Structural and functional analyses reveal the contributions of the C- and N-lobes of Argonaute protein to selectivity of RNA target cleavage

Abstract

Some gene transcripts have cellular functions as regulatory noncoding RNAs. For example, ∼23-nucleotide (nt)–long siRNAs are loaded into Argonaute proteins. The resultant ribonucleoprotein assembly, the RNA-induced silencing complex (RISC), cleaves RNAs that are extensively base-paired with the loaded siRNA. To date, base complementarity is recognized as the major determinant of specific target cleavage (or slicing), but little is known about how Argonaute inspects base pairing before cleavage. A hallmark of Argonaute proteins is their bilobal structure, but despite the significance of this structure for curtailing slicing activity against mismatched targets, the molecular mechanism remains elusive. Here, our structural and functional studies of a bilobed yeast Argonaute protein and its isolated catalytic C-terminal lobe (C-lobe) revealed that the C-lobe alone retains almost all properties of bilobed Argonaute: siRNA-duplex loading, passenger cleavage/ejection, and siRNA-dependent RNA cleavage. A 2.1 Å–resolution crystal structure revealed that the catalytic C-lobe mirrors the bilobed Argonaute in terms of guide-RNA recognition and that all requirements for transitioning to the catalytically active conformation reside in the C-lobe. Nevertheless, we found that in the absence of the N-terminal lobe (N-lobe), target RNAs are scanned for complementarity only at positions 5–14 on a 23-nt guide RNA before endonucleolytic cleavage, thereby allowing for some off-target cleavage. Of note, acquisition of an N-lobe expanded the range of the guide RNA strand used for inspecting target complementarity to positions 2–23. These findings offer clues to the evolution of the bilobal structure of catalytically active Argonaute proteins.

Introduction

Some transcripts remain as functional RNAs instead of being translated into protein (1). These noncoding RNAs (ncRNAs)² are classified into two types based on their physiological structures. The first type, including tRNAs, ribosomal RNAs, and riboswitches, folds into unique tertiary structures that are stabilized by employing hydrogen bonds through their 2′-hydroxyl groups. They build characteristic shapes complementary to their cognate substrates, which bolsters their extremely high specificity. Another type of ncRNA, exemplified by CRISPR-targeting RNAs (crRNAs), siRNAs, microRNAs (miRNAs), and Piwi-interacting RNAs (piRNAs), uses a linear single-stranded RNA (ssRNA) region to hybridize with target sequences (2). This type achieves high substrate specificity for DNA and RNA targets by forming ∼23 nucleotides (nt) of RNA-DNA heteroduplexes (or in some cases even longer than 30 base pairs) and double-stranded RNAs (dsRNAs), respectively. However, there is a trade-off between the length of ssRNAs and their specificity. Although their thermal stability may increase with guiding sequence length, their chance of binding to similar sequences also increases due to the possibility of multiple nucleation events occurring independently at different positions (3, 4). To avoid such off-target hybridization, linear ncRNAs are incorporated into large cognate proteins composed of multiple domains. The loaded ncRNAs are preorganized for target recognition within the ribonucleoprotein complex (5–7), yet the molecular basis of how the proteinaceous component inspects the bound target for its complementarity with the loaded linear ncRNA remains to be investigated.

In the case of miRNAs and siRNAs, they alone are not functional and need to be loaded into Argonaute proteins to form effector ribonucleoprotein complexes referred to as RNA-induced silencing complexes (RISCs) (9, 10, 33). These small RNAs serve as guides to take the RISCs to the target mRNAs. Canonical Argonaute proteins seen in both eukaryotic and prokaryotic clades adopt bilobal structures. The N-terminal lobe (N-lobe) includes the N and PAZ domains. The N domain serves as a wedge to pry apart the loaded duplex during passenger ejection (11) while the PAZ domain captures the 3′ end of the guide strand (12). The C-terminal lobe (C-lobe) comprises the MID and PIWI domains. The Rossmann-like MID domain is responsible for recognition of the 5′ end of the guide (30, 32). The PIWI domain adopts an RNase H fold and houses the conserved four residues tasked with endonucleolytically cleaving the targets (13–15). RNase H functions as a stand-alone nuclease that directly recognizes preformed RNA-DNA heteroduplexes and slices the RNA strands (16). Unlike Argonaute, this enzyme does not make use of any auxiliary domains to scrutinize its substrate and facilitate target specificity, which results in nonspecific cleavage of RNAs (17). Given that RNase H is one of the oldest protein folds (18), it is plausible that Argonaute proteins have evolved in an RNase H–centered manner by acquiring each of the above mentioned functional domains. These two lobes are connected through the L2 linker, shaping an intervening nucleic acid-binding channel that accommodates both guide and target strands.

Previous structures of RISC revealed that the loaded guide RNA is exposed at positions 2–6 (g2–g6) to the solvent, whereas the remainder of the guide is sequestered within the channel (8, 19–23). These structural observations proposed a model that the exposed g2–g6 is used to scan target mRNAs. This idea was supported by a single-molecule approach showing the unidirectional propagation of the guide-target duplex from its 5′ side of the guide strand toward its 3′ end (24). It is well-known that catalytically active Argonaute proteins can modulate their slicing activity against mismatched targets (8, 25), and a previous study reported that an N-lobe–deleted construct of fly Ago1 cleaved mismatched targets (26). Meanwhile, a recent phylogenetic study reported that an N-lobe is naturally missing in some prokaryotic Argonaute proteins, such as Archaeoglobus fulgidus Argonaute (AfPIWI) (27). Interestingly, all such unilobed Argonaute proteins are catalytically inactive. These results collectively imply that the bilobal structure is critical for Argonaute proteins to inspect the bound target for complementarity with the guide before cleavage. Despite its significance, the molecular basis for the mismatch sensor of bilobed Argonaute proteins remains unknown. Molecular dissection of a bilobed Argonaute and its C-lobe is essential toward understanding the mechanism of mismatch sensing. However, the loaded guide sews all Argonaute domains and stabilizes the entire ribonucleoprotein complex. Due to such an inseparable relationship between Argonaute and guide, it was difficult to design a stable, catalytically active C-lobe, which has hampered research exploring the mechanisms by which Argonaute proteins optimize the use of linear ncRNAs for target recognition.

In this study, we set out to reveal how the N-lobe enhances the activity of AGO proteins with respect to target specificity. More specifically, we wanted to understand why the absence of the N-lobe allows AGO to slice off-targets and to raise appreciation for the cooperativity between AGO's two lobes for target slicing beyond simple guide-target complementarity. To this end, we successfully designed a competent C-lobe of a yeast Argonaute and investigated its functional capabilities by comparing with its bilobal parent construct. Our structural and functional studies revealed that the bilobal structure is indispensable for catalytically active Argonaute proteins.

Results

N-terminal eukaryote-specific motif is critical for target cleavage

Compared with their prokaryotic counterparts, eukaryotic Argonaute proteins expanded their molecular weight with insertion of specific segments. A previous study revealed an N-terminal fragment that is conserved throughout eukaryotic Argonaute proteins and thus named it conserved segment 1 (cS1) (Fig. 1A) (8). We aligned the corresponding fragment from Argonaute and PIWI clades, revealing that the Argonaute clade possesses a characteristic sequence, RXXXGXXG, whereas the PIWI clade retains an alternative motif, GXXG (Fig. 1B). The crystal structure of Kluyveromyces polysporus Ago1 shows that cS1 reinforces the local structure composed of the L2 linker and the PIWI domain (Fig. S1A) (8), although these elements are far away from one another in the primary sequence (Fig. 1A). A similar network is seen in the crystal structure of silkworm PIWI-clade Argonaute, SIWI, showing that the PIWI clade-specific GXXG motif and a relatively conserved lysine residue, instead of the arginine, harden the local structure (Fig. S1B) (28).

