Engineered base editors with reduced bystander editing through directed evolution

Abstract

Base editors enable precise genome modification but are constrained by bystander edits that limit their applicability. Existing strategies to enhance precision often compromise efficiency and remain highly sequence dependent. Here we present a parallel engineering approach that optimizes both guide RNAs and the deaminase enzyme to minimize bystander editing without sacrificing activity. We designed a library of 3′-extended guide RNAs and identified context-dependent variants that improved specificity. Using a precision-driven phage-assisted evolution system and protein language models, we evolved adenine base editor variants two- to threefold more precise than adenine base editor ABE8e while maintaining high efficiency across a library of thousands of human pathogenic contexts in vitro. Our findings establish a scalable framework for precision engineering of base editors, addressing a major challenge in genome editing.