|Home | About | Journals | Submit | Contact Us | Français|
In Drosophila, small interfering RNAs (siRNAs), which direct RNA interference through the Argonaute protein Ago2, are produced by a biogenesis pathway distinct from microRNAs (miRNAs), which regulate endogenous mRNA expression as guides for Ago1. Here, we report that siRNAs and miRNAs are sorted into Ago1 and Ago2 by pathways independent from the processes that produce these two classes of small RNAs. Such small RNA sorting reflects the structure of the double-stranded assembly intermediates—the miRNA/miRNA* and siRNA duplexes—from which Argonaute proteins are loaded. We find that the Dcr-2/R2D2 heterodimer acts as a gatekeeper for the assembly of Ago2 complexes, promoting the incorporation of siRNAs and disfavoring miRNAs as loading substrates for Drosophila Ago2. A separate mechanism acts in parallel to favor miRNA/miRNA* duplexes and exclude siRNAs from assembly into Ago1 complexes. Thus, in flies small RNA duplexes are actively sorted into Argonaute-containing complexes according to their intrinsic structures.
Small interfering RNAs (siRNAs) and microRNAs (miRNAs) play an unexpectedly large role in regulating plant and animal gene expression (Kloosterman and Plasterk, 2006). Twenty-one to 23 nucleotides long, these two classes of small silencing RNAs repress the expression of specific genes through mechanistically similar RNA silencing pathways (Baulcombe, 2004; Du and Zamore, 2005; Kim, 2005; Sontheimer, 2005; Tomari and Zamore, 2005). siRNAs are produced by the endonucleolytic cleavage of long, double-stranded RNA (dsRNA) by members of the Dicer family of dsRNA-specific endonucleases (Bernstein et al., 2001). When extensively complementary to their mRNA targets, siRNAs direct cleavage of the phosphodiester bond between the target nucleotides paired to siRNA bases 10 and 11 (Elbashir et al., 2001a; Elbashir et al., 2001b). All known plant miRNAs and at least eight mammalian miRNAs similarly guide cleavage of the mRNAs they regulate (reviewed in Du and Zamore, 2005). In contrast, most animal miRNAs lack sufficient complementarity to guide endonucleolytic cleavage of their regulatory targets. Instead, they promote sequence-specific repression of mRNA translation or accelerate mRNA decay, perhaps by recruiting components of more general mRNA turnover pathways (Valencia-Sanchez et al., 2006).
miRNAs reside in discrete genes and are produced by the sequential processing of long transcripts—pri-miRNAs—by the RNase III enzyme Drosha into pre-miRNAs and of pre-miRNAs by Dicer into miRNA-containing RNA duplexes (Cullen, 2004; Kim, 2005). More than 4,000 miRNAs have been reported (Griffiths-Jones et al., 2006), many of which are evolutionally conserved, whereas others are restricted to primates or even to humans (Bentwich et al., 2005; Berezikov et al., 2005; Berezikov et al., 2006). miRNA are proposed to regulate diverse cellular functions, including developmental timing, cell proliferation, cell death, and fat metabolism. They may also act to make biological regulatory circuits more robust (Stark et al., 2005). miRNA-regulated genes typically contain in their 3′ untranslated regions (UTRs) several partially complementary binding sites for one or more miRNAs (Lewis et al., 2003; Krek et al., 2005; Lewis et al., 2005).
Members of the Argonaute (Ago) family of small RNA-binding proteins lie at the core of all known RNA silencing effector complexes, collectively called RNA-induced silencing complexes (RISCs). RISC variants are distinguished by their Argonaute protein. In Drosophila, miRNAs partition between Ago1- and Ago2-RISC (Förstemann et al., accompanying manuscript), whereas siRNAs associate almost exclusively with Ago2-RISC (Hammond et al., 2001; Okamura et al., 2004). Ago1 and Ago2-RISC are functionally distinct, silencing different types of target RNAs by different mechanisms (Förstemann et al., accompanying manuscript).
