A Protein Sensor for siRNA Asymmetry

See allHide authors and affiliations

Science  19 Nov 2004:
Vol. 306, Issue 5700, pp. 1377-1380
DOI: 10.1126/science.1102755


To act as guides in the RNA interference (RNAi) pathway, small interfering RNAs (siRNAs) must be unwound into their component strands, then assembled with proteins to form the RNA-induced silencing complex (RISC), which catalyzes target messenger RNA cleavage. Thermodynamic differences in the base-pairing stabilities of the 5′ ends of the two ∼21-nucleotide siRNA strands determine which siRNA strand is assembled into the RISC. We show that in Drosophila, the orientation of the Dicer-2/R2D2 protein heterodimer on the siRNA duplex determines which siRNA strand associates with the core RISC protein Argonaute 2. R2D2 binds the siRNA end with the greatest double-stranded character, thereby orienting the heterodimer on the siRNA duplex. Strong R2D2 binding requires a 5′-phosphate on the siRNA strand that is excluded from the RISC. Thus, R2D2 is both a protein sensor for siRNA thermodynamic asymmetry and a licensing factor for entry of authentic siRNAs into the RNAi pathway.

In Drosophila lysates, siRNAs are loaded into the RISC by an ordered pathway in which one of the two siRNA strands, the guide strand, is assembled into the RISC, whereas the other strand, the passenger strand, is excluded and destroyed (114). A central step in RISC assembly is formation of the RISC-loading complex [RLC, previously designated complex A (13)], which contains double-stranded siRNA, the double-stranded RNA binding protein R2D2, and Dicer-2 (Dcr-2), as well as additional unidentified proteins. The function of Dicer in loading siRNA into the RISC is distinct from its role in generating siRNA from long double-stranded RNA (dsRNA) (10, 15). Both R2D2 and Dcr-2 are required to form RLC (13) and to unwind siRNA (Fig. 1, A and B), but recombinant Dcr-2/R2D2 heterodimer or Dcr-2 alone cannot catalyze siRNA unwinding (fig. S1). Thus, the Dcr-2/R2D2 heterodimer is necessary but not sufficient to unwind siRNA.

Fig. 1.

The RISC-loading complex (RLC) initiates siRNA unwinding. (A) r2d2 mutant ovary lysates, which cannot assemble RLC, do not unwind siRNA (wt, wild type). (B) siRNA unwinding was defective in ovary lysate from a mutation that disrupts all known Dcr-2 functions (L811fsX), including RLC assembly (10, 12, 18), but was normal for a point mutation in dcr-2 (Gly31 → Arg, G31R) that disrupts its function in dicing long dsRNA into siRNA, but not RISC assembly (10, 12). (C) The RLC, which is composed largely of double-stranded siRNA, also contains single-stranded siRNA. Note the different scales for the relative amounts of single- and double-stranded siRNA.

If siRNA unwinding is initiated in the RLC, then the RLC should contain some single-stranded siRNA. To test this idea, we briefly incubated siRNA duplex in lysate, resolved the complexes formed by native gel electrophoresis, divided the gel into 11 parts, and analyzed the structure of the siRNA in each gel slice (fig. S2A). Consistent with our previous findings, a peak of double-stranded siRNA comigrated with both the RLC and complex B, which is thought to be a precursor to RLC (13) (Fig. 1C). A small peak of single-stranded siRNA also comigrated with the RLC, but not with complex B (Fig. 1C), which suggests that the RLC initiates siRNA unwinding. Similar peaks of single-stranded siRNA comigrated with the RLC for the passenger strand of this siRNA and for the guide and passenger strands of a second siRNA (fig. S2B). We conclude that the RLC initiates siRNA unwinding.

