AI-designed Cas9-like genome editing nuclease

Characteristics

Retains the prototypical Type II Cas9 RuvC/HNH domain architecture despite 403 amino acid substitutions from SpCas9 and 182 from any known natural CRISPR protein. Uses canonical SpCas9 single-guide RNA scaffolds and requires an NGG PAM, with reduced editing at non-NGG sites. Across 28 endogenous loci in HEK293T cells, produces a median on-target indel rate of 56.4% versus 47.1% for SpCas9, with off-target indel rates of 0.32% versus 6.1%, a 95% reduction. Genome-wide Digenome-seq profiling found 25.6 ± 13.3 potential off-target sites per guide versus 88.5 ± 33.2 for Cas9, with up to 553-fold higher specificity at validated sites. Lacks immunodominant SpCas9 T-cell epitopes for HLA-A*02:01 and shows lower iELISA signal than SpCas9, suggesting reduced immunogenicity.

Applications: Demonstrated genome editing in HEK293T cells, human iPSCs, and MRC-5 fibroblasts. Converted into prime editors OpenCRISPR-PE2 and OpenCRISPR-PE7 using standard and engineered pegRNAs, matching Cas9-based prime editor on-target activity with fewer predicted and validated off-target sites, including delivery as an engineered virus-like particle. Compatible with base editing architectures using canonical SpCas9 guide RNA scaffolds.

Limitations: Requires an NGG PAM, restricting targetable sites, with reduced editing at non-NGG PAMs. Research and commercial use requires accepting Profluent's ethical-use license agreement rather than being freely open-source. As a recently designed protein, long-term immunogenicity and in vivo delivery performance have not yet been reported.

References

1. Ruffolo et al. (2025). Design of highly functional genome editors by modelling CRISPR–Cas sequences. Nature - Ruffolo 2025 OpenCRISPR
2. Hwang et al. (2026). High-fidelity genome and prime editing enabled by the AI-designed openCRISPR-1. Genome Med - Hwang 2026 openCRISPR-1 Fidelity