Both siRNAs and miRNAs are proposed to be loaded into Argonaute protein-containing RISCs from double-stranded intermediates generated by Dicer: siRNA duplexes and miRNA/miRNA* duplexes (Figure 1A). In flies, loading of double-stranded siRNAs into Ago2-RISC is facilitated by the RISC-loading complex (RLC) (Liu et al., 2003; Pham et al., 2004; Tomari et al., 2004a; Tomari et al., 2004b; Pham and Sontheimer, 2005; Kim et al., 2006; Liu et al., 2006). The RLC comprises several proteins, including Dicer-2 and its dsRNA-binding partner protein, R2D2. Which strand of the siRNA duplex is assembled into Ago2-RISC is thought to be determined by the orientation of the Dicer-2/R2D2 heterodimer on the siRNA duplex (Tomari et al., 2004a). The strand loaded, the guide strand, typically has a 5′ end less tightly base paired in the duplex than the passenger strand, which is destroyed during the loading process (Khvorova et al., 2003; Schwarz et al., 2003). Passenger-strand destruction and RISC maturation are initiated for Ago2-RISC assembly by guide-strand directed endonucleolytic cleavage of the passenger strand by Ago2, as if the passenger strand were an mRNA target (Matranga et al., 2005; Rand et al., 2005; Kim et al., 2006; Leuschner et al., 2006). One strand—the miRNA strand—of a miRNA/miRNA* duplex is similarly selectively loaded into Ago1-containing RISC but the proteins facilitating Ago1 loading remain to be identified (Okamura et al., 2004). Both siRNA and miRNA/miRNA* duplexes contain a ~19 base pair double-stranded core flanked by ~2 nt single-stranded 3′ overhanging ends (Figure 1A). However, the guide and passenger strands of an siRNA duplex are complementary throughout its ~19 bp central domain, whereas the miRNA and miRNA* strands invariably contain G:U wobble pairs, mismatches, and internal loops in this region.
In flies, distinct Dicer complexes produce siRNAs and miRNAs (Lee et al., 2004). miRNAs are cleaved from pre-miRNA by Dicer-1 (Dcr-1), acting with its dsRNA-binding protein partner, Loquacious (Loqs) (Forstemann et al., 2005; Jing et al., 2005; Saito et al., 2005). siRNAs are produced from long dsRNA by Dicer-2 (Dcr-2), which partners with the dsRNA-binding protein R2D2 (Liu et al., 2003). Thus, the different origins of miRNAs and siRNAs might direct them to distinct Argonaute proteins, with Dcr-1/Loqs recruiting Ago1 to miRNAs and Dcr-2/R2D2 directing siRNAs to Ago2. Alternatively, the specific structural differences between a miRNA/miRNA* duplex and an siRNA duplex (Figure 1A) might promote their sorting into Ago1- and Ago2-containing RISC, respectively. Here, we report that the Dcr-2/R2D2 heterodimer acts as a gatekeeper for the assembly of Ago2-RISC, promoting the incorporation of siRNAs and disfavoring the use of miRNAs as loading substrates for Drosophila Ago2. An independent mechanism acts in parallel to favor assembly of miRNA/miRNA* duplexes into Ago1-RISC and to exclude siRNAs from incorporation into Ago1. These two pathways compete for loading small RNA duplexes with structures intermediate between that of an siRNA and a typical miRNA/miRNA* duplex, and such small RNAs partition between Ago1 and Ago2. Thus, small RNA duplexes are actively sorted into Argonaute-containing complexes according to their intrinsic structures, rather than as a consequence of their distinct biogenesis pathways.
The structure of a small RNA duplex could determine into which Argonaute paralog it is loaded. To test this hypothesis, we synthesized ten small RNA duplexes: an authentic let-7/let-7* duplex, a functionally asymmetric let-7 siRNA, in which the guide and passenger strands were fully paired except at guide position 1 (mm1 siRNA duplex); and eight let-7 siRNA duplex derivatives incorporating one additional mismatch between the guide and passenger strands, at guide position 3, 5, 7, 9, 11, 13, 15, or 17 (mm3–mm17 siRNAs) (Figure S1). Each small RNA duplex, which contained a 5′ 32P-radiolabel on the let-7 (guide) strand and a 5′, non-radioactive phosphate on the miRNA* or passenger strand, was incubated in Drosophila embryo lysate, then photocrosslinked with 254 nm UV light and analyzed by SDS-PAGE to identify small RNA-bound proteins. The identity of crosslinked proteins was assigned by their immunoprecipitation with specific antibodies and their loss in lysate prepared from mutant ovaries or embryos.
The authentic let-7/let-7* duplex crosslinked only to Ago1, whereas the let-7 mm1 siRNA duplex crosslinked predominantly to Ago2 (Figure 1B). Introducing a position 9 mismatch into the siRNA (mm9) shifted the balance in favor of Ago1, while retaining significant Ago2 association. Quantitative analysis of the ratio of Ago1 to Ago2 crosslinking for the entire series of mismatched let-7 siRNA derivatives revealed that central mismatches direct small RNA duplexes into Ago1 rather than Ago2 (Figure 1C).
