Structural and functional bases of F. rodentium Cas9 provide insights into CRISPR-Cas protein engineering

Summary

The Faecalibaculum rodentium (Fr) CRISPR-Cas9 system exhibits enhanced gene-editing precision and efficiency compared to SpCas9, with distinctive advantages in targeting the TATA box in eukaryotic promoters. However, the underlying molecular mechanisms remained unexplored. Here, we present cryo-electron microscopy structures of the FrCas9-single guide RNA (sgRNA)-DNA complex in both the R-loop expansion and pre-catalytic states, shedding light on its specialized recognition of the 5′-NRTA-3′ protospacer adjacent motif (PAM) and the unusual overwinding of the sgRNA-DNA heteroduplex. Our investigations into the structure and extensive mutational analyses reveal that the phosphate lock loop plays a pivotal role in finely adjusting FrCas9’s off-target sensitivity and catalytic efficiency. Remarkably, targeted residue substitutions in the phosphate lock loop and the PAM-distal region were found to synergistically enhance both the editing precision and efficiency of FrCas9. These findings advance our understanding of Cas9’s accuracy and potency mechanisms while providing a molecular foundation for the rational design and development of next-generation CRISPR technologies.

Graphical abstract

Highlights

• •Two cryo-EM structures of F. rodentium Cas9s show distinct states
• •The recognition of the 5′-NRTA-3′ PAM by PI domain
• •The V1103K variant has hyper-accuracy and high efficiency in vivo
• •Engineering synergy between REC3 domain and phosphate lock loop

Yang et al. determined the structures of F. rodentium Cas9 ternary complexes in two conformational states, revealing the distinctive structural features of the sgRNA-DNA heteroduplex. Through in vitro and in vivo validation of critical amino acid residues and combinatorial mutagenesis, they developed an improved FrCas9 variant with enhanced properties.

Introduction

The advent of gene-editing technologies, especially the CRISPR-Cas9 system, has marked the beginning of a new era in biotechnology.¹ Perhaps the most immediate impact of gene editing lies in its potential to revolutionize medicine. It offers hope for curing genetic disorders, such as cystic fibrosis, sickle cell anemia, and Huntington’s disease, by directly repairing the genetic abnormalities at their source.^2,3,4 Beyond rare diseases, it also has the potential to contribute to cancer therapy, personalized medicine, and the prevention of hereditary conditions, paving the way for a healthier global population.^5,6,7

Currently, the application of gene editing in the biomedical field, while promising, still faces several limitations and challenges.⁸ First, the risk of off-target effects, where editing tools unintentionally alter DNA sequences other than the intended ones, can lead to unwanted mutations, necessitating improved accuracy and specificity in gene-editing technologies.^8,9 Second, the effectiveness of SpCas9, a popular CRISPR-Cas9 enzyme, is limited by its need for a specific protospacer adjacent motif (PAM) sequence “NGG” near the target site.^10,11 This requirement restricts the enzyme’s ability to edit certain genomic regions, complicating efforts to correct mutations associated with genetic diseases that do not lie close to an “NGG” sequence.¹² Researchers are working on developing Cas9 variants with flexible PAM requirements and improving the precision of gene editing to overcome these obstacles, aiming for safer and more versatile applications.^{13,14,15,16,17,18}

FrCas9, a CRISPR-Cas9 variant derived from Faecalibaculum rodentium, presents several advantages that enhance the versatility and applicability of CRISPR-based gene-editing technologies.¹⁹ FrCas9 has been observed to exhibit high specificity and efficiency in editing its target DNA sequence, ensuring both the safety and efficacy of gene editing. This is particularly important in therapeutic applications. Second, FrCas9 can recognize and bind to an “NRTA” PAM, allowing researchers to address genes and mutations that were previously inaccessible with SpCas9, especially the TATA box for the initiation of gene transcriptions.²⁰ In addition, FrCas9’s NRTA PAM differs from SpCas9’s NGG PAM in genomic distribution and accessibility. AT-rich regions tend to adopt more compact chromatin conformations, making them less accessible for protein binding.²¹ This PAM-mediated specificity advantage, in conjunction with FrCas9’s inherent characteristics, synergistically contributes to its exceptional fidelity profile.

Despite these advantages, the molecular mechanisms underlying FrCas9’s gene-editing capabilities have remained elusive. In this study, we leveraged cryo-electron microscopy (cryo-EM) to capture the structures of two FrCas9-sgRNA-DNA ternary complexes. Detailed structural data from cryo-EM revealed critical interactions between FrCas9 and the sgRNA-DNA heteroduplex. These interactions highlight the importance of specific amino acid residues in domain configurations, PAM recognition, and R-loop stability. Based on structural insights, we introduced targeted amino acid mutations to investigate their effects on FrCas9’s gene-editing efficiency and fidelity. Mutations in areas such as the phosphate lock loop (PLL) or the PAM-distal interface can reveal their roles in enhancing or diminishing editing performance. Further, the rationally engineered FrCas9 (eFrCas9) was tested in the treatment of an inherited genetic disease, Duchenne muscular dystrophy (DMD), in immortalized human skeletal muscle myoblast (HsKMM) cells.

The integration of structural and functional data in this study provides important insights into the gene-editing mechanisms of FrCas9. These mechanisms unlock new opportunities for researchers to design hyper-active enzymes with high fidelity.²² This can potentially lead to the development of more effective and safer gene-editing tools, further broadening the applicability of CRISPR technologies in research and therapy.²²

Results

Compared with SpCas9, FrCas9 exhibits significantly low off-target effects

In our research, we applied the adaptor-mediated off-target identification by sequencing (AID-seq)²³ method to perform a detailed comparative analysis of FrCas9 and SpCas9 across 1,478 sgRNAs. Our findings revealed that FrCas9 not only showed a higher average on-target read count of 2,129.58 compared to SpCas9’s 1,750.93 but also significantly reduced off-target activities (Figures 1A and 1B; Table S1). Specifically, FrCas9 averaged 58.57 off-target events per sgRNA, markedly lower than the 386.31 events observed with SpCas9, highlighting its enhanced specificity (Figure 1B). This was further corroborated by analyzing the top three off-target sites for each, where FrCas9 showed substantially fewer off-target reads (32.99) compared to SpCas9 (56.33; Figure 1C).

Figure 1.

Structural basis of high-efficiency and high-fidelity FrCas9

(A–D) AID-seq-based comparative analyses of FrCas9 and SpCas9 for 1,478 sgRNAs in parallel (two-tailed t test, ∗∗∗∗p < 0.0001). Figures show on-target reads (A), off-target site numbers (B), average reads of top 3 off-target sites (C), and on-target:off-target ratios (D). On-target:off-target ratios were calculated as log2 (on-target reads + 1)/(total off-target reads + 1).

(E and F) Off-target effects in the top 3 sgRNAs with the highest target cleavage activity of FrCas9 (E) and SpCas9 (F), respectively. The top 5 highest off-target read sequences were selected for display. The summarized results could be found in descending order in Table S1. Mismatched positions with off-target sites are highlighted in color, the gray lines represent the missing base in the off-target sites, and AID-seq read counts are shown to the right of the sequences.

(G) Domain composition of FrCas9. BH, bridge helix; PI, PAM-interaction domain.

