Molecular Requirements for RNA-induced Silencing Complex Assembly in the Drosophila RNA Interference Pathway*

Complexes in the Drosophila RNA-induced silencing complex (RISC) assembly pathway can be resolved using native gel electrophoresis, revealing an initiator called R1, an intermediate called R2, and an effector called R3 (now referred to as holo-RISC). Here we show that R1 forms when the Dicer-2/R2D2 heterodimer binds short interfering RNA (siRNA) duplexes. The heterodimer alone can initiate RISC assembly, indicating that other factors are dispensable for initiation. During assembly, R2 requires Argonaute 2 to convert into holo-RISC. This requirement is reminiscent of the RISC-loading complex, which also requires Argonaute 2 for assembly into RISC. We have compared R2 to the RISC-loading complex and show that the two complexes are similar in their sensitivities to ATP and to chemical modifications on siRNA duplexes, indicating that they are likely to be identical. We have examined the requirements for RISC formation and show that the siRNA 5′-termini are repeatedly monitored during RISC assembly, first by the Dcr-2/R2D2 heterodimer and again after R2 formation, before siRNA unwinding. The 2′-position of the 5′-terminal nucleotide also affects RISC assembly, because an siRNA strand bearing a 2′-deoxyribose at this position can inhibit the cognate strand from entering holo-RISC; in contrast, the 2′-deoxyribose-modified strand has enhanced activity in the RNA interference pathway.

In many eukaryotic organisms, double-stranded RNAs elicit an innate gene-silencing phenomenon known as RNAi 2 (1). The RNAi response occurs in two phases. In the first phase, the double-stranded RNAs are cleaved by the RNase III enzyme Dicer into double-stranded fragments called siRNAs (2)(3)(4). In the second phase, these siRNAs are conveyed into a large ribonucleoprotein complex known as the RISC where they act as guides, base-pairing to complementary mRNAs and triggering RISC-mediated mRNA cleavage (2, 4 -10) and destruction (11)(12)(13). Exactly how the double-stranded RNAs make the transition from the targets of the first phase to the triggers of the second phase is the subject of recent and continuing investigation.
The siRNAs produced by Dicer may determine the outcome of this transition. Newly generated siRNAs have characteristic 5Ј-phosphoryl-ated ends that are recognized by the RNAi machinery (8) and facilitate their entry into the RISC (14 -16). To get there, the siRNA duplexes have to be separated into their constituent strands (8), and only one of the two strands is assembled into the RISC (10). When synthetic siRNAs are used to initiate RNAi, the siRNA strand whose 5Ј-end is less stably base-paired is more frequently incorporated into the RISC (17,18), whereas the cognate strand is more frequently degraded.
We showed previously that siRNAs are channeled into the RISC in an ordered pathway that links the early and late phases of RNAi (14). The pathway begins when a group of pre-associated factors binds the siRNA to form an initiation complex known as R1. R1 is an ATP-independent complex that cannot form on siRNAs that lack 5Ј-phosphorylated ends. It contains at least two protein factors: Dicer-2 (Dcr-2), the Drosophila enzyme responsible for generating siRNAs, and presumably R2D2, a Dcr-2-associated factor that is a key player in bridging the initiation and effector phases of RNAi (19). The R1 complex is converted into the intermediate complex R2 by the ATP-independent addition of unidentified factors. Eventually the siRNA in R2 is unwound, leading to the formation of a large, ATP-dependent, ϳ80 S complex known as holo-RISC (formerly R3). This complex can cleave targeted mRNAs, likely acting as the RNAi effector in Drosophila (14).
