Crystal Structure of Cas9 in Complex with Guide RNA and Target DNA

Overall structure of the Cas9–sgRNA–DNA ternary complex

We solved the crystal structure of full-length S. pyogenes Cas9 (residues 1–1368; D10A/C80L/C574E/H840A), in complex with a 98-nt sgRNA and a 23-nt target DNA, at 2.5 Å resolution by the SAD (single-wavelength anomalous dispersion) method, using a SeMet-labeled protein (Figure 1, Figure S1 and Table S1). To improve the solution behavior of Cas9, we replaced two less conserved cysteine residues (Cys80 and Cys574) with leucine and glutamic acid, respectively. This C80L/C574E mutant retained the ability to efficiently cleave genomic DNA in human embryonic kidney 293FT (HEK293FT) cells, confirming that these mutations have no effects on the Cas9 nuclease function (Figure S2). Additionally, to prevent target DNA cleavage during crystallization, we replaced two catalytic residues, Asp10 from the RuvC domain and His840 from the HNH domain, with alanines.

The crystallographic asymmetric unit contained two Cas9–sgRNA–DNA ternary complexes (Mol A and Mol B). Although there are conformational differences between the two complexes, the sgRNA and the DNA are recognized by Cas9 in similar manners. Most notably, while the HNH domain in Mol A is connected to the RuvC domain by a disordered linker, the HNH domain in Mol B is not visible in the electron density map, indicating the flexible nature of the HNH domain. Thus, we will first describe the structural features of Mol A unless otherwise stated, and then discuss the structural differences between the two complexes, which suggest the conformational flexibility of Cas9.

The crystal structure revealed that Cas9 consists of two lobes, a recognition (REC) lobe and a nuclease (NUC) lobe (Figures 1A–D). The REC lobe can be divided into three regions, a long α-helix referred to as the Bridge helix (residues 60–93), the REC1 (residues 94–179 and 308–713) domain, and the REC2 (residues 180–307) domain (Figures 1A–D). The NUC lobe consists of the RuvC (residues 1–59, 718–769 and 909–1098), HNH (residues 775–908), and PAM-interacting (PI) (residues 1099–1368) domains (Figures 1A–D). The negatively-charged sgRNA:target DNA heteroduplex is accommodated in a positively-charged groove at the interface between the REC and NUC lobes (Figure 1E). In the NUC lobe, the RuvC domain is assembled from the three split RuvC motifs (RuvC I–III), and interfaces with the PI domain to form a positively-charged surface that interacts with the 3′ tail of the sgRNA (Figure 1E). The HNH domain lies in between the RuvC II–III motifs and forms only a few contacts with the rest of the protein.

The REC lobe interacts with the repeat:anti-repeat duplex

The REC lobe includes the REC1 and REC2 domains. REC1 adopts an elongated, α-helical structure comprising 25 α-helices (α2–α5 and α12–α32) and two β-sheets (β6 and β10, and β7–β9), whereas REC2 adopts a six-helix bundle structure (α6–α11) (Figure 2A and Figure S3). A Dali search (Holm and Rosenstrom, 2010) revealed that the REC lobe does not share structural similarity with other known proteins, indicating that it is a Cas9-specific functional domain.

The REC lobe is one of the least conserved regions across the three Cas9 families within the Type II CRISPR system (IIA, IIB and IIC), and many Cas9 orthologs contain significantly shorter REC lobes (Figures S4, S5). In the present structure, the REC2 domain does not contact the bound guide:target heteroduplex. We thus hypothesized that truncations in the REC lobe could be tolerated. As expected, a Cas9 mutant lacking the REC2 domain (Δ175–307) retained ~50% of the wild-type Cas9 activity (Figure 2B), indicating that the REC2 domain is not critical for DNA cleavage. The lower cleavage efficiency may be attributed in part to the reduced expression levels of the Δ175–307 mutant relative to that of the wild-type protein (Figure 2C). In striking contrast, the deletion of either the repeat-interacting region (Δ97–150) or the anti-repeat-interacting region (Δ312–409) of the REC1 domain abolished the DNA cleavage activity (Figure 2B), indicating that the recognition of the repeat:anti-repeat duplex by the REC1 domain is critical for the Cas9 function.