(H) The front cylinder representation of RE complex. Domains are colored as in (G). The HNH domain has a weak density and has not been included in the atomic models. The red circle represents the PAM DNA double-stranded structure that is different from the PreC complex.

(I) Schematic diagram of Cas9 conformational change. The schematic diagram in the outer ring illustrates the reported key conformational states of SpCas9 during DNA cleavage. The highlighted structural models in the middle represent the two FrCas9-sgRNA-DNA complexes solved in this study. Black arrows: the dynamic transition pathway of SpCas9 structural change. Blue arrows: the conformational change states of FrCas9 structure. The figure was drawn by Figdraw.

(J) The front cylinder representation of PreC complex. Domains are colored as (G). The red circle represents the full-length sgRNA-DNA heteroduplex structure that is different from the RE complex.

See also Figures S1–S9.

Moreover, by comparing the on-target:off-target ratios, FrCas9 demonstrated superior performance over SpCas9 (Figure 1D). Data from the three sgRNAs with the highest on-target reads further illustrated that FrCas9 achieved comparable on-target reads (11,712, 8,995, and 8,951) and substantially fewer numbers of off-target sites (17, 14, and 39) than SpCas9, which had moderately higher on-target reads (10,424, 10,033, and 9,647) and markedly more off-target sites (331, 746, and 1,230) (top 5 off-targets presented; Figures 1E and 1F; Table S1). This superior specificity profile was confirmed in live-cell validation studies using 11 guide RNAs in HEK293T cells. FrCas9 achieved on-target efficiency comparable to that of SpCas9 while maintaining significantly higher on-target:off-target ratios (Figure S1; Tables S2 and S3).

These results underscore FrCas9’s exceptionally lower off-target effects than SpCas9, prompting further investigation into the structural mechanisms that contribute to FrCas9’s enhanced specificity.

FrCas9-sgRNA-dsDNA ternary complex elucidates two distinct conformations

Studies have shown that the precise cleavage of the target by the Cas9-sgRNA complex is a highly collaborative process triggered by PAM recognition, R-loop expansion (RE), target DNA cleavage, and product release.²² To elucidate the recognition mechanism of FrCas9 in detail, we solved the cryo-EM complex structure of a catalytically inactive FrCas9 (H877A) bound to sgRNA with a 43-bp double-stranded DNA (dsDNA) substrate in the presence of Mg²⁺, at a 2.89 Å resolution (Figures 1G, 1H, S2, and S3; Table 1). Cryo-EM reconstruction revealed that the majority of the FrCas9-sgRNA-DNA complex could be unambiguously solved, except for the HNH domain (Figures S4A, S4B, and S5). Although the sequence homology between FrCas9 and SpCas9 is only 29.21%, the structural alignment revealed that the domains of FrCas9 and SpCas9 exhibit conservation and divergence features (PDB: 7Z4H, root-mean-square deviation [RMSD] of 2.97 Å for 758 equivalent Cα atoms; Figure S6A). The FrCas9-sgRNA-DNA ternary complex exhibited sgRNA-DNA hybridization, forming an incomplete R-loop structure (Figure S6B). To determine the captured state of the FrCas9-sgRNA-DNA ternary complex, we calculated the angle between the PAM duplex in linear conformation and the sgRNA-target sequence (TS) heteroduplex in the PAM-distal region. The kinked gRNA-TS heteroduplex was bent by 30° relative to the PAM duplex in linear conformation, suggesting that this state was reminiscent of RE intermediates^24,25 (RE complex; Figures 1I and S6B).

Table 1.

Cryo-EM data collection, refinement, and validation statistics

	RE complex (43-bp dsDNA) (EMDB: 61854, PDB: 9JWN)	PreC complex (26-nt TS) (EMDB-61853, PDB: 9JWK)
Data collection and processing
Magnification	105,000	105,000
Voltage (kV)	300	300
Electron exposure (e–/Å2)	50	50
Defocus range (μm)	−2.0 to −1.0	−2.0 to −1.0
Pixel size (Å)	0.8192	0.83
Symmetry imposed	C1	C1
Initial particle images (no.)	1,398,644	2,939,319
Final particle images (no.)	684,922	219,843
Map resolution (Å)	2.89	3.46
FSC threshold	0.143	0.143
Map resolution range (Å)	2.0–7.5	3.0–7.0
Refinement
Initial model used (PDB code)	6O0Y	6O0Y
Model resolution (Å)	3.1	3.6
FSC threshold	0.5	0.5
Map sharpening B factor (Å2)	−111.09	−141.0
Model composition
Non-hydrogen atoms	11,887	12,409
Protein residues	1,063	1,164
Nucleotides	152	139
Ligands	0	0
B factors (Å2)
Protein	104.35	73.75
Nucleotides	88.06	80.51
Ligands	0	0
RMSDs
Bond lengths (Å)	0.003	0.003
Bond angles (°)	0.535	0.592
Validation
MolProbity score	2.12	1.90
Clashscore	11.00	11.49
Poor rotamers (%)	2.41	0.48
Ramachandran plot
Favored (%)	95.99	95.32
Allowed (%)	4.01	4.68
Disallowed (%)	0	0

Subsequently, we reconstituted another complex containing sgRNA with a 22-nt target single-strand DNA (ssDNA) sequence and a 4-nt PAM ssDNA sequence at a 3.46 Å resolution (Figures 1J, S4C, S4D, S7, and S8; Table 1). This complex compensated for the density weakness in the PAM-distal region of the RE complex and had partial density for the HNH domain (Figure S9A). Through structural analysis, the overall conformation of this complex represented a transient, discrete intermediate between checkpoint conformation (PDB: 6O0Z) and catalytic state (PDB: 6O0Y and 7Z4J)^24,26 (PDB: 6O0Z, an RMSD of 3.4 Å for 780 equiv Cα atoms; PDB: 6O0Y, an RMSD of 2.6 Å for 709 equiv Cα atoms; and PDB: 7Z4J, an RMSD of 2.7 Å for 755 equiv Cα atoms). Furthermore, the REC2 and REC3 domains of FrCas9 have adopted conformations allosteric to the catalytic state (Figures S9B and S9C). Therefore, the complex is in the pre-catalytic (PreC) state (PreC complex; Figures 1I, S9B, and S9C). Collectively, these data provide the framework for FrCas9-sgRNA-DNA complex formation, revealing its unique stepwise domain rearrangements in RE and PreC states.

The 5′-NRTA-3′ PAM recognition pattern of FrCas9

Our previous work demonstrated that FrCas9 utilizes a TA-rich palindromic sequence in the PAM region to target specific genes.¹⁹ To understand the recognition patterns, we examined key residues in the RE complex for PAM interaction (PI). The overall structure of the FrCas9 RE complex revealed that the PAM duplex was nested in a positively charged groove of the PI domain, recognizing 5′-TGTA-3′, which was different from the SpCas9 GC-rich PAM²⁷ (Figures 2A, S4B, S10A, and S10B). The dT1∗ of the 5′-NRTA-3′ PAM had a single contact with the FrCas9 protein, established by the hydrogen bond between the phosphate backbone and the Y1198 side chain (Figure 2B). Concurrently, constraints on the phosphate backbone by Y1198 and R1213 together stabilized the recognition of the PAM by the PI domain (Figure 2B). The dG2∗ of the 5′-NRTA-3′ PAM was recognized by the side chain of N1137. Previously, we found the second position of the PAM region to be rich in purine nucleotides.¹⁹ Here, we showed that purine nucleotides (adenine and guanine) at the second position of the PAM region were recognized by N1137 via hydrogen bonds, whereas pyrimidine nucleotides were positioned too distally for such interactions (Figure 2C). This observation highlights FrCas9’s preference for purine nucleotides at the second PAM position. The dT3∗ was recognized by the amino group of K1321 through a water-mediated hydrogen bond. Additionally, the N6 and N7 of dA-3 contacted the amino group and the hydroxyl group of Q1334, respectively, forming bidentate hydrogen bonds (Figure 2D). This structure aligns with the known involvement of glutamine residues in the major-groove recognition of adenines.²⁸ The N6 of dA4∗ formed hydrogen bonds with the hydroxyl group of T1214 (Figure 2E), while the dT-4 formed a non-base-specific π-π stacking interaction with Q1334 (Figure 2E).