How does a central mismatch influence Argonaute loading? Such a disruption to siRNA structure might disfavor its association with the RLC, reducing Ago2 loading, favor its association with the Ago1-loading machinery, or both. To test the idea that central mismatches reduce the association of a small RNA duplex with the RLC, we incubated a let-7 siRNA bearing a mismatch at position 11 (mm11) (Figure 2A) with embryo lysate supplemented with purified recombinant Dcr-2/R2D2 heterodimer or Dcr-2 alone. In the absence of supplemental recombinant protein, the mm11 duplex partitioned between Ago1 (~60%) and Ago2 (~40%). Increasing the concentration of the Dcr-2/R2D2 heterodimer, the core constituent of the RLC, increased the amount of duplex crosslinked to Ago2 (Figure 2B). In contrast, increasing the concentration of Dcr-2 alone did not enhance crosslinking of the duplex to Ago2, consistent with earlier observations that R2D2 is required to recruit Dcr-2 to siRNA for RISC loading (Liu et al., 2003; Liu et al., 2006). Moreover, in the absence of R2D2, Dcr-2 reduced Ago2 crosslinking to siRNA (Figure 2B), suggesting that Dcr-2 forms a complex with siRNA that cannot load Ago2 (see below and Figure S2). Together, the data in Figures Figures1C1C and and2B2B suggest that a central mismatch weakens the binding of the Dcr-2/R2D2 heterodimer to a small RNA duplex, disfavoring its assembly into Ago2-RISC; increasing the concentration of the Dcr-2/R2D2 heterodimer increases loading of the small RNA into Ago2 by overcoming its reduced affinity for the RLC.
Although increasing Dcr-2/R2D2 concentration promoted loading of the mm11 duplex into Ago2, the crosslinking assay cannot determine whether the Ago2 and Ago1-loading pathways compete for loading of an siRNA, because the majority of the 20 nM RNA duplex remained unassociated with the Ago2-loading machinery (Schwarz et al., 2003; Haley and Zamore, 2004). This free RNA creates a reservoir of duplex that can, in principle, be loaded into Ago1. Unfortunately, reducing the concentration of small RNA in the crosslinking assay caused the RNA-crosslinked proteins to become undetectable.
To test if the Ago2- and Ago1-loading pathways compete for loading small RNA duplexes, we used a lower concentration of small RNA and a more sensitive assay—immunoprecipitation—to measure the association of a small RNA with Ago1. (The assay cannot currently measure small-RNA association with Ago2, because no suitable anti-Ago2 antibody exists.) We incubated 1 nM 5′ 32P-radiolabeled mm11 duplex in embryo lysate with increasing concentrations of Dcr-2/R2D2, immunoprecipitated Ago1 using a monoclonal anti-Ago1 antibody, and measured the concentration of Ago1-associated small RNA duplex by scintillation counting. Increasing the concentration of Dcr-2/R2D2 decreased the amount of siRNA associated with Ago1 (Figure 2, A and C), indicating that Ago1 loading competes with Dcr-2/R2D2-mediated loading of Ago2.
To test directly the idea that the affinity of the Dcr-2/R2D2 heterodimer for a small RNA duplex determines the extent of its loading into Ago2-RISC, we used a gel-mobility shift assay to measure the affinity of recombinant Dcr-2/R2D2 heterodimer for the series of ten let-7 small RNA duplexes. Purified recombinant Dcr-2/R2D2 and 5′ 32P-radiolabeled small RNAs bearing a 5′, non-radioactive phosphate on the passenger or miRNA* strand were incubated for 30 min, then free siRNA resolved from protein:siRNA complexes by native gel electrophoresis in the presence of Mg2+. Figure S3A shows a representative assay for the let-7 mm1 siRNA duplex. With increasing concentration of Dcr-2/R2D2, we detected two distinct complexes: complex 1 (C1) peaked at ~20 nM Dcr-2/R2D2, whereas complex 2 (C2) appeared at higher concentration of Dcr-2/R2D2, apparently replacing C1 (Figures S3A and and3A3A).