Figure 1.

Domain architecture of Argonaute proteins and contribution of RXXXGXXG motif to substrate binding and cleavage. A, domain architecture of Argonaute proteins and K. polysporus Ago1 variants used in this study showing conserved domains and linkers (conserved segment cS1 (blue), N domain (cyan), L1 linker (yellow), PAZ domain (magenta), L2 linker (gray), MID domain (orange), and PIWI domain (green)). Dashed lines indicate long-range interactions observed in the available crystal structure (PDB code 4F1N). B, highly conserved RXXXGXXG motif across eukaryotic Argonaute proteins showing residues conserved across AGO-clade and PIWI-clade proteins colored in blue and those conserved only within each clade colored in red. C and D, analytical gel-filtration chromatography binding experiments with evaluating AGO236 interaction with a 5′-monophosphorylated 23-nt miR-20a guide RNA (C) or a 5′-hydroxylated 23-nt miR-20a guide RNA (D). E, guide RNA–mediated cleavage assay with AGO207 or AGO236. Guide (red) and target (blue) are shown on the left. The circled phosphate indicates radiolabel. Black arrowhead, cleavage site. Cleaved ³²P-cap–labeled products are resolved from intact substrates by denaturing PAGE (16%). F, nuclease sensitivity of nucleic acid extracted from either AGO207 or KpC-lobe. Mock extraction is shown in the three left lanes. Extracted nucleic acid was dephosphorylated, 5′-end–labeled with ³²P, and treated with either RNase (R) or DNase (D) before resolving on denaturing PAGE (16%) alongside base-hydrolyzed 45-nt polyuridine RNA.

These structural observations prompted us to investigate the significance of cS1. We deleted cS1 from a previously reported construct encompassing K. polysporus Ago1(207–1251) (hereafter referred to as AGO207) that retains the properties of its full-length counterpart (8). The cS1-truncated construct was named AGO236 (Fig. 1A), and the recombinant protein could be purified in its RNA-free form (Fig. S1C). To test its guide-RNA–binding activity, purified AGO236 was incubated with a synthetic single-stranded miR-20a, and their interaction was investigated by gel filtration analysis. AGO236 bound a 5′-monophosphorylated miR-20a but not a 5′-hydroxylated RNA of the same sequence (Fig. 1, C and D), indicating that cS1 is dispensable for monophosphate-dependent guide-RNA binding, a hallmark of Argonaute proteins (29–32). Next, we tested whether AGO236 retains guide-dependent RNA slicing activity. AGO207 and AGO236 were loaded with the synthetic single-stranded miR-20a, followed by incubation with a 60-nt target RNA containing a sequence that is fully complementary to miR-20a. As expected, AGO207 cleaved the cognate target, but AGO236 failed to produce a cleavage product (Fig. 1E), demonstrating that cS1 is indispensable for guide-dependent RNA cleavage. These results suggest that cS1 plays an important role in a step after 5′-end capture of the guide RNA by the MID domain.

cS1 is critical to form a stable ribonucleoprotein complex

A previous study designed an isolated C-lobe of QDE-2 (29) from Neurospora crassa, a homologue of catalytically active Argonaute proteins (34). This C-lobe was composed of only the MID and PIWI domains without the corresponding cS1 (Fig. 1A). Although this earlier study did not investigate slicing activity of QDE-2, the recombinant protein displayed 5′-monophosphate–dependent RNA binding, albeit with lower affinity. Another recent study reported that fly Ago1 C-lobe retained guide-dependent RNA cleavage (Fig. 1A) (26). Based on K. polysporus Ago1, we designed a corresponding C-lobe that encompassed only the 768–1251 and named it AGO768 (Fig. 1A), to determine whether it retains the functional properties of Argonaute proteins. The recombinant AGO768 purified from Escherichia coli weakly bound to nucleic acids and readily degraded during the purification (data not shown). Given that cS1 interacts with the L2 linker and the PIWI domain (Fig. S1A), the composite local structure seems to be critical to form a stable ribonucleoprotein complex. To design a stable C-lobe construct, AGO768 was extended by fusing cS1 and fragments of the N domain and L2 linker (Fig. S1D). The resultant construct was named KpC-lobe (Fig. 1A). Unlike AGO768, the recombinant KpC-lobe purified from E. coli tightly bound to nucleic acids (Fig. S1E), which were resistant to DNase but susceptible to RNase (Fig. 1F), demonstrating that KpC-lobe preferentially bound cellular RNAs over DNAs. The observation that the reinforced C-lobe significantly increases RNA-binding affinity proved that KpC-lobe is capable of forming a ribonucleoprotein complex.

RNAs fortuitously bound to KpC-lobe were arranged as guides

The interaction between KpC-lobe and the co-purified RNAs was too strong to be dissociated during the purification. Such strong RNA binding was reported previously in the preparation of bilobed Argonaute proteins encoded by yeast (8) and human (19–23). We attempted to remove the bound RNA from KpC-lobe by purifying the protein under denaturing conditions; however, the protein failed to properly refold, yielding an insufficient amount of soluble sample for experimentation (data not shown). To see how KpC-lobe recognizes the co-purified RNAs, we crystallized the purified protein with the tightly bound small RNAs and determined the 2.1 Å crystal structure by molecular replacement (Table 1). The asymmetric unit contained four KpC-lobe molecules, and the initial F_o − F_c map showed clear and continuous electron density on each (Fig. S2). Because KpC-lobe co-purified with tightly bound endogenous E. coli RNA, we surmised that the observed electron density reflects a composite signal corresponding to a complex mixture of sequences. Extraction and analysis of the RNA from the crystallized sample validated the presence of RNA in the crystal (Fig. S1F), further suggesting that the observed strong density was derived from the bound RNA. Following previous methods (8, 35), we modeled the bound RNA as 5′-uridine followed by six adenines (Fig. S3) and refined the final structure as KpC-lobe–pUAAAAAA binary complex (Fig. 2A). The backbones of RNAs bound to the four molecules in the asymmetric unit were structurally indistinguishable when the protein chains were superposed, but slight variations can be observed (Fig. S4, A–C, right). Each RNA bound to the different asymmetric KpC-lobe molecules showed similar but discernibly different thermal stability (Fig. S4C, left).

Table 1.

Crystallographic data collection and refinement statistics for KpC-lobe in complex with endogenous E. coli RNA

Data collection
Space group	P2
Cell dimensions
a, b, c (Å)	119.6, 85.6, 127.9
α, β, and γ (°)	90.0, 89.8, 90.0
Resolution (Å)	50.00–2.10 (2.14–2.10)a
Rsym	15.0 (58.4)
I/σ(I)	19.4 (2.9)
Completeness (%)	99.9 (99.4)
Redundancy	3.8 (3.7)
Refinement
Resolution (Å)	47.13–2.10 (2.14–2.10)
No. reflections	150,594
Rwork/Rfree (%)	15.76/20.58 (19.29/25.11)
No. atoms
Protein	16,180
RNA	628
Water	1423
B factors
Protein	32.5
RNA	43.5
Water	40.7
r.m.s. deviations
Bond lengths (Å)	0.008
Bond angles (°)	0.914

^a Values in parentheses are for the highest-resolution shell.

Figure 2.

Crystal structure of KpC-lobe and conformational change to a catalytically active form. A, crystal structure of KpC-lobe shown in a ribbon representation. Domains are colored as in Fig. 1A. Bound RNA is shown as red sticks, and backbone phosphates are colored yellow. B, left, expanded view of the PIWI domain of superposed AGO207 (PDB code 4F1N) and KpC-lobe (PDB code 5THE). AGO207-bound RNA is colored cyan, and KpC-lobe–bound RNA is colored red. Right, expanded view of the PIWI domain of superposed KpC-lobe (green) and RNA-free C-terminal lobe of N. crassa QDE-2 (pink, PDB code 2YHA). C, left, AGO207 MID-PIWI lobe (blue) and KpC-lobe (colored as in Fig. 1A) are superposed on their PIWI domains. Right, NcQDE-2 (pink) and KpC-lobe are superposed on their PIWI domains.