The PAM-interacting (PI) domain confers PAM specificity

The NUC lobe contains the PI domain, which forms an elongated structure comprising seven α-helices (α46–α52), a three-stranded antiparallel β-sheet (β18–β20), a five-stranded antiparallel β-sheet (β21–β23, β26 and β27), and a two-stranded antiparallel β-sheet (β24 and β25) (Figure 2D and Figure S3). Similar to the REC lobe, the PI domain also adopts a novel protein fold unique to the Cas9 family.

The locations of the bound complementary DNA strand and the active site of the RuvC domain in the present structure suggested that the PI domain is positioned to recognize the PAM sequence on the non-complementary DNA strand. We tested whether the replacement of the PI domain of S. pyogenes Cas9 (SpCas9; Cas9 in this study) with that of an orthologous Cas9 protein, recognizing a different PAM, would be sufficient to alter their PAM specificities. The Streptococcus thermophilus CRISPR-3 Cas9 (St3Cas9) shares ~60% sequence identity with SpCas9. While their crRNA repeats and tracrRNAs are interchangeable (Fonfara et al., 2013), SpCas9 and St3Cas9 require different PAM sequences (5′-NGG-3′ for SpCas9 and 5′-NGGNG-3′ for St3Cas9) for target DNA cleavage (Fonfara et al., 2013).

We swapped their PI domains to generate two chimeras, Sp-St3Cas9 (SpCas9 with the PI domain of St3Cas9) and St3-SpCas9 (St3Cas9 with the PI domain of SpCas9), and examined their cleavage activities for target DNA sequences bearing 5′-NGG-3′ PAM (5′-GGGCT-3′) or 5′-NGGNG-3′ PAM (5′-GGGCG-3′) (Figure 2E). SpCas9 and St3-SpCas9, but not St3Cas9, cleaved the target DNA with 5′-NGG-3′ PAM (Figure 2E), indicating that the PI domain of SpCas9 is required for the recognition of 5′-NGG-3′ PAM and is sufficient to alter the PAM specificity of St3Cas9. Sp-St3Cas9 retained the cleavage activity for the target DNA with 5′-NGG-3′ PAM, albeit at a lower level than that of SpCas9 (Figure 2E). Additionally, the deletion of the PI domain (Δ1099–1368) abolished the cleavage activity (Figure 2E), indicating that the PI domain is critical for the Cas9 function. These results revealed that the PI domain is a major determinant of the PAM specificity.

The RuvC domain has an RNase H fold

The RuvC domain consists of a six-stranded mixed β-sheet (β1, β2, β5, β11, β14 and β17) flanked by α-helices (α33, α34 and α39–α45) and two additional two-stranded antiparallel β-sheets (β3/β4 and β15/β16) (Figure 3A and Figure S3). It shares structural similarity with the retroviral integrase superfamily members characterized by an RNase H fold, such as Escherichia coli RuvC (Ariyoshi et al., 1994) (PDB code 1HJR, 14% identity, root-mean-square deviation (rmsd) of 3.6 Å for 126 equivalent Cα atoms) and Thermus thermophilus RuvC (Gorecka et al., 2013) (PDB code 4LD0, 12% identity, rmsd of 3.4 Å for 131 equivalent Cα atoms) (Figure 3B). The RuvC nucleases have four catalytic residues (e.g., Asp7, Glu70, His143 and Asp146 in T. thermophilus RuvC), and cleave Holliday junctions through a two-metal mechanism (Ariyoshi et al., 1994; Chen et al., 2013; Gorecka et al., 2013). Asp10 (Ala), Glu762, His983 and Asp986 of the Cas9 RuvC domain are located at positions similar to those of the catalytic residues of T. thermophilus RuvC (Figures 3A, B), consistent with previous results showing that Asp10 is critical for the cleavage of the non-complementary DNA strand, and that Cas9 requires Mg²⁺ ions for the cleavage activity (Gasiunas et al., 2012; Jinek et al., 2012). Moreover, the alanine substitution of Glu762, His983 or Asp986 also converted Cas9 into a nickase (Figures 3C, D). Each nickase mutant was able to facilitate targeted double strand breaks using a pair of juxtaposed sgRNAs (Figures 3C, D, Table S2), as previously demonstrated with the D10A nickase (Ran et al., 2013). This combination of structural observations and mutational analyses suggested that the Cas9 RuvC domain cleaves the non-complementary strand of the target DNA through the two-metal mechanism, as previously observed for other retroviral integrase superfamily nucleases.

