miRTRAP, a computational method for the systematic identification of miRNAs from high throughput sequencing data

Abstract

MicroRNAs (miRs) have been broadly implicated in animal development and disease. We developed a novel computational strategy for the systematic, whole-genome identification of miRs from high throughput sequencing information. This method, miRTRAP, incorporates the mechanisms of miR biogenesis and includes additional criteria regarding the prevalence and quality of small RNAs arising from the antisense strand and neighboring loci. This program was applied to the simple chordate Ciona intestinalis and identified nearly 400 putative miR loci.

Figure 1

Outline of the miRTRAP program, Ciona abundance versus conservation, neighbor window. (a) Schematic illustration of the miRTRAP program. The algorithm first identifies read regions that do not overlap repeats or tRNAs. The genomic region up to 150 nucleotides around the individual read is folded using RNAfold. Then, all read products within the hairpin window are identified as 5p-miR/3p-miR, 5p-moR/3p-moR or loop based on their positions relative to the hairpin and loop. Each read region is then evaluated by a set of filters to remove those incompatible with the biochemical rules of miR biogenesis. All the rejected read regions are used to filter the initial set of candidate loci to produce a list of positive predictions. (b) Average antisense product displacement (AAPD) score distribution from the Ciona dataset shows that the majority of known miRs have an AAPD score of zero, while non-miR loci have a broad distribution and peaks at 8 and 10. (c) The difference between the non-miR neighbor counts within windows centered at known miRs and non-miR loci in Ciona. Whereas non-miR neighbor counts centered around non-miR loci increases sharply as window sizes expand, all known miR loci have non-miR neighbor counts equal or fewer than 10.

Figure 2

Comparison of miRTRAP with miRDeep. (a) miRTRAP out-performed miRDeep for the Ciona library data set, identifying approximately five times more miRs. In addition, it identified 11 mirtron/half-mirtrons, while miRDeep found only 1. (b) For the Drosophila small RNA data set, miRTRAP identified 25 more known fly miRs than miRDeep. In particular, miRTRAP found 12 out of 14 mirtrons, while miRDeep identified only 3. (c) Example of a novel Drosophila mirtron predicted by miRTRAP. (d) A novel Drosophila miR/miR* containing locus predicted by miRTRAP.

Figure 3

Phylogeny of Ciona miRNA families in the deuterostome lineage. Newly identified conserved Ciona miR families (shaded circles) and previously known Ciona miRs (dark circles) are grouped with homologous miR families from representative deuterostome species (echinoderm, Strongylocentrotus purpuratus; hemichordate, Saccoglossus kowalevskii; Amphioxus, Branchiostoma floridae). Missing miRs are shown as empty circles. It is evident from the phylogenetic tree that the miR repertoire from Ciona is closely related to the vertebrate miRs.

Figure 4

Prevalence of antisense miRs in Ciona. (a) In the scaffold 20 11-miR cluster, three miR loci have antisense reads that exactly match the sense miR/miR* products. (b) Secondary structures of Ci-mir-2217-1 and its associated antisense locus, miR-2217-1-as, both form highly symmetric hairpins, on which the miR and miR* products are indicated as lines. (c) In one case, we observed the antisense locus of Ci-mir-2246 produces not only miR/miR*, but also a 5p-moR product. (d) Secondary structures and product distribution of Ci-mir-2246 and Ci-mir-2246-as.

Figure 5

Non-canonical miR examples from Ciona. (a) A classic example of mirtron Ci-mir-2219-2 shows the miR and miR* products are produced from the precisely spliced intron from gene ci100134440. (b) In some cases, only one of the miR/miR* products abuts the splice junction, while the other product is fully inside the intron. A so-called half-mirtron example, Ci-mir-2227, is represented. (c) Ci-mir-2233 produces a miR/miR* pair from a perfectly structured hairpin, which overlaps with a protein coding exon in the gene ci0100152310.