As a result of entirely lacking the N-lobe, KpC-lobe completely exposes its nucleic acid–binding channel that anchors guide nucleotides 1–7 (g1–g7). The g1–g6 were superposed well on those of the AGO207-bound guide RNA (Fig. 2B, Fig. S4D, Table S1), but g7 was arranged differently, which will be discussed later. Thus, KpC-lobe basically mirrors bilobed Argonaute proteins (19–23) in terms of recognition of the seed region of the guide RNA. Although the electron density map of the bound RNA discontinued post-g7 (Fig. S3), the extracted RNAs from the purified KpC-lobe solution and crystalline samples were longer than 7 nt (Fig. 1F and Fig. S1F). These results suggest that the unanchored nucleotides of the bound guide strand are free to move in the bulk solvent of the crystal structure.

Previous structural studies revealed that Argonaute proteins convert into a catalytically active conformation upon incorporation of a guide strand (8, 29, 36). Specifically, a conserved glutamate is known to be rearranged to complete the catalytic tetrad in the RISC (8). The guide-bound structures of KpC-lobe and AGO207 were well-aligned (r.m.s. deviation = 0.38 Å per 451 Cα) (Fig. 2C, left), and both completed the catalytic tetrad (Fig. 2B, left). In addition, the current structure showed that cS1 of KpC-lobe interacts with the L2 linker and the PIWI domain, as does AGO207 (Figs. S1A and S5A) (8). Supporting the significance of cS1 for structural stability, either of the single point mutations of the conserved arginine or glycines compromised solubility (Fig. S5B). On the other hand, superposition of the guide-bound KpC-lobe and the apo-QDE-2 C-lobe, the latter of which lacks cS1 and remains in the inactive conformation (Fig. 2B, right inset) (29), showed different MID-domain arrangements relative to their PIWI domains (Fig. 2C, right). These findings suggest that all requirements for RISC assembly reside in the KpC-lobe and that the transition can be triggered solely by binding of guide RNA.

KpC-lobe retains guide-dependent RNA cleavage activity

A previous study showed that fly Ago1 C-lobe loaded with a 23-nt guide RNA cleaved target RNAs (26), indicating that the N-lobe is dispensable for guide-dependent RNA cleavage by fly Ago1. To test KpC-lobe for slicing activity, the protein was incubated with a single-stranded miR-20a guide, followed by the addition of a cap-labeled 60-nt target RNA harboring a perfectly matched sequence. We were able to detect a robust cleavage product, indicating that KpC-lobe is catalytically competent (Fig. 3A). This result also indicates that although most of the purified KpC-lobe molecules were pre-occupied with the fortuitously bound E. coli-endogenous RNAs (Fig. S1E), a small population remained RNA-free, as reported in the previous studies of eukaryotic bilobed Argonaute proteins (8, 19–23). In addition, single-nucleotide resolution of the endonucleosis products revealed that AGO207 and KpC-lobe both cleaved the target at the same positon (Fig. S6). These findings suggest that hybridization of a target strand with the protein-anchored g2–g7 can autonomously propagate the duplex toward the 3′-end of the guide as long as the two strands are complementary to each other (Fig. 3B). These processes occur entirely independent of the N-lobe.

Figure 3.

Guide-mediated target RNA cleavage by KpC-lobe. A, guide-mediated target cleavage assay using a perfectly complementary guide (red) and target (blue) pair. The circled phosphate indicates radiolabel. Either AGO207 or KpC-lobe was incubated with single-stranded guide RNA, followed by the addition of a ³²P-cap–labeled RNA target. Substrates and products were resolved by denaturing PAGE (16%). B, schematic of target RNA recognition by KpC-lobe and autonomous duplex propagation. Guide positions anchored by the protein are shown in boldface type. Cleavage occurs on the target strand at the t10-t11 phosphodiester bond indicated by a star.

KpC-lobe retains in vitro RNAi activity

Our structural and functional data demonstrate that KpC-lobe is a competent construct in terms of guide-dependent target cleavage when loaded with a single-stranded guide RNA. This prompted us to test KpC-lobe for the other properties of bilobed Argonaute proteins: loading of siRNA duplexes, cleavage and ejection of the passenger strand, and siRNA-dependent target cleavage. First, we investigated whether KpC-lobe can load an siRNA-like duplex and cleave its passenger strand in vitro (Fig. 4A), comparing its activity to that of AGO207 (8). Unlike human and other Argonautes (37–39), budding yeast Argonaute is known to load siRNA duplexes efficiently even in the absence of any chaperone machinery (8). To examine passenger-strand cleavage, a miR-20a siRNA duplex whose passenger strand is 5′-end–labeled was incubated with either AGO207 or KpC-lobe (Fig. 4, A–C). KpC-lobe cleaved the 5′-end–labeled passenger strand at the expected position, as did AGO207 (Fig. 4, D–F). However, in this experiment, cleaved passenger strands were detected by denaturing PAGE, so the experiment is unable to detect whether cleaved passenger strands are released after cleavage or remain in complex with the protein and never actually assemble a functional RISC. To examine whether KpC-lobe can eject the cleaved passenger and subsequently use the retained guide to recognize and cleave targets, we reconstituted in vitro RNAi; the protein was preincubated with an unlabeled miR-20a siRNA duplex, followed by the addition of a cap-labeled target RNA containing a sequence perfectly complementary to the siRNA guide (Fig. 4, G and H). Appearance of a target RNA–derived product would demonstrate that KpC-lobe is able to assemble a RISC and then recognize and cleave its complementary target. Our assay showed that KpC-lobe generated a cleavage product of expected size, diagnostic of RNAi activity in vitro (Fig. 4, I–K). These data offer evidence that the isolated KpC-lobe retains the functional properties of RISC assembly and guide-dependent target cleavage, which define catalytically active Argonaute proteins.

Figure 4.

In vitro RNAi by KpC-lobe. A, schematic of duplex loading and passenger cleavage assay for KpC-lobe. Appearance of a cleavage product demonstrates that KpC-lobe is able to load the siRNA duplex and cleave the passenger strand. Yellow circle, ³²P-end label on passenger strand. B, substrate preparation for passenger-strand cleavage assay. Guide (red) and ³²P-end–labeled passenger strand (green; circled phosphate indicates radiolabel) were annealed and shown to be dsRNA (C) compared with migration of the corresponding single-stranded end-labeled passenger by native PAGE (20%). This demonstrates that the substrate used in the experiment is properly annealed and is double-stranded in solution. D, passenger-strand cleavage assay described in A. Either AGO207 or KpC-lobe was incubated with siRNA, and passenger-strand cleavage was monitored by denaturing PAGE (16%). Quantification of passenger-strand cleavage by AGO207 (E) or KpC-lobe (F) is shown. Black solid lines, average mean passenger cleavage of three replicates, each replicate indicated by gray dots. A dashed line shows that 2.6% percent of the end-labeled passenger strand is single-stranded in solution and was not annealed to the guide (see C). Passenger-strand cleavage percentage is above the 2.6% contamination value, indicating that cleaved passengers are loaded into KpC-lobe as duplexes rather than ssRNA. G, schematic of complete in vitro RNAi pathway for KpC-lobe including duplex loading, passenger cleavage, and target recognition and cleavage. Detection of a cleavage product by denaturing PAGE would demonstrate that KpC-lobe is able to load the duplex, cleave and release the passenger to assemble a functional RISC, and subsequently recognize and cleave the target strand. A yellow circle indicates a ³²P-cap label on target strand. H, unlabeled siRNA preparation for assay as described in G. The same guide and passenger strands shown in B were used except that the unlabeled passenger and guide were annealed and shown to be dsRNA compared with migration of the single-stranded passenger strand by native PAGE (20%). Gel was stained with SYBR Gold stain and visualized by fluorescence. I, evaluation of in vitro RNAi by KpC-lobe. Either AGO207 or KpC-lobe was preincubated with the unlabeled siRNA duplex shown in H to undergo RISC assembly, followed by the addition of a ³²P-cap–labeled RNA target that perfectly matched the guide strand of the siRNA. Products and substrates were resolved by denaturing PAGE (16%). Shown is quantification of target RNA cleaved following RISC assembly by AGO207 (J) or KpC-lobe (K). Black lines, average mean target cleavage of three replicates, with each replicate indicated by gray dots.

