The CRISPR RNA-guided endonuclease Cas9 cleaves double-stranded DNA targets complementary to the RNA guide (1) (fig. S1) and has been harnessed for genome editing in eukaryotic cells (2). However, the widely used Cas9 from S. pyogenes (SpCas9) strictly recognizes an NGG sequence (where N is any nucleobase) as the protospacer adjacent motif (PAM) (3), thereby restricting the targetable genomic loci. Structure-guided directed evolution approaches to address this limitation yielded several SpCas9 variants with altered PAM specificities, such as the SpCas9 VQR and VRER variants, which recognize the NGA and NGCG PAMs, respectively (4). In addition, Cas9 and Cas12a (also known as Cpf1) enzymes with distinct PAM specificities, such as Staphylococcus aureus Cas9 (SaCas9) (5), Acidaminococcus sp. Cas12a (AsCas12a), and Lachnospiraceae bacterium Cas12a (LbCas12a) (6), have extended the targeting range in CRISPR-Cas-mediated genome editing.

To expand the targeting range of CRISPR-Cas9, we sought to engineer a SpCas9 variant with relaxed preferences for the third nucleobase of the PAM. Previous studies revealed that the second and third nucleobases in the NGG PAM are recognized by Arg¹³³³ and Arg¹³³⁵ of SpCas9, respectively (7) (fig. S2). We thus hypothesized that the PAM constraint can be reduced by eliminating the base-specific interaction between Arg¹³³⁵ and the third G, and compensating for the loss of this base-specific interaction by introducing non-base-specific interactions with the PAM duplex. We first measured the in vitro cleavage activities of purified wild-type SpCas9 and the R1335A (Arg¹³³⁵ → Ala) mutant toward a target plasmid with the TGG PAM and confirmed that, whereas wild-type SpCas9 efficiently cleaves the TGG target, R1335A has almost no activity (fig. S3, A to C).

We next examined whether the R1335A activity is restored by the substitution of residues surrounding the PAM duplex, and found that the replacements of Leu¹¹¹¹, Gly¹²¹⁸, Ala¹³²², and Thr¹³³⁷ with Arg partially restored the activity of the R1335A mutant (fig. S3, A to C). Furthermore, the R1335A/L1111R/G1218R/A1322R/T1337R variant (referred to as ARRRR) efficiently cleaved the TGG target (fig. S3, A to C). However, the cleavage kinetics of ARRRR was slower than that of wildtype SpCas9 (fig. S3, D and E). In the previously reported VQR (D1135V/R1335Q/T1337R) and VRER (D1135V/G1218R/R1335E/T1337R) variants, the D1135V mutation provides interactions with the sugar-phosphate backbone of the PAM duplex (8,9). In addition, molecular modeling suggested that the E1219F mutation forms hydrophobic interactions with the ribose moiety of the second G, and that the R1335V mutation stabilizes Arg¹³³³ and Phe¹²¹⁹ (E1219F) (fig. S3F). Indeed, the addition of the D1135V, E1219F, and R1335V mutations enhanced the cleavage activity (fig. S3, D and E). We designated the R1335V/L1111R/D1135V/G1218R/E1219F/A1322R/T1337R variant as VRVRFRR.

We next measured the cleavage activities of VRVRFRR toward the target plasmid with TGN PAMs. Relative to wild-type SpCas9, VRVRFRR slowly but more efficiently cleaved the TGA, TGT, and TGC targets (Fig. 1, A to C, and fig. S4). Although VRVRFRR is less active than wild-type SpCas9 and SaCas9, its cleavage activity was comparable to those of AsCas12a and LbCas12a (5, 6) (fig. S5). Using an in vitro PAM identification assay (10), we confirmed that whereas wildtype SpCas9 is specific to NGG PAMs, VRVRFRR preferentially recognizes NG PAMs (Fig. 1D and fig. S6). Although the PAM identification assay revealed that VRVRFRR slightly recognizes NAN PAMs (Fig. 1D), in vitro cleavage experiments demonstrated that VRVRFRR is less active toward TAN PAMs than toward TGN PAMs (fig. S7). Thus, we concluded that VRVRFRR recognizes a relaxed PAM, and we refer to this variant as SpCas9-NG, as it has increased activity on NGH (H = A, T, or C) PAMs, albeit with reduced relative activity on NGC.