To determine if each complex contained Dcr-2, R2D2, or both, we repeated the assay using a let-7 siRNA bearing a 5-iodo uracil at guide-strand position 20; 5-iodo U at this position allows the siRNA to be site-specifically photocrosslinked to Dcr-2 or R2D2 upon irradiation with 302 nm light (Tomari et al., 2004a). The let-7 siRNA was incubated with 20 nM (for C1) or 100 nM Dcr-2/R2D2 (for C2), photocrosslinked, the complexes resolved by native gel electrophoresis, and then C1 and C2 were excised from the gel, and the crosslinked proteins in each complex were separated by SDS-PAGE (Figure S3B). Both C1 and C2 contained Dcr-2 and R2D2 crosslinked to siRNA (Figure S3C). Thus both C1 and C2 reflect binding of the Dcr-2/R2D2 heterodimer to siRNA.
Which complex then corresponds to the active form of siRNA-bound Dcr-2/R2D2 heterodimer competent to load Ago2? We added increasing concentration of recombinant Dcr-2/R2D2 heterodimer to lysate prepared from dcr-2L811fsX mutant embryos, which lack both Dcr-2 and R2D2 (TD and PDZ, unpublished). At each concentration of heterodimer, we measured the relative amount of Ago2-RISC activity assembled by determining the extent of cleavage after 15 min incubation with target RNA (Figure 3,B-D) when the reaction was linear (Figure S4). The Dcr-2/R2D2 concentration producing half-maximal target cleavage in this assay coincided with the apparent dissociation constant (Kapp) for C1 production, indicating that C1 is the active complex for RISC loading (compare Figure 3, A and D). Interestingly, at high concentrations of Dcr-2/R2D2 heterodimer, which favor the production of C2 (Figure 3A), target cleavage was inhibited (Figure 3D), reinforcing the view that complex C1 is the active, Ago2-loading form of siRNA-bound heterodimer and suggesting that C2 corresponds to a higher order, inactive aggregate of Dcr-2/R2D2 heterodimers.
Next, we examined the affinity of the Dcr-2/R2D2 heterodimer for various small RNA duplexes. We measured the Kapp of the heterodimer for formation of complex C1, the species active for Ago2-loading. Figure S5 (A and B) shows representative binding curves for the mm1 siRNA duplex, mm9 duplex and let-7/let-7* duplex, and Table 1 summarizes the Kapp for each determined in three independent trials. The Dcr-2/R2D2 heterodimer bound the mm9 duplex about half as tightly as it bound the let-7 mm1 siRNA duplex, whereas the heterodimer bound the let-7/let-7* duplex about five-fold less tightly than it bound the corresponding siRNA. Although previous studies concluded that Dcr-2 does not detectably bind siRNA in the absence of R2D2 (Liu et al., 2006), we found that purified recombinant Dcr-2 alone readily bound the mm1 siRNA duplex, with a Kapp of 94.6 ± 6.4 nM (average of four trials ± standard deviation; Figure S2). Thus, the apparent lack of Dcr-2 binding to siRNA reported previously likely reflects the ~12-fold lower affinity for siRNA of Dcr-2 alone compared to the intact heterodimer.
For the Dcr-2/R2D2 heterodimer, the order of relative affinities of Dcr- 2/R2D2 for the three small RNA duplexes correlated well with their extent of incorporation into Ago1- and Ago2-RISC: the greater the strength of binding of the heterodimer for a small RNA, the greater its association with Ago2 and the more reduced its association with Ago1. To further test this idea, we determined the fraction of small RNA duplex bound to 8 nM Dcr-2/R2D2 heterodimer for all ten let-7 small RNA duplexes (Figure 3E). The amount of small RNA associated with Dcr-2/R2D2 in this assay correlated well with the amount of the small RNA assembled into Ago2 relative to Ago1 (Figure 1, B and C).
Clearly, the affinity of the Dcr-2/R2D2 heterodimer for a small RNA duplex is an important determinant of the extent to which the small RNA is loaded into Ago2. Our data also suggest that Ago1 and Ago2 compete for loading with a small RNA duplex (Figure 2C). In theory, small RNAs whose structure disfavors their loading into Ago2 pathway, might enter the Ago1-loading pathway simply by default. To test this idea, we examined the loading of the mm1 siRNA duplex, mm9 duplex and let-7/let-7* duplex into Ago1 in lysate prepared from dcr-2L811fsX and from ago2414 mutant embryos. In the dcr-2L811fsX and ago2414 lysates, where Ago2 is not loaded, the relative amount of each small RNA duplex loaded into Ago1, measured by photocrosslinking, remained essentially unchanged from that observed in wild-type lysate (Figure 4A). Even in the absence of Ago2-loading machinery or Ago2 itself, Ago1 was preferentially loaded with the let-7/let-7* duplex, largely rejected the mm1 siRNA duplex, and accepted some of the mm9 duplex. Thus, both the Ago1- and the Ago2-loading pathways are selective, with each favoring a small RNA structure disfavored by the other.
