De Novo Structural Mining and Engineering of New TadA Deaminases for Highly Efficient Adenine Base Editors

Protein function is primarily determined by tertiary structure, yet enzyme discovery has relied heavily on sequence homology. Here, we present a structure-guided mining framework integrating AlphaFold2-predicted structures with Foldseek-based 3D alignment to identify novel TadA deaminases with low sequence similarity to ecTadA. Through rational engineering, two deaminases, eMaTadA and eErTadA, were developed into functional adenine base editors (ABEs) when fused with nCas9. Both editors matched ecTadA-8e in A-to-G editing efficiency and product purity while substantially reducing bystander editing and DNA off-target activities. In human CD34+ HSCs, eMa-ABE achieved up to 80% editing at the HBG1/2 promoter with robust fetal hemoglobin restoration. LNP-mediated delivery enabled efficient correction of the ATP7B P992L mutation in murine hepatocytes with no detectable off-target effects. Additionally, an eMaTadA variant with an expanded editing window enabled high-efficiency gene silencing via start codon targeting. This work establishes a structure-guided approach for rapid protein discovery and engineering, expanding the ABE repertoire for clinical applications.