Of all the RNA silencing pathways, RNAi has been the most thoroughly studied, using a combination of biochemical, genetic, and bioinformatic approaches. Increased understanding of the mechanism of RNAi has led directly to improvements in scientists’ ability to harness this pathway for reverse genetics in eukaryotic systems. In particular, design algorithms for siRNAs that are used to artificially induce the RNAi pathway in mammalian systems have been significantly improved through research on the mechanism of RNAi.

The mechanism of RNAi can be divided conceptually into a processing step where a dsRNA “trigger” is processed into siRNA and assembled with other components to form mature RISC, and an effector/silencing step in which mature RISC cleaves the target mRNA. A brief overview of what is known about dsRNA triggers and the mechanism of RNAi follows.

Double-stranded RNA Triggers of RNAi
The RNAi pathway can be artificially induced in cells of interest by delivery of a dsRNA trigger. In non-mammalian systems, long dsRNA (typically >200 bp) homologous to the target gene effectively triggers RNAi. In mammalian cultured cells, short synthetic siRNAs are typically used, because introduction of dsRNA longer than ~30 bp induces an antiviral interferon response. Alternatively, DNA constructs that express short hairpin RNAs (shRNAs) can also be used to induce RNAi. Careful design of the siRNA or shRNA is required to maximize silencing of the target gene while minimizing off-target effects.

Processing of dsRNA Triggers and Assembly with RISC

•ATPase/RNA helicase domain
•conserved PAZ domain that is shared with Argonaute (another component of silencing machinery)
•two catalytic RNase III domains
•C-terminal dsRNA binding domain (dsRBD)

The PAZ domain of Dicer is thought to bind the end of the dsRNA and may specifically recognize 3' overhangs. Dicer processively cleaves dsRNA at 21–25 bp intervals beginning at one end of the dsRNA. The resulting siRNAs have dinucleotide 3' overhangs and phosphorylated 5' ends.

When shRNAs are used as triggers, the length of the stem of the hairpin appears to affect how the shRNA enters the RNAi pathway. In vivo or in vitro, Dicer processes shRNAs with 29 bp stems to 21–22 nt siRNAs, but shRNAs with 19 bp stems are not effectively cleaved, although these shRNAs enter RISC and induce RNAi at rates comparable to siRNAs. It appears that exogenously supplied siRNAs and shRNAs with short stems enter the RNAi pathway at a point after initial Dicer processing [43].

RISC Assembly
Once a dsRNA trigger has been processed into siRNA, or siRNA has been introduced directly into the cytoplasm, the guide strand of the siRNA is assembled with RISC components to form mature RISC. The main features of this process have now been broadly outlined, based on in vitro studies of RISC assembly using Drosophila extracts and siRNAs as starting substrate (Figure 2). In this system, the Dicer, Dcr-2, works with a small dsRNA-binding partner, R2D2, to direct the siRNA to RISC components. See Orthologs of RNAi Machinery (below) for some of the orthologous genes that code components of the RNAi machinery in several animal systems and in Arabidopsis.

Figure 2. Assembly and mRNA Cleavage Activity. This is a general summary of current thinking about assembly of siRNA with RISC and mRNA cleavage activity, based primarily on in vitro studies of RISC assembly using Drosophila extracts and using the component names from that system. Only core components of the RISC Loading complex (RLC) and RISC are depicted. RLC contains a Dcr-2/R2D2 heterodimer which binds the siRNA containing dinucleotide 3’ overhangs. Core RISC component Ago2 displaces Dcr-2/R2D2. This schematic depicts recent findings that indicate transfer of duplex siRNA to Ago2 with either concurrent or immediate Ago2-mediated cleavage of the passenger strand [44–46]. ATP hydrolysis is required for RISC maturation and has been postulated to accelerate release of the cleaved passenger strands as it does for cleaved mRNA. Mature RISC includes only the guide strand of the siRNA, and can cleave multiple mRNA targets. Details of the models are described in the text and the literature [2, 7, 44–46].

In current models, Dcr-2 and R2D2 form a heterodimer that binds siRNA, with R2D2 binding the thermodynamically more stable end of the siRNA and Dcr-2 binding the less stable end. These thermodynamic and binding asymmetries appear to determine which strand is loaded into RISC: the strand whose 5' terminus is at the thermodynamically less stable end of the duplex siRNA is the final guide strand. The presence of a phosphate on the 5' end of an siRNA strengthens R2D2 binding, providing a mechanism for R2D2 to authenticate siRNAs as the products of Dcr-2. The Dcr-2/R2D2 heterodimer participates in a RISC loading complex (RLC), an intermediate where Dcr-2/R2D2 is gradually displaced by the Argonaute protein Ago2, the core RISC component. Initial models postulated that the guide strand is unwound from the passenger strand either in the RLC or concurrent with transfer to Ago2. Recent studies indicate that the siRNA is transferred from Dcr-2/R2D2 to Ago2 as a duplex, and that Ago2 cleaves the passenger strand in a manner analogous to cleavage of target mRNAs [44–46]. After discharge of the passenger strand, possibly in an ATP-dependent step, the guide strand remains in contact with Ago2 in fully mature RISC.