Figure 2.

5′-NRTA-3′ protospacer adjacent motif recognition by FrCas9

(A) Binding of the protospacer adjacent motif (PAM) duplex to the groove of PI domain. Left, stereo view. Right, detailed view.

(B–E) Zoomed-in view of 5′-NRTA-3’ (R is A or G) PAM duplex recognition by PI domain. Images show interactions between the FrCas9 protein and PAM at the +1 position (B), +2 position (C), +3 position (D), and +4 position (E). Hot pink: PAM in the NTS DNA. Gold: TS DNA. Residues are color coded by domain as in (A). Red spheres denote water molecules.

(F) On-target editing efficiency of WT-FrCas9 and FrCas9 mutants, assayed by amplicon sequencing in three sites (Dunnett’s test, ∗∗∗p < 0.001, n = 3, mean ± SEM).

(G) GUIDE-seq genome-wide profiles of on-target and off-target sites for the WT-FrCas9 and FrCas9-N1137A mutant in 3 sgRNAs. GUIDE-seq read counts of each site are shown on the right side of the sequence. Mismatched positions with off-target sites are highlighted in color, and the gray lines indicate the missing base in the off-target sites.

See also Figure S10.

To verify the PAM recognition patterns in vivo, we next conducted both amplicon sequencing to assess on-target editing efficiency and a GUIDE-seq (genome-wide, unbiased identification of DSBs enabled by sequencing) assay. Amplicon sequencing revealed that all mutations (Y1198R, N1137A, T1214W, T1214F, T1214Y, Q1334R, Q1334T, K1321A, and R1213A) significantly reduced the on-target cleavage efficiency of FrCas9, highlighting the importance of these residues (Figure 2F). Specifically, the asparagine-to-alanine protein mutant (N1137A) responsible for recognition at the second site of the 5′-NRTA-3′ PAM reduced the on-target cleavage efficiency, proving that N1137 plays an important role in recognizing the second position of the PAM (Figures 2F and 2G). This functional importance was further validated biochemically, with the N1137A mutant showing reduced recognition of pyrimidines (C/T) at the +2 PAM position, supporting our structural model (Figures S10C–S10E). Collectively, our data present combined structural and functional insights into how FrCas9 recognizes the 5′-NRTA-3′ PAM sequence, uncovering a molecular mechanism that sets it apart from SpCas9.

The PLL of FrCas9 has a unique recognition mechanism

Previous studies have shown that the PLL element in Cas9 protein is responsible for PAM-dependent target DNA melting and RNA-DNA hybrid formation.²⁹ Our structural observations indicated that among the contacts between the PLL element of FrCas9 and the +1 phosphate of the target DNA strand, residues D1101 and S1102 had the major interactions, while residue V1103 had less contact with the +1 phosphate (RE complex; Figures 3A and 3B). This is different from the contact mechanism of the SpCas9 1107-KES-1109 PLL element (Figures S11A and S11B). In accordance with this, the cleavage activity of the S1102P mutant was almost completely lost and that of the D1101A mutant was greatly reduced (Figure S11C). Notably, there is a certain electron density cavity at V1103 in the complex structure, which allows more engineering possibilities for the 1103 residue site at PLL elements to obtain FrCas9 variants with enhanced fidelity and efficiency (Figure 3B). Therefore, we generated saturated mutants with 19 amino acid substitutions and measured their on-target cleavage efficiency in three sites by amplicon sequencing (Figure 3C). Amino acids with linear side chains—leucine (L), lysine (K), arginine (R), cysteine (C), asparagine (N), and glutamine (Q)—improved the cleavage efficiency of FrCas9 to various extents (Figure 3C). In contrast, mutations to proline (P) and aromatic amino acids significantly reduced the on-target cleavage efficiency of FrCas9. This was attributed to mutations causing a severe steric clash within the ribonucleoprotein (RNP) complex, negatively impacting the enzyme’s activity (Figure 3D). Additionally, mutations to aspartic acid and glutamic acid reduced efficiency due to their negative charges repelling nucleic acids and weakening essential enzyme-DNA interactions (Figure S11D).

Figure 3.

Mutagenesis in PLL element reveals a FrCas9-V1103K variant with hyper-accurate and high-activity properties in human cells

(A) Phosphate lock loop element (D1101-V1103) near the +1 TS DNA position in RE complex. Left, a close-up view. Right, detailed view.

(B) Model and density map of RE complex show a certain electron density cavity at V1103 position. The density map is shown as colored surfaces.

(C) On-target editing efficiency of 19 amino acid substituted mutants in V1103 site by amplicon sequencing with three sgRNAs. Error bars represent mean ± SEM of n = 3 independent biological replicates with individual values shown as dots.

(D) Four classes of amino acid simulations performed at position 1103 using ChimeraX (v.1.4). Residues colored gray are the mimic amino acids. The pink dotted lines represent spatial clashes in the structure.

(E) GUIDE-seq genome-wide profiles of on-target and off-target sites for WT-FrCas9, FrCas9-V1103K, and FrCas9-V1103R mutants in 3 sgRNAs. GUIDE-seq read counts of each site are shown on the right side. Mismatched positions with off-target sites are highlighted in color, and the gray lines represent the missing base in the off-target sites.

(F) GUIDE-seq genome-wide profiles of on-target and off-target sites for WT-FrCas9 and FrCas9-V1103K mutants in 6 sgRNAs.

(G–J) GUIDE-seq-based comparative analyses of FrCas9-V1103K for 9 sgRNAs (two-tailed t test, ∗p < 0.05 and ∗∗p < 0.01, n = 3, mean ± SEM). Images show the GUIDE-seq number of each off-target site (G), the number of on-target reads at all sites (H), the total number of off-target sites (I), and the number of off-target reads at all sites (J).

(K) The on-target:off-target ratios of GUIDE-seq reads relative to the 9 sites. Black lines indicate the median (two-sided paired Wilcoxon signed-rank test, ∗∗p < 0.01, n = 9). On-target:off-target ratios were calculated as log10 (on-target reads + 1)/(total off-target reads + 1).

See also Figures S11 and S12.