It is important to note that there are key structural dissimilarities between the Cas9 RuvC domain and the RuvC nucleases, which explain their functional differences. Unlike the Cas9 RuvC domain, the RuvC nucleases form dimers and recognize Holliday junctions (Gorecka et al., 2013) (Figure 3B). In addition to the conserved RNase H fold, the Cas9 RuvC domain has other structural elements involved in interactions with the guide:target heteroduplex (an end-capping loop between α42 and α43) and the PI domain/stem loop 3 (β-hairpin formed by β3 and β4) (Figure 3A).

The HNH domain has a ββα-metal fold

The HNH domain comprises a two-stranded antiparallel β-sheet (β12 and β13) flanked by four α-helices (α35–α38) (Figure 3E). It shares structural similarity with the HNH endonucleases characterized by a ββα-metal fold, such as phage T4 endonuclease VII (Endo VII) (Biertümpfel et al., 2007) (PDB code 2QNC, 20% identity, rmsd of 2.7 Å for 61 equivalent Cα atoms) and Vibrio vulnificus nuclease (Li et al., 2003) (PDB code 1OUP, 8% identity, rmsd of 2.7 Å for 77 equivalent Cα atoms) (Figure 3F). HNH nucleases have three catalytic residues (e.g., Asp40, His41, and Asn62 in Endo VII), and cleave nucleic acid substrates through a single-metal mechanism (Biertümpfel et al., 2007; Li et al., 2003). In the structure of the Endo VII N62D mutant in complex with a Holliday junction, a Mg²⁺ ion is coordinated by Asp40, Asp62, and the oxygen atoms of the scissile phosphate group of the substrate, while His41 acts as a general base to activate a water molecule for catalysis (Figure 3F). Asp839, His840, and Asn863 of the Cas9 HNH domain correspond to Asp40, His41, and Asn62 of Endo VII, respectively (Figure 3E), consistent with the observation that His840 is critical for the cleavage of the complementary DNA strand (Gasiunas et al., 2012; Jinek et al., 2012). The N863A mutant functions as a nickase (Figure 3C, D), indicating that Asn863 participates in catalysis. These observations suggested that the Cas9 HNH domain may cleave the complementary strand of the target DNA through a single-metal mechanism, as observed for other HNH superfamily nucleases. However, in the present structure, Asn863 of Cas9 is located at a different position from that of Asn62 in Endo VII, whereas Asp839 and His840 (Ala) of Cas9 are located at positions similar to those of Asp40 and His41 in Endo VII, respectively (Figures 3G). This might be due to the absence of divalent ions, such as Mg²⁺, in our crystallization solution, and Asn863 may point towards the active site and participate in catalysis. Although the HNH domain shares a ββα-metal fold with other HNN endonucleases, their overall structures are distinct (Figures 3E, F), consistent with the differences in their substrate specificities.

The sgRNA:DNA complex adopts a T-shaped architecture

The sgRNA consists of crRNA- and tracrRNA-derived sequences connected by an artificial tetraloop (Figure 4A). The crRNA sequence can be divided into guide (20-nt) and repeat (12-nt) regions, while the tracrRNA sequence can be divided into anti-repeat (14-nt) and three tracrRNA stem loops (Figure 4A). The crystal structure revealed that the sgRNA binds the target DNA to form a T-shaped architecture comprising a guide:target heteroduplex, a repeat:anti-repeat duplex, and stem loops 1–3 (Figures 4A, B). The repeat:anti-repeat duplex and stem loop 1 are connected by a single nucleotide (A51), while stem loops 1 and 2 are connected by a 5-nt single-stranded linker (nucleotides 63–67).