The RLC also senses siRNA thermodynamic asymmetry, thereby determining which strand enters the RISC. siRNA containing 5-iodouracil at the 20th nucleotide (p20) can be photocrosslinked to R2D2 and Dcr-2 (13). Photocrosslinking is position-specific: An siRNA containing 5-iodouracil at position 12 was not cross-linked to R2D2 or Dcr-2 (13). Photocrosslinking attaches the radiolabel of the siRNA to the protein, identifying proteins that lie near p20 of the substituted siRNA strand. We evaluated the relative efficiency of photocrosslinking to R2D2 and Dcr-2 for three types of siRNA (Fig. 2) (table S1): a luciferase-specific siRNA whose sequence makes the 5′ end of the antisense strand less thermodynamically stable than the 5′ end of the sense strand; a nearly symmetric siRNA targeting human Zn, Cu superoxide dismutase 1 (sod1), in which the stabilities of the 5′ ends are essentially the same; and a series of highly asymmetric sod1-directed siRNAs in which the first nucleotide of the guide strand is mismatched to the passenger strand, causing the guide strand to be loaded into the RISC almost exclusively. When we used the partially asymmetric luciferase-specific siRNA, R2D2 was more efficiently photocrosslinked when the 5-iodouracil was on the strand more frequently incorporated into the RISC, whereas when the 5-iodouracil was on the strand less often incorporated into the RISC, Dcr-2 was more efficiently photocrosslinked (Fig. 2, A and B). Because a 5-iodouracil at p20 of one siRNA strand is near the 5′ end of the other strand, Dcr-2 must lie near the 5′ end of the strand entering the RISC (the guide strand), whereas R2D2 binds near the 5′ end of the strand destined for destruction.

Fig. 2.

Asymmetric binding of the Dcr-2/R2D2 heterodimer to siRNA duplexes in Drosophila embryo lysate. (A) Structure of the fully base-paired but asymmetric luciferase siRNA duplex used in (B) for protein-siRNA photocrosslinking. The local thermodynamic stability of the yellow highlighted base pairs is indicated here and in (D). Stars denote 5-iodouracil; an asterisk denotes a 5′-[32P]phosphate. (B) The blue strand generates about 5 times as much RISC as the red; more Dcr-2 photocrosslinks to the 5-iodouracil nearest the less thermodynamically stable end, whereas more R2D2 photocrosslinks to the 5-iodouracil nearest the more thermodynamically stable end. (C) The series of 5-iodouracil substituted siRNAs used in (D). siRNA a is thermodynamically symmetric. siRNAs b to g contain a single unpaired nucleotide at the 5′ end of one strand, making them highly asymmetric (7). For siRNAs b to d, the red strand serves as the guide strand, whereas the blue strand serves as the guide for siRNAs e to g. (D) For the functionally asymmetric siRNAs, Dcr-2 binding was detected nearest the 5′ end of the guide strand, whereas R2D2 was detected near the 5′ end of the passenger strand, the more thermodynamically stable end. The lower panel shows the siRNA strands at the bottom of the same gel. Single-stranded siRNA was detected only when the labeled strand served as a guide strand and entered the RISC. The asterisk indicates a photocrosslink to Ago2.

When we used the symmetric sod1 siRNA (Fig. 2, C and D, siRNA a), Dcr-2 and R2D2 were photocrosslinked with nearly equal efficiency to the 5-iodouracil strand (Fig. 2D, siRNA a); this finding suggests that each protein binds about half the time to one or the other end of the siRNA. In contrast, when we used derivatives of this siRNA that contained single-nucleotide mismatches that made them highly asymmetric, the 5-iodouracil strand was photocrosslinked to either Dcr-2 or R2D2, but not to both (Fig. 2, C and D, siRNAs b and c). With the asymmetric siRNA sequence, the photocrosslinking data suggest that Dcr-2 is almost always near the 5′ end of the guide strand and R2D2 near the 5′ end of the passenger strand. As expected when both siRNA strands contained p20 5-iodouracil and 5′-[32P]phosphate groups, both proteins were photocrosslinked (Fig. 2, C and D, siRNA d). When we used a reciprocal series of siRNAs in which the strands assembled into and excluded from the RISC were reversed (Fig. 2, C and D, siRNAs e, f, and g), Dcr-2 was again found near the 5′ end of the guide strand and R2D2 near the 5′ end of the passenger strand.