While the extent of Ago1 loading was essentially the same in the wild-type and mutant lysates, the rate at which the three small RNA duplexes associated with Ago1 was accelerated in both the ago2414 and dcr-2L811fsX lysates (Figure 4B). This effect was most pronounced for the let-7/let-7* duplex, which was loaded twice as fast in the dcr-2L811fsX mutant lysate, which lacks the Ago2 loading machinery. The finding that, in the absence of the Ago2-loading machinery, Ago1 is more rapidly loaded with its authentic substrate, the let-7/let-7* duplex, suggests that miRNA/miRNA* duplexes bind the Dcr-2/R2D2 heterodimer transiently, even when they ultimately make little or no Ago2-RISC.
Conversely, a small RNA duplex favored to produce Ago2-RISC associated with Ago1 in both the absence and presence of Ago2 (Figure 4A and B). But does this Ago1-associated small RNA correspond to mature RISC, which contains only the miRNA or guide strand of the original duplex, or pre-RISC, a RISC-assembly intermediate in which the double-stranded miRNA/miRNA* or siRNA is bound to Argonaute (Matranga et al., 2005; Rand et al., 2005; Kim et al., 2006; Leuschner et al., 2006)? We determined if the let-7 strand was bound to Ago1 or Ago2 as single-stranded or double-stranded RNA. For the mm1, the mm9 and the let-7/let-7* duplexes, each 5′-32P-radiolabeled small RNA duplex was incubated with wild-type or ago2414 mutant lysate to assemble RISC, photocrosslinked to identify siRNA-associated proteins, and then single-stranded RNA-crosslinked proteins captured using an immobilized 2′-O-methyl antisense oligo (ASO) complementary to let-7 (Figure 5A); in this assay, proteins crosslinked to double-stranded siRNA or miRNA/miRNA* remain in the supernatant. (Dcr-1, Dcr-2, and R2D2 were never recovered with the immobilized ASO, consistent with previous observations that they bind only double-stranded small RNAs (Tomari et al., 2004a).) As expected, the majority of the crosslinked Ago2 was recovered with the immobilized ASO for the let-7 mm1 siRNA duplex, whereas most of the crosslinked Ago1 was recovered with the immobilized ASO for the let-7/let-7* duplex. We conclude that the let-7 mm1 siRNA duplex efficiently assembled mature Ago2-RISC, whereas the let-7/let-7* duplex efficiently assembled mature Ago1-RISC. The mm9 duplex also efficiently assembled mature Ago2-RISC.
Much of the Ago1-associated let-7 loaded from the mm1 siRNA duplex or the mm9 duplex, however, remained double-stranded, suggesting that the Ago1-loading machinery or Ago1 itself cannot efficiently dissociate the passenger strand from a highly base-paired duplex (Figure 5A). In contrast, little double-stranded, Ago2-associated let-7 was observed for the mm1 siRNA duplex or mm9 duplex in the wild-type lysate, likely reflecting the rapid cleavage of the passenger strand by Ago2. This is consistent with our findings that Ago1 is not an efficient endonuclease (Förstemann et al., accompanying manuscript).
We note that in the absence of Ago2, some let-7-programmed Ago1-RISC was formed from the mm1 siRNA duplex. The low efficiency of incorporation of the let-7 siRNA guide strand into mature Ago1-RISC, together with the reduced endonuclease activity of Ago1 compared to Ago2, likely explains the small amount of siRNA-directed target cleavage observed in vitro in lysate prepared from ago2414 (Okamura et al., 2004) and r2d21 mutant embryos (Liu et al., 2006).
Immunoprecipitation experiments confirmed these photocrosslinking and ASO-binding studies. (Figure 5B). RISC was assembled with 5′ 32P-radiolabeled mm1 siRNA duplex, the mm9 duplex, or the let-7/let-7* duplex, immunoprecipitated with anti-Ago1 monoclonal antibody, immunoprecipitated proteins removed by digestion with protease at room temperature, and then the 32P-radiolabeled small RNAs resolved by native gel electrophoresis to assess if they were single- or double-stranded. For both the mm1 siRNA duplex and the mm9 duplex, most of the Ago1-associated let-7 was double-stranded. In contrast, essentially all of the Ago1-associated let-7 loaded from the let-7/let-7* duplex was single-stranded, indicating it had been successfully assembled into functional Ago1-RISC. Our data suggest that the conversion of pre-Ago1-RISC to mature Ago1-RISC requires additional structural features that help separate the two siRNA strands, such as mismatches in the siRNA seed region. Such features might act in a pathway similar to the ‘bypass’ mechanism that facilitates the conversion of pre-RISC to mature RISC for Ago2 when passenger-strand cleavage is blocked (Matranga et al., 2005). In fact, when miRNA* cleavage by human Ago2 is blocked, seed mismatches between the miRNA and its miRNA* accelerate separation of the two strands (Matranga et al., 2005). We note that Drosophila Ago1 is more closely related to human Ago2 than to the conspecific Ago2 protein.