The guide (nucleotides 1–20) and target DNA (nucleotides 1′–20′) form the guide:target heteroduplex via 20 Watson-Crick base pairs (Figures 4A, B). The conformation of the heteroduplex is distorted from that of the canonical A-form RNA duplex (Figure S6). The repeat (nucleotides 21–32) and the anti-repeat (nucleotides 37–50) form the repeat:anti-repeat duplex via nine Watson-Crick base pairs (U22:A49–A26:U45 and G29:C40–A32:U37) (Figures 4A, B). Within this region, G27, A28, A41, A42, G43 and U44 are unpaired, with A28 and U44 flipped out from the duplex (Figure 4C). The G27 and A41 nucleobases stack with the A26:U45 and G29:C40 pairs, respectively, and the 2-amino group of G27 interacts with the backbone phosphate group between G43 and U44, stabilizing the duplex structure (Figure 4C). G21 and U50 form a wobble base pair at the three-way junction between the guide:target heteroduplex/repeat:anti-repeat duplexes and stem loop 1, contributing to the formation of the T-shaped architecture (Figure 4C).

As expected from the RNA-fold predictions based on the nucleotide sequence, the tracrRNA 3′ tail (nucleotides 68–81 and 82–96) forms stem loops 2 and 3 via four and six Watson-Crick base pairs (A69:U80–U72:A77 and G82:C96–G87:C91), respectively (Figures 4A, B). In addition, nucleotides 52–62 form the newly detected stem loop (stem loop 1) via three Watson-Crick base pairs (G53:C61, G54:C60 and C55:G58), with U59 flipped out from the stem (Figures 4A, B). Stem loop 1 is stabilized by the G62–G53:C61 stacking interaction and the G62–A51/A52 polar interactions (Figure 4C).

The guide:target and repeat:anti-repeat duplexes are deeply buried in a positively-charged groove at the interface of the two lobes, while the rest of the sgRNA extensively interacts with the positively-charged surface on the back side of the protein (Figure 1E). In Mol A, the three nucleotides at the 5′ end of the target DNA (3′-ACC-5′, complementary to the PAM) are not visible in the electron density map. Although the two adjacent nucleotides (3′-AC-5′) in Mol B are structurally ordered due to the crystal packing interactions, and are visible in the electron density map, these nucleotides are not recognized by Cas9 (data not shown). These observations suggested that the 3′-NCC-5′ sequence complementary to the 5′-NGG-3′ PAM is not recognized by Cas9, and are consistent with previous biochemical data showing that Cas9-catalyzed DNA cleavage requires the 5′-NGG-3′ PAM on the non-complementary strand but not the 3′-NCC-5′ sequence on the complementary strand (Jinek et al., 2012).

Previous studies showed that, although sgRNA with a 48-nt tracrRNA tail [referred to as sgRNA(+48)] is the minimal region for the Cas9-catalyzed DNA cleavage in vitro (Jinek et al., 2012), sgRNAs with extended tracrRNA tails, sgRNA(+67) and sgRNA(+85), dramatically improved the Cas9 cleavage activity in vivo (Hsu et al., 2013). The present structure revealed that sgRNA(+48), sgRNA(+67) and sgRNA(+85) contain stem loop 1, stem loops 1–2 and stem loops 1–3, respectively (Figures 4A, B). These observations indicated that, whereas stem loop 1 is essential for the formation of the functional Cas9–sgRNA complex, stem loops 2 and 3 further support the stable complex formation and enhance the stability of the sgRNA, thus improving the in vivo activity.

