Bioinformatic discovery of microRNA precursors from human ESTs and introns

Abstract

Background: MicroRNAs (miRNAs) function in many physiological processes, and their discovery is beneficial for further studying their physiological functions. However, many of the miRNAs predicted from genomic sequences have not been experimentally validated to be authentic expressed RNA transcripts, thereby decreasing the reliability of miRNA discovery. To overcome this problem, we examined expressed transcripts - ESTs and intronic sequences - to identify novel miRNAs as well as their target genes.

Results: To facilitate our approach, we developed our scanning method using criteria based on the features of 207 known human pre-miRNAs to discriminate miRNAs from random sequences. We identified 208 candidate hairpins in human ESTs and human reference gene intronic sequences, 52 of which are known pre-miRNAs. The discovery pipeline performance was further assessed using 130 newly updated pre-miRNA and randomly selected sequences. We achieved sensitivity of 85% (110/130) and overall specificity of 49.7% using this method. Because miRNAs are evolutionarily conserved regulators of gene expression, it is expected that their host genes and target genes should have respective phylogenetic orthologs. Our results confirmed that, in certain mammals, the host genes carrying the same miRNAs are orthologs, as previously reported. Moreover, this observation is also the case for some of the miRNA target genes.

Conclusion: We have predicted 208 human pre-miRNA candidates and over 10,000 putative human target genes. Using sequence information from ESTs and introns ensures that the predicted pre-miRNA candidates are expressed and the combined expression transcription information from ESTs and introns makes our prediction results more decisive with regard to expressed pre-miRNAs.

Figure 1

Illustration of how to infer putative miRNAs. After applying Srnaloop, we noted the positions of the first nucleotide (Pfn) and the terminal nucleotide (Ptn) of the terminal loop. We elongated each putative miRNA by two nucleotides at each end. By doing so, we acquired 26-nt putative miRNAs, each of which is located between (Pfn-24, Pfn+1) or (Ptn+1, Ptn+26) within each candidate hairpin.

Figure 2

Evaluation of inferring putative miRNAs and sequence comparison among mir-192 orthologs from distinct species. (a) We inferred putative miRNAs from 195 known pre-miRNAs (detected by Srnaloop). We then compared the sequences of the known mature miRNAs with those of the putative miRNAs. The results show that 93.8% of the known miRNAs were almost entirely included in the putative miRNAs inferred from their corresponding precursors. This high level of coverage enabled us to use the putative miRNA sequences for the conservation examination. (b) Mir-192 distributes in human (hsa), mouse (mmu) and rat (rno). Using ClustalW [30], we compared sequences of mir-192 orthologs. The alignment shows that most of the mismatches occur in the terminal loop, the opposite arm and the external portion of the hairpins. Besides the mature functional sequences, the entire pre-miRNA sequences are also highly conserved.

Figure 3

Illustration of TDL_miRBase dataset report interface. (a) The complete information report for a candidate hairpin includes host gene, host gene NM accession number (for intronic candidates), genomic location, expression level and match to known miRNAs. The score and minimum free energy (mfe) are the output results from Srnaloop and RNAfold, respectively. (b) Target gene information for a candidate from hairpin Ih788. Target genes were discovered by RNAhybrid and pre-defined conserved motif seeds as described in the text. Optimal free energy and RNA duplex mfe are the output values of RNAhybrid. The GO information of the Ih788 host gene and one of its target genes are displayed in (c). (d) Orthologous target genes report. Some of the target genes were found to be orthologous pairs according to Ensembl gene information. They are displayed as human-mouse or human-rat pairs.