The multiMiR R package and database: integration of microRNA-target interactions along with their disease and drug associations

Abstract

microRNAs (miRNAs) regulate expression by promoting degradation or repressing translation of target transcripts. miRNA target sites have been catalogued in databases based on experimental validation and computational prediction using various algorithms. Several online resources provide collections of multiple databases but need to be imported into other software, such as R, for processing, tabulation, graphing and computation. Currently available miRNA target site packages in R are limited in the number of databases, types of databases and flexibility. We present multiMiR, a new miRNA-target interaction R package and database, which includes several novel features not available in existing R packages: (i) compilation of nearly 50 million records in human and mouse from 14 different databases, more than any other collection; (ii) expansion of databases to those based on disease annotation and drug microRNAresponse, in addition to many experimental and computational databases; and (iii) user-defined cutoffs for predicted binding strength to provide the most confident selection. Case studies are reported on various biomedical applications including mouse models of alcohol consumption, studies of chronic obstructive pulmonary disease in human subjects, and human cell line models of bladder cancer metastasis. We also demonstrate how multiMiR was used to generate testable hypotheses that were pursued experimentally.

Figure 1.

multiMiR components and data workflow. multiMiR components, including R functions and database are highlighted in grey. Data analysis flow is denoted by arrows.

Figure 2.

R commands and results for examples 1 (A) and 2 (B). (A) Parameter ‘summary’ was set to TRUE (to summarize the result) and other parameters were as default. By default, get.multimir searches the validated miRNA–target interactions in human. MTI, miRNA–target interaction.

Figure 3.

(A) Diagram of 3’UTR of Gnb1. Graphic was from Saba et al. (41) and generated using tools available at UCSC Genome Browser (http://genome.ucsc.edu/). Among four different Gnb1 transcripts, there are two versions of the 3′ UTR region of Gnb1 (blue bars bottom). The diagram also depicts the differentially expressed Affymetrix probe set for Gnb1 (bright blue), polyadenylation sites (green), predicted miRNA target sites (red) and the SNPs predicted to differ between ISS and three other inbred strains (purple). The entire length of the 3’UTR is 1809bp and the longer version extends from 254 to1809bp. mmu-miR-218 had predicted target sites by three databases (DIANA-microT, ElMMo and PITA) and mmu-miR-101a/b had predicted target sites by four databases (ElMMo, miRanda, PicTar and TargetScan). By checking location information from the original databases, we found that in the long form of the 3’UTR, the mmu-miR-218 predicted target site was located at 727–734bp and the mmu-miR-101a/b predicted target site were located at 1765–1771bp. In the shorter version of the 3’ UTR, mmu-miR-218 also had a predicted target site at 150–178bp. (B) RT-PCR results for mmu-miR-101a and mmu-miR-218 in ILS and ISS mice. Strain effects determined by ANOVA are suggestive of mmu-miR-101a (p = 0.058, fold-change FC = 1.94) and mmu-miR-218 (p = 0.069, FC = 1.37) expression increases in ISS. Conditions for each strain were combined to create the boxplot.

Figure 4.

Results of luciferase reporter assay for hsa-miR-429 and CERS6. The first construct (A) contains the hsa-miR-429 predicted binding site and the second construct (B) contains site directed mutations of the predicted binding site. Plasmids were treated with control (con) or hsa-miR-429 mimics. Bars indicate the standard error. The wild type CERS6 target site is AAC AGT ATT TGC ATT and the mutant target site is AAC AaT gTT TGC ATT, where lower case indicates the variants.