To determine the significance of each sgRNA structural component on the Cas9 function, we tested a number of sgRNAs with mutations in the repeat:anti-repeat duplex, stem loops 1–3, and the linker between stem loops 1 and 2. Our results revealed that, whereas stem loops 2 and 3 as well as the linker region can tolerate a large number of mutations, the repeat:anti-repeat duplex and stem loop 1 are critical for the Cas9 function (Figure 4D). Moreover, the sgRNA sequence can tolerate a large number of mutations (Figure 4D, reconstructed sgRNA). These results highlight the functional significance of the structure-dependent recognition of the repeat:anti-repeat duplex by Cas9.

The conserved arginine cluster on the Bridge helix is critical for sgRNA:DNA recognition

The sgRNA guide region is primarily recognized by the REC lobe (Figure 5). The backbone phosphate groups of the guide region (nucleotides 2, 4–6 and 13–20) interact with the REC1 domain (Arg165, Gly166, Arg403, Asn407, Lys510, Tyr515 and Arg661) and the Bridge helix (Arg63, Arg66, Arg70, Arg71, Arg74 and Arg78) (Figure 6A). The 2′-hydroxyl groups of G1, C15, U16 and G19 hydrogen bond with Val1009, Tyr450, Arg447/Ile448 and Thr404, respectively (Figure 6A). These observations suggested that the Watson-Crick faces of eight PAM-proximal nucleotides in the Cas9-bound sgRNA are exposed to the solvent, thus serving as a nucleation site for pairing with the complementary strand of the target DNA. This is consistent with previous reports that the 10–12 bp PAM-proximal “seed” region is critical for the Cas9-catalyzed DNA cleavage (Cong et al., 2013; Fu et al., 2013; Hsu et al., 2013; Jinek et al., 2012; Mali et al., 2013a; Pattanayak et al., 2013).

A mutational analysis demonstrated that the R66A, R70A and R74A mutations on the Bridge helix markedly reduced the DNA cleavage activities (Figure 6B), highlighting the functional significance of the recognition of the sgRNA “seed” region by the Bridge helix. Although Arg78 and Arg165 also interact with the “seed” region, the R78A and R165A mutants showed only moderately decreased activities (Figure 6B). These results are consistent with the fact that Arg66, Arg70 and Arg74 form multiple salt bridges with the sgRNA backbone, while Arg78 and Arg165 form a single salt bridge with the sgRNA backbone (Figure 6A). The cluster of arginine residues on the Bridge helix is highly conserved among the Cas9 proteins in the Type II-A–C systems (Figures S4, S5), suggesting that the Bridge helix is a universal structural feature of the Cas9 proteins. This notion is supported by the previous observation that a strictly conserved arginine residue, equivalent to Arg70 of S. pyogenes Cas9, is essential for the function of Francisella novicida Cas9 in the Type II-B system (Sampson et al., 2013). Moreover, the alanine mutations of the repeat:anti-repeat duplex-interacting residues (Arg75 and Lys163) and the stem loop 1-interacting residue (Arg69) resulted in decreased DNA cleavage activity (Figure 6B), confirming the functional importance of the recognition of the repeat:anti-repeat duplex and stem loop 1 by Cas9.

The sgRNA guide region is recognized by Cas9 in a sequence-independent manner, except for the U16–Arg447 and G18–Arg71 interactions (Figures 5 and 6A). This base-specific G18–Arg71 interaction may partly explain the observed preference of Cas9 for sgRNAs with guanines in the four PAM-proximal guide region (Wang et al., 2014).

