De Novo Structural Mining and Engineering of TadA Deaminases for High Efficiency Adenine Base Editors

Conventional enzyme discovery relies primarily on sequence homology and phylogenetic relationships, limiting the identification of functionally related but sequence-divergent proteins. Here, we developed a structure-guided mining framework integrating AlphaFold2-predicted structures with Foldseek-based three-dimensional alignment to identify previously uncharacterized TadA scaffolds with conserved folds and divergent sequences. Two candidates, eMaTadA and eErTadA, were engineered into functional adenine base editors (ABEs) by fusion to Cas9 nickase. The resulting editors mediated robust A-to-G editing with high product purity and restricted bystander editing. Under expression-stratified and activity-matched conditions, eMa-ABE and eEr-ABE showed reduced guide-dependent DNA off-target editing relative to ABE8e, while orthogonal R-loop assays revealed lower SpCas9-independent deamination of exposed ssDNA. In human CD34⁺ hematopoietic stem and progenitor cells, eMa-ABE achieved up to approximately 80% editing at established HBG1/2 promoter targets and induced fetal hemoglobin expression following erythroid differentiation. Lipid nanoparticle delivery corrected approximately 50% of ATP7B P992L alleles in mouse liver, with no detectable increase in editing at the examined computationally predicted off-target loci. Furthermore, embedding eMaTadA within Cas9 expanded the editing window from A3-A9 to A3-A14 and enabled efficient start-codon-mediated gene inactivation. Collectively, this study establishes a structure-guided strategy for discovering and engineering sequence-divergent deaminases and expands the molecular diversity and therapeutic potential of the ABE toolkit.