Vertebrate microRNA genes.

MicroRNAs (miRNAs) are an abundant class of 22-nucleotide (nt) noncoding RNAs, some of which are known to control the expression of other genes at the posttranscriptional level (1–4). We developed a computational procedure (MiRscan) to identify miRNA genes (5) and apply it here to identify most of the miRNA genes in vertebrates. MiRscan relies on the observation that the known miRNAs derive from phylogenetically conserved stem loop precursor RNAs with characteristic features. MiRscan evaluates conserved stem loops as miRNA precursors by passing a 21-nt window along each conserved stem loop, assigning a log-likelihood score to each window that measures how well its attributes resemble those of the first 50 experimentally verified C. elegans miRNAs with C. briggsae homologs (2, 3, 5). Folding of aligned regions of the human and mouse genomes, with subsequent comparison to the pufferfish Fugu rubripes genome, identified 15,000 human genomic segments that fell out-side of predicted protein coding genes, were predicted to form stem loops, and were at least loosely conserved among the three vertebrate species (6). MiRscan evaluation revealed a high-scoring set of 188 human loci, using a natural cutoff score of 10, defined by a dip in the distribution at this point (Fig. 1). This set included 81 of the 109 members of a reference set of known human miRNA loci, for a sensitivity of 0.74. The fact that a procedure developed and trained solely using nematode miRNAs could also identify most of the vertebrate miRNAs shows that the generic features of the miRNAs and their precursors are conserved broadly among diverse animals, even though the sequences of most miRNAs are not as broadly conserved. Our analysis can be used to calculate an upper bound on the number of human miRNA genes. If all 188 candidates were authentic miRNA genes and these represented 74% of the total miRNA genes, then there are no more than 255 miRNA genes in the genome. Note that this calculation assumes that rare miRNAs—those expressed at low levels or in a limited set of conditions or cell types, which would be underrepresented in our reference set of cloned miRNAs—will have a distribution of scores and degree of conservation similar to the cloned miRNAs. This assumption is supported by our finding that in nematodes, there is no correlation between the number of times an miRNA was cloned and its MiRscan score (5). Furthermore, a tissue such as mouse brain, which might be