N-lobe heightens the sensitivity to guide-target duplex length

The data presented thus far have identified no biochemical distinction between AGO207 and KpC-lobe and indicate that KpC-lobe seems to be a fully functional bona fide Argonaute, which raised the question of what the contribution of the N-lobe is. Previous studies reported that the guide-target duplex adopts a helical conformation closer to the ideal A-form duplex (40, 41). Given that KpC-lobe lacks the N-lobe and loses any corresponding interactions with the 3′ end of the guide, we wondered whether AGO207 and KpC-lobe would display any observable differences in how they utilize their guide RNAs. We began by checking to see whether the length of the guide would affect target cleavage. To determine the minimum length of A-form guide-target duplex required for target cleavage, different lengths of single-stranded miR-20a variants were loaded into either AGO207 or KpC-lobe, followed by the addition of the 60-nt matched target (Fig. 5A). The slicing activity of KpC-lobe was detected with 11-nt or longer guides (Fig. 5, B and D), which led us to speculate that the target can be cleaved once duplex propagation reaches the catalytic site despite not being anchored by the protein in the crystal structure. The activity was dramatically enhanced when the guide was elongated to 14 nt. However, 16- and 23-nt miR-20a variants did not increase the slicer activity any further, suggesting that a 14-nt guide is long enough for the catalytic activation of KpC-lobe. These results indicate that a consecutive 3 nt of base pairing at g12–g14 beyond the cleavage site are essential for sufficient target cleavage by KpC-lobe. In contrast, the slicing activity of AGO207 was enhanced further as the guide strand was elongated from 14 nt to 16 and 23 nt, suggesting that sufficient activation of AGO207 required a longer guide compared with KpC-lobe (Fig. 5, B and C). Therefore, extensive base pairing in the seed, central region, and 3′-supplementary site is required for the catalytic activation of bilobed Argonaute proteins.

Figure 5.

Evaluating the effect of guide RNA length on target cleavage. A, schematic of miR-20a guide RNAs varying in length are shown colored red, and perfectly matched ³²P-cap–labeled target is colored blue. The circled phosphate indicates radiolabel. B, representative gel for cleavage assay guided by different length guides shown in A. Products were resolved from substrates by denaturing PAGE (16%). C and D, guide-mediated cleavage assays using either AGO207 (C) or KpC-lobe (D). Proteins were programmed with guide RNAs shown in A followed by the addition of the cap-labeled target. Bars, average mean of three independent replicates; gray dots, values of each replicate. p values (Student's t test) were calculated for each guide (10, 11, 12, 13, 14, and 16 nt) compared with the 23-nt guide RNA (*, p < 0.0001; **, p < 0.001; ***, p < 0.01).

KpC-lobe cleaves targets using only a middle region of the guide

AGO207 significantly decreased target RNA cleavage if the bases break Watson–Crick pairing to the guide strand at g10 and g11 (t10 and t11 on the target) (Fig. 6, A and B, left) (8), indicating that AGO207 can recognize the mismatches and attenuate its slicing activity. Hur et al. (26) reported that deletion of the N-lobe resulted in the constitutive activation of the fly Ago1 PIWI domain. Using KpC-lobe, we tested whether deletion of the N-lobe also converts yeast Argonaute protein to such a promiscuous slicer. KpC-lobe was able to cleave the mismatched target efficiently (Fig. 6B, right), which was consistent with the reported cleavage activity of fly Ago1 C-lobe (26). These results indicated that the N-lobe is indispensable for modulating target cleavage in response to mismatch(es), another feature of catalytically active Argonaute proteins to avoid promiscuous RNA cleavage.

Figure 6.

Absence of N-lobe drives nonspecific cleavage by KpC-lobe. A, schematic of miR-20a guide (red) with either ³²P-cap–labeled perfectly matched target or a target harboring a dinucleotide mismatch at the nonpermissive t10 and t11 site. The circled phosphate indicates radiolabel. Arrowhead (black), cleavage site (t10-t11). B, cleavage assay evaluating match and mismatch target cleavage by AGO207 and KpC-lobe. Proteins were preincubated with guide RNA, followed by the addition of either the matched or mismatched target shown in A. Products and substrates were resolved by denaturing PAGE (16%). C–E, cleavage assay guided by 14-nt guide RNA either perfectly matched to the target or with systematic dinucleotide mismatches along the guide. Representative gels of assay for AGO207 (C) and KpC-lobe (D) are shown. Products were resolved from substrates by denaturing PAGE (16%). E, guide positions anchored by AGO207 in the crystal structure are colored red and shown in boldface type, and onset of nucleotides not anchored by the protein is colored magenta. Cleavage by AGO207 or KpC-lobe using the variable guides is plotted relative to each protein's target cleavage guided by the perfectly matched 14-nt guide. Error bars, S.D.

How does the acquisition of the N-lobe make Argonaute proteins sensitive to mismatches? It is well-known that the nucleotides of guide RNA have distinct roles depending on their positions from the 5′ end (5, 42–45). Our current structure has shown that KpC-lobe recognizes the g1–g6 as does AGO207 (Fig. 2B). These structural observations prompted us to raise the question whether the role and significance of each nucleotide of guide RNA are different in the presence or absence of the N-lobe. To answer this question, dinucleotide mismatches were systematically incorporated into the 14-nt miR-20a, which was selected based on our earlier observation that it showed comparable cleavage levels as the 23-nt guide loaded into KpC-lobe (Fig. 5D). The influence of the mismatches on target cleavage was compared between KpC-lobe and AGO207. The slicing activity of KpC-lobe was more tolerant to the dinucleotide mismatches at every position than that of AGO207 (Fig. 6, C–E). Especially, the dinucleotide mismatches within the g2–g4 had no effect on the slicing activity of KpC-lobe at all, implying a possibility that KpC-lobe cleaves target RNAs without hybridization with g2–g4. To test this idea, KpC-lobe was loaded with either of 14-nt miR-20a variants that mismatch to the target at g1-g2, g1–g3, g1–g4, g1–g5, or g1–g6, and their impact on cleavage was tested using the same 60-nt target (Fig. 7, A and B). Although the target mismatched to g1-g2, g1–g3, or g1–g4, KpC-lobe cleaved it as efficiently as when loaded with the perfectly matched guide (Fig. 7, A (lanes 10–13) and B). However, KpC-lobe did not efficiently cleave targets that mismatched to g1–g5 and g1–g6 (Fig. 7, A (lanes 14 and 15) and B). These results indicate that KpC-lobe cleaves targets that fully match to g5–g14, suggesting that the lack of an N-lobe shortens the region on the guide that is used to inspect the complementarity with target RNA, before cleavage. Our current structure showed that KpC-lobe anchors only g1–g7, whereas the remainder of the guide is not anchored (Fig. 2B). Thus, in the absence of the N-lobe, even nucleation at the g5–g7 enables the bound target RNA to continuously hybridize with the subsequent unanchored g8–g14 (Fig. 7C). In contrast, the slicing activity of AGO207 significantly decreased in response to contiguous target mismatches at g1-g2 to g1–g6 (Fig. 7, A (lanes 4–8) and B), indicating that AGO207 uses the whole g2–g14 to inspect the complementarity with targets. However, AGO207 cleaved to some extent targets that include a dinucleotide mismatch against g1–g4 (Fig. 6E, left). This observation is consistent with the previous discovery of miRNA-directed cleavage at centered sites that include at least 11 nt of contiguous Watson–Crick base pairing to the central region of miRNA at either g4–g14 or g5–g15 (46).

