MicroRNA binding sites in C. elegans 3' UTRs

Abstract

MicroRNAs (miRNAs) are post-transcriptional regulators of gene expression. Since the discovery of lin-4, the founding member of the miRNA family, over 360 miRNAs have been identified for Caenorhabditis elegans (C. elegans). Prediction and validation of targets are essential for elucidation of regulatory functions of these miRNAs. For C. elegans, crosslinking immunoprecipitation (CLIP) has been successfully performed for the identification of target mRNA sequences bound by Argonaute protein ALG-1. In addition, reliable annotation of the 3' untranslated regions (3' UTRs) as well as developmental stage-specific expression profiles for both miRNAs and 3' UTR isoforms are available. By utilizing these data, we developed statistical models and bioinformatics tools for both transcriptome-scale and developmental stage-specific predictions of miRNA binding sites in C. elegans 3' UTRs. In performance evaluation via cross validation on the ALG-1 CLIP data, the models were found to offer major improvements over established algorithms for predicting both seed sites and seedless sites. In particular, our top-ranked predictions have a substantially higher true positive rate, suggesting a much higher likelihood of positive experimental validation. A gene ontology analysis of stage-specific predictions suggests that miRNAs are involved in dynamic regulation of biological functions during C. elegans development. In particular, miRNAs preferentially target genes related to development, cell cycle, trafficking, and cell signaling processes. A database for both transcriptome-scale and stage-specific predictions and software for implementing the prediction models are available through the Sfold web server at http://sfold.wadsworth.org.

Keywords: GO analysis; developmental stage; microRNA; prediction; target binding site.

Figure 1. Enrichment of representative site features: (A) site accessibility for seed sites; (B) upstream accessibility (window size of 10 nt) for seed sites; (C) type of miRNA target seed sites; (D) percentage (Y-axis) of sites/seed/off-seed regions with conservation scores greater than or equal to a pre-specified threshold (X-axis), in the IP+ seed set or the IP- seed set (dashed line corresponding to a threshold of 0.57 previously used for defining conservation^25,41); (E) site accessibility for seedless sites; (F) percentage (Y-axis) of seedless sites with conservation scores greater than or equal to a pre-specified threshold (X-axis), in the IP+ set or the IP- set.

Figure 2. Performance comparison of logistic models with three established algorithms for site predictions in 3′ UTRs (dashed diagonal line for random predictions). ROC curve and Youden’s J statistic are shown for the predictions of seed sites (A and B), seedless sites (C and D), and seedless site with one G•U pair or one mismatch within seed complementary region (E and F). The color-matched dots on ROC curves correspond to a logistic probability threshold of 0.5. The rectangle, triangle and square correspond to the best-performing score threshold (according to Youden’s J statistic) for TargetScan, PITA and miRanda, respectively.

Figure 3. True positive rate comparison for top-ranked (from top 1% to top 50%) predictions of seed sites (A), seedless sites (B), seedless sites with one G•U pair or one mismatch in the seed complementary region (C) in 3′ UTRs.

Figure 4. (A) STarMirDB search result for binding sites of cel-miR-796 on ncs-2 3′ UTR isoforms, with “3′ UTR-seedless” option selected for output display; (B) hybrid diagram of a seed site; (C) hybrid diagram of a seedless site.