Zamore and co-workers (15) have proposed a similar yet distinct assembly pathway that begins when siRNAs first bind to a complex known as B. Complex B contains neither Dcr-2 nor R2D2 but is proposed to be a direct precursor to the RISC-loading complex (RLC; formerly complex A), which harbors both of these factors and reportedly requires ATP to form (15). Dcr-2 and R2D2 cross-link to opposite ends of the siRNA duplex in the RLC, and the orientation of this interaction is strongly influenced by the relative thermodynamic stabilities of the two siRNA ends (16). R2D2 prefers to bind the more stably base-paired end, whereas Dcr-2 typically binds to the less stably base-paired of the two ends. Because the siRNA strand that enters RISC is the one whose 5Ј-end is less stably base-paired, R2D2 is thought to sense siRNA asymmetry, distinguishing the strand that enters the RISC from the one destined for degradation (16).
Although there are parallels between the two proposed RISC assembly models, there are also key differences. For example, it is unclear how complex B and the RLC fit into the R1 3 R2 3 holo-RISC assembly scheme. We aim to resolve the discrepancies in the two models and develop a coherent, unified understanding of RISC assembly in Drosophila. To that end, we have defined in greater detail the molecular events of the assembly pathway, providing a model that includes multiple 5Ј-recognition events and reconciles the reported differences in the R2 and RLC complexes. land Biolabs) according to the manufacturer's instructions. The singlestranded siRNAs were annealed as described previously (14). Chromatography-200-l standard RNAi reaction mixtures (20) containing either Drosophila embryo extract prepared as described (20) (15%) or recombinant R2D2/Dcr-2 heterodimer (107 nM) and 32 P-radiolabeled Photinus pyralis luciferase (Pp-luc) siRNA were incubated at 25°C for 40 min. The purified, baculovirus-expressed recombinant R2D2/Dcr-2 heterodimer was generously provided by Y. Tomari and P. Zamore. The mixtures were loaded onto a Superdex-200 HR 10/30 gel filtration column (Amersham Biosciences) and fractionated as described previously (14). 200-l fractions were collected and analyzed by Cerenkov counting.

EXPERIMENTAL PROCEDURES
To partially purify R2D2/Dcr-2 fractions, 200 l of embryo extract was loaded directly onto a Superdex 200 column and chromatographed as described above. The R1-forming fractions were collected (fractions 18 -28), supplemented with carrier insulin (final concentration 0.25 mg/ml) (Sigma), and concentrated to ϳ180 l in a Microcon YM-30 spin column (Millipore).
Native Gel Electrophoresis-RNAi reaction mixtures for native gel experiments were assembled as described previously (14), except that reaction volumes were 5 l rather than 10 l and the mixtures contained 15% embryo lysate. The mixtures were quenched with 1 l of a 4 mg/ml heparin mix and analyzed on 4% polyacrylamide native gels (10.8 mm ϫ 17 cm ϫ 23 cm) that were run at 10 watts at 4°C in 1ϫ Tris borate-EDTA as described previously (14). Agarose native gels were prepared and run as reported previously (15) with modifications (21). ATP depletion was performed by pre-incubating reaction mixtures with glucose (20 mM) and hexokinase (Calbiochem, 0.5 units) for 20 min at 25°C. The concentration of ATP after depletion was measured using an ATP bioluminescence assay kit (Sigma) and was confirmed in all ATP-depleted samples to be Ͻ100 nM.
RNA Unwinding-Standard unwinding assays were performed as described previously (8). The zero time point samples were treated with proteinase K and stop buffer just prior to the addition of radiolabeled siRNA. The amount of double-stranded and single-stranded siRNA detected at time 0 was set to 100 and 0%, respectively.

Dcr-2 and R2D2 Comprise R1 and Can Initiate RISC Assembly-
RISC assembly in Drosophila proceeds in a stepwise manner and is first detected when proteins bind siRNA within the R1 complex (14). To examine whether R1 contains any factors other than R2D2 and Dcr-2, we incubated radiolabeled siRNA with purified recombinant R2D2/ Dcr-2 heterodimer and compared it to the R1 complex formed in embryo lysates. When analyzed by gel filtration (Fig. 1a) and native gel electrophoresis (Fig. 1b), the recombinant R2D2/Dcr-2/siRNA complex was indistinguishable from R1. These results indicate that R1 is not likely to contain any additional factors, so we now refer to it as the R2D2/Dcr-2 initiator (RDI) complex to distinguish it from the unbound Dcr-2/R2D2 heterodimer.