Figure 7.

KpC-lobe reassigns the position of its seed region in the absence of the N-lobe. A and B, cleavage assay guided by 14-nt guide RNAs with diminished primary seed pairing. The availability of guide positions matching the target was systematically reduced from the 5′-end, and they were assayed for their ability to drive target cleavage. A, representative gel showing products resolved from substrates by denaturing PAGE (16%). Guide naming reflects positions available to match with the target. B, guide-target pairing schematic and quantification of cleavage products shown in A. C, model of target recognition by KpC-lobe using noncanonical primary seed positions when g1–g4 are mismatched. Bases not used for seed pairing are colored in black, and noncanonical primary seed positions used for target recognition are colored in red. Guide nucleotides anchored by KpC-lobe in the crystal structure are shown in boldface type. Onset of nucleotides not anchored by the protein is colored in magenta. D–F, cross-sections (black) of the surface models (white) of AGO207 (D) and KpC-lobe (E) along the nucleic acid–binding channel. The bound guide RNA is colored red. Scissors (yellow) indicate the catalytic site. Inset in D, expanded view of the AGO207 nucleic acid–binding channel. α16 in the L2 linker and α28 in the PIWI domain are involved in the constriction of the nucleic acid–binding channel. F, mismatch sensor model of bilobed Argonaute. Guide positions anchored (red boldface type) are used to nucleate with the target (blue) and drive propagation of the growing duplex within the narrow nucleic acid–binding channel. Extensive complementarity between the guide and target strands results in a robust and rigid duplex that struggles to fit within the nucleic acid–binding channel and creates outward pressure on the N-lobe. Movement of the N-lobe allows the duplex to snugly fit in the channel with the cleavage site on the target strand poised for catalysis by AGO's catalytic residues in the PIWI domain (blue star). Mismatches between the guide and target would not propagate as extensively nor have the rigidity to push the N-lobe, resulting in rejection of the weaker duplex rather than catalysis.

Bilobal structure completes a mismatch sensor

The recent crystal structures of human Argonaute2 in complex with a short target pairing to the g2–g9 showed that the nucleic acid-binding channel is tapered from the 5′ end toward the central region of guide (40). A cross-section of AGO207 uncovered that the nucleic acid-binding channel is constricted between the two lobes (Fig. 7D). In contrast, our current structure revealed that KpC-lobe does not have any constriction due to the lack of the N-lobe (Fig. 7E). As a result, AGO207 positions guide nucleotides 7 and 8 closer to the C-lobe, compared with KpC-lobe (Fig. 2B, left inset). These observations prompted us to hypothesize that the constriction applies pressure on the A-form guide-target duplex from both sides and thus serves as a physical barrier for duplex propagation within the nucleic acid-binding channel. Presumably, only perfectly complementary guide-target duplexes could retain a robust structure during propagation, thereby widening the channel and positioning the target strand properly on the catalytic site (Fig. 7F). This idea was supported by the observation that any dinucleotide mismatches in the post-seed region extremely lowered the slicing activity of AGO207 (Fig. 6, C and E, left). On the other hand, KpC-lobe showed tolerance to 2-nt mismatches in the g8–g11 window to some extent (Fig. 6, D and E, right). The observed difference in their sensitivity to the mismatches in the window of g8–g11 supports our model that the constriction between the two lobes serves as a mismatch sensor.

Acquisition of the N-lobe involves more guide nucleotides in target inspection

Our data showed that KpC-lobe efficiently cleaves targets RNAs when they match to the guide at g5–g14 (Fig. 7, A (lanes 10–13) and B). We tested whether KpC-lobe can sense mismatches and modulate the slicing activity when g5–g14 is the only available pairing site. To this end, KpC-lobe was loaded with a 14-nt miR-20a variant that mismatches at g2–g4, followed by the addition of a 60-nt target that adds additional mismatches at g10-g11 (t10-t11 mismatched target) (Fig. 8A, far left panel). As a result, target pairing only to g5–g9 and g12–g14 avoided cleavage by KpC-lobe (Fig. 8, B and C), indicating that using g5–g14, KpC-lobe could inspect the complementarity of a 10-nt stretch on targets with a high specificity. Next, we expanded the available pairing site from the g5–g14 to g4–g14, g3–g14, or g2–g14 and tested KpC-lobe sensitivity to mismatches at g10-g11 (Fig. 8). KpC-lobe modulated slicing activity against the mismatched target when only g4–g14 was available for hybridization with targets. However, KpC-lobe cleaved the t10-t11 mismatched target when either g3–g14 or g2–g14 was the only available pairing site (Fig. 8). In contrast, AGO207 could attenuate the slicing activity when g2–g23 on the 23-nt miR-20a was fully available for base pairing (Fig. 6B). Given that AGO207 exposes mainly g2–g4 whereas KpC-lobe does all guide nucleotides post-g1, the observed off-target cleavage by KpC-lobe indicates that solvent exposure of the entire seed region results in reduced target specificity. These findings demonstrate that the N-lobe is indispensable to expand the effective pairing site without losing its ability to sense mismatches. Therefore, it is sensible that bilobed Argonaute proteins expose g2–g4 as the primarily available site (primary seed) when scanning for substrates to maximize target specificity before the cleavage.

Figure 8.

Expansion of nucleotides available for pairing drives promiscuous slicing. A, schematic of 14-nt guide RNAs (matched nucleotides (red) and mismatched nucleotides (black)) with diminished complementarity to the target in their primary seed shown pairing with either the perfectly matched target (blue) or mismatches at t10 and t11 (black). B, cleavage assay by KpC-lobe using guides with diminished seed pairing for targeting the matched or t10-t11 mismatched target shown in A. Guide naming reflects positions available to match with the target except for t10 and t11 for the mismatched target. Products were resolved from substrates by denaturing PAGE (16%). C, cleavage of the mismatched target is plotted relative to cleavage of the matched target by the indicated guide RNA. Red bars, average mean of five replicates; gray dots, individual replicates.

Discussion

It is well-known that there is a trade-off relationship between the length of nucleotide hybridization and the sensitivity to mismatches (3). This paradox is a serious problem for nucleases such as Argonaute proteins that make maximum use of the loaded linear ncRNA as a guide to search for the cognate targets. The inspection of base complementarity before target cleavage needs to be quite strict because, physiologically, target cleavage is executed only in special cases, whereas most gene-silencing events occur in a slicer-independent manner (47–51). Our structural and functional studies revealed that KpC-lobe retains all requirements for catalytically active Argonaute proteins with the exception of mismatch sensing (Figs. 3, 4, and 6, A and B). We revealed that the failure in mismatch recognition is attributed to the drastic change of the roles and significance of each guide nucleotide in the absence of the N-lobe (Figs. 5 and 6E). We also discovered that the lack of an N-lobe shortens the length of target nucleotides that can be tested for complementarity with the guide before target cleavage (Fig. 7B). These results indicate that the N-lobe is indispensable to exploit the guide RNA without risking hybridization at numerous, off-target sites (Fig. 9, A and B).

Figure 9.

