Sharpening AI-Designed Base Editors: OpenABE Improves Nuclear and Mitochondrial DNA Editing Efficiency

The challenge in gene editing is often not simply whether DNA can be rewritten, but whether a single, clean change can be made at the correct location. OpenABE1.1 and OpenABE1.2, developed by a South Korean research team, aim to improve efficiency, product purity, and delivery simultaneously, bringing AI-designed adenine base editors closer to becoming practical biomedical tools.

Adenine base editors can convert A to G at specific positions without directly cutting both strands of DNA, and may therefore be used to correct certain disease-causing variants. Starting with AI-derived editors, the researchers combined protein structure prediction with experimental screening, introducing amino acid substitutions including L78F, P103S, and A105R. They then separately attached two carboxyl-terminal extension sequences derived from ABE8e to create the two OpenABE versions.

The research was published in the peer-reviewed journal *Nucleic Acids Research*. According to the cellular experiments reported in the paper, the degree of improvement varied across targets, but OpenABE significantly increased A-to-G conversion at some nuclear DNA and mitochondrial DNA sites, reaching as much as 36 times the efficiency of the original AI-derived editor. This result shows that AI-generated protein sequences can still be further refined through structure-guided engineering, rather than being put directly into use once their design is complete.

Beyond efficiency, the team also examined whether the editors inadvertently altered nearby bases. At a FANCF target containing an ATC sequence, unintended cytosine bystander editing with OpenABE1.1 and 1.2 was 7.0- to 12.2-fold lower than with the highly active ABE8e, while maintaining similar A-to-G activity. This difference may improve the purity of the final product, but the current data mainly come from specific sequences and experimental conditions, so the same effect cannot be assumed at all genomic locations.

To avoid the risks that may arise from sustained editor expression, the researchers also packaged OpenABE into engineered virus-like particles for more transient delivery into cells. The paper states that this delivery method can improve product purity; in some analyses, specificity ratios reached as high as 181.26 for OpenABE1.1 and 69.59 for OpenABE1.2, outperforming plasmid delivery. However, these figures depend on the targets, doses, and calculation methods used in the study and cannot yet be directly translated into the degree of safety for human treatment.

Mitochondrial DNA editing has particular potential value because many mitochondrial diseases arise from single-base variants, while the available tools for precise repair remain limited. However, moving from cultured cells to treatment will still require addressing tissue delivery, immune responses, long-term off-target effects, and the proportion of editing within different cells. The study has not yet provided animal efficacy or human data, so OpenABE is currently better regarded as a candidate platform than as a proven treatment.

The research team has made its sequencing data and supplementary methods publicly available to facilitate external replication and comparison, while the authors also disclosed patent applications related to the technology. The key next steps include not only identifying more targets that show large fold improvements, but also demonstrating in disease models that the edits can restore biological function and assessing the trade-offs using consistent genome-wide and transcriptome-wide methods.