If R2D2 and Dcr-2 are the only protein components of the RDI, then the recombinant R2D2/Dcr-2 heterodimer should suffice for initiating RISC assembly. To test this hypothesis, we performed a pulse-chase experiment with pre-formed RDI prepared from recombinant R2D2/ Dcr-2 ( Fig. 1c). Radiolabeled siRNAs were incubated with recombinant R2D2/Dcr-2 and allowed to form the RDI complex. An excess of unlabeled siRNA competitor was then added, and the mixture was "chased" with wild-type embryo extract. The radiolabeled siRNA in the RDI complex was rapidly conveyed into R2 and later into holo-RISC, indicating that recombinant R2D2/Dcr-2 alone can initiate RISC assembly (Fig. 1c, Argonaute 2 (Ago2) Affects RISC Assembly at the R2 to Holo-RISC Transition-Ago2 is a core RISC constituent (5) that acts at several stages in the RNAi pathway, both early and late, to promote siRNA unwinding (21) and, ultimately, to cleave targeted mRNAs as the "slicer" endonuclease (22)(23)(24)(25). Holo-RISC formation requires Ago2 because the assembly pathway stalls at the RLC in ago2 null lysate (21). To determine whether the RDI or R2 complexes are functionally similar to the RLC, we used the native polyacrylamide gel system to examine the complexes that form on radiolabeled siRNA incubated in an ago2 null lysate (ago2 414 ) (21). In this lysate, RISC assembly arrested at R2 (Fig. 2, lane 1) but proceeded to holo-RISC both in wild-type lysates and in those prepared from ago2 null embryos transformed with an AGO2 transgene FIGURE 1. R2D2 and Dcr-2 comprise R1 and can initiate RISC assembly. a, 32 P-radiolabeled Pp-luc siRNAs were incubated with embryo lysate (triangles) or purified, recombinant R2D2/Dcr-2 (rR2D2/Dcr-2) heterodimer (squares) and fractionated through a Superdex-200 gel filtration column. b, 32 P-radiolabeled Pp-luc siRNAs were incubated as described for panel a and analyzed by native polyacrylamide gel electrophoresis. c, 32 Pradiolabeled siRNAs were incubated with purified, recombinant R2D2/Dcr-2 heterodimer and then mixed with a 10,000-fold excess of unlabeled competitor siRNA. The reaction mixture was then split into aliquots and chased with Drosophila embryo lysate. The lysate was first added to the 60-min sample, and the experiment was performed as a reverse time course. Reactions were quenched with heparin on ice and analyzed by native gel electrophoresis. In two control experiments the R2D2/Dcr-2 initiator (RDI) was chased with buffer instead of lysate (lane 1) and, in lane 8, the excess unlabeled competitor was added at the outset. FIGURE 2. The R2 complex accumulates in the absence of Ago2. 32 P-radiolabeled Pp-luc siRNA was incubated with lysates from embryos as shown, quenched with heparin, and analyzed by native gel electrophoresis. WT, wild-type.
( Fig. 2, lanes 2 and 3). These results indicate that R2 may be similar to the RLC (see below). Furthermore, because Ago2 is required for siRNA unwinding (21), these data indicate that R2 contains double-stranded siRNA, an indication that is consistent with our previous observation that R2 does not require ATP to form (14).
