Structural basis for overhang-specific small interfering RNA recognition by the PAZ domain

siRNA is a 19–23-base-paired (bp) duplex with 2-nucleotide (nt) 3′ overhangs at both ends, containing 5′-phosphates and 3′-hydroxyls^7,15. siRNA is not merely a consequence of Dicer processing, as both the length and ends are important for mediating RNA interference (RNAi)^7,10,15, whereby target messenger RNA is sequence-specifically degraded in an siRNA-programmed effector complex termed the RNA-induced silencing complex (RISC)^8,9,16. This suggests that specific structural features of siRNA constitute recognition targets for the protein components within the RNAi machinery. The PAZ domain can bind to single-stranded (ss) RNAs and siRNA duplexes^12–14, and requires the 2-nt 3′ overhang for efficient complex formation^13,14. We have determined a co-crystal structure of PAZ domain and a 9-mer RNA (5′-CGUGACUCU-3′) in order to illustrate the details of PAZ–RNA interaction and gain functional insights into the recognition process.

The structure of the PAZ–RNA complex reveals an unanticipated arrangement in which the 9-mer RNA, initially designed as a ssRNA ligand, forms a self-complementary siRNA-like A-form duplex, which is bound by the PAZ domain at each end (Fig. 1). The 3′-shifted pairing of each strand results in 2-nt, single-stranded 3′ overhangs at the duplex ends, which are characteristic of siRNA architecture. However, the short 7-bp duplex is unstable, due to the presence of three non-canonical pairs (Fig. 1a), raising the concern that the observed duplex formation could result from crystal-packing interactions. This is not the case, as the PAZ domain and the 9-mer RNA form a stable 2:2 complex in solution, as judged by gel filtration (Supplementary Fig. S1). Thus, PAZ binding apparently stabilizes siRNA-like duplex formation. Moreover, the absence of contacts between the two PAZ domains in the complex indicates independent binding of PAZ domains to the siRNA-like duplex, and suggests that such an arrangement should also hold for complex formation to the ends of a typical 21-nt siRNA.

The PAZ domain in the complex adopts a heart-shaped globular topology (Fig. 2a), with a twisted β-barrel consisting of six β-strands (β1–β3, β6–β8), capped by two amino-terminal α-helices (α1, α2) on one side and connected to αβ module (β4–β5–α3) on the other side. The two strands of the siRNA-like duplex interact with a specific PAZ domain in a highly asymmetric manner. The strand bound with its 3′ end contacts the PAZ domain along its full 9-nt length, whereas the complementary strand makes contacts only with the 5′-terminal residue. The first 7 nucleotides constituting the RNA duplex structure, proceeding in the 5′ to 3′ direction, interact with the protein's carboxy-terminal tail and the positively charged surface formed by the strands β2, β3 and the β6–β7 loop (Fig. 2b). The bound RNA adopts a stacked 5′ to 3′ helical trajectory, except for a sharp turn (clockwise rotation of ~110° along the helical axis) in the phosphodiester backbone between the duplex and 2-nt 3′-overhang segments, thereby inserting the 2-nt ends into a central protein pocket formed between the barrel and αβ module. The same central RNA-binding pocket has been proposed previously from an evaluation of NMR chemical shift perturbations, mutagenesis and sequence conservation analysis^12–14. The free^12–14 and RNA-bound PAZ structures superimpose well around the RNA-binding pocket (Supplementary Fig. S2), indicating that the pocket is largely preformed.

The elongated pocket, which is lined by a large number of aromatic and hydrophobic residues, is generated by helix α3 and strand β4 along one face, and strand β7 and loop β2–β3 along the opposite face (Fig. 2a). The two nucleotides at the 3′ end retain a stacked A-form conformation (Fig. 3a), with the 3′-terminal residue buried deep within the pocket and targeted by numerous protein interactions (Fig. 3b). The non-bridging oxygens of the phosphodiester backbone linking the last two residues make four hydrogen bonds with Y309, Y314 and H269 and a water molecule stabilized by Y277. The sugar ring of the terminal residue is further anchored in place by van der Waals packing against L337 and T335, whereas its 2′- and 3′-hydroxyl groups form hydrogen bonds to the backbone amide and carbonyl groups of Y336, respectively. In addition, portions of the sugar ring and base of the terminal nucleotide are stacked over aromatic residue F292.