The REC1 and RuvC domains facilitate RNA-guided DNA targeting

Cas9 recognizes the 20-bp guide:target heteroduplex in a sequence-independent manner (Figure 5). The backbone phosphate groups of the target DNA (nucleotides 1′, 9′–11′, 13′ and 20′) interact with the REC1 (Asn497, Trp659, Arg661 and Gln695), RuvC (Gln926), and PI (Glu1108) domains. The C2′ atoms of the target DNA (nucleotides 5′, 7′, 8′, 11′, 19′ and 20′) form van der Waals interactions with the REC1 domain (Leu169, Tyr450, Met495, Met694 and His698) and the RuvC domain (Ala728) (Figure 5). These interactions are likely to contribute toward the ability of Cas9 to discriminate between DNA and RNA targets. The terminal base pair of the guide:target heteroduplex (G1:C20′) is recognized by the RuvC domain via end-capping interactions (Figure 6C); the sgRNA G1 and target DNA C20′ nucleobases interact with the Tyr1013 and Val1015 side chains, respectively, whereas the 2′-hydroxyl and phosphate groups of sgRNA G1 interact with Val1009 and Gln926, respectively. These end-capping interactions are consistent with the previous observation that Cas9 recognizes a 17–20-bp guide:target heteroduplex, and that extended guide sequences are degraded in cells and do not contribute to improving sequence specificity (Ran et al., 2013). Taken together, these structural findings explain the RNA-guided DNA targeting mechanism of Cas9.

The repeat:anti-repeat duplex is recognized by the REC and NUC lobes in a sequence-dependent manner

In contrast to the sequence-independent recognition of the sgRNA guide region, sequence-dependent interactions exist between Cas9 and the repeat:anti-repeat duplex (Figures 5, 6D and 6E). The nucleobases of U23/A49 and A42/G43 hydrogen bond with the side chain of Arg1122 and the main-chain carbonyl group of Phe351, respectively (Figure 6D). The nucleobase of the flipped U44 is sandwiched between Tyr325 and His328, with its N3 atom hydrogen bonded with Tyr325, while the nucleobase of the unpaired G43 stacks with Tyr359 and hydrogen bonds with Asp364 (Figure 6E).

The present structure revealed that the repeat:anti-repeat duplex is recognized by the REC lobe, which is divergent in both sequence and length among the Cas9 orthologs within the Type II-A–C systems (Figures S4, S5). This structural finding explains the previous observation that Cas9 and sgRNA are interchangeable only between closely related Type II systems (Fonfara et al., 2013). The three PAM-distal base pairs (C30:G39–A32:U37) are not recognized by Cas9 and protrude from the complex (Figure 5), consistent with a proposed model in which the Cas9-bound repeat:anti-repeat duplex is processed by the host RNase III enzyme (Deltcheva et al., 2011).

The nucleobases of G21 and U50, in the G21:U50 wobble pair, stack with the terminal C20:G1′ pair in the guide:target heteroduplex and Tyr72 on the Bridge helix, respectively, with the U50 O4 atom hydrogen bonded with Arg75 (Figure 6D). Notably, A51 adopts the syn-conformation, and is oriented in the direction opposite to U50 (Figures 4C and 6F). The nucleobase of A51 is sandwiched between Phe1105 and U63, with its N1, N6 and N7 atoms hydrogen bonded with G62, Gly1103 and Phe1105, respectively (Figure 6F). Whereas the repeat:anti-repeat duplexes have diverse sequences and lengths among the Type II-A–C systems, the G:U base pair at the three-way junction is highly conserved among the repeat:anti-repeat duplexes in these three systems (Fonfara et al., 2013), suggesting that this wobble pairing is a universal structural feature involved in the three-way junction formation.

To verify the importance of the sequence-dependent recognition of the repeat:anti-repeat duplex by Cas9, we evaluated the effects of repeat:anti-repeat mutations on the Cas9-mediated DNA cleavage (Figure 6B). The replacement of G43, which forms base-specific hydrogen bonds with Phe351 and Asp364, with cytosine reduced the Cas9 activity by over 3-fold. In addition, the replacement of the flipped U44 with guanine resulted in an over 5-fold drop in the cleavage activity, whereas the replacement of U44 with another pyrimidine base (cytosine) did not significantly affect the cleavage activity (Figure 6B). These results confirmed the functional importance of the base-specific recognition of G43 and U44 by Cas9.