Open in a separate window

Fig. 1.

In vitro cleavage activity.

(A) SDS-polyacrylamide gel electrophoresis analysis of wild-type SpCas9, SpCas9-NG, and xCas9. (B, C, and E) In vitro DNA cleavage activities of wild-type SpCas9 (B), SpCas9-NG (C), and xCas9 (E) toward the TGN PAM targets. Data are means ± SD (n = 3). (D) PAM preference of SpCas9-NG.

We compared SpCas9-NG with the xCas9 enzyme (a SpCas9 variant with the A262T/R324L/S409I/E480K/E543D/M694I/E1219V mutations), which was engineered via directed evolution and recognizes NG PAMs (11) (fig. S8). We measured their in vitro cleavage activities toward the target plasmid with the TGN PAMs. Under our assay conditions (50 nM Cas9), xCas9 showed almost no activity toward the TGA, TGT, and TGC targets, and it cleaved the TGG target with lower efficiency than did SpCas9-NG (Fig. 1, A to C and E, and fig. S9A). At a higher concentration (200 nM Cas9), xCas9 cleaved the TGA, TGT, and TGC targets (fig. S9, D and G); nonetheless, the cleavage kinetics of xCas9 was slower than that of SpCas9-NG (fig. S9, C, D, F, and G), and the cleavage activity of xCas9 toward the TGA, TGT, and TGC targets was comparable to or lower than that of wild-type SpCas9 toward the TGA target (fig. S9, B, D, E, and G). These results demonstrate that SpCas9-NG outperforms xCas9 in recognizing NGH PAMs in vitro.

To clarify the NG PAM recognition mechanism, we determined the crystal structure of SpCas9- NG, in complex with a single guide RNA and its target DNA containing the TGG PAM, at 2.7 Å resolution (fig. S10A and table S1). The second G in the PAM (dG2*) forms bidentate hydrogen bonds with Arg¹³³³ (Fig. 2A), which is stabilized by Val¹³³⁵ (R1335V). The third G (dG3*) forms a hydrogen bond with Arg¹³³⁷ (T1337R) (Fig. 2A), consistent with the preference of SpCas9-NG for the third G (Fig. 1C). Arg¹¹¹¹ (L1111R), Val¹¹³⁵ (D1135V), and Phe¹²¹⁹ (E1219F) interact with the sugar-phosphate backbone of the PAM duplex, whereas Arg¹²¹⁸ (G1218R) does not directly interact with the DNA backbone (Fig. 2B). A structural comparison with the SpCas9 R-loop complex (12) suggested that Arg¹³²² (A1322R) interacts with the nontarget DNA strand (fig. S10B).

Open in a separate window

Fig. 2.

Crystal structure of SpCas9-NG.

(A) Recognition of the PAM duplex. Arg1333 and the substituted residues are shown as stick models. (B) Non-base-specific interactions between the PAM duplex and the substituted residues. TS, target strand; NTS, nontarget strand.

To assess the activity of SpCas9-NG in mammalian cells, we next measured the formation of indels (insertions or deletions) induced by wildtype SpCas9, SpCas9-NG, and xCas9 at 69 endogenous target sites with NGN PAMs in human embryonic kidney (HEK) 293FT cells (table S2). As expected, wild-type SpCas9 induced indels predominantly at the NGG sites, with some recognition of NGA sites (Fig. 3, A and B). Wild-type SpCas9 achieved a >20% indel rate at NGG (17 of 17 sites) and NGA sites (3 of 19 sites), but not NGT (0 of 18 sites) or NGC sites (0 of 15 sites) (Fig. 3, A and B). In contrast, SpCas9-NG edited NGA, NGT, and NGG sites, with lower activity at NGC (Fig. 3, A and B), consistent with the in vitro cleavage preference for the third D (D = A, T, or G) PAM nucleotide (Fig. 1C). SpCas9-NG achieved a >20% indel rate at 13 NGG, 13 NGA, 15 NGT, and 5 NGC sites (Fig. 3, A and B). Relative to SpCas9-NG, xCas9 had lower editing efficiency at all of the NGH sites (Fig. 3, A and B), consistent with our in vitro cleavage data (Fig. 1E). xCas9 induced a >20% indel rate at 15 NGG sites, five NGA sites, and four NGT sites, but none at NGC sites (Fig. 3, A and B).