Therefore, we next selected two positively charged amino acid mutants from long-side-chain amino acid mutants and evaluated these mutants by GUIDE-seq. Surprisingly, both the V1103K and V1103R mutants increased on-target reads at the three target sites, and no off-target sites were generated except at the HEK293 SITEx2 site (Figure 3E). The V1103K variant exhibited better on-target performance than V1103R. We further assessed the V1103K variant at six additional target sites and found that it enhanced the overall fidelity of FrCas9 by 30.15% across nine targets (Figures 3F–3K and S11E). The positive charge and the long side chains with amino and guanidine groups of K1103 and R1103, respectively, may be better suited to the minor groove of the heteroduplex than V1103. These results suggest that the PLL element of FrCas9 could be the key site for engineering modification to improve both efficiency and specificity. We found that this engineering strategy was also applicable to other newly identified Cas9 proteins and could improve their target cleavage activity (Figure S12).

Recognition of the sgRNA-DNA heteroduplex and rational engineering of FrCas9

Structural comparison between the FrCas9 RE complex and the PreC complex revealed repositioning of the heteroduplex mainly occurring at bp 8–22 (Figure 4A). The 744-750 loop mediated an upward movement of the REC3 domain toward the REC2 domain by approximately 30°, while the REC2 domain moved slightly upward to facilitate domain rearrangement (Video S1). The structure of the RE complex showed a 15-bp heteroduplex formed by the sgRNA and TS DNA, as well as an 8-nt seed region, even though the sgRNA-TS heteroduplex was partially visible in the electron density map (Figures S4A, S4B, and S13A). Notably, during FrCas9 R-loop propagation, until 15 bp formed, the heteroduplex remained lodged between the REC2 and REC3 domains rather than extending along the positively charged groove between the REC3 and HNH domains (like the R-loop propagation of SpCas9), indicating that FrCas9 has a different R-loop propagation identification mechanism from SpCas9 (Figure S13B). In the PreC complex, the FrCas9 heteroduplex was overwinding at bp 8–9 and 15–16 compared to the SpCas9 sgRNA-TS heteroduplexes, unexpectedly breaking the almost regular A-type conformation (Figure 4B). Specifically, a sharp increase in the helical inclination (by 40°) at bp 6–7 and 16–17 compressed the major grooves at bp 7 and 13 to 15.7 and 15.5 Å, respectively, coupled with concomitant distortion of base-pair geometry (Table S4). As a result, the deformation caused by this overwinding allowed FrCas9 to accommodate a longer spacer (22 bp) within almost the same length as that of SpCas9 (spacer: 20 bp) (Figure 4B). Furthermore, AlphaFold3 simulations revealed that overwinding is an intrinsic structural property of FrCas9, independent of specific target sequences (Figure S13C). Thus, FrCas9’s overwinding mechanism necessitates a tighter and more selective heteroduplex pairing that reduces the tolerance for mismatches, fostering more precise and reliable gene targeting. This mechanism of FrCas9 may be an important reason for its increased fidelity while still retaining the efficiency.

Figure 4.

Recognition of sgRNA:DNA heteroduplex and engineering of FrCas9

(A) Zoomed-in view of structural comparison between REC lobe and heteroduplex in RE complex and PreC complex. NUC lobe and dsDNA are not included in the atomic models for clarity. Color codes are the same as those shown in the figure.

(B) Superimposition of the heteroduplex of SpCas9 (PDB: 6O0Y) and FrCas9 (PreC complex).

(C) A magnified view of D660, W692, N532, and Y763 residues of PreC complex in PAM-proximal region. Domains and TS are colored as in Figures 1G and S8.

(D) Zoomed-in view of conserved arginine residues in BH of PreC complex. Domains and sgRNA are colored as in Figures 1G and S8.

(E) On-target editing efficiency of WT-FrCas9 and FrCas9 mutants in (C) and (D) was measured by amplicon sequencing (Dunnett’s test, ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001, n = 3, mean ± SEM).

(F and G) A magnified view of amino acids at PAM-distal region interacting with TS DNA (F) and sgRNA (G), respectively.

(H) On-target editing efficiency of WT-FrCas9 and FrCas9 mutants in (F) and (G) was measured by amplicon sequencing (Dunnett’s test, ∗p < 0.05 and ∗∗p < 0.01, n = 3, mean ± SEM).

(I–K) GUIDE-seq-based comparative analyses of WT-FrCas9 and FrCas9 mutants in (F) and (G) for 3 sgRNAs. Images show GUIDE-seq number of on-target reads at all sites (I), the total number of off-target sites (J), and the number of off-target reads at all sites (K) (Dunnett’s test, ∗p < 0.05 and ∗∗p < 0.01, n = 3, mean ± SEM).

(L) Schematic illustration of structure-guided combination engineering of FrCas9. The PreC complex model is displayed with a transparent surface. Silver: FrCas9 protein. Gold: TS DNA. Dodger blue: sgRNA.

(M) GUIDE-seq genome-wide profiles of on-target and off-target sites for WT-FrCas9 and FrCas9 mutants in 3 sgRNAs. GUIDE-seq read counts of each site are shown on the right side of the sequence. Mismatched positions with off-target sites are highlighted in color, and the gray lines indicate the missing base in the off-targets sites.

(N) On-target and off-target indel formation efficiency of FrCas9, eFrCas9, and SpCas9. Error bars represent mean ± SEM of n = 3 independent biological replicates, with individual values shown as dots.

See also Figures S13 and S14.

Video S1. Structural comparison analysis of REC lobe and heteroduplex in RE complex and PreC complex, related to Figure 4

Blue: RE complex. Burlywood: PreC complex. Reconstruction of 744–750 loop shows the movement of the REC3 domain.

The DNA near the turning point 8 bp was recognized by D660, W692, N532, and Y763; the RNA was mainly recognized by the conserved bridge helix (BH) domain (Figures 4C and 4D). For the tail of the seed region, W692A (which interacts with TS) and R64A/R68A/R72A (which interact with gRNA) greatly reduced target substrate cleavage in vivo, indicating that anchorage of the seed tail is important in the conformational changes of FrCas9 (Figures 4E and S13D). In comparison, the loss of the other seed region part interactions caused less reduction in editing efficiency (R61A/R69A). Amplicon sequencing and GUIDE-seq results showed that the N532A mutant improved the on-target cleavage efficiency of FrCas9, while the increase in the off-target rate was minimal (Figures 4E and S13D). The PAM-distal region of the heteroduplex was mainly recognized by the REC lobe and RuvC domain (Figures 4F and 4G). In detail, the REC3 domain (T615, Q728, and N732) recognized the backbone phosphate groups of the target DNA in the PAM-distal region (dT18, dG20, and dC21), and the side chain of N725 formed a hydrogen bond interaction with the deoxyribose of dT19 (Figure 4F). The phosphate backbone of sgRNA in the PAM-distal region (G2, G6, A7, and C8) interacted with the REC3 domain (K545, Y550, N694, and K986) (Figure 4G). In addition, the 2′-OH groups of G2 and A5 in sgRNA hydrogen bonded with K986 and Y536, respectively (Figure 4G). Multiple studies have reported that relieving the interaction between the PAM-distal region and the heteroduplex can improve the specificity of SpCas9 without affecting on-target cleavage efficiency.³⁰ To determine the impact of FrCas9 PAM-distal interaction amino acids on cleavage efficiency and specificity, we replaced the main interacting amino acids with alanine and found that these amino acids can be divided into 4 categories through GUIDE-seq and amplicon sequencing (Figures 4H–4K and S13E). In category 1, the N725A and Q728A mutants showed increased on-target efficiency with reduced off-target effects. In category 2, the Y550A/N694A mutant demonstrated improved on-target cleavage efficiency but with increased off-target effects. In category 3, the T615A mutant showed enhanced on-target efficiency with a minimal impact on off-target effects. In category 4, mutants such as N732A, K986A, and K545A exhibited a minimal effect on on-target efficiency while significantly reducing off-target effects. Taken together, these findings not only provide a deeper understanding of FrCas9’s unique off-target identification mechanism but also offer a foundation for engineering more specific and efficient CRISPR-Cas9 systems by manipulating these critical interactions within the heteroduplex.