Stem loops 1–3 reinforce the interaction between Cas9 and sgRNA

Stem loop 1 is primarily recognized by the REC lobe, together with the PI domain (Figure 5). The backbone phosphate groups of stem loop 1 (nucleotides 52, 53 and 59–61) interact with the REC1 domain (Leu455, Ser460, Arg467, Thr472 and Ile473), the PI domain (Lys1123 and Lys1124), and the Bridge helix (Arg70 and Arg74), with the 2′-hydroxyl group of G58 hydrogen bonded with Leu455 (Figure 6G). A52 interacts with Phe1105 through a face-to-edge π–π stacking interaction (Figure 6F), and the flipped U59 nucleobase hydrogen bonds with Asn77 (Figure 6G).

The single-stranded linker and stem loops 2 and 3 are primarily recognized by the NUC lobe (Figure 5); in contrast, stem loop 1 and the guide:target/repeat:anti-repeat duplexes are recognized by both the REC and NUC lobes. The backbone phosphate groups of the linker (nucleotides 63–65 and 67) interact with the RuvC domain (Glu57, Lys742 and Lys1097), the PI domain (Thr1102), and the Bridge helix (Arg69), with the 2′-hydroxyl groups of U64 and A65 hydrogen bonded with Glu57 and His721, respectively (Figure 6H). The C67 nucleobase forms two hydrogen bonds with Val1100 (Figure 6H).

Stem loop 2 is recognized by Cas9 via the interactions between the NUC lobe and the non-Watson-Crick A68:G81 pair, which is formed by direct (between the A68 N6 and G81 O6 atoms) and water-mediated (between the A68 N1 and G81 N1 atoms) hydrogen-bonding interactions (Figure 6I). The A68 and G81 nucleobases contact Ser1351 and Tyr1356, respectively, while the A68:G81 pair interacts with Thr1358 via a water-mediated hydrogen bond (Figure 6I). The 2′-hydroxyl group of A68 hydrogen bonds with His1349, while the G81 nucleobase hydrogen bonds with Lys33 (Figure 6I).

Stem loop 3 interacts with the NUC lobe more extensively, as compared to stem loop 2 (Figure 5). The backbone phosphate groups of C91 and G92 interact with the RuvC domain (Arg40 and Lys44), while the G89 and U90 nucleobases hydrogen bond with Gln1272 and Glu1225/Ala1227, respectively (Figure 6J). The A88 and C91 nucleobases are recognized by Asn46 via multiple hydrogen-bonding interactions (Figure 6J).

Structural flexibility of Cas9 and sgRNA

Although the HNH domain cleaves the complementary strand of the target DNA at a position three nucleotides upstream of the PAM sequence (Gasiunas et al., 2012; Jinek et al., 2012), in the present structure, the HNH domain is located away from the scissile phosphate group of the bound complementary strand (Figure 7A). A structural comparison of Mol A and Mol B provided mechanistic insights into complementary strand cleavage by the HNH domain. In Mol A, the HNH domain is followed by the α39 helix of the RuvC domain, which is connected to the α40 helix by the α39–α40 linker (residues 919–925) (Figure 7A). In Mol A, residues 913–925 form the C-terminal portion of the α39 helix and the α39–α40 linker, while in Mol B, these residues form an extended α-helix, which is directed toward the cleavage site of the complementary strand (Figure 7A). These observations suggested that the HNH domain can approach and cleave the target DNA through conformational changes in the segment connecting the HNH and RuvC domains.

The structural comparison further revealed the conformational flexibility between the REC and NUC lobes (Figure 7B). As compared to Mol A, Mol B adopts a more open conformation, in which the two lobes are rotated by 15° at a hinge loop between the Bridge helix and strand β5 in the RuvC domain (Figure 7B). The bound sgRNA also undergoes an accompanying conformational change at the linker, which interacts with the hinge loop (Figure 7C). We also observed the concomitant displacement of the β17–β18 loop of the PI domain, which interacts with the repeat:anti-repeat duplex and the α2–α3 loop of the REC1 domain (Figure 7B). Notably, there is no direct contact between the two lobes in the present structure, except for the interactions between the α2–α3 and β17–β18 loops (Figure 7D), suggesting that Cas9 is highly flexible in the absence of the sgRNA. The flexible nature of Cas9 is likely to play a role in the assembly of the Cas9–sgRNA–DNA ternary complex.