In silico characterization of miRNA and long non-coding RNA interplay in multiple myeloma (95 MM lncRNA data sets)

The identification of deregulated microRNAs (miRNAs) and long non-coding RNAs (lncRNAs) in multiple myeloma (MM) has progressively added a further level of complexity to MM biology. In addition, the cross-regulation between lncRNAs and miRNAs has begun to emerge, and theoretical and experimental studies have demonstrated the competing endogenous RNAs (ceRNAs) activity of lncRNAs as natural miRNA decoys in pathophysiological conditions, including cancer. Currently, information concerning lncRNA and miRNA interplay in MM is virtually absent. Herein, we investigated in silico the lncRNA and miRNA relationship in a representative datasets encompassing 95 MM and 30 plasma cell leukemia patients at diagnosis and in four normal controls, whose expression profiles were generated by a custom annotation pipeline to detect specific lncRNAs. We applied target prediction analysis based on miRanda and RNA22 algorithms to 235 lncRNAs and 459 miRNAs selected with a potential pivotal role in the pathology of MM. Among pairs that showed significant correlation between lncRNA and miRNA expression levels, we identified 10 lncRNA-miRNA relationships suggestive of novel ceRNA network with relevance in MM. We searched for putative significant correlation between lncRNAs and miRNAs expression level in paired samples, with the support of target prediction analysis. Wilcoxon rank-sum test was applied using standard functions in R base package. The Benjamini-Hochberg method was applied for multiple testing correction. miRNA targets custom predictions were obtained for lncRNA sequences by merging the results of two different algorithms: RNA-22 version 2.0 prediction algorithm and miRanda, which both allow customizing input sequences and parameters. The RNA-22 perl script was run on the combination of the 459 miRNA and 1546 (number of different transcripts corresponding to 235 lncRNAs) lncRNA FASTA sequences selected as relevant among all the miRNAs and lncRNAs respectively detected by the arrays. Default sensitivity/specificity ratio of 1.032 was chosen, provided a minimum seed size of 7 bases with only one mismatch exception, a minimum number of 12 paired-up bases in heteroduplex with -12 Kcal/mol maximum folding energy, and no more than one G-U wobble in the seed region. The miRanda algorithm has been run under default conditions, without any a-priori restrictions on score, energy and trimming parameters.