To investigate whether V1103K and PAM-distal mutations exhibit synergistic effects, we analyzed combined variants of FrCas9 using amplicon sequencing and GUIDE-seq (Figure 4L). All combinatorial mutants demonstrate increased catalytic efficiencies compared to the wild-type (WT)-FrCas9 (Figure S14A). For fidelity, we found that the V1103K/N532A, V1103K/T615A, and V1103K/N545A mutants improved the specificity of the V1103K variant at the HEK293 SITEx2 site (Figures 3E and 4M). The introduction of the D660A mutation further reduced the off-target effect of the variant (V1103K/N532A) (Figure 4M). Notably, the V1103K/N732A variant (referred to as eFrCas9) exhibited no detectable off-target effects while maintaining the high activity brought by V1103K (Figures 3E, 4M, and S14B). Comprehensive biochemical and cellular validation further confirmed eFrCas9’s exceptional performance. In brief, eFrCas9 demonstrated superior on-target editing efficiency compared to WT-FrCas9 and SpCas9 while maintaining exceptional specificity with no detectable off-target activity (Figure 4N). In vitro kinetics analysis confirmed eFrCas9’s enhanced performance, achieving 92.08% cleavage efficiency in 0.25 min (versus 2 min for WT-FrCas9 and 5 min for SpCas9) with reduced off-target activity (9.58% versus 13.81% for WT-FrCas9 and 81.97% for SpCas9) (Figures S14C and S14D).

Based on these findings, our research highlights the potential of eFrCas9 variants to achieve high specificity and activity in gene editing. The ability of eFrCas9 to minimize off-target effects while maintaining high editing efficiency paves the way for its application in more challenging contexts. We next investigate the application of eFrCas9 in genetic diseases and gene therapy.

Highly efficient and precise deletion of large-fragment DMD exons 45–55 using eFrCas9

Large deletions of exons 45–55 in the DMD gene address approximately 60% of DMD cases³¹ (Figure 5A). The primary challenges with this approach include the inefficiency of large-fragment deletions and potential off-target effects. From the pooled AID-seq analysis (Figure 1), we identified and selected 10 sgRNAs each for eFrCas9 and SpCas9 that exhibited high on-target:off-target ratios. We subsequently employed GUIDE-seq to validate these sgRNAs in immortalized HsKMM cells, aiming to identify the best sgRNA with high on-target activity and low off-target effects. Our results showed that the on-target efficiency of eFrCas9 across the 10 sgRNAs was comparable to that of SpCas9 (Figure 5B). Notably, eFrCas9 exhibited significantly fewer off-target effects than SpCas9 (Figure 5C). No off-target effects were detected for eFrCas9 with 5 out of the 10 sgRNAs, while each of the remaining five sgRNAs exhibited only one off-target site (Figure 5C). In contrast, SpCas9 showed significantly higher numbers of off-target sites and reads, with multiple off-targets detected per sgRNA (Figure 5C). The on-target:off-target ratios of eFrCas9 were significantly higher than those of SpCas9 (Figure 5D).

Figure 5.

Comparative analysis of large-fragment deletion efficiency and off-target effects in DMD exons 45–55 using eFrCas9 and SpCas9

(A) Schematic diagram showing the deletion pattern of DMD exons 45–55. This range covers a significant proportion of mutations leading to Duchenne muscular dystrophy.

(B) GUIDE-seq on-target read counts of 10 sgRNAs each for eFrCas9 and SpCas9, with no significant difference between the two enzymes (Student’s t test, p = 0.68, n = 10).

(C) Number of off-target sites detected by GUIDE-seq (Student’s t test, ∗∗p < 0.01, n = 10).

(D) Comparison of on-target:off-target ratios for 10 sgRNAs each in eFrCas9 and SpCas9 (Student’s t test, ∗∗∗p < 0.001, n = 10). On-target:off-target ratios were calculated as log2 (on-target reads + 1)/(total off-target reads + 1).

(E) Fragment sizes for large deletions created by introducing sgRNAs in pairs. The deletions ranged from 436,216 to 621,421 bp.

(F) PCR gel electrophoresis showing the target bands resulting from successful large-fragment deletions. Schematic shows DMD exons 44–56, sgRNA target sites (scissors), ∼700 kb deletion region (exons 45–55), and PCR primers (arrows). Successful deletions reduce the amplicon size from ∼700 to ∼10 kb, producing detectable bands of <1 kb. Gel electrophoresis shows deletion products from different sgRNA pairs (e.g., Fr-DMD-163-208 indicates sgRNAs at positions 163 and 208).

(G) Sanger sequencing results for sgRNA pairing Fr-H45-g-447 and Fr-H55-g-87 for eFrCas9, demonstrating fewer indels and cleaner editing.

(H) Sanger sequencing results for sgRNA pairing Sp-H45-g-587 and Sp-H55-g-163 for SpCas9, showing a higher frequency of indels.

(I) Large-fragment deletion efficiency measured by PEM-seq. Blue bar chart: precise deletions. Red bar chart: indel-containing deletions.

See also Figure S15.

To investigate the precise efficiency of large-fragment deletions in DMD exons 45–55, we selected six eFrCas9 sgRNAs and three SpCas9 sgRNAs, which were introduced in pairs into HsKMM cells. The sizes of the deleted fragments ranged from 436,216 to 621,421 bp (Figure 5E). PCR results confirmed the presence of small target bands resulting from large deletions (Figure 5F), and Sanger sequencing corroborated these findings (Figure 5G). All sgRNA pairs facilitated successful large deletions with anticipated bands. Moreover, Sanger sequencing peaks indicated that eFrCas9 sgRNA pairs produced purer deletion products, whereas SpCas9 sgRNA pairs exhibited more insertions or deletions (indels) (Figure 5H). To further confirm these results, we employed primer extension-mediated sequencing (PEM-seq)³² to characterize the editing events induced by the paired sgRNAs. Precise large-fragment deletion was defined as the editing product in which the cut ends from the two sgRNAs directly join, thereby producing a repair product without any indels. The precise large-fragment-deletion efficiency of eFrCas9 reached up to 17.25% with the Fr-DMD-447-208 sgRNA pair. In contrast, no precise large deletions were detected in two SpCas9 groups (Figure 5I). Additionally, we calculated the proportion of large deletions accompanied by indels, which were also considered useful and safe for DMD exon 45–55 deletion therapy. eFrCas9 achieved a 26.66% large-fragment-deletion efficiency with indels by the Fr-DMD-447-208 pair, while SpCas9 reached a 19.21% efficiency with the Sp-DMD-587-163 pair. In total, the highest efficiency for deleting DMD exons 45–55 by eFrCas9 was 43.91%, combining both precise and indel deletions, whereas the highest efficiency by SpCas9 was only 19.21% with indels (Figure 5I). Indel size distribution analysis further revealed that eFrCas9 not only achieved higher precision in large deletion events but also generated fewer undesired indels at editing sites, supporting its superior performance for therapeutic applications requiring precise genomic modifications (Figure S15).