The 5Ј-Ends of siRNA Duplexes Are Monitored at Multiple Stages of RISC Assembly-siRNAs that lack 5Ј-phosphate groups on both strands are unable to enter the RISC assembly pathway (8,14,15), obscuring the potential 5Ј-recognition events that may occur at later stages of assembly and catalysis. To address these possible recognition events, we examined the native gel complexes that formed on siRNA lacking a 5Ј-phosphate on one of the two strands. The radiolabeled 5Ј-phosphorylated strand was annealed to an unlabeled cognate strand bearing either a 5Ј-hydroxyl or 5Ј-phosphate terminus (Fig. 3a). When incubated in embryo lysate, the singly phosphorylated siRNA showed reduced levels of holo-RISC compared with the doubly phosphorylated control (Fig. 3b, lanes 1 and 2), even though there is a kinase in the lysate that phosphorylates most of the siRNA (8). Our observations suggest that a fraction of the siRNA rapidly enters the RISC assembly pathway, escaping the kinase activity. These siRNAs appear to be shielded (at least partially) from the kinase but are defective at some stage of RISC assembly, presumably because they remain singly phosphorylated. Similar effects were observed with three other siRNA sequences (supplemental Fig. 1, available in the on-line version of this article). As an initial test to identify the source of the defect, we incubated the singly and doubly phosphorylated siRNAs in ago2 null lysate and found that they formed comparable levels of the RDI and R2 complexes (Fig. 3b, lanes 3 and 4).
Because singly phosphorylated siRNA can assemble into these early complexes, but not into holo-RISC, these data indicate that a second 5Ј-recognition event occurs after formation of both the RDI and R2 complexes.
We performed a more rigorous analysis to test our initial observations, examining the formation of each complex in the assembly pathway. To assess how singly phosphorylated siRNAs fare in RDI formation, we incubated siRNA with partially purified R2D2/Dcr-2 fractions depleted of ATP. These fractions are capable of forming the RDI but cannot form R2. Furthermore, ATP depletion prevents phosphorylation of the siRNA 5Ј-end. Under these conditions, the singly and doubly phosphorylated siRNAs formed comparable levels of RDI (Fig. 3c), indicating that singly phosphorylated siRNAs can initiate RISC assembly. Similar results were obtained when the siRNAs were incubated with recombinant R2D2/Dcr-2 instead of partially purified R2D2/Dcr-2 fractions (data not shown).
To see how well singly phosphorylated siRNAs assemble into R2, we incubated siRNA with ago2 null lysate depleted of ATP. When incubated in this lysate, the singly and doubly phosphorylated siRNAs formed similar levels of R2 (Fig. 3d), consistent with our initial observations (Fig. 3b).
We next examined the requirement for 5Ј-phosphates in holo-RISC formation. To do this, we prepared Pp-luc siRNA bearing a 5Ј-methoxy modification on the non-radiolabeled strand. This was necessary because, unlike both the RDI and R2 complexes, holo-RISC requires ATP to form (14,15). Therefore, we could not use 5Ј-hydroxyl siRNAs in extracts depleted of ATP for our analysis. The 5Ј-methoxy-modified FIGURE 3. Singly phosphorylated siRNAs can assemble into the RDI and R2 complexes but can neither unwind nor enter holo-RISC. a, radiolabeled siRNA duplex used in panels b-d. The 32 P-radiolabel is in red, the radiolabeled strand is black, and the unlabeled strand is purple. The unlabeled strand was prepared with either a 5Ј-terminal 5Ј-hydroxyl or 5Ј-phosphate, shown in parentheses. b, the siRNA from panel a was incubated with wild-type (WT) or ago2 null (ago 414 ) lysates for 40 min, quenched with heparin, and analyzed by native gel. c, the radiolabeled siRNA shown in panel a bearing a 5Ј-phosphate (magenta) or 5Ј-hydroxyl (blue) on the non-radiolabeled strand was incubated with partially purified R2D2/Dcr-2 fractions depleted of ATP. At time points the aliquots were removed, quenched with heparin, and analyzed by native gel. d, the time course experiment was performed