediction, and the negative slope might be related to direct targeting effect between miRNAs and mRNAs.

Figure 3. Histogram of P values from the WRS test for each of the miRNAs.

The WRS test compared the correlations of the predicted targets with that of the non-predicted ones for each of the 91 miRNAs. As the correlation went from negative to positive the percentage of predicted target pairs decreased. Interestingly, some highly positively correlated miRNAs had high P values, which suggests an indirect relationship with the predicted target.

Figure 4. Distribution of correlations between miRNAs and proximate mRNAs.

This figure showed the distributions for 3 groups of miRNA-mRNA pairs, classified by the distance between a miRNA and an mRNA on the same chromosome. The y-axis shows the percentage of miRNA-mRNA pairs in each range of correlation (−1 to 1 with binsize 0.1). An intronic miRNA had a distance zero to its host gene. The average length of a human gene, 55Kb (http://www.ncbi.nlm.nih.gov/Web/Newsltr/Spring03/human.html), was chosen as one distance boundary. As the distance between the correlated pair increased, the distribution increasingly shifted away from positive correlations to a normal distribution.

Figure 5. Experimental Validation of Highly correlated miRNA/transcript pairs.

The above figure represents Ct_scrambled-Ct_{pre-mir/2′-O-Me} from three individual transfection experiments of pre-miR-181c, pre-miR-182, 2′-O-Me-miR-19a, 2′-O-Me-scrambled and pre-mir-scrambled. PCAF, ASCC1, FADS2, HPS1, ANXA4, GSTO1, IFITM1, DDX11, TTC3 and PRR13 levels were determined after transfection.

Figure 6. Immunoblot analysis of miR-181c treated cells.

Immunoblots of U251 and U87 transfected cells transfected with either miR-181c pre-miRNA or a scrambled sequence. PCAF (A) and p53 (C) were immunodetected and actin was used as a loading control. Quantification of the immunoblots showed a significant difference in B and C (p<0.01, t-test in all three comparisons).

Figure 7. Cell Growth and Apoptosis Assays.

U251 cells were treated with pre-miR-181c and negative control scramble. 24 hrs post-transfection, a) cell growth (p<0.01, t-test) and b) apoptosis were assayed (p<0.05, t-test).

Figure 8. Tumor suppressive pathway involving miR-181c.

Figure 9. Mechanisms through which high miRNA-mRNA correlation pairs could operate.