In cells, small RNA duplexes are produced from longer precursors by dicing. How faithfully do our studies of small RNA sorting, which bypass this step, reflect the cellular pathway? To answer this question, we programmed Drosophila embryo lysate with three different dicer substrates: a short-hairpin RNA designed to generate an asymmetric let-7 siRNA after dicing (mm1 shRNA), the same shRNA, but also containing a mismatch at let-7 position 9 (mm9 shRNA), and authentic pre-let-7 RNA. (As reported previously (Hutvágner and Zamore, 2002), less active RISC was produced in vitro from hairpin substrates than when siRNAs are used directly.)
We first incubated each precursor with embryo lysate to generate let-7–programmed RISC, and then we added a target RNA containing a site complementary to let-7 and monitored target cleavage (Figure 6). Of the three precursor RNAs, mm1 shRNA produced the most active RISC. To determine the degree to which the target cleavage observed for each precursor RNA reflected Ago1-RISC programmed with let-7, we immunodepleted Ago1 after the RISC assembly step but before adding target RNA. Our immunodepletion strategy removed more than 98% of the Ago1 protein (Figure S6). Depletion of Ago1 reproducibly enhanced to a small extent the rate of target cleavage for mm1 shRNA, but had little effect on mm9 shRNA. In contrast, most of the RISC activity produced by pre-let-7 was removed when Ago1 was immunodepleted. These results are consistent with mm1 shRNA loading Ago2 and pre-let-7 loading Ago1.
To determine the degree to which the target cleavage observed for each precursor RNA reflected Ago2-RISC programmed with let-7 (Figure 6), we compared the amount of let-7–directed target cleaving activity generated from each precursor in wild-type lysate to that generated in ago2414 lysate, which lacks Ago2 protein. Little or no RISC activity was detected for mm1 shRNA in the ago2414 mutant lysate. In contrast, for pre-let-7 more RISC activity was detected for the ago2414 mutant than for the wild-type lysate, presumably because the loss of competition with the Ago2 pathway resulted in more Ago1-RISC. As for the Ago1 immunodepletion experiment, mm9 shRNA produced less active RISC than the other two substrates. This RISC activity was reduced in the ago2414 mutant lysate, consistent with our finding (Figure 5) that most of the Ago1-RISC produced by a mm9 siRNA was inactive because the siRNA remained double-stranded. We conclude that dicing has little or no influence on the subsequent partitioning of a small RNA duplex between Ago1- and Ago2-RISC.
Here we show that in Drosophila the structure of a small RNA duplex determines its partitioning between Ago1- and Ago2-RISC. Our data suggest a simple model for this partitioning (Figure 7), with a central unpaired region serving as both an anti-determinant for the Ago2-loading pathway and a preferred binding substrate for the Ago1 pathway. Supporting this view, miRNAs that contain central mismatches, such as let-7 and bantam, assemble primarily into Ago1-RISC (Okamura et al., 2004). The accompanying manuscript (Förstemann et al.) shows that miR-277, whose central region is base paired, partitions between Ago1 and Ago2 in vivo.
Both the Ago2- and Ago1-loading pathways are selective. For Ago2, the affinity of the Dcr-2/R2D2 heterodimer for a small RNA duplex provides the primary source of small RNA selectivity. In the absence of either the Ago2-loading machinery or Ago2 itself, Ago1 is nonetheless preferentially loaded with a miRNA/miRNA* duplex; an siRNA duplex still loads poorly into Ago1. Thus, the Ago1-loading pathway is also inherently selective, and not a default pathway that assembles small RNAs rejected by the Ago-2 pathway. We do not yet know if this selectivity is a direct property of Ago1, of an Ago-1 loading machinery that remains to be identified, or both.