as for panel c, but the siRNAs were incubated in ago 414 extract depleted of ATP instead of R2D2/Dcr-2 fractions. e, Pp-luc siRNA was modified on the non-radiolabeled strand as shown, incubated in embryo lysate, and analyzed by native gel electrophoresis. The radiolabel is shown as a red asterisk. The modified nucleotide is denoted with an X. The blue box indicates the more thermodynamically stable of the two siRNA ends. f, the Pp-luc siRNA shown in panel a was incubated in embryo lysate. At various time points aliquots were removed, deproteinized, and precipitated with ethanol. The recovered RNAs were separated on a 15% non-denaturing gel and quantified with a PhosphorImager. The results are plotted as the percentage of the total RNA that was single-stranded. g, as in panel f, but the results are plotted as the fraction of double-stranded siRNA remaining (relative to the amount of double stranded siRNA at time 0). RNA is commercially available only for 2Ј-deoxythymidine nucleotides (dT-OCH 3 ). To control for the 2Ј-deoxyribose modification, we prepared an analogous siRNA bearing a 5Ј-phosphate group (dT-PO 4 ). We also prepared an siRNA bearing a terminal 5Ј-phosphorylated ribouridine (rU-PO 4 ) as an unmodified control. When incubated in embryo lysate and analyzed by native gel electrophoresis, the rU-PO 4 siRNA formed holo-RISC as expected (Fig. 3e, lane 1). However, the 5Ј-phosphorylated, 2Ј-deoxythymidine control siRNA (dT-PO 4 ) formed unexpectedly low levels of holo-RISC relative to the rU-PO 4 siRNA control (Fig. 3e, lane 2). This defect was also observed with an siRNA containing a 5Ј-terminal 2Ј-deoxyuridine nucleotide (dU-PO 4 ), indicating that the 2Ј-deoxyribose, not the thymine base, was responsible for the defect (data not shown). The singly phosphorylated dT-OCH 3 siRNA formed little (if any) holo-RISC (Fig. 3e, lane 3), confirming our earlier observations (Fig. 3b). Similar patterns of holo-RISC formation were observed using different siRNA sequences (supplemental Fig. 2, available in the on-line version of this article) at levels that were consistent with the prevailing asymmetry rules (17,18). Interestingly, we observed elevated levels of both the RDI and R2 complexes with the singly phosphorylated dT-OCH 3 siRNA (Fig. 3e, lane 3). Because these early complexes contain duplex siRNA, the singly phosphorylated siRNAs may be inhibited from unwinding.
The siRNA 5Ј Termini Regulate siRNA Unwinding-To test this hypothesis, we assayed the unwinding of Pp-luc siRNAs that were chemically modified on the non-radiolabeled strand. When we examined the rate and extent of single-stranded siRNA production, significant levels of unwound siRNA accumulated for the rU-PO 4 Pp-luc siRNA but not for the analogous Pp-luc siRNAs that were modified on the non-radiolabeled strand (Fig. 3f), consistent with the levels of holo-RISC formed on these siRNAs (Fig. 3e). When we quantitated the data as the fraction of double-stranded siRNA that persisted with time, approximately half of the rU-PO 4 Pp-luc siRNA remained after 2 h (Fig.  3g). The discrepancy between the amount of single-stranded siRNA produced (ϳ10%; Fig. 3f) and the amount of double-stranded siRNA lost over time (ϳ45%; Fig. 3g) is likely due to degradation of the radiolabeled strand, which in this case is thermodynamically disfavored for entry into the RISC and, relative to the unlabeled strand, more frequently destroyed. In contrast, the level of duplex dT-OCH 3 siRNA remained relatively unchanged (Fig. 3g). Strikingly, the level of duplex dT-PO 4 siRNA dropped precipitously after just 30 min (Fig. 3g), even though we observed no accumulation of single-stranded RNA (Fig. 3f). Similar patterns of unwinding were observed for a different siRNA duplex (supplemental Fig. 3, available in the on-line version of this arti-cle). Together, these data indicate that siRNAs must be doubly phosphorylated to efficiently unwind. Furthermore, an siRNA strand bearing a 5Ј-terminal 2Ј-deoxyribose modification somehow inhibits the cognate strand from entering the RISC.
