Design of Safe and Efficient Adenine Base Editors via Protein Language Model Screening for Osteoarthritis Treatment

ABSTRACT Base editors enable precise genome modification and have emerged as a promising therapeutic approach for correcting diseases caused by single‐nucleotide variants. While the current efficient version of adenine base editors (ABEs), such as ABE8e, exhibits exceptional efficiency for A‐to‐G conversions, their clinical translation is hindered by persistent high off‐target editing effects. Here, we applied artificial intelligence‐assisted design a safe ABE variant, RDLot‐ABE, with a narrow(4 nt) editing window and substantially lower DNA off‐target editing activity compared to ABE8e. Moreover, targeted knockdown of Fscn1 for osteoarthritis treatment using RDLot‐ABE alleviates cartilage degradation in explants derived from human patients. Notably, intra‐articular delivery of the RDLot‐ABE to reduce Fscn1 effectively arrests disease progression in a murine osteoarthritis model. These findings establish RDLot‐ABE as a safe and precise tool, expanding the clinical potential of gene editing therapies.