Among residues recognizing the 3′-terminal nucleotide, F292, Y309 and L337 are invariant, and H269, Y277, K313 and Y314 are conserved among PAZ homologues (Supplementary Fig. S3). These residues seem to be critical for efficient binding, as single mutations F292A and L337A cause a 615-fold and 155-fold reduction in binding affinity, and double mutation Y309F/Y314F causes a 46-fold reduction (Fig. 4a). Other conserved residues surrounding the pocket, such as V306, F310, P338 and V341, although not directly contacting the RNA, probably maintain the hydrophobic environment of the pocket. The strong binding of the 3′-terminal nucleotide, as reflected in numerous intermolecular contacts, explains the previous observation that single nucleotide uridine 5′ monophosphate (UMP) can also bind in the same pocket¹². By contrast, the penultimate nucleotide from the 3′ end makes only minimal contacts with PAZ—that is, a hydrogen bond between its 2′-hydroxyl group and Y277, and an electrostatic interaction with K313—and is largely exposed to solvent.

The base edges of the 2-nt 3′-overhang residues are facing outwards towards solvent and have no contact with the protein. The absence of specific hydrogen-bonding recognition of base edges as well as the available space along the base edges suggest that the pocket could accommodate 3′-overhang segments containing all nucleotide combinations. This feature is consistent with the potential function of PAZ, which is to recognize siRNAs of random sequence.

The PAZ domain laterally secures the siRNA-like duplex through hydrogen-bond and electrostatic interactions between five phosphodiester groups of the strand bound at the 3′ end, and the basic residues R275, K333, R278, K264 and the C-terminal residues (Fig. 3c, d). The side chain of R278, which is buttressed in place by G262, forms two hydrogen bonds with adjacent phosphodiester groups (Fig. 3d). The importance of this recognition is manifested in the 17-fold loss in binding affinity observed for the R278A mutant (Fig. 4a).

No specific interactions are directed to the 2′-OH groups, which line the minor groove of the RNA helix, consistent with RNAi not generally being affected by 2′-OH group modifications within the siRNA duplex segment¹⁷. Nevertheless, the observed contacts between the PAZ domain and adjacent non-bridging phosphate oxygens (Fig. 3d) may provide a mechanism for discriminating the interphosphate separations, which are different between RNA and DNA helices. The PAZ domain also does not contact base edges within either groove of the RNA helix, indicative of non-sequence-specific molecular recognition.

The 5′ end of the complementary strand is capped by loop β2–β3 centred about residue M273 (Fig. 3c). This docking constitutes the only interaction involving the complementary strand, burying 100Ų of the protein surface, in contrast to 900Ų of total buried protein surface, upon complex formation. The aliphatic side chain of M273 extends towards the U7 base (Fig. 3c), but seems not to contribute to RNA binding, as no loss of binding was observed for mutant M273A (Fig. 4a). Furthermore, the PAZ domain does not contact the 5′-OH (and presumably 5′-phosphate, were it to be present) of the complementary strand.

To define more accurately the RNA structural determinants for PAZ recognition, we measured apparent dissociation constants (K_d) for RNA ligands with variant 3′ ends, using surface plasmon resonance (Fig. 4b). RNAs containing both UU and AG 3′ overhangs bind PAZ with the highest affinities (K_d values of 0.89–2.18 nM) among all tested constructs (Fig. 4b). The overhang contributes critically to the binding, as removal of one or two 3′-nucleotides reduces the binding affinity by 85- or >5,000-fold, consistent with the crystal structure of the complex, in which the PAZ domain optimally fits the duplex end with 2-nt 3′ overhang in a sequence-independent manner.

In addition, the pocket favours the 2-nt overhang as ribonucleotides, because replacement with deoxyribonucleotides, which eliminates the two hydrogen bonds to the 2′-OH groups of the overhang segment, reduces the binding by 100-fold (Fig. 4c). The pocket is incompatible with bulky 3′-end modifications, such as propanediol, fluorescein and puromycin, which decrease affinity by >10⁴-fold (Fig. 4c), but is fairly tolerant towards accommodation of a smaller 2′-o-methyl group at the terminal nucleotide (Fig. 4c).