We wondered if the deoxyribose-modified strand itself could enter the RISC. To address this question, we radiolabeled the 2Ј-deoxyribosemodified Pp-luc siRNA strand, annealed it to the unlabeled, unmodified, cognate RNA, and tracked its progress through the RISC assembly pathway. The modified strand (dT) assembled into holo-RISC at higher levels than the unmodified control (rU) (Fig. 4, lanes 3 and 4). One possible explanation for our observations is that the 5Ј-terminal 2Ј-deoxyribose modification destabilizes the RNA helix at the modified end. This would shift the siRNA asymmetry, resulting in enhanced selection of the modified strand and reduced selection of the unmodified strand, as we observed. We consider this scenario unlikely, because thermal melting analysis revealed no significant differences in the melting temperatures and free energies of the rU-PO 4 and dT-PO 4 siRNAs (supplemental Fig. 4, available in the on-line version of the article). This result indicates that the terminal 2Ј-deoxyribose modification does not significantly destabilize the RNA duplex.
To directly address how the modification affects RISC activity, we examined the cleavage of both sense and antisense Pp-luc target mRNAs guided by the Pp-luc siRNA (Fig. 5a). The control rU-PO 4 siRNA was able to guide cleavage of both the sense and antisense Pp-luc mRNAs, as expected (Fig. 5, b and c, lanes 4). In contrast, the singly phosphorylated dT-OCH 3 siRNA was unable to guide significant cleavage of either the sense or the antisense Pp-luc mRNAs (Fig. 3, b and c,  lanes 6), consistent with its inability to unwind (Fig. 3, f and g). When we assayed cleavage guided by the siRNA bearing a single 2Ј-deoxyribose modification on the antisense strand (dT-PO 4 ), we observed distinct activity with respect to the two target mRNAs. The unmodified sense siRNA strand did not promote significant RISC-mediated cleavage (Fig.  5b, lane 5). In contrast, the 2Ј-deoxyribose-modified antisense strand generated robust cleavage of the sense mRNA (Fig. 5c, lane 5). When we performed the experiment as a time course, the dT-PO 4 -modified siRNA clearly cleaved more sense mRNA than the rU-PO 4 control (Fig.  5d). These results confirm that a 5Ј-terminal 2Ј-deoxyribose modifica- . An siRNA strand modified with a 5Ј-terminal 2Ј-deoxyribose assembles into the RISC more efficiently than the cognate strand. Duplex Pp-luc siRNAs bearing 5Ј-terminal ribose (rU) or 2Ј-deoxyribose (dT) modifications on the antisense strand were prepared. The siRNAs were labeled either on the sense strand (black) or on the modified antisense strand (purple). The 5Ј-modified nucleotides are denoted with an X, and the radiolabel is shown as a red asterisk. The radiolabeled siRNAs were incubated in embryo extract, and the mixture was analyzed by native gel. tion enhances the RNAi activity of the modified strand while simultaneously suppressing the activity of the cognate strand. Interestingly, mRNA cleavage guided by single-stranded siRNA was also enhanced by a 5Ј-terminal 2Ј-deoxyribose modification (supplemental Fig. 5, available in the on-line version of this article).
R2 Is Similar to the RLC-We have shown that the R2 complex does not require ATP to form (14) and, further, that it can assemble on singly phosphorylated siRNA (Fig. 3e). It is therefore unlike the RLC, which reportedly depends on ATP as well as both 5Ј-phosphates (15). Although the two complexes differ in these regards, they share common features. They both directly precede holo-RISC in the assembly pathway, and they both accumulate in the absence of Ago2 (Fig. 2) (16,21). To see if these complexes are truly distinct, we compared them directly. 5Ј-Radiolabeled Pp-luc siRNA was incubated in embryo lysates that were either depleted of ATP or supplemented with ATP. These reaction mixtures were split and loaded onto either a polyacrylamide or an agarose native gel to detect R2 and the RLC, respectively. Surprisingly, in our hands the RLC formed in extracts depleted of ATP, albeit at lower levels than the ATP-supplemented control (Fig. 6a, lanes 3 and 4). ATP depletion was successful because the ATP-dependent RISC complex did not form in either gel system after depletion (Fig. 6a, lanes 1 and 3). Furthermore, the residual ATP concentration after depletion was Ͻ100 nM (data not shown), a level reported previously to block RLC formation (15).