Previous bioinformatic analyses noted that a central region of thermodynamic instability was a common feature of miRNA/miRNA* duplexes (Khvorova et al., 2003; Han et al., 2006). Our data ascribe a function in flies to this common miRNA/miRNA* structural feature: directing the miRNA into Ago1 and away from Ago2. Mammalian miRNA/miRNA* duplexes also typically contain a central unpaired region, but it is not yet known if they are preferentially loaded into one of the four mammalian Ago-subclade Argonaute proteins.
What is the biological significance in flies of sorting miRNAs into Ago1 and siRNAs into Ago2? One idea, supported by the accompanying manuscript (Förstemann et al.), is that Ago1 and Ago2 are functionally distinct, with only Ago2 silencing targets that possess extensive complementarity to the small RNA guide and only Ago1 directing repression of targets that contain multiple but only partially complementary miRNA-binding sites. Sorting small RNAs between Ago1 and Ago2 may also prevent miRNAs from saturating the Ago2 machinery, which might compromise Ago2-mediated anti-viral defense (Galiana-Arnoux et al., 2006; Obbard et al., 2006; Wang et al., 2006; Zambon et al., 2006). Conversely, excluding from Ago1 siRNAs produced in response to viral infection may minimize competition between such anti-viral siRNAs and endogenous miRNAs, protecting flies from misregulation of gene expression during a viral infection. Restricting a robust RNAi—i.e., target cleavage—response to siRNAs loaded into Ago2 may also minimize undesirable, miRNA-like regulation of cellular genes by virally derived siRNAs. Thus, small RNA sorting ensures that miRNAs are largely restricted to Ago1, whose relaxed requirement for complementarity between a miRNA and a regulated mRNA target allows each miRNA to control many different mRNAs, and that siRNAs are restricted to Ago2, whose silencing activity requires more extensive complementarity between the target and the siRNA guide.
Nonetheless, a final question remains unanswered: why do some iconoclastic miRNA/miRNA* duplexes contain features that favor their loading into Ago2?
Preparation of 0-2 h embryo lysate, lysis buffer, and 2x PK buffer , in vitro assembly of RISC, inactivation of RISC assembly by NEM treatment, in vitro RNAi reactions, purification of recombinant Dcr-2/R2D2 purification, and UV photocrosslinking of proteins to 5-iodo-uracil-containing siRNAs were performed as described previously (Nykanen et al., 2001; Haley et al., 2003; Tomari et al., 2004a). In vitro RNAi target cleavage was performed with 20 nM siRNA and 10 nM 32P-cap radiolabeled target RNA for Figure Figure3C3C and S4 and 0.5 nM target in Figure 6.
20 nM 5′-32P-labeled small RNA duplex was incubated with lysate in a standard RNAi reaction (Haley et al., 2003) and then irradiated with 254 nM ultraviolet light for 5 min using a Stratalinker (Stratagene) with the sample ~3 cm below the UV bulbs. The photocrosslinked proteins were then resolved by 4–20% gradient SDS-polyacrylamide gel electrophoresis (Criterion pre-cast gels; BioRad). 2′-O-methyl ASO were used to isolate proteins photocrosslinked to single-stranded let-7 as described previously (Tomari et al., 2004a).
1 nM 5′-32P-radiolabeled let-7 mm11 duplex (Figure 2C) or 20 nM 5′-32P-radiolabeled mm1, mm9 and let-7/let-7* duplexes (Figure 5B) were incubated for 1 h with wild-type embryo lysate. The reactions were then incubated with anti-Ago1 mouse monoclonal antibody (Okamura et al., 2004) tethered to Dynabeads protein G paramagnetic beads (Invitrogen) for 1 h. The beads were washed by lysis buffer three times, and the radioactivity of the bound RNA was measured by scintillation counting (Figure 2C), or the beads were deproteinized with 2 mg/ml (f.c.) proteinase K in 2x PK buffer at room temperature for 30 min, the supernatant precipitated with 2.5 volumes of absolute ethanol, and the precipitate resolved by electrophoresis in a 20% native polyacrylamide gel (19:1) containing 1x TBE and 3 mM MgCl2 (Figure 5B). Control experiments demonstrated that the let-7/let-7* duplex remains double-stranded under these gel conditions.
Anti-Ago1 antibody beads were prepared by incubating 5 μl of tissue culture supernatant from the anti-Ago1 antibody producing cells for every 5 μl protein G beads for 1 h on ice, and then washing the beads three times. 5 μl of these beads bearing the Ago1 antibody were used per 10 to 20 μl reaction.
