To exert their functions, siRNAs and miRNAs need to form an effector ribonucleoprotein complex, termed RNA-induced silencing complex (RISC). Because small RNAs only act as guide molecules to direct RISC to target mRNAs in a sequence-dependent manner, they can regulate the expression of their targets only after they form RISC. Thus, RISC assembly is a key process in small RNA-mediated gene silencing, and understanding of the assembly mechanisms is important for the development and improvement of therapeutic and research tools. The core component of RISC is a member of the Argonaute (Ago) subfamily of proteins, of which there are four in mammals (Ago1–4) and two in flies (Ago1 and 2). Each of the four mammalian Ago proteins can induce translational repression and/or deadenylation that leads to mRNA decay, but only Ago2 can endonucleolytically cleave target mRNAs when the sequence of the small RNA is extensively complementary to the target mRNA (Hutvagner and Simard, 2008; Carthew and Sontheimer, 2009; Ghildiyal and Zamore, 2009; Siomi and Siomi, 2009).

Initial understanding of the mechanism of guide strand selection came from studies of fly Ago2, where the Dicer-2/R2D2 heterodimer binds small RNA duplexes asymmetrically. R2D2 binds at the most stable end of the duplex, while Dicer-2 is positioned at the less stable end. In flies, such binding of Dicer-2/R2D2 to siRNA duplexes is a prerequisite for Ago2-RISC assembly (Liu et al., 2003; Pham et al., 2004; Tomari et al., 2004a,b). In mammals, Dicer, with the aid of its binding partners (TRBP or PACT), also binds RNA duplexes asymmetrically (Noland et al., 2011). However, in contrast to fly Ago2-RISC assembly, it has been recently shown that, Dicer is dispensable for the asymmetric assembly of fly Ago1-RISC and mammalian RISC (Kawamata et al., 2009; Betancur and Tomari, 2012; Figure ​Figure1A1A).

Although the precise mechanism of strand selection in mammals remains obscure, the crystal structures of prokaryotic and eukaryotic Ago proteins have hinted that Ago proteins themselves might be able to sense the asymmetry of the duplex. The crystal structures suggest that when an RNA duplex is loaded in Ago and the 5′-phosphate of the guide strand is docked at the phosphate-binding pocket between the MID and PIWI domains, the base pair between the first nucleotide of the guide and the complementary base of the passenger strand needs to be broken. Such structural conformation should favor the incorporation of the RNA duplex with the end that is more easily wedged (i.e., the less stable end) toward the phosphate-binding pocket (Parker et al., 2005; Wang et al., 2008a; Boland et al., 2011; Elkayam et al., 2012; Nakanishi et al., 2012; Schirle and MacRae, 2012;Figure ​Figure1A1A), which is in agreement with the well known asymmetry rule that is widely applied to the design of functional siRNAs (Khvorova et al., 2003; Schwarz et al., 2003). In the case of the fly Ago2-RISC assembly pathway, Dicer-2/R2D2 might double-check the duplex asymmetry prior to loading the duplex into Ago2.

There are additional structural requirements for RISC loading of small RNA duplexes, that we were able to reveal using an agarose native gel system (Kawamata and Tomari, 2011). According to the results, central mismatches (position 8–11 at guide strand) in small RNA duplexes are preferred for RISC loading in human Ago1–4. Duplexes containing mismatches only at non-central regions are disfavored for loading into Ago1–4. At the same time, Ago1–4 can also incorporate perfectly complementary, siRNA duplexes (Yoda et al., 2010; Gu et al., 2011; Figure ​Figure1B1B). How can human Ago proteins accommodate both siRNA duplexes and miRNA/miRNA* duplexes with apparently distinct structural properties? We have recently shown that fly Ago1 inspects the authenticity of miRNAs by utilizing multiple anchoring points (Kawamata et al., 2011). Given that fly Ago1 shares many features with mammalian Ago1–4, they should also employ a similar strategy. We envision that, while central mismatches act as an anchor to Ago, seed or 3′-mid mismatches antagonize Ago anchoring. Because most miRNA/miRNA* duplexes bear mismatches in the seed and/or 3′-mid region as well as in the central region (Kawamata et al., 2009), the destabilization effect by seed/3′-mid mismatches, which is later required for efficient slicer-independent unwinding (see below), is neutralized by the anchoring effect by the central mismatches. On the other hand, siRNA duplexes have neither the anchoring effect by central mismatches nor the debilitation effect by seed/3′-mid mismatches. This model explains how mammalian Ago1–4 selectively load authentic miRNA/miRNA* duplexes and siRNA duplexes. By contrast, in flies, Dicer-2/R2D2 acts as a gatekeeper to sort miRNA/miRNA* duplexes and siRNA duplexes into Ago1 and Ago2, respectively (Forstemann et al., 2007; Tomari et al., 2007).

Based on these recent findings, we can now describe a strategy to rationally design highly efficient and specific, artificial miRNA/miRNA* duplexes from conventional siRNA duplexes. First, a functionally asymmetric siRNA duplex can be designed by following the widely used thermodynamic asymmetry rule (Khvorova et al., 2003; Schwarz et al., 2003; Reynolds et al., 2004). Most conveniently, this can be achieved by introducing a mismatch at the 5′-end of the guide strand into a conventional, fully complementary siRNA, by modifying the facing nucleotide (position 19) on the passenger strand (Figures 2A,B). Because it has been shown that a 5′ U on the guide strand is preferred for both RISC loading and unwinding (Frank et al., 2010; Kawamata et al., 2011), and because that nucleotide is not involved in target RNA recognition (Ma et al., 2005; Parker et al., 2005), it is also reasonable to modify the 5′-end nucleotide of the guide strand to U to introduce a mismatch. Then, mismatches should be introduced in the central region (position 8–11 of the guide; to promote duplex loading) and the seed region (position 2–7 of the guide; to promote unwinding) to create a miRNA/miRNA*-like duplex (Figure ​Figure2C2C), by modifying the passenger strand sequence. Usually, introduction of a single seed mismatch and a single central mismatch is sufficient to promote duplex loading and unwinding (Yoda et al., 2010), but combination of mismatches and wobble base pairs is also possible. For example, we have successfully designed an artificial miRNA/miRNA* duplex by introducing a mismatch at position 11 and wobble base pairs at positions 9 and 10; and a mismatch at position 4 and a wobble base pair at position 3 (Yoda et al., 2010; Betancur and Tomari, 2012). Note that seed mismatches (and potentially wobble base pairs) not only promote unwinding but also enhance the thermodynamic asymmetry of the duplex, reducing the unwanted off-target effect from the passenger strand. Although mismatches in the 3′-mid region also promote unwinding, they may negatively affect the functional asymmetry of the duplex (Khvorova et al., 2003; Schwarz et al., 2003) and therefore seed mismatches are preferred for practical use. The above-described guideline, based on the mechanistic understanding of the RISC assembly pathway, is simple, efficient and cost-effective, and should be widely applicable to general gene silencing purposes.