Argonaute proteins are characterized by two conserved domains: PAZ, also found in Dicer, and PIWI. It is thought that the PIWI domain of Argonaute binds the 5' phosphorylated end of the guide strand of the siRNA, while the PAZ domain contacts the 3' end. This highly conserved family of proteins can be subdivided into two subclasses, based on a higher degree of homology to either Arabidopsis AGO1 or to Drosophila Piwi. Structural studies of prokaryotic Ago-like proteins complexed with siRNA mimics indicate that the sugar phosphate backbone of nucleotides 2–5 of the guide strand contact the PIWI domain, presenting the corresponding bases on the surface for base pairing with the target mRNA. This is consistent with bioinformatics analysis of effective siRNAs and miRNAs, which has defined a “seed” region 2–8 nucleotides from the 5' end of the guide strand. This seed region is thought to be a nucleation region for pairing with target mRNAs [2, 5].

Presumably, the thermodynamic stability profile of the mature siRNA, which determines the guide strand, would be unrecognized in long dsRNA precursors during processing by Dicer before RISC assembly. It is unknown whether the siRNA that is generated by Dicer excision is released and must rebind Dicer for RISC assembly or if it remains bound. Mechanistic models must account for the potential conflict that could arise between the relative stabilities of the siRNA ends and the direction of Dicer processing with regard to strand selection for RISC assembly.

The Effector/Silencing Step
Mature RISC, fully assembled with the guide strand of the siRNA, is sometimes called siRISC, and the analogous complex assembled with miRNA, miRISC. Early characterization of the siRNA- and miRNA-mediated silencing pathways indicated that siRISCs direct mRNA cleavage, while miRISCs direct mRNA cleavage in plants but repression of translation in animals. It now appears that a more general principle applies: perfect complementarity of an siRNA or miRNA with its target results in mRNA cleavage, whereas partial base pairing results in translational repression.

In the RNAi pathway, siRISC operates as a multiple turnover enzyme: once the mRNA target is cleaved, siRISC dissociates from the cleaved mRNA to repeat the cleavage cycle with other targets. ATP is not required for target cleavage, but does increase the rate of enzyme turnover. The Argonaute component of RISC has been shown to be the core catalytic site for mRNA cleavage [47]. Other proteins participate in RISC but their roles remain unclear (see Sontheimer, 2005 [7], Tomari and Zamore, 2005 [5], or Filipowicz 2005 [2] for more details). A recent study of human miRISC (47a) indicates that components orthologous to Dicer/R2D2 remain associated with the Argonaute component in miRISC. In this study, Dicer processing of pre-miRNAs increased target mRNA cleavage by RISC; ATP and other nucleotides stimulated RISC activity but hydrolysis was not required.

The details of the catalytic mechanism remain to be defined, but all models share the following characteristics. The PAZ domain of Argonaute contacts the 3' end of the guide strand. Argonaute’s PIWI domain, which is structurally similar to RNase H and is a strong candidate for the catalytic site, contacts the 5' end of the guide strand [2, 5, 6]. A single cleavage of the target mRNA occurs across from nucleotides 10 and 11 (with respect to the 5' end) of the guide strand [6]. It has been proposed [2] that perfect base pairing in the central part of the siRNA-mRNA duplex allows formation of an A-form helix across from a conserved triad, Asp-Asp-His, in the PIWI domain of Argonaute. This motif is thought to be catalytically important for mRNA cleavage. In this scenario, mismatches or other bulges which interfere with A-helix formation would prevent mRNA cleavage, instead causing translational repression.

Release of the two resulting fragments of mRNA from RISC requires ATP. In vivo, the 3' fragment of the cleaved mRNA is degraded in the cytoplasm by exonuclease Xrn1, while the 5’ fragment is degraded by a complex of exonucleases that directs 3'–5' degradation of RNA, the exosome [6].

The mechanism(s) of translational repression by imperfectly paired miRNA- or siRNA-guided silencing complexes remains ill-defined. Multiple target sites in the 3' untranslated region (UTR) of the message are usually required to see measurable translational repression. Recent studies have shown that small RNAs, when bound to mammalian Argonaute Ago2, can sequester target mRNAs from the cytosol to “P-bodies,” which are thought to be sites of mRNA destruction. This would explain the findings that some animal miRNAs reduce the stability of their target mRNAs, even when only partial complementarity exists and the mRNA is not cleaved [6]. Whether translational repression is a consequence of relocation to P-bodies, or vice versa, is a fundamental question. Pillai et al. (2005) recently presented evidence that miRNA let7-triggered translational repression occurs at initiation of translation, and suggest that localization of repressed transcripts to P-bodies in mammalian cells is a consequence of translational repression [48].