miRNAs are short, endogenously expressed RNA molecules of approximately 21 nucleotides in length. They do not encode peptides, but instead regulate the expression of proteins encoded by their mRNA targets. With few exceptions, miRNAs share a common biogenesis pathway (7
). Their primary transcripts (‘pri-miRNAs’) are transcribed from the genome in the ordinary way, usually by RNA polymerase II (Pol II). Some miRNAs share a transcript with a protein-coding gene, appearing in an intron or exonic 3′ untranslated region (UTR) of a mRNA, while other pri-miRNAs have dedicated promoters and contain no functional open reading frames. miRNAs often appear in ‘clusters’ in the genome. Though appearance in a cluster does not gaurantee co-regulation of their expression, these miRNAs are often transcribed together in a single pri-miRNA. Like mRNAs, pri-miRNAs may be spliced or left unspliced, and those transcribed by Pol II are polyadenylated at their 3′ end.
After this point, however, mRNAs and pri-miRNAs diverge in their path to maturity and function in the cytoplasm. Stem-loop structures containing the mature miRNA sequence are cropped out of pri-miRNAs by the Microprocessor complex, which consists of the RNAse III enzyme Drosha and its obligate RNA-binding protein partner Dgcr8. This yields ‘pre-miRNAs’ that are exported to the cytoplasm, where they are further processed by the related enzyme Dicer. Dicer removes the loop from pre-miRNAs, yielding a short double-stranded RNA very similar to the exogenous short-interfering RNAs (siRNAs) that mediate RNAi when transfected into mammalian cells. Indeed, one strand of this final miRNA precursor is loaded onto an Argonaute (Ago) protein, the core functional component of the RNA induced silencing complex (RISC). Association with Ago stabilizes the miRNA, whereas the ‘passenger’ strand is short lived. miRNAs derived from the 5′ end of the stem-loop pre-miRNA are termed -5p (e.g. miR-142-5p), and those derived from the 3′ end are termed -3p (e.g. miR-142-3p). Thermodynamic features of the RNA duplex strongly influence strand selection, and most miRNA precursors display strong strand bias that results in only one functionally relevant miRNA. However, both strands of some precursors are loaded into Ago at similar rates, and strand selection can be modulated in a context-dependent fashion in cells of the immune system (12
Ago proteins programmed with a miRNA form the ‘miRISC’, and this complex is responsible for miRNA-directed gene repression (13
). Ago proteins interact with factors that repress translation and accelerate mRNA decay, while the miRNA component provides sequence-specific target mRNA recognition through complementary base pairing (14
). Targeted sequences may reside in any part of the mRNA, but bioinformatic, biochemical, and mechanistic studies indicate that target sequences in the 3′ UTR are the most common and most effective sites of miRISC action. Within the miRISC structure, nucleotides in positions 2–7 from the 5′ end of the miRNA are particularly accessible for base pairing (15
), and this ‘seed’ sequence is a major determinant of targeting. Therefore, miRNA are classified into ‘families’ that share a common seed sequence and presumably a large fraction of their mRNA targets. However, targeting does not require perfect complementarity with the seed sequence, and further pairing in the 3′ half of the miRNA can enhance or even be sufficient alone to direct repression of some mRNAs.
The complex degeneracy of miRNA target sequences and the redundancy provided by large seed families complicate target prediction and validation. Several target prediction algorithms have been devised and in some cases continuously refined using empirical data (16
). These programs are very useful as a starting point for hypothesis generation in functional studies. However, they are generally limited either by their dependence on seed pairing and evolutionary conservation of miRNA and target sequences or by the large number of false positive predictions that result when these constraints are removed. Validation of these predictions is further complicated by the relatively modest repression mediated by miRNAs on each of their direct targets – almost always less than twofold at both the mRNA and protein level (17
). Sensitive luciferase-based reporter assays are the current gold standard to test candidate miRNA targets, but these experiments often fail to preserve the cellular and sequence context in which the biologically relevant interaction is proposed to take place. Biochemical methods that detect Ago protein association with mRNAs have been developed and employed in the immune system (18
). Assays like these will likely become part of the standard for evidence of direct miRNA targeting, as chromatin immunoprecipitation has become necessary to validate direct gene activation or repression by transcription factors.
So how do such small regulators with such modest activity on each target have meaningful biological effects? This question underlies the important challenge of integrating miRNAs into our models of gene expression networks that govern cell identity and behavior. In some cases, a single target has been identified that can account for a significant proportion of the functional effect of a miRNA. ‘Fine tuning’ of gene expression matters, and small changes in the abundance of a limiting factor can trigger threshold effects and feedback loops that drive big changes in cell behavior. However, it is a common misconception that a key dominant target must exist. Like transcription factors, miRNAs function through their combined action on a large number of targets. Evolutionarily conserved miRNA target sequences occur in over half of all mRNAs, and each miRNA interacts with dozens to hundreds of mRNAs (22
). The conservation of miRNA binding sequences in the untranslated regions of so many target mRNAs indicates that the modest effects mediated by many of these interactions are important enough to exert selection pressure in some evolutionarily relevant context. miRNAs often amplify their effects by targeting several genes that participate in a common pathway. This principle can be exploited for pathway discovery when the biological function and bona fide
targets of a miRNA are known (23