Structure-guided engineering of AI-derived adenine base editors for nuclear and mitochondrial DNA editing

Traditional adenine base editors (ABEs), primarily derived from TadA, suffer from bystander editing and limited diversity, while artificial intelligence (AI)-designed deaminases such as Deam-P32 exhibit lower efficiency and precision than state-of-the-art editors such as ABE8e. Here, we report the OpenABE variants through the structure-guided evolution of Deam-P32 by optimizing DNA engagement and the base-contacting pocket and incorporating C-terminal extensions from ABE8e. These variants achieved a 16-36-fold increase in A-to-G conversion efficiency over Deam-P32, matching the editing efficiency of ABE8e across 29 endogenous nuclear loci while mitigating bystander cytosine editing. OpenABEs also reduced ATC motif bystander editing and produced fewer guide RNA-independent off-target effects on DNA and RNA. Further, we adapted these variants to mitochondrial DNA editing by designing OpenABE-TALEDs, which yielded editing efficiencies comparable or superior to those of ABE8e-TALEDs. Delivery through engineered virus-like particles further enhanced specificity and product purity. These results demonstrate that structure-guided refinement of AI-designed deaminases can produce precise, versatile base editors for nuclear and mitochondrial genomes, expanding the research and therapeutic applications of genome editing toolkits.