In the mid-1990s, microRNAs took the RNA world by storm. These small noncoding RNA molecules can specifically down-regulate the expression of target genes by a natural process called RNA interference. Harnessing RNA interference with synthetic small RNAs has revolutionized the experimental manipulation of gene expression. In parallel, RNA interference as a natural phenomenon continues to surprise researchers as its unsuspected and amazingly wide scope is revealed. To fully comprehend microRNAs, however, we need more functional studies and additional tools.

Unheard of ten years ago, microRNAs have now been found virtually everywhere—in worms, in all vertebrates tested and even in the human virus Epstein-Barr. In this issue (p. 269) Tuschl and colleagues report the analysis of about 30 human viruses and the identification or prediction of new viral microRNAs, suggesting that some viruses could use RNA interference to regulate their own gene expression or even interfere with that of the cells they infect.

This prevalence raises a fundamental question: what are these viral and the more than 200 animal micro-RNAs really doing? A specific genetic function has been assigned to only a handful of them. The majority of animal microRNAs were found by a combination of computational predictions and cDNA cloning, not based on a phenotype. Finding their targets in messenger RNAs is the first step to understanding microRNA function, and for this, investigators have naturally turned to computational predictions.

Over the past few months, new predictions have emerged and the numbers are stunning. The boldest published estimates talk of as many as 12,000 targets in flies (Brennecke et al., PLoS Biol. 3, e85; 2005) and about 5,300 target genes in humans (Lewis et al., Cell 120, 15–20; 2005). These predictions do not come as a total surprise to those working in the field. Although all are cautious about going on the record with precise numbers, in whispers they acknowledge that even these staggering numbers may be underestimates.

The main challenge with these computational predictions, however, is that recognition of a messenger RNA target by a microRNA occurs in the context of the protein complex responsible for inhibiting gene expression. Structure-function studies indicate that target recognition does not necessarily correlate with the overall predicted thermodynamic stability of a microRNA-target pair. MicroRNAs are 21–22 nucleotides long but can recognize a target with as few as eight perfectly complementary nucleotides, and both a short core stretch of complementary bases and individual residues outside that stretch may be important for recognition. Several algorithms have been developed to tackle this challenge, but they all rely on observations made on the very few microRNA-target pairs that have been experimentally validated. Thus there is an urgent need for more validation studies to gain insight into the rules of microRNA-target recognition, which in turn will allow refinement of prediction algorithms.

Why are these validation studies lacking? An important reason is that the techniques are cumbersome. Typically, the candidate target is fused to a reporter gene and coexpressed, in cells or transgenic animals, with an excess of the microRNA. Moreover, these experiments do not necessarily reflect the reality. Controlling for specificity, for example, is not always possible. Other factors, such as accessibility of the target and endogenous levels of microRNA and target, must be taken into account. With these drawbacks in mind, one understands that biologists are not eager to start validation experiments when they receive a list of 100 computationally predicted targets, even with the assurance that “all of them have a very good statistical score”.

The field needs a streamlined, practical, high-throughput methodology for performing target validation—ideally, one that would make biological sense by factoring in biochemical information, such as the structure of the RNA interference complex, as well as the likelihood that microRNA and target coexist in a relevant cellular situation. Microarray data on gene and microRNA expression may help in this regard. Such screens would not obviate the need for in vivo validation but would reduce the scope of required testing. Methods development, it seems, is a limiting step in our understanding of this newly discovered biological phenomenon.

The new predictions, alluding to 100 targets per microRNA, suggest a picture of combinatorial, microRNA-based regulatory effects as complex as those of transcription factors. We have no idea how these regulatory networks could be involved in diseases, but the discovery of microRNAs in human viruses already points to opportunities for therapeutic intervention. To understand the regulation and find these opportunities, however, it will be necessary to develop new methods to match microRNAs and their targets.