Plausible model of evolutionary maturation of bilobed AGO proteins. A, bilobal architecture allows AGO to sequentially utilize its guide segments after initial nucleation occurs at the primary seed. Only extensively matched targets are cleaved, whereas mismatches disrupt pairing and AGO avoids target cleavage. The red region of the guide strand contributes to the inspection of target complementarity before cleavage. The pink region of the guide strand does not contribute the target inspection before cleavage. The paired and unpaired nucleotides of the target strand are colored cyan and black, respectively. B, a unilobed AGO-like protein exposes its entire guide to the solvent simultaneously, allowing it to recognize any target as long as some portion of it is complementary, reducing specificity and driving promiscuous activity. The color codes of guide and target nucleotides are the same as in A. The pink region of the guide strand pairs to targets but does not contribute the target inspection before cleavage. C, model of molecular evolution of AGO proteins showing concomitant loss of catalytic machinery and the N-terminal lobe. Bilobed AGO proteins retained their catalytic residues and continued to evolve into both slicer-competent and slicer-deficient types. Gray circle, catalytic site.

Ancestral bilobed Argonaute proteins have undergone a wide array of evolutionary changes that have modified their structure and function from the progenitor prokaryotic Argonaute proteins to the more recent counterparts found in higher eukaryotes (8, 19, 20, 27, 40, 52, 53). A previous phylogenetic analysis suggested an evolutionary pattern whereby bilobed prokaryotic Argonaute may have undertaken several truncation events that gave rise to a class of unilobed prokaryotic Argonaute proteins, such as AfPIWI, a natural C-lobe Argonaute (27). Intriguingly, all unilobed prokaryotic Argonaute proteins, including AfPIWI, lack endonuclease activity (27), but little has been known about why evolution has not allowed Argonaute proteins to convert to the unilobed structure retaining catalytic activity. A reason could be offered by our current study that KpC-lobe failed to extend the effective pairing site on the guide strand and, as a result, indiscriminately cleaved the mismatched target (Fig. 9B). Such a unilobed Argonaute would have become a fatal cellular nuisance because it would promiscuously cleave undesired target RNAs. Therefore, we believe that the N-lobe and Argonaute's cleavage activity have co-evolved together (Fig. 9C). On the other hand, bilobed Argonaute proteins could have continuously evolved as slicers until having their catalytic tetrad discontinued (Fig. 9C). Eventually, this path of evolution may have given rise to bilobed, slicer-deficient Argonaute proteins, such as Rhodobacter sphaeroides Argonaute (54), as well as slicer-deficient human Argonaute paralogues (55, 56).

Our findings demonstrate that bilobed Argonaute proteins exploit the primary seed to scan RNA substrates. This two-step target recognition prevents bilobed Argonaute proteins from RNA cleavage caused by target matching to only g5–g14, unlike KpC-lobe (Figs. 7B (right) and 9). Our results propose that the nucleotides targeted by bilobed Argonaute proteins could be split into two segments, a 3-nt unit and the following ∼20-nt region, which are complementary to the primary seed (g2–g4) and the post-primary seed (g5 onward), respectively (Fig. 10A). This molecular mechanism is reminiscent of the strategy employed by the CRISPR/Cas systems that use crRNAs. crRNA is another type of linear ncRNA that hybridizes with a specific target site with the help of a trans-activating crRNA (57–60). Cas proteins search for a 2–5-nt sequence called a protospacer adjacent motif (PAM) on the target site before the loaded crRNA starts pairing to the target sequence (Fig. 10B) (57, 60). This strategy enables Cas proteins to distinguish exogenous target DNAs from bacterial self-DNAs. To recognize the PAM sequence, for instance, Cas9 adopts a proteinaceous part called the C-terminal domain (Fig. 10B), whereas Argonaute uses the g2–g4 of the loaded linear ncRNA. Both Argonaute and Cas proteins search for a few nucleotides preceding the target sequence. Their similar strategies would allow them to readily release the bound substrates if the subsequent target sequence is not the desired one. Such regulation enables Argonaute and Cas proteins to scan the vast number of substrates efficiently and precisely. Their similar strategies provide an example of convergent functional evolution wherein proteinaceous nucleases incorporate linear ncRNAs as guides to recognize particular nucleic acids.

Figure 10.

Mechanism of target recognition and cleavage by AGO and Cas9. A, RNA guide-bound AGO exposes the primary seed to the solvent to bind to the complementary sequence in the target RNA strand (orange). Nucleation drives propagation of the two strands placing the remainder of the complementary target sequence (cyan) in proximity to AGO's cleavage site (scissors), leading to cleavage of the RNA target. B, Cas9 searches for a 2–5-nt sequence near the target site called the PAM by employing a C-terminal domain (yellow). After finding the PAM, Cas9 uses the loaded crRNA to start pairing to the target of dsDNA, leading to endonucleosis of both strands.

Experimental procedures

Expression and purification of K. polysporus AGO207

K. polysporus Ago1 Thr²⁰⁷–Ile¹²⁵¹ (referred to as AGO207) was expressed and purified as described previously (8). The concentration of purified AGO207 was determined by a Bradford assay (Bio-Rad), and stock aliquots were stored at −80 °C.

Expression and purification of K. polysporus AGO236

The gene for AGO236 was generated by site-directed mutagenesis using the gene for AGO207 as a template to generate a new construct of AGO236 starting with residue lysine 236 just after the conserved RXXXGXXG motif. The polypeptide was overexpressed in E. coli BL21 (DE3) Rosetta2 (Novagen) with an N-terminal Ulp1-cleavable His₆-SUMO tag, which leaves an N-terminal serine remaining from the SUMO tag. Cell extract was prepared by homogenization in Buffer A (10 mm phosphate buffer, pH 7.3, 2 m NaCl, 25 mm imidazole, 10 mm β-mercaptoethanol, 1 mm phenylmethylsulfonyl fluoride) and clarified by centrifugation. The supernatant was loaded onto a nickel column (GE Healthcare), washed with Buffer A, and eluted with a linear gradient to 100% Buffer B (10 mm phosphate buffer, pH 7.3, 1 m NaCl, 750 mm imidazole, 10 mm β-mercaptoethanol). Fractions containing the target protein were mixed with Ulp1 protease and dialyzed overnight against Buffer C (10 mm phosphate buffer, pH 7.3, 500 mm NaCl, 20 mm imidazole, 10 mm β-mercaptoethanol), and the digested protein was loaded onto a nickel column (GE Healthcare) to remove the cleaved His₆-SUMO tag. The flow-through sample containing AGO236 was dialyzed against Buffer D (10 mm phosphate buffer, pH 7.3, 10 mm β-mercaptoethanol), loaded onto a SP column (GE Healthcare), washed with 7% Buffer E (10 mm phosphate buffer, pH 7.3, 2 m NaCl, 10 mm β-mercaptoethanol) to remove loosely bound contaminants, followed by elution from 7% Buffer E to 100% Buffer E over a linear gradient. Fractions containing AGO236 were brought to 0.8 m AmSO₄ by gradual addition for purification by hydrophobic interaction chromatography. The precipitated protein was separated from the soluble population by centrifugation, and the supernatant was loaded onto a HiTrap Phenyl HP column (GE Healthcare) equilibrated with 50% Buffer G (2.5 mm phosphate buffer, pH 7.3, 10 mm β-mercaptoethanol) and 50% Buffer H (10 mm phosphate buffer, pH 7.3, 2 m AmSO₄, 10 mm β-mercaptoethanol). The bound sample was eluted linearly from 50% Buffer H to 20% Buffer H to remove aggregated contaminants, followed by elution from 20% Buffer H to 0% Buffer H to elute the target protein. The sample was dialyzed against 20 mm phosphate buffer, pH 7.3, and 10 mm β-mercaptoethanol followed by concentration by ultrafiltration. The protein was loaded onto a HiLoad 16/600 Superdex 200 column (GE Healthcare) equilibrated with 20 mm phosphate buffer, pH 7.3, and 5 mm DTT. The RNA-free protein concentration was determined by absorbance and stored at −80 °C.