Discussion

This study presents the cryo-EM structures of the FrCas9-sgRNA-DNA ternary complex in both the RE and PreC states, offering insights into its unique gene-editing capabilities. The structures reveal FrCas9’s specialized recognition of the 5′-NRTA-3′ PAM, the PLL element, and an unusual overwinding of the sgRNA-DNA heteroduplex, which are pivotal for its enhanced specificity and efficiency. These distinct structural features of FrCas9 provide new insights into the CRISPR family.³³ By training artificial intelligence (AI) on the features, researchers can develop algorithms that are more effective at identifying valuable Cas proteins from large-scale datasets.^34,35 For example, the distinctive structural features of FrCas9, especially its protein-DNA interaction mechanisms, can serve as training templates for AI algorithms to identify better and novel Cas systems. In addition, the unique overwinding characteristics of the sgRNA-DNA heteroduplex in FrCas9 provide insights that can be used by AI models to screen for superior Cas systems.

The identification of key structural elements and their roles in FrCas9’s mechanism of action opens avenues for rational design and optimization of Cas9 variants with improved characteristics.^25,29,36 Our findings demonstrate that the PLL plays a crucial role in tuning FrCas9’s off-target sensitivity and catalytic efficiency. Consistently, a recent study demonstrated that the FnCas9 variant en31 (G1243T/E1369R/E1603H), engineered with the PLL element and the WED-PI domain, exhibited significantly enhanced activity and specificity.³⁷ The success of PLL engineering across multiple Cas9 orthologs suggests that it represents a scalable optimization strategy.

Building upon these intrinsic advantages of FrCas9, we further enhanced its performance through rational protein engineering. Targeted mutations in the PLL and PAM-distal region synergistically improve both the editing precision and efficiency of FrCas9, providing a foundation for the development of next-generation CRISPR technologies with increased fidelity and targeting ability.^38,39 The therapeutic potential of this engineered variant, eFrCas9, is exemplified by our demonstration of highly efficient and precise deletion of a large fragment in DMD exons 45–55, which could cure approximately 60% of patients with this genetic disorder. Notably, eFrCas9 showed fewer off-target effects and higher precise deletion efficiency compared to SpCas9, validating the structure-guided engineering approach and highlighting its value for therapeutic applications requiring large genomic modifications.

The structural and functional insights gained from this study advance our understanding of the mechanisms underlying FrCas9’s high fidelity and efficiency. FrCas9, with its unique NRTA PAM requirement, is able to precisely target promoter regions containing the TATA box. Combined with the optimized properties of eFrCas9, it has great potential for safe and efficient epigenome editing and is also a promising candidate for the development of promoter base editors and prime editors.

Limitations of the study

While our findings provide comprehensive insights into FrCas9’s structural and functional properties, several limitations warrant acknowledgment and suggest avenues for future investigation. First, given that the cryo-EM structures represent conformational states captured at specific time points, time-resolved structural studies could further elucidate the dynamic mechanistic transitions underlying catalysis. Second, although our engineered eFrCas9 variants exhibit robust performance in cell culture models, comprehensive in vivo validation in diverse animal models is currently lacking. Such validation is crucial for advancing therapeutic applications.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Zheng Hu (huzheng1998@163.com).

Materials availability

This study did not generate new unique reagents. Requests for the generated plasmids and strains in this study should be directed to the lead contact, Zheng Hu (huzheng1998@163.com).

Data and code availability

Atomic coordinates, maps, and structure factors of the reported cryo-EM structures have been deposited in the Protein Data Bank under accession numbers PDB: 9JWN (RE complex) and 9JWK (PreC complex) and in the Electron Microscopy Data Bank under accession codes EMD-61854 (RE complex) and EMD-61853 (PreC complex). All sequencing data generated in this study are available through the NCBI Sequence Read Archive (SRA) database under the accession code SRA: PRJNA1061652.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (grant nos. 32171192 and 92269106 to S.C., 32171465 and 32371541 to Z.H., 82102392 to R.T., and 82502718 to J.Z.), the National Key Research and Development Program of China (grant no. 2022YFC2304200 to S.C.), the Guangdong-Hong Kong-Macao University Joint Laboratory of Interventional Medicine Foundation of Guangdong Province (2023LSYS001 to S.C.), the China Postdoctoral Science Foundation (no. 2023M734090 to R.T.), and the Basic and Applied Basic Research Foundation of Guangdong Province (no. 2023A1515110777 to M.Y.).

Author contributions

M.Y. and S.L. contributed to protein purification and cryo-EM sample preparation; S.C., D.S., and Y.W. performed structural determination and validation; M.Y., S.L., and G.C. drew the figures; G.C., X.L., and R.T. performed the experiments; M.Y., S.L., and R.T. drafted the manuscript. J.Z., Z.H., and S.C. revised the manuscript; and M.Y., S.L., and G.C. contributed equally to this work.

Declaration of interests

R.T. is an employee of Generulor Co., Ltd.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
Escherichia coli Rosetta 2 (DE3)	This manuscript	Novagen
Critical commercial assays
SF cell Line 4D-NucleofectorTM X kit	This manuscript	V4XC-2024, Lonza, Germany
Cu R1.2/1.3 holey carbon grids	This manuscript	Quantifoil
Deposited data
GUIDE-seq	This manuscript	SRA: PRJNA1061652
Amplicon sequencing	This manuscript	SRA: PRJNA1061652
RE complex structure	This manuscript	PDB: 9JWN; EMDB: 61854
PreC complex structure	This manuscript	PDB: 9JWK; EMDB: 61853
Experimental models: Cell lines
HEK293T cell line	ATCC	ATCC CRL-3216
Oligonucleotides
RNA transcription template	This manuscript	TAATACGACTCACTATAG GTGCAAGACATAGAAATAA CCTGGTTTGAGTGTCTTGTT AATTgaaaAATTAACAAGATG AGTTCAAATCAGGCTCCTAG AGAGATCCGAACUTACCTTCA TGGCGGGCATTGTGCCCTT TTTTTT
sgRNA	This manuscript	GGUGCAAGACAUAGAAAUA ACCUGGUUUGAGUGUCUUG UUAAUUgaaaAAUUAACAAG AUGAGUUCAAAUCAGGCUC CUAGAGAGAUCCGAACUUA CCUUCAUGGCGGGCAUUGU GCCCUUUUUUUU
TS DNA for RE complex	This manuscript	TGCAATATACACAGGTTATT TCTATGTCTTGCAGTGAAGTGTT
NTS DNA for RE complex	This manuscript	AACACTTCACTGCAAGACATAGAA ATAACCTGTGTATATTGCA
TS DNA for PreC complex	This manuscript	TACACAGGTTATTTCTATGTCTTGCA
NTS DNA for PreC complex	This manuscript	TGTA
In vitro cleavage substrate	This manuscript	TGGCCATGCGACCCTCTTCATAGNTA
In vitro on-target cleavage substrate	This manuscript	TGGCCATGCGACCCTCTTCATAGGTA
In vitro off-target cleavage substrate	This manuscript	GGGCCAACCGACCCTCTTCATAGGTA
Software and algorithms
cryoSPARC (v.3.3.1)	Punjani et al.40	http://www.cryosparc.com
RELION (v.3.0_beta)	Scheres et al.41	https://github.com/3dem/relion
Chimera (v.1.16)	Pettersen et al.42	https://www.cgl.ucsf.edu/chimera/
COOT	Emsley et al.43	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
Phenix	Adams et al.44	https://phenix-online.org/download
ResMap (v1.1.4)	Swint-Kruse and Brown45	https://resmap.sourceforge.net/
ChimeraX (v.1.4)	Pettersen et al.46	https://www.cgl.ucsf.edu/chimerax/
FLASH	Magoč and Salzberg47	https://github.com/krishnaroskin/FLASH
BWA-MEM	Li and Durbin48	https://github.com/topics/burrows-wheeler-transform
GUIDESEQ (v.1.1)	Tsai et al.49	https://github.com/aryeelab/guideseq

Experimental model and subject details

The primary experimental model in this study was the human female embryonic kidney-derived cell line HEK293T. The cells were cultured in Dulbecco’s modified Eagle medium (DMEM) (Gibco, catalog number 11885084) at 37°C.