Purified, recombinant Dcr-2/R2D2 heterodimer alone can also sense the thermodynamic stabilities of the ends of an siRNA duplex. At physiologically relevant concentrations of the proteins (16), photocrosslinking reflected siRNA asymmetry (Fig. 3A). Like heterodimer binding to an siRNA (17), differential photocrosslinking of recombinant Dcr-2/R2D2 heterodimer to an siRNA (18) did not require adenosine triphosphate (ATP). In contrast, formation of the RLC requires ATP (13). The orientation of Dcr-2 and R2D2 on the siRNA duplex was less asymmetric for the recombinant heterodimer than for embryo lysate (compare Figs. 2D and 3A). We propose that siRNA asymmetry is initially sensed by the Dcr-2/R2D2 heterodimer in an ATP-independent manner but is later amplified by the ATP-dependent action of other proteins.

Fig. 3.

The Dcr-2/R2D2 heterodimer alone can sense the asymmetry of an siRNA. (A) Photocrosslinking of recombinant Dcr-2/R2D2 heterodimer to the series of asymmetric siRNAs in Fig. 2C. (B) Structure of the siRNAs used in (C). The siRNAs all contained a single 5′-[32P]phosphate on one strand and either a 5′-hydroxyl or 5′-phosphate group on the other. The siRNA sequences were as in Fig. 2C. (C) R2D2 senses the presence of a 5′-phosphate on the passenger strand. R2D2 photocrosslinking to the 5-iodouracil nearest the 5′ end of the passenger strand was reduced when the 5′ end of the passenger strand contained a 5′-hydroxyl rather than a 5′-phosphate group (siRNAs c and e); photocrosslinking of Dcr-2 was unaltered by the presence or absence of a 5′-phosphate on the guide strand (siRNAs b and f).

Photocrosslinking of R2D2, but not Dcr-2, to the two ends of an siRNA duplex was influenced by the presence of a 5′-phosphate group on the siRNA. We prepared a series of highly asymmetric siRNAs in which the strand containing the p20 5-iodouracil was radiolabeled with 32P at the 5′ end and the other strand contained either a 5′-hydroxyl or 5′-phosphate group (Fig. 3B). In four trials, R2D2 photocrosslinking to the nearby p20 5-iodouracil of the guide strand was greater by a factor of 4.6 ± 0.4 (average ± SD) when the passenger strand contained a 5′-phosphate rather than a hydroxyl group (Fig. 3C, left, siRNAs c and e). R2D2 photocrosslinking in ATP-depleted embryo lysate likewise required a 5′-phosphate at the more thermodynamically stable siRNA end (Fig. 3C, right, siRNAs c and e). Thus, R2D2 can sense two aspects of siRNA structure: the stability of an siRNA 5′ end, and the presence of a 5′-phosphate group. In contrast, Dcr-2 photocrosslinking was unperturbed by a 5′-hydroxyl group on the guide strand, both for the purified protein and in ATP-depleted lysate (Fig. 3C, siRNAs b and f).

Active siRNAs contain 5′-phosphate groups on both strands (3, 11, 1921). A 5′-phosphate on the guide strand is essential for siRNA function, but blocking 5′-phosphorylation of the passenger strand impairs rather than eliminates siRNA activity (11). Our results suggest a molecular explanation for this observation: A 5′-phosphate on the passenger strand enhances R2D2 binding, thereby facilitating efficient incorporation of an siRNA into the RLC and consequently into the RISC. Thus, R2D2 is a licensing factor that ensures that only authentic siRNAs enter the RNAi pathway in Drosophila.

