Applications of SpCas9-induced genome editing are often limited by off-target effects or insufficient on-target editing. Several high-fidelity variants (eSpCas9(1,1), Cas9-HF1, HypaCas9, Cas9 R63A/Q768A, evoCas9, HiFi Cas9, and Sniper-Cas9 (Sniper1)) have been developed to address these issues. However, the modifications designed to reduce off-target cleavage often compromise on-target activity, leading to a trade-off between activity and specificity. A high-fidelity variant with SpCas9-like activity would significantly enhance the applicability of SpCas9-based genome editing in gene therapy and genetic screening. This study aimed to develop such a variant.

The authors cite previous research demonstrating the trade-off between specificity and activity in high-fidelity SpCas9 variants. They also reference studies that detail the mechanisms of SpCas9 activity, focusing on DNA unwinding and the role of mismatches in hindering this process. Existing high-fidelity variants, such as Sniper1, show improved specificity but at the cost of reduced on-target efficacy. The authors' previous work on directed evolution of SpCas9 using a Sniper screen is also referenced as a foundation for this study.

The researchers employed directed evolution of Sniper1, a previously identified high-fidelity Cas9 variant. The Sniper screen, a method combining positive and negative selection pressures in E. coli, was used, employing a different sgRNA and target sequence pair with a mismatch at position 13 to identify a mutational hotspot. Saturation mutagenesis at this hotspot (amino acid 1007) generated various Sniper1 variants, and their activities at matched and mismatched target sequences were assessed using plasmid delivery and subsequent deep sequencing. Two promising variants, Sniper2L (E1007L) and Sniper2P (E1007P), were selected for further high-throughput evaluation. High-throughput evaluations utilized lentiviral delivery and electroporation of ribonucleoprotein (RNP) complexes, a clinically relevant delivery method. Three lentiviral libraries (A, B, C) with varying sgRNA-target pairs were used. Library A assessed PAM compatibility and mismatch tolerance. Library B validated variant activity on numerous target sequences with NGG PAMs. Library C used perfectly matched tRNA-N20 sgRNAs. The effects of different delivery methods and various types of mismatches (single-base, wobble, transversion, two-base, and three-base) on activity and specificity were systematically assessed. Single-molecule fluorescence resonance energy transfer (smFRET) was utilized to investigate the sequence specificity of DNA unwinding by different SpCas9 variants. Finally, deep learning models (DeepSniper) were developed to predict Sniper2L and Sniper1 activities at matched and mismatched target sequences.

Sniper2L demonstrated significantly higher specificity than Sniper1 while retaining high on-target activity comparable to SpCas9, overcoming the trade-off observed in other high-fidelity variants. This superior specificity was confirmed using both lentiviral and RNP delivery methods. The high specificity of Sniper2L stemmed from its superior ability to avoid unwinding target DNA with even a single mismatch, as evidenced by smFRET assays. Sniper2L showed significantly reduced activity at mismatched target sequences compared to Sniper1 and Sniper2P, particularly in PAM-proximal and -distal regions. The DeepSniper model, a deep learning tool developed using the data from the high-throughput evaluations, accurately predicted Sniper2L and Sniper1 activities with high correlation coefficients (r = 0.96, R = 0.94 for matched targets; r = 0.92, R = 0.90 for mismatched targets). A novel high-throughput method was developed to assess sgRNA activity using RNP delivery, which is highly relevant to ex vivo gene editing therapies.

The study's findings directly address the challenge of balancing high specificity and activity in high-fidelity Cas9 variants. Sniper2L stands out as an exception to the previously observed trade-off, achieving both high specificity and activity, making it a more effective tool for precise genome editing. The development of DeepSniper further enhances the utility of Sniper2L by providing a predictive tool for sgRNA design. The high-throughput RNP-based method addresses a practical limitation in sgRNA screening for clinical applications. The improved specificity of Sniper2L is likely due to its superior ability to discriminate against mismatched target DNA during the DNA unwinding phase of the editing process, as revealed by smFRET analysis. Although the single molecule unwinding activity of Sniper2L is lower than that of other variants, this does not fully correlate to its on-target editing activities.

Sniper2L, a novel high-fidelity Cas9 variant, exhibits high on-target activity and exceptional specificity, overcoming the limitations of existing high-fidelity variants. Its superior performance, confirmed through various delivery methods and supported by a predictive model (DeepSniper), positions it as a valuable tool for efficient and precise genome editing applications. Future studies could explore additional modifications to further enhance Sniper2L's properties or adapt the directed evolution strategy to create variants with alternative specificity profiles.

The study primarily focused on HEK293T cells. Although previous research suggests similar relative activities across cell types, testing Sniper2L in diverse cell lines would strengthen the findings. The 6-TG selection in the RNP-based high-throughput method reduced library coverage, which could be improved by using highly active sgRNAs targeting HPRT. The single-molecule unwinding data did not fully capture the on-target gene editing activities of Sniper2L.