Method details

FrCas9 expression and purification

The sequence encoding Faecalibaculum rodentium Cas9 (FrCas9) was a gift from the First Affiliated Hospital of Sun Yat-Sen University. The FrCas9-H877A mutant (residues 1–1372) introduced by site-directed PCR was cloned into a pSumo expression vector between Xho1 and BamH1 sites, followed by a 6x-His tag and a ULP1 protease cleavage at N-terminal. The mutant protein expressed in Escherichia coli Rosetta 2 (DE3) (Novagen), which was cultured in TB medium at 37°C until cells reached OD₆₀₀ of 0.8–1.0. Then cells were grown at 20°C for 20 h and 0.1 mM IPTG was added into the medium for protein expression. Bacterial pellets were resuspended and lysed in lysis buffer (40 mM Tris-HCl, 500 mM NaCl, 20 mM Imidazole, 5% glycerol, pH 8.0). The Clarified lysate centrifuged and bound in batch to Ni Sepharose (Novagen), washed with the same buffer as lysis buffer. The His₆-Sumo-FrCas9-H877A protein was eluted with elution buffer (40 mM Tris-HCl, 500 mM NaCl, 500 mM Imidazole, 5% glycerol, pH 8.0). The His₆-Sumo tag was then incubated overnight with ubiquitin-like modifier protease (ULP1) at 4°C with gentle spinning. After purification by HiTrap Heparin HP affinity column (GE Healthcare) in ion exchange buffer (20 mM Tris-HCl (pH 8.0), 375 mM/1 M NaCl, 5% glycerol, 5 mM DTT), the purified protein was further loaded on Superdex200 16/600 column (GE Healthcare) in size-exclusion buffer (20 mM Tris-HCl, 150 mM NaCl, 5 mM DTT, pH 8.0). The final FrCas9 protein was concentrated to ∼30.5 mg/mL, flash-frozen in liquid nitrogen and stored at −80°C.

Nucleic acid preparation

The sgRNA was transcribed from a dsDNA template in a 5 mL transcription reaction (100 mM HEPES-K, pH 7.9, 12 mM MgCl₂, 2 mM spermidine, 2 mM of each nucleoside triphosphate, 30 mM DTT, PCR-assembled DNA template, T7 RNA polymerase). The transcription reaction was incubated at 37°C for 4 h. The re-dissolved RNA was purified by 8% denaturing TBE-urea PAGE, ethanol precipitated and dissolved in DEPC-treated water. The DNA oligonucleotides were commercially synthesized (Sangon Biotech) and were constructed in dsDNA. Target ssDNA and non-target ssDNA were mixed in a 1:1 ratio and hydrated in buffer (20 mM Tris-HCl, 150 mM NaCl, 5 mM DTT, pH 8.0). Then dsDNA was heated to 95°C for 5min and cooled at room temperature at leisure.

Reconstitution of the Cas9-sgRNA-dsDNA complex

For recombination of the RE complex, purified sgRNA with a final concentration of 1 mM EDTA pre-annealed at 94°C 5min, cooling to room temperature slowly. Then sgRNA was incubated with 10 mM MgCl₂ at room temperature for 5 min. After the FrCas9-H877A protein melted, MgCl2 with a final concentration of 10 mM was added and rewarming for 15 min. Next, the FrCas9-H877A protein and sgRNA were preincubated at 25°C 20 min at a 1.75:1 molar ratio. The ternary complex was constituted of FrCas9-H877A-sgRNA and dsDNA in a 1:1.2 ratio and allowed to incubate at 25°C 20min. Finally, the complex was purified by Superdex200 10/300 GL column (GE Healthcare) in buffer (20 mM Tris-HCl, 150 mM NaCl, 5 mM DTT, 1% Glycerol, pH 8.0). Peak fractions were collected every 50 μL for preparing frozen samples.

For recombination of the PreC complex, purified sgRNA was pre-annealed at 94°C 5min, then cooled to room temperature slowly. The FrCas9-H877A protein and sgRNA were preincubated in buffer (20 mM Tris-HCl, 150 mM NaCl, 5 mM DTT, pH 8.0) for 25°C 20 min at a 1:1.5 ratio. The ternary complex was constituted of Cas9, sgRNA and dsDNA in a 1:1.5:2.3 molar ratio and allowed to incubate at 25°C 20min. Finally, the complex was purified by Superdex200 10/300 GL column (GE Healthcare) in size-exclusion buffer (20 mM Tris-HCl, 150 mM NaCl, 5 mM DTT, pH 8.0). Then the MgCl₂ with a final concentration of 10 mM was added into fractions and incubated for 3h on ice. The ternary complex was rewarmed at room temperature for 20 min before preparing frozen samples.

Cryo-EM sample preparation and data collection

The RNA concentration of the ternary complex was measured by Nanodrop and diluted to 0.138–0.169 mg/mL. Cu R1.2/1.3, 300 mesh (or 200 mesh) holey carbon grids (Quantifoil) were glow-discharged in a PELCO easiGlow for 40s at 25mA. A volume of 4μL sample was loaded on grids and allowed to absorb for 15 s before blotted for 1–3 s with a blot force 1–3 s, then plunged freezing into liquid nitrogen-cooled ethane with a FEI Vitrobot equilibrated to 100% humidity at 8°C.

Data of FrCas9-sgRNA-DNA ternary complex were collected using an FEI Titan Krios cryo-electron microscope equipped with a Gatan K3 Summit direct electron detector operating at 300 kV. Images were automatically recorded with EPU with the pixel size of 0.8192 Å for RE complex datasets (8670 micrographs) and 0.83 Å for PreC complex datasets (2956 micrographs), at a defocus range between −1.0 and −2.0 μm in super-resolution mode, giving a total dose of 50 electrons per 32-frame.

Cryo-EM image processing

For PreC complex, images processing is firstly performed within RELION (v.3.0_beta). Movies underwent motion correction and CTF estimation, followed by multiple rounds of 3D classification, 3D auto-refine and maskcreate. Finally, 219,843 particles were transferred to the cryoSPARC (V.3.3.1).⁴⁰ Then particles were 2D classified (50 classes) in cryoSPARC without image alignment. Sixteen of the fifty classes showing blurred and damaged were excluded from further process. The particles (207,734 particles) from remain classes were used for ab-initio reconstruction (1 number), followed by a non-uniform (NU) refinement to achieve higher resolution and map quality. Then the refined particles were further classified into 4 number of ab-initio reconstructions followed by conformational states evaluation. Later, non-uniform refinement was performed for class 2 (85,201 particles) and class 3 (62,466 particles) respectively. The optimized 147,667 particles from class 2 and class 3 and the volume from class 3 were used for final non-uniform refinement. The overall resolution was estimated based on the gold-standard criterion of Fourier shell correlation (GSFSC)⁵⁰ of 3.46 Å.