To test the 5Ј-phosphate requirements for RLC formation, we examined the complexes that formed on singly phosphorylated Pp-luc siRNA that were modified on one strand with a 5Ј-methoxy group. Unexpectedly, the singly phosphorylated siRNA assembled into the RLC but did not proceed to the RISC in the agarose native gel (Fig. 6b, lane 3). We observed similar results using the 3Ј-cordycepin-labeled duplexes employed previously (15) (data not shown). There was some signal in the agarose gel that co-migrated with RISC when we used the dT-PO 4 siRNA (Fig. 6b, lane 2). Because the radiolabeled, unmodified strand of this siRNA can neither assemble into RISC (Fig. 3e, lane 2) nor cleave targeted mRNAs (Fig. 5b, lane 5), some of the signal in this region of the gel is not RISC or represents a nonfunctional form of RISC. Based on these results, we conclude that R2 and the RLC are similar, probably identical, complexes.

DISCUSSION
To act as silencing triggers, double-stranded siRNAs must be channeled through an ordered RISC assembly pathway that results in the selection of one strand and the destruction of the other. Our results indicate that siRNA ends are recognized at multiple steps in the pathway and that these recognition events determine whether the siRNAs will become incorporated into the RISC or blocked from further assembly.
The first recognition event occurs at initiation, when Dcr-2 and R2D2 bind the siRNA to form the RDI complex. RDI formation requires 5Ј-phosphorylated siRNA; however, the siRNA need not be phosphorylated on both strands. Nykänen et al. (8) showed previously that siR-NAs bearing a single 5Ј-phosphate group are ineffective at guiding mRNA cleavage. This was true for siRNAs phosphorylated only on the guiding strand as well as those phosphorylated only on the non-guiding strand. To explain these observations, Tomari et al. (16) invoke a model in which R2D2 acts as a sensor for siRNA asymmetry. They suggest that a 5Ј-phosphate on the thermodynamically disfavored "passenger" strand is required for stable R2D2 binding, facilitating siRNA incorporation into the RLC and the RISC. Because the non-guiding strand of the siRNA used by Nykänen et al. lacked this phosphate, they suggest that it was excluded from the RISC at the level of R2D2 binding. Consequently, they propose that R2D2 is a licensing factor, allowing only authentic siRNAs to enter the RNAi pathway (16). If the basic features of their model are correct, then our data indicate that R2D2 does not act alone in distinguishing authentic from inauthentic (non-phosphorylated) siRNA.
When only one of the two siRNA strands is phosphorylated, R2D2 and Dcr-2 can still avidly engage the siRNA, probably with R2D2 located at the phosphorylated end as predicted (16). Despite this observation, singly phosphorylated siRNA duplexes are still ineffective silencing triggers. Our data indicate that a second 5Ј-recognition event occurs after R2D2 binding. RNA unwinding and activation can proceed, but only if the siRNA bears the second 5Ј-phosphate at the end not occupied by R2D2. Because Ago2 is required both for siRNA unwinding and target mRNA cleavage, it may be the factor that recognizes the second 5Ј-phosphate. Two groups have recently shown that an archaebacterial PIWI protein has conserved residues that interact with the 5Ј-phosphate group of siRNA-like duplexes (26,27). Furthermore, when analogous residues were mutated in the human Ago2 PIWI domain, target mRNA cleavage was compromised (26). In light of our own observations, these results seem to implicate Ago2 in the second 5Ј-recognition event. Alternatively, Ago2 may not engage the 5Ј-phosphate until later in the assembly pathway. In this case, some other factor may interact with the second 5Ј-phosphate upstream of Ago2, causing the RISC assembly defect that we observed with singly phosphorylated siRNA duplexes.
