Swanhild U Meyer1, Steffen Sass2, Fabian J Theis2, Michael W Pfaffl1
1Physiology Weihenstephan, ZIEL Research Center for Nutrition and Food Sciences; 2MIPS, Institute for Bioinformatics and System Biology, Helmholtz Center Munich, German Research Center for Environmental Health, Neuherberg, Germany
Background – In silico target prediction of miRNAs often reveals several thousand possible mRNA interactions for an individual miRNA. Elucidating the function of miRNAs by target prediction algorithms is biased by false positives and negatives as well as miRNA targets which might not be relevant in a specific cellular context. However, holistic miRNA-target interaction analysis by cross-linking and pull-down experiments is time consuming and expensive. Using miRNA and mRNA expression data in combination with in silico predictions allows improving the chance of predicting functional relevant miRNA-mRNA interactions.
Methodology / Principal Findings – We used samples of in vitro myoblast differentiation and differentiation with TNFalpha or IGF1 treatment for profiling analysis of miRNA (Agilent microarray, Life Technologies qPCR card) and mRNA (Affymetrix array) expression. For integrated miRNA-mRNA analysis 21 miRNAs were selected. We used the data of each miRNA profiling platform as well as a joint dataset of microarrays and qPCR cards representing the intersection of both platforms. Target prediction was based on TargetScan (www.targetscan.org) and miRanda (www.microrna.org) data. Integrated miRNA-mRNA was performed by miRLastic, a negative multiple linear regression analysis. miRLastic analysis including the intersection miRNA dataset (individual miRNA datasets) revealed around 3,800 (6,700) putative miRNA target interactions in total and about 180 (320) targets per miRNA on average. Target prediction solely based on in silico data by e.g. miRanda would have resulted in more than 127,000 total target predictions and about 6,000 targets per miRNA. For focusing on predicted miRNA-mRNA interactions with a high likelihood of being functional in myogenic differentiation we applied filter criteria such as the number of inverse miRNA-target correlations of a given miRNA, number of transcription factors targeted per miRNA, number of different miRNAs targeting a gene transcript, as well as enrichment of targets by cocitation or in GO terms and signalling pathways.
Conclusions – To narrow down the complexity of possible miRNA and mRNA interactions in myogenic differentiation we integrated the analysis of miRNA and mRNA profiling data and target prediction algorithms. The resulting number of possible miRNA target interactions was further reduced by the sequential application of specific selection criteria to generate a dataset of high putative functional significance.
|Back to microRNA, siRNA and long non-coding RNAs|