Substrate preparation

A list of RNA and DNA oligonucleotides used in this study is provided (Table S2). 5′-Phosphorylated guide RNAs were chemically synthesized (Dharmacon), deprotected, and gel-purified. The sequences encoding target RNAs were cloned into pUC19 vector and transcribed in vitro using T7 RNA polymerase. DNase-treated transcripts were gel-purified, capped using the ScriptCap m⁷G Capping System (CellScript) either with GTP for unlabeled targets or with [α-³²P]GTP (3000 Ci mmol⁻¹) for cap-labeled target RNAs, and gel-purified again. For passenger-strand cleavage assays, 5′-OH RNA was phosphorylated using OptiKinase (Affymetrix) either with ATP for unlabeled passenger strands or with [γ-³²P]ATP (3000 Ci mmol⁻¹) for 5′-³²P-end–labeled passenger strands. siRNA duplexes were prepared as described previously (8).

Cloning, expression, and purification of KpC-lobe

DNA encoding the designed KpC-lobe from K. polysporus Ago1 was generated by first amplifying the MID-PIWI lobe from K. polysporus Ago1 using Primer Set I (Forward1 (GACATTTTGACAGGTTCAGGTAGAGTACCATCTCGTATTCTAGATGCCCC) and Reverse1 (GCGCGCCTCGAGTCAAATGTAATACATTACGGATTTAATGTTATCG)) by using PrimeSTAR Max DNA polymerase (Takara) followed by a second PCR amplification using Primer Set II (Forward2 (GCGCGCGGATCCATCTATAAAGTTGAAAATAGACATGATTATGGTACTAAAGGTACTAAAGTTGAC) and Reverse2 (GGTACTCTACCTGAACCTGTCAAAATGTCAACTTTAGTACCTTTAGTACCATAATCATGTC)) to fuse two Ago1 fragments, Ile²²¹–Thr²⁴¹ and Arg⁷²⁸–Ile¹²⁵¹ with a Gly-Ser-Gly linker. Following restriction enzyme treatment with BamHI and XhoI, the gene was cloned by using T4 DNA ligase (Roche Applied Science) into a modified pRSF Duet vector (Novagen) containing an N-terminal Ulp1-cleavable His₆-SUMO tag and transformed into E. coli DH5α competent cells. Positive transformants were validated by DNA sequencing, and KpC-lobe was overexpressed in E. coli BL21 (DE3) Rosetta2 (Novagen). Cell extract was prepared by homogenization in Buffer A (10 mm phosphate buffer, pH 7.3, 2 m NaCl, 25 mm imidazole, 10 mm β-mercaptoethanol, 1 mm phenylmethylsulfonyl fluoride) and clarified by centrifugation. The supernatant was loaded onto a nickel column (GE Healthcare), washed with Buffer A, and eluted with a linear gradient to 100% Buffer B (10 mm phosphate buffer, pH 7.3, 1 m NaCl, 750 mm imidazole, 10 mm β-mercaptoethanol). Fractions containing KpC-lobe were mixed with Ulp1 protease and dialyzed overnight against Buffer C (10 mm phosphate buffer, pH 7.3, 500 mm NaCl, 20 mm imidazole, 10 mm β-mercaptoethanol), and the digested protein was loaded onto a nickel column (GE Healthcare) to remove the cleaved His₆-SUMO tag. The flow-through sample containing KpC-lobe was dialyzed against Buffer D (10 mm phosphate buffer, pH 7.3, 10 mm β-mercaptoethanol), loaded onto a SP column (GE Healthcare), and eluted with a linear gradient to 70% Buffer E (10 mm phosphate buffer, pH 7.3, 2 m NaCl, 10 mm β-mercaptoethanol). Fractions containing KpC-lobe were dialyzed against Buffer D, loaded onto a MonoQ column (GE Healthcare), and eluted with a linear gradient to 100% Buffer E. KpC-lobe was again dialyzed against Buffer D and loaded onto a MonoS column (GE Healthcare) and eluted with a linear gradient to 14% Buffer E. The eluted protein was dialyzed against Buffer F (10 mm Tris-HCl, pH 7.5, 200 mm NaCl, 5 mm DTT), concentrated by ultrafiltration, and loaded onto a HiLoad 16/600 Superdex 200 column (GE Healthcare) equilibrated with Buffer F. Purified KpC-lobe was concentrated to ∼40 mg ml⁻¹ measured by Bradford Assay (Bio-Rad), and stored at −80 °C.

Expression of RXXXGXXG point mutants for solubility assay

Point mutations at Arg²²⁷, Gly²³¹, or Gly²³⁴ were introduced by PCR-based mutagenesis to generate vectors encoding mutant KpC-lobe by using PrimeSTAR Max DNA polymerase (Takara) and the following DNA oligonucleotides: R227A Forward, GATTGGTGGATCCATCTATAAAGTTGAAAATGCACATGATTATGGTAC; R227A Reverse, GTCAACTTTAGTACCTTTAGTACCATAATCATGTGCATTTTCAACTTTATAG; G231A Forward, TCCATCTATAAAGTTGAAAATAGACATGATTATGCGACTAAAGGTA; G231A Reverse, CCTGTCAAAATGTCAACTTTAGTACCTTTAGTCGCATAATCATGTCTATTTTC; G234A Forward, CTATAAAGTTGAAAATAGACATGATTATGGTACTAAAGCGACTAAAGT; G234A Reverse, CCTGAACCTGTCAAAATGTCAACTTTAGTCGCTTTAGTACCATAATCATGTC. For each site-directed mutagenesis experiment, 4 pmol of forward and reverse oligonucleotide were mixed with PrimeSTAR Max DNA Polymerase and underwent five PCR cycles under the following protocol: 98 °C for 10 s, 55 °C for 15 s, 72 °C for 2 s. 500 ng of plasmid DNA template encoding the gene for KpC-lobe was added to the mixture, and 30 PCR cycles were performed under the following protocol: 98 °C for 10 s, 55 °C for 15 s, 72 °C for 60 s. DpnI-treated PCR products were transformed into E. coli DH5α competent cells, and positive transformants were validated by DNA sequencing. The mutants were overexpressed in E. coli BL21 (DE3) Rosetta2 (Novagen). After ultrasonication, the cell lysate was centrifuged to separate the soluble fraction from the pellet. The pellet was resuspended in the original volume using Buffer A. Representative samples of the supernatant and pellet for each construct were resolved by SDS-PAGE.