For RE complex, images were processed in Relion (v.3.1.0)⁴¹ unless indicated otherwise. Movies first underwent motion correction and CTF estimation, followed by particle auto-picking (4,987,028 particles) and two rounds of 3D classification. Volume generated from class 4 was used as a template in the following particle picking (1,398,644 particles), which went through 2D classification subsequently. Then refined and recentered particles (1,385,148 particles) underwent one round of 3D classification and multiple rounds of 3D auto-refine with maskcreate. Finally, three rounds of CTF refinement and post-processing were performed on the most optimized volume (684,922 particles). The overall resolution was estimated based on the gold-standard criterion of Fourier shell correlation (GSFSC)⁵⁰ of 2.89 Å.

Model building, refinement and structure analysis

The initial model was derived from the high-resolution structures of PDB model 6O0Y. Manual Cas9 domain and nucleic acid placement and adjustment were completed using Chimera (v.1.16),⁴² COOT⁴³ and phenix.⁴⁴ Final models were validated with statistics from MolProbity scores, EMRinger scores and Ramachandran plots (Table 1). The local resolution of ternary complexes was evaluated by ResMap (v1.1.4).⁴⁵ Protein-nucleic acid interactions were analyzed using the PISA web sever,⁵¹ Characterization of the guide-protospacer duplex was performed using the w3DNA web sever.⁵² Structural figures were generated using ChimeraX.⁴⁶

AID-seq

Firstly, the genomic DNA was cut into approximately 500-bp fragments, followed by joining with hairpin i7 adaptors at the end. Then the free DNA fragments were digested using a mixture of three different exonuclease in 2–4 rounds to maximize the removal of false positives. The remaining double-ended “dumbbell” DNA were further incubated with the Cas9 ribonucleoprotein complex for new DNA ends exposure, which was then attached to the biotin-modified i5 adaptors. Afterward, the dual adaptor-ligation DNA was enriched by streptavidin magnetic beads, followed by linearization. Finally, the linearized DNA was used as a template for nested PCR to construct the next-generation sequencing library.⁵³ The summarized results can be found in Table S1.

Amplicon sequencing

Targeted pooled-amplicon deep sequencing libraries were generated by PCR and 150-bp paired-end sequencing data were produced by Illumina Hiseq2500 platform. Forward and reverse pair-end reads were merged into extended amplicons by FLASH software.⁴⁷ Then, all merged reads were aligned to the genomic reference using BWA-MEM⁴⁸ algorithm with standard parameters. The calculation of indels (insertion and deletions) in each read for the considered region was determined by extracting sequences of the targets ±10bp flanking regions with CRISPRMatch package.⁵⁴ The summarized results can be found in Table S2.

GUIDE-seq

The GUIDE-seq is an unbiased manner that could identify translocations and breakpoint hotspots. Firstly, HEK293T cell line using SF cell Line 4D-Nucleofector X kit (V4XC-2024, Lonza, Germany) were transfected with double-stranded oligonucleotide (dsODN), PX330-FrCas9 plasmid and PX330-sgRNA plasmid. Approximately 72h after nucleofection, cell genomic DNA was extracted. To better evaluate editing efficiency of off-targets, we executed dsODN-PCR experiment. Library construction were performed as the previous study described.⁵⁵ The DNA were sequenced using MGISEQ-2000RS sequencer after sheared, end-repaired and adding adapters. Finally, GUIDE-seq Data were analyzed using open-source GUIDESEQ software (version 1.1).⁴⁹ The off-targets were identified using original standards with mismatches ≤7.⁵⁶ The summarized results can be found in Table S3.

PEM-seq

PEM-seq (Primer Extension-Mediated Sequencing) is a method used to detect and characterize genomic DNA editing events, such as those induced by CRISPR/Cas9. This technique is essential for identifying double-strand breaks (DSBs), small insertions and deletions (indels), and large genomic rearrangements. To perform PEM-seq, genomic DNA is extracted from edited cells and fragmented with a sonicator to obtain DNA pieces of appropriate length. The fragmented DNA undergoes end-repair and adapter ligation to attach a known adapter sequence to the DNA ends. PCR amplification is then performed with specific primers targeting the adapter sequence and the desired genomic region, ensuring that only DNA fragments containing the targeted genomic region are amplified. The amplified products are subjected to high-throughput sequencing in MGISEQ-2000RS, and the data are analyzed using PEM-Q to determine the type of editing events.

Statistical analysis

GraphPad Prism v8.0 (GraphPad, Inc., USA) was used to carry out the statistical analysis. p < 0.05 was statistically significant (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). The summarized results can be found in Table S5.

In vitro cleavage assay (IVC)

For the in vitro DNA cleavage assay, the target sequence containing the 5′-GGTA-3′ PAM was first cloned into the pUC119 plasmid vector. Following sequence verification, the plasmid was linearized by PCR amplification to serve as the linear DNA substrate for the in vitro cleavage assay. Cas9 protein and sgRNA were pre-incubated at a 1:1 molar ratio in NEB reaction buffer at 25°C for 10 min. Subsequently, the linearized DNA substrate (3 nM final) was incubated with the Cas9-sgRNA complex (30 nM final) in 20 μL of reaction buffer at 37°C for durations ranging from 0.25 to 120 min. Reactions were terminated by the addition of 2 μL of 250 mM EDTA and 1 μL of 20 mg/mL Proteinase K, followed by incubation at 55°C for 10 min. Reaction products were separated by electrophoresis on a 1% agarose gel stained with GelRed and visualized. The cleavage efficiency in the IVC assays was quantified from the agarose gel electrophoresis images using ImageJ. Kinetic data were fitted with a one-phase exponential association curve using GraphPad Prism.

For the in vitro detection of the nickase activity of the FrCas9-H877A mutant, the linearized DNA substrate described above was substituted with supercoiled pUC119 plasmid DNA. This substrate was incubated with the corresponding Cas9-sgRNA complex at 37°C for 30 min. All subsequent steps were performed identically to the aforementioned in vitro cleavage assay protocol.

For the in vitro detection of the PAM preference of the FrCas9-N1137A mutant, the substrate described above was substituted with linearized DNA containing GGTA, GATA, GCTA and GTTA PAM. This substrate was incubated with the corresponding Cas9-sgRNA complex at 37°C for 5 min. All subsequent steps were performed identically to the aforementioned in vitro cleavage assay protocol.

Quantification and statistical analysis

Statistical analysis

The statistical details for any given analyses are provided in the corresponding figure legend. Additional information about statistical analysis can be found in the relevant method details sections.

Software

cryoSPARC (v.3.3.1).

RELION (v.3.0_beta).

Chimera (v.1.16).

COOT.

Phenix.

ResMap (v.1.1.4).

ChimeraX (v.1.4).

FLASH.

BWA-MEM.

GUIDESEQ (v.1.1).

Published: October 16, 2025