Predicting and Validating microRNA Targets at GSK

By Laurie Sullivan

Accurate prediction of miRNA targets is central to the understanding of miRNA biology. Yet predicting and validating those targets has its challenges, leaving many questions as-yet unanswered. “We must accept a great deal of uncertainty as we move forward,” acknowledged Michael R. Barnes, head of molecular discovery and pharmacogenetic bioinformatics at GlaxoSmithKline. Speaking last week at CHI’s microRNA conference in Boston, Barnes cited computational goals of target prediction as a relatively unobtainable wish-list. 

Case in point: A comparison of three of the many in silico target prediction methods available found 53,233 hits in common from a beginning data set of 146,292 targets (miRanda), 655,409 (RNAhybrid), and 502,369 (TargetScan). miRanda-RNAhybrid shared the most targets (226,579 hits) and miRanda-TargetScan shared the least targets (64,166 hits). Such small overlap makes it hard to ascertain which method works best. 

Target prediction is further complicated since the region of binding is debatable. Generally target binding is believed to take place in the 3' UTR; however, this view may be somewhat dogmatic as there have still been no systematic surveys of miRNA binding in other areas of the transcript, such as the open reading frame. Problems with target prediction are further compounded by the fact that only 706 human miRNA targets have been laboratory validated to date, and these validated binding sites do not closely follow the 5' seed complementarity expected from canonical target binding. “We really don’t know yet how complementary a target has to be,” said Barnes. “The gold standard for miRNA target validation is conservation.” 

Even so, the value of target conservation is context dependent. For one thing, many miRNA targets rapidly evolve and therefore are not conserved. Adding another layer of complexity, species- or order-specific miRNA may be critical in disease. For example, human/primate specific miRNA may be linked to brain development and psychiatric disorders. 

Evaluating Impact of Variation on miRNA 

Mutations in the miRNA genes and mRNA target sites have very different implications. Mutations in genes have broad implications, but miRNA redundancy may influence the impact of variation in miRNA genes. While one miRNA usually targets many genes, negative selection should prevent mutation from occurring frequently. Of the 147 dbSNP SNPs known to be located in miRNA genes, only 12 of them were validated. The extensive redundancy of miRNA may also ameliorate the impact of variation. 

“How promiscuous are our target prediction methods?” Barnes asked, describing an analysis of all exonic SNPs for impact on target sites for 452 miRNAs. The study found that an astonishing 65% of SNPs analyzed potentially impact 1.89 million predicted miRNA target sites. Stringent filtering was subsequently applied to the results. The new findings showed that only about 21% of SNPs potentially impacted miRNA binding. “It’s possible that most SNPs at miRNA binding sites have only a modest biological effect on mRNA translation,” Barnes concluded. 

This information was used to re-evaluate published UTR disease associations in a miRNA context. After curating literary associations, 40 robust SNP associations in the UTR binding regions of 33 genes were evaluated. Biological rationale of the gene in the disease was evaluated to predict creation or disruption of miRNA target sites. Largely in line with the earlier observations, it was discovered that 12 of these disease-associated SNPs impacted miRNA binding.

 miRNA… and Systems Biology? 

Can miRNA help improve our understanding of the biology of genes and the organization of pathways? “It’s a bit more speculative,” Barnes concedes. But given accurate prediction, it may be possible to infer gene function from the miRNA that target it and vice versa. 

For example, might miRNA play a role in GABAA receptor subunit composition? GABA receptors are composed of a combination of five out of 18 different subunits. Studies have demonstrated that the subunit composition determines affinity for GABA, trafficking, and the specific effect of small-molecule modulators. 

Barnes purports GABRA3 is a particularly interesting subunit: Examining co-regulated miRNA in introns of coding genes (e.g., hsa-miR-105 in GABRA3), only seven targets -- including GABRB3 -- are predicted for hsa-miR-105. Theoretically, if hsa-miR-105 is expressed from GABRA3, then the GABRA3 subunit will be favored and the GABRB3 subunit suppressed. 

Barnes left his audience with this proposal: Could it be that new miRNAs are an evolutionary proving ground, with deleterious targets quickly selected out? As new miRNAs arise, perhaps those targets that are neutral or advantageous are proving themselves by natural selection, resulting in markedly higher expression vis-à-vis their harmful counterparts.

4-04- 07 - Biomarker Breakthroughs