The PAZ domain does not offer a strong binding site for the RNA 5′ end, because elongation of the 3′ overhang to 4 nt, which would disrupt the protein interaction with the 5′-terminal residue of the complementary strand, only slightly decreases the binding affinity by 5-fold (Fig. 4b). No binding occurs for ssRNA with a free 5′ end (Fig. 4c), reinforcing this conclusion.

The PAZ domain is able to bind with the 10-nt overhang, in essence an ssRNAwith a free 3′ end, with a K_d of ~100 nM (Fig. 4b). Because of the predominant interaction of PAZ with the 3′-end-bound strand, the ssRNA might conceivably bind the pocket of PAZ by its 3′ end and adopt an extended conformation along the β-barrel surface. The 50-fold reduction in affinity for ssRNA (10-nt overhang), relative to the siRNA end (2-nt overhang), may result from the lack of interaction with the 5′-terminal residue of the complementary strand and the increased flexibility of ssRNA due to the absence of duplex pairing. Together, the structure of the complex and the binding assays establish that the PAZ domain binds to the 3′ end of RNA, requiring an essential 2-nt, single-stranded segment, and has a preference for the remaining RNA in the duplex over the single-stranded form.

The PAZ domain adopts a topology similar to the oligonucleotide/oligosaccharide-binding (OB) fold, despite different connections linking secondary structure elements^13,14. The 2-nt overhangs are anchored within a pocket unique to the PAZ domain, with the remainder of the RNA positioned along a surface formed by the β2–β3 strands. This contrasts with OB fold RNA recognition, where the RNA targets the opposite face of the β-barrel¹⁸, corresponding to β6–β7 strands of the PAZ domain.

The conservation among PAZ homologues, especially for the RNA-binding residues (Supplementary Fig. S3), predicts that all PAZ family members should similarly recognize siRNA ends, satisfying their exclusive distribution in the RNAi-related proteins Dicer and Argonaute. As an siRNA binder, Dicer–PAZ could hold the product siRNA at one end, after dsRNA cleavage, and transfer the other siRNA end to the PAZ domain of Argonaute in the RISC. Such a transfer would lead to a direct loading of siRNA into RISC, consistent with a previous observation that the activities of Dicer and RISC are coupled through transfer of siRNA⁷. Because some Dicer proteins, such as Drosophila Dicer-2, do not contain a PAZ domain, alternative pathways are likely to exist, that mediate transfer of siRNA from Dicer to RISC. Drosophila R2D2, which binds to Dicer-2, has been proposed to facilitate loading of siRNA into RISC¹⁹.

In view of the predominant interaction of PAZ with one strand of siRNA, and the associated residual binding to ssRNA, the PAZ domain might continue to secure the guide strand following siRNA unwinding, and present it for scanning and pairing with the target sequence in the activated RISC.

PAZ–siRNA recognition seems to provide the molecular basis for previous observations that RNAi efficiency was reduced when the 3′ overhangs of synthetic siRNAs deviated from 2 nt in length^7,15, or when they were substituted by deoxyribonucleotide residues²⁰, or chemically modified at the 3′-OH position²¹. The PAZ domain represents the best candidate to sense these 3′-overhang modifications, which would result in reduced binding as shown experimentally in Fig. 4b, c, and consequently might impair the PAZ-mediated recruitment of siRNA into the RISC. By contrast, many other studies have reported that 3′ modifications of siRNA^11,22–24 and commonly used deoxynucleotide-substituted 3′ overhangs¹⁵ do not interfere with RNAi, thereby questioning the importance of the PAZ–siRNA interaction in the RNAi pathway. One explanation is that PAZ may not be the only siRNA recognition domain mediating RNAi, with its contribution potentially compensated for by alternative high-affinity binding events during siRNA recruitment. Blunt-end siRNA can induce RNAi in some cases^15,25, probably by also by-passing PAZ–siRNA recognition. In addition, the nuclease stability of chemically modified siRNAs may partially compensate for impaired PAZ recognition. Further studies are needed to understand the full diversity of siRNA recognition events in the RNA silencing pathway.