Open in a separate window

Fig. 3.

Gene editing in human cells.

(A) Indel formation efficiencies of wild-type SpCas9, SpCas9-NG, and xCas9 at the 69 endogenous target sites in HEK293FT cells. Data are means ± SD (n = 3). (B) Summary of the editing efficiencies of SpCas9, SpCas9-NG, and xCas9. Medians and first and third quartiles are shown. In (A) and (B), 20% indel frequency is indicated by dashed lines. (C) Specificities of wild-type SpCas9, SpCas9-NG, and the enhanced-specificity versions of SpCas9 (ES) and SpCas9-NG (NG-ES). The off-target cleavages were evaluated by GUIDE-seq.

Using GUIDE-seq (genome-wide, unbiased identification of double-stranded breaks enabled by sequencing) (13), we examined the specificity of wild-type SpCas9 and SpCas9-NG in human cells at two previously characterized target sites (EMX1 and VEGFA). We found that wild-type SpCas9 and SpCas9-NG have comparable numbers of off-target sites for the EMX1 target, and that the K848A/K1003A/R1060A mutations in the high-fidelity SpCas9 variant [eSpCas9(1.1)] (14) substantially reduced the off-target cleavage by SpCas9-NG (Fig. 3C and fig. S11). These results show that SpCas9-NG has cleavage specificity comparable to that of wild-type SpCas9, and that its specificity can be enhanced by the high-fidelity mutations. As expected, SpCas9-NG had a different off-target profile with off-target sites harboring NGH PAMs (fig. S11), further confirming the relaxed PAM recognition.

The nuclease-inactive version of SpCas9 can be applied to numerous technologies such as base editing (15,16). We examined whether the SpCas9-NG D10A nickase fused to the activation-induced cytidine deaminase (nSpCas9-NG-AID, referred to as Target-AID-NG) mediates C-to-T conversion at 32 endogenous target sites with NG PAMs in human cells (table S3). nSpCas9-AID (Target- AID) efficiently induced C-to-T conversion at the NGG sites, with lower activity at NGA sites and no activity at NGT or NGC sites (Fig. 4A and figs. S12A and S13). In contrast, Target- AID-NG showed base-editing activity toward all PAMs assessed, albeit with lower efficiency at NGC sites (Fig. 4A and figs. S12A and S13), consistent with the indel data (Fig. 3A). In addition, Target-AID-NG showed higher base-editing efficiencies than did the xCas9 D10A nickase fused to the APOBEC1 cytidine deaminase (xCas9-BE4) at most of the tested poly-C–containing target sites (Fig. 4B and figs. S12B and S13). These results demonstrate that the catalytically inactive version of SpCas9-NG can serve as a useful RNA-guided DNA targeting platform.

Open in a separate window

Fig. 4.

Base editing in human cells.

(A and B) C-to-T conversion efficiencies at the 20 endogenous target sites (Target-AID and Target-AID-NG) (A) and at the 12 poly-C-containing target sites (Target-AID, Target-AID-NG, and xCas9-BE4) (B) in HEK293Tcells. The experiments were performed at least twice, and similar results were obtained.

In this study, we engineered an SpCas9-NG variant with an increased targeting range (fig. S14). Nonetheless, the cleavage activity of SpCas9-NG is lower than that of wild-type SpCas9 at NGG sites, and SpCas9-NG is less active at NGC sites relative to NGD sites. Thus, it will be important to improve the activity of SpCas9-NG by further molecular engineering. Overall, the rationally designed SpCas9-NG and its high-fidelity variants can serve as useful genome-editing tools with increased versatility across genomes.