Structure determination and refinement

Crystal trays for KpC-lobe bound to a mixture of endogenous E. coli RNA that co-purifies with the protein during overexpression were set up using 10 mg ml⁻¹ protein. Initial protein crystals were obtained by sitting-drop vapor diffusion at 20 °C in 100 mm sodium citrate, pH 5.5, and 15% PEG 6000 and optimized in 100 mm sodium citrate, pH 5.5, and 18% PEG 4000. Crystals were soaked in collection buffer containing 1.1-fold reservoir buffer and cryoprotected with 25% glycerol. Diffraction data sets were collected at the NE-CAT beamlines (Advanced Photon Source, Chicago) at 0.97918-Å wavelength and processed with HKL2000 (61). Data collection and refinement statistics are listed in Table 1. Initial scaling of the diffraction data showed that the protein obeyed either primitive or orthorhombic crystallographic symmetry based on the scaling statistics in both P2 and P222 space groups (Table S3). To determine the correct space group assignment of the crystal, we scaled the experimental data in all six possible space groups and performed molecular replacement with PHASER (62) on each using the C-terminal lobe of AGO207 (PDB code 4F1N, residues 728–1251) (8) as the search model. Whereas all six space groups yielded a similar model of the protein structure, only the P2 space group showed clear and continuous electron density for the KpC-lobe–bound RNA (Fig. S7A), whereas neither of the possible orthorhombic space groups showed electron density for the bound RNA following molecular replacement phasing. Despite the β angle of the unit cell measuring slightly smaller than conventional 90° for the P2 space group, the presence of electron density for bound RNA only in this space group prompted us to refine the data as P2. Additionally, significant improvement of the initial refinement statistics for the P2 space group over the others further suggested that the RNA-bound KpC-lobe should follow P2 assignment (Fig. S7B). The remainder of the model (Ile²²¹-Thr²⁴¹-Gly-Ser-Gly) and bound nucleic acid were built manually with COOT (63). Following previous methods (8, 35), we modeled the first nucleotide as uridine and the next six as adenine. Then the entire model of the KpC-lobe–pUAAAAAA binary complex was improved with iterative cycles of refinement with Phenix (64). Ramachandran plot analysis was performed by PROCHECK (CCP4)(65) and showed 90.3 and 9.8% of the protein residues in the favored and allowed regions, respectively, with no residues in disallowed regions. Simulated annealing OMIT maps were generated by Phenix (64), B-factors were calculated with BAVERAGE (CCP4), and hydrogen-bond interactions were identified by CONTACT (CCP4) (65). All figures of structures were generated using PyMOL (66) and Chimera (67).

Gel-filtration binding experiments

Analytical gel-filtration chromatography binding experiments were performed in 1× binding buffer (20 mm Tris-HCl, pH 7.5, 150 mm NaCl, and 1 mm DTT). Either 5′-phosphorylated or 5′-hydroxylated miR-20a single-stranded RNA guide (2.5 μm), AGO236 (5 μm), or a mixture of RNA and protein was mixed in 1× binding buffer on ice for 30 min in a 200-μl reaction before loading to a Superdex 200 Increase 10/300 GL, 24-ml column (GE Healthcare) equilibrated with 1× binding buffer.

Cleavage assays

For all biochemical assays, stock AGO207 and KpC-lobe were diluted and stored in dilution buffer (Buffer F + 0.5 mg ml⁻¹ Ultrapure BSA (Ambion)) at −80 °C. All assays were performed in 1× reaction buffer (20 mm Tris-HCl, pH 7.5, 150 mm NaCl, 1 mm MgCl₂, 1 mm DTT, 5% glycerol), 0.05 mg ml⁻¹ BSA (Ambion), and 4 units of RiboLock RNase inhibitor (Thermo Scientific). For all guide-mediated cleavage assays, a 1 μm concentration of either AGO207 or KpC-lobe was mixed with 50 nm guide RNA and incubated at 25 °C for 30 min to form the RISC. For the cleavage assay shown in Fig. 1E, a gradient of protein concentration of AGO236 was used (2.5–10 μm). Cleavage was initiated by adding 1 μl of 5′-capped target RNA (final concentration, 25 nm) with trace amounts of ³²P-cap–labeled target in a 10-μl reaction and incubated at 30 °C for 20 min before quenching with 10 μl of formamide loading buffer (95% formamide, 18 mm EDTA, 0.025% SDS, 0.025% bromphenol blue, 0.025% xylene cyanol). For time-course reactions, 10-μl reactions were prepared similarly, except 3-μl aliquots were removed at the indicated time points and quenched by the addition of 10 μl of formamide loading buffer. For passenger-strand cleavage assays (shown in Fig. 4D), 1 μl of 10 nm ³²P-passenger-strand–labeled siRNA duplex was added to a 9-μl mixture containing 1× reaction buffer and a 1 μm concentration of either AGO207 or KpC-lobe. Reactions were quenched at the indicated time points by the addition of formamide loading buffer. For siRNA-mediated target cleavage (shown in Fig. 4I), a 1 μm concentration of either AGO207 or KpC-lobe was preincubated with an unlabeled siRNA duplex at 30 °C for 30 min to allow for passenger-strand cleavage and RISC maturation, followed by the addition of 10 nm cap-labeled target RNA. Reactions were quenched at the indicated time points by the addition of formamide loading buffer. For all guide-mediated cleavage assays, AGO207 or KpC-lobe was preincubated with a single-stranded synthetic guide RNA at 25 °C for 30 min before the addition of cap-labeled target RNAs at 30 °C for 20 min. Reactions were quenched with formamide loading dye, resolved by 16% denaturing PAGE. All gels were dried and visualized by phosphorimaging following overnight exposure. Data were quantified by densitometry using ImageQuantTL software (GE Healthcare).

All cleavage percentages were calculated using the equations listed below and averaged over three independent experiments.

Cleavage by k-nt guide was calculated using the following equation.

$$P_k=100\times \left(C_k/\left(C_k+U_k\right)\right)$$ (1)

where C_k and U_k are the intensities of the cleaved and uncleaved bands, respectively.

The relative cleavage by k-nt guide was calculated using the following equation,

$$R_k=100\times \left(C_k/\left(C_k+U_k\right)\right)/\left(C_{23}/\left(C_{23}+U_{23}\right)\right)$$ (2)

The relative cleavage by guides with variable sequences was calculated using the following equation,

$$R{\prime }_k=100\times \left(C_{\text{var}}/\left(C_{\text{var}}+U_{\text{var}}\right)\right)/\left(C_{14}/\left(C_{14}+U_{14}\right)\right)$$ (3)

where C_var and U_var are the intensities of the cleaved and uncleaved bands when using guides with variable sequences, respectively.

The relative cleavage percentage of the t10-t11 mismatch target was calculated using the following equation,

$$R″=100\times \left(C_{\text{mis}}/\left(C_{\text{mis}}+U_{\text{mis}}\right)\right)/\left(C_{\text{match}}/\left(C_{\text{match}}+U_{\text{match}}\right)\right)$$ (4)

where C_mis and U_mis are the intensities of the cleaved and uncleaved bands derived from the t10-t11 mismatch target, respectively, whereas C_match and U_match are the intensities of the cleaved and uncleaved bands derived from the match target, respectively.

Analysis of co-purifying nucleic acid

Polynucleotides were extracted from either AGO207, KpC-lobe, or water (for mock) by phenol/chloroform and dephosphorylated with alkaline phosphatase (Roche Applied Science) by incubation at 37 °C for 30 min. Reactions were quenched by the addition of EDTA to a final concentration of 10 mm followed by inactivation of phosphatase by incubation at 70 °C for 30 min. Before 5′-labeling, samples were supplemented with 10 mm MgCl₂. 5′-End–labeling reactions were performed in a 30-μl reaction containing 3 μl of heat-inactivated dephosphorylation reaction, 3 μl of 10× OptiKinase buffer (USB), 2 μl of OptiKinase (USB), and 0.5 μl of [γ-³²P]ATP (3000 Ci mmol⁻¹). End-labeling reactions were incubated at 37 °C for 40 min before aliquotting into three equal volumes and treating with either RNase A (USB), RQ-1 RNase-free DNase I (Promega), or neither for 20 min at 37 °C. Samples were resolved by 16% denaturing PAGE alongside a base-hydrolyzed 45-nt polyuridine ladder. Gels were visualized by phosphorimaging (Typhoon, GE Healthcare). Analysis of RNA that crystalized with KpC-lobe was performed similarly, except RNA was extracted from ∼30 crystals. Each crystal was individually harvested, rinsed in crystallization buffer (100 mm sodium citrate, pH 5.5, and 18% PEG 4000) three times, and then dissolved in water before dephosphorylation and end-labeling.

Author contributions

D. M. D., B. C. K., and K. N. formal analysis; D. M. D. and K. N. writing-original draft; K. N. conceptualization; K. N. supervision; K. N. funding acquisition; K. N. validation; K. N. project administration; K. N. writing-review and editing.