Like the 5Ј-phosphate group, the 2Ј-hydroxyl of the 5Ј-terminal nucleotide can influence RISC assembly. A 5Ј-terminal 2Ј-deoxyribose modification on one siRNA strand inhibits the cognate strand from entering the RISC. Because the unmodified cognate strand does not accumulate in an unwinding assay, it is probably degraded. In contrast, the modified strand has enhanced activity and is apparently favored for FIGURE 6. The R2 complex is similar to the RLC. a, 5Ј-radiolabeled Pp-luc siRNA was incubated in embryo lysates that were either depleted of ATP or supplemented with ATP. After 40 min, the reactions were adjusted to 6% (v/v) glycerol and loaded on either polyacrylamide or agarose native gels. The samples loaded onto polyacrylamide were quenched with heparin before loading. b, chemically modified siRNAs were used in standard RNAi reactions that were supplemented with ATP as described for panel a and analyzed by agarose gel electrophoresis. c, integrated pathway for RISC assembly based on current data. Complex B is not directly required for siRNA binding in the RDI but may channel siRNAs into the RDI, the RLC, or both. entry into the RISC. There are at least two (non-mutually exclusive) explanations for these observations. The 2Ј-deoxyribose may be a negative determinant for R2D2 binding. In support of this possibility, the dT-PO 4 -modified Pp-luc-switch siRNA forms very little of the RDI and R2 complexes (supplemental Fig. 2, lane 2), even though it has a relatively stable and phosphorylated terminus that R2D2 would be expected to bind. Decreased R2D2 binding would bias the "sensor" of siRNA asymmetry (16), leading to increased selection of the modified strand and increased degradation of the unmodified strand, as we observed. Alternatively, the 2Ј-deoxyribose-modified strand may have an increased affinity for some downstream RNAi factor, thereby enhancing its entry into the RISC. This may also explain why a single-stranded dT-PO 4 -modified siRNA is better at guiding target mRNA cleavage than the analogous rU-PO 4 control (supplemental Fig. 5). In either case, our observations are surprising because the RNAi apparatus generally tolerates limited modifications at the 2Ј-position of siRNA duplexes, both on the sense and antisense strands (28 -30). However, completely substituting siRNAs with 2Ј-deoxy modifications on either the sense or antisense strand abolishes their ability to guide target mRNA cleavage (31), perhaps by altering the helical geometry of the duplex. Because there does not seem to be a general requirement for 2Ј-hydroxyls along the length of the siRNA non-guiding strand, the defect we observe for 5Ј-terminal 2Ј-deoxyribose modifications could reflect a specific recognition event that we cannot yet fully define.
Because a 5Ј-terminal 2Ј-deoxyribose modification on one siRNA strand can inhibit the activity of the cognate strand, it should reduce the off-target effects that can complicate RNAi-mediated gene silencing. Thus, this simple and inexpensive modification should be considered in the design of effective siRNA reagents for targeted gene knock-down.
Finally, we have made progress in our goal of developing a more unified understanding of RISC assembly in Drosophila (Fig. 6c). We have shown that the R2D2/Dcr-2 heterodimer alone can initiate RISC assembly by binding siRNAs to form the RDI complex. Like the RDI, complex B has been proposed to act at initiation (15). Because this complex can form in the absence of R2D2 (15), it is upstream of the RDI and may play an auxiliary role in RISC assembly. Complex B may transfer bound siRNA directly to the RDI in addition to (or instead of) the RLC. Continuing investigation will be required to resolve this issue.
Our results suggest that the RLC and R2 complexes have more in common than previously thought. Based on our side-by-side comparisons, both complexes can form in extracts that lack Ago2 and in extracts depleted of ATP. Furthermore, both complexes can assemble on siR-NAs that lack a phosphate on one of the two siRNA strands. The ATP and 5Ј-phosphate requirements that we observed for the RLC are distinct from those reported previously (15). We do not know the reason for these discrepancies. Based on our observations, R2 and the RLC are probably identical complexes, and we now refer to R2 as the RLC.