Dcr-2 alone does not efficiently bind siRNA (17), nor can Dcr-2 alone be photocrosslinked to any of the siRNAs in this study (18). Taken together, these results and the data presented here suggest that orientation of the Dcr-2/R2D2 heterodimer is determined largely by R2D2 binding to the siRNA end with the most double-stranded character. This binding is presumably mediated by one or both of the R2D2 double-stranded RNA binding domains. A 5′ mismatch on an siRNA strand may therefore be an antideterminant for R2D2 binding, acting to direct the R2D2 protein to the 5′ end of the passenger strand and positioning Dcr-2 near the 5′ end of the strand to be loaded into the RISC. In this model, R2D2, as a component of the Dcr-2/R2D2 heterodimer, is the primary protein sensor of siRNA thermodynamic asymmetry.

How does the RLC, with the Dcr-2/R2D2 heterodimer positioned asymmetrically on the siRNA, progress to the RISC? Argonaute 2 (Ago2) is a ∼130-kD protein that is a core component of the RISC (22) and is required for siRNA unwinding (14). We found that a ∼130-kD protein was crosslinked to siRNA when the guide strand contained 5-iodouracil at p20 (asterisk in Fig. 2C, siRNAs c, d, e, and g). The ∼130-kD protein was photocrosslinked only to the guide strand of the siRNA (Fig. 4), which suggests that this protein is a component of the RISC. The ∼130-kD protein was immunoprecipitated with antibodies to Ago2 but not to Ago1 (fig. S3A) and was not observed in embryos lacking both maternal and zygotic Ago2 (ago2414, fig. S3B). Thus, the ∼130-kD protein is Ago2. When R2D2 and Ago2 were photocrosslinked to siRNAs b or e (which contain 5-iodouracil at p20 of the passenger or the guide strand), R2D2 was bound to the 3′ end of the guide strand and Dcr-2 to the 3′ end of the passenger strand at early times in the reaction (Fig. 4A). Later, binding of R2D2 and Dcr-2 decreased concurrently, accompanied by a corresponding increase in binding of Ago2 to the 3′ end of the guide strand. In ago2414 lysates, R2D2 binding to the 3′ end of the guide strand and Dcr-2 binding to the 3′ end of the passenger strand did not decrease with time (fig. S4A); this finding suggests that binding of Ago2 facilitates the release of the heterodimer from siRNA.

Fig. 4.

Exchange of R2D2 for Ago2 at the 3′ end of the siRNA guide strand. (A) 32P-radiolabeled siRNAs b and e (Fig. 2B) were incubated with embryo lysate in the presence of ATP for the times indicated, and binding of proteins near the 3′ end of the siRNA guide and passenger strands was monitored by photocrosslinking. Indicated times include the 4 min during which the sample was exposed to ultraviolet irradiation at room temperature. (B) 32P-radiolabeled siRNA c was incubated with embryo lysate and photocrosslinked. The photocrosslinked proteins were then captured with a 2′-O-methyl oligonucleotide complementary to the siRNA guide strand. T, total reaction before incubation with the tethered oligonucleotide; S, supernatant after incubation; B, siRNA-photocrosslinked proteins bound to the tethered oligonucleotide.

The siRNA bound by Ago2 is single-stranded, because Ago2, when photocrosslinked to siRNA, was captured by a tethered 2′-O-methyl oligonucleotide complementary to the siRNA guide strand (Fig. 4B) (23), as has been observed for the RISC (7, 2325). R2D2 was not captured by the 2′-O-methyl oligonucleotide, but was instead recovered in the supernatant, consistent with R2D2 binding of double-stranded siRNA.

Our data suggest a model for RISC assembly. First, R2D2 orients the Dcr-2/R2D2 heterodimer on the siRNA within the RLC. As siRNA unwinding proceeds, the heterodimer is exchanged for Ago2, the core component of the RISC. Indeed, we cannot detect single-stranded siRNA in the RLC assembled in ago2414 lysate (fig. S4, B and C). We hypothesize that unwinding occurs only when Ago2 is available, so that siRNA in the RLC is unwound only when the RISC can be assembled.

Supporting Online Material

Materials and Methods

Table S1

Figs. S1 to S4


References and Notes

View Abstract

Navigate This Article