~100 pM 5′-32P-labeled small RNA duplexes were incubated for 30 min with recombinant Dcr-2/R2D2 heterodimer or Dcr-2 alone in lysis buffer containing 5 mM DTT, 0.1 mg/ml BSA, 3% (w/v) ficoll-400 and 5% (v/v) glycerol and then resolved by electrophoresis on a 5.25 % native polyacrylamide gel (37.5:1) containing 0.5x TBE and 1.5 mM MgCl2. RNA and complexes were detected by phosphorimagery, quantified using an FLA-5000 image analyzer and ImageGuage 4.22 software (Fujifilm), and fit to the Hill equation with IGOR Pro 5 software (WaveMetrics).
For immunodepletion, 120 μl Dynabeads Protein G paramagnetic bead suspension (Invitrogen) was incubated overnight with 120 μl anti-Ago1 mouse monoclonal antibody (1B8) (Okamura et al., 2004) at 4°C with gentle agitation. Next, the magnetic beads were washed 3 times with lysis buffer and then split among three tubes. Each precursor RNA was incubated in 100 μl standard RNAi reaction at room temperature for 1 h. Subsequently, 60 μl of the reaction was added to the anti-Ago1 magnetic beads, and the mixture agitated gently at 4 °C overnight. The supernatant was removed, and the beads washed three times with lysis buffer. The input, supernatant, and beads (the immunoprecipitate) were subsequently analyzed by Western blotting to confirm Ago1 depletion, by native gel analysis to measure the amount of Ago1-associated single-stranded let-7, and by a target cleavage assay to measure RISC activity.
Figure S1. Small RNA duplexes used in this study. The guide or miRNA strand (red) of each small RNA duplex corresponded to the sequence of the Drosophila miRNA let-7. Guide strands were 5′ 32P-radiolabeled; the passenger or miRNA* strands all contained a non-radioactive 5′ phosphate group.
Figure S2. Dcr-2 binds siRNA with nM affinity. (A) Representative native gel analysis of 32P-radiolabled siRNA incubated with increasing concentrations of purified, recombinant Dcr-2. (B) Quantitative analysis of the data in (A). The average ± standard deviation of four such trials, 94.6 ± 6.4 nM, is reported in the text as the Kapp for Dcr-2 binding siRNA.
Figure S3. (A). Native gel analysis of recombinant Dcr-2/R2D2 heterodimer binding siRNA. Quantitative analysis appears in Figure 3A. (B) Strategy for identifying the constituent proteins for complexes C1 and C2. (C) SDS-PAGE analysis of proteins crosslinked to 5′-32P-radiolabeled, iodo-uracil substituted siRNA as in (B) Complexes C1 and C2 contain both Dcr-2 and R2D2.
Figure S4. (A) Experimental strategy for (B). (B) Target RNA cleavage reactions in dcr-2 mutant embryo lysate supplemented with recombinant Dcr-2/R2D2 were linear with respect to time at 15 min, the end time point used in Figure 3, B–D, even with 25 nM Dcr-2/R2D2 the concentration of heterodimer that produced the maximum amount of RISC in Figure 3, B–D. (C) Experimental strategy for (D). (D) The 15 min end-point assay showed a near-linear correlation between the fraction of target RNA cleaved and the relative RISC concentration (varied by stopping RISC assembly with N-ethylmaleimide (NEM), then diluting the siRNA-programmed lysate in NEM-pre-treated naïve lystate). The fitted curve was used to calculate the RISC concentrations reported in Figure 3D.
Figure S5. A central mismatch in a small RNA duplex impairs Dcr-2/R2D2 binding. (A) An example of native gel analysis of exemplary small RNA duplexes. (B) Quantification of native gel analysis for each small RNA duplex in (A). The average ± standard deviation for three independent trials is reported as Kapp in Table 1.
Figure S6. Depletion of Ago1-RISC using anti-Ago1 monoclonal antibody. + Ago1, input embryo lysate; − Ago1, supernatant after immunodepletion.
We thank Mikiko and Haruhiko Siomi for anti-Ago1 antibody and ago2414 flies, Richard Carthew for dcr-2L811fsX flies, Qinghua Liu for Dcr-2- and R2D2-expressing baculoviruses, Alicia Boucher for assistance with fly husbandry, Gwen Farley for technical assistance, and members of the Zamore lab for advice, suggestions, and critical comments on the text. PDZ is a W.M. Keck Foundation Young Scholar in Medical Research. This work was supported in part by grants from the National Institutes of Health to PDZ (GM62862 and GM65236) and a post-doctoral fellowship from the Human Frontier Science Program to YT.