Fig. 2: AI-Assisted design of programmable APOBEC platform for engineering ProAPOBEC enzymes.

a Crystal structure of Human APOBEC3A (PDB: 4XXO), AlphaFold2-predicted structures of native APOBEC proteins: Mouse APOBEC3, Human APOBEC1, Mouse APOBEC1, and Rat APOBEC1, showcasing the NTD, C-to-U deaminase domain, and CTD. b Phylogenetic tree of the AID/APOBEC family, with major clades highlighted in different colors. c Schematic representation of the assembly of ProAPOBECs. d Estimated number of potential ProAPOBEC candidates based on AI-predicted AID/APOBEC deaminase domains. e AlphaFold2-predicted structures of AID/APOBEC proteins. Top: AI-predicted structures of native AID/APOBECs. Bottom: AI-predicted structures of ProAPOBECs. f The programmable CU-REWIRE5 system with ProAPOBEC variants. Top: Upgraded CU-REWIRE5, featuring integration of ProAPOBEC variants for targeted C-to-U editing. Middle: Schematic depiction of the modular assembly process of CU-REWIRE5. The ProAPOBECs domain, constructed from a selection of natural APOBEC deaminase domains (APOBEC_DD) derived from various sources, is engineered for specific C-to-U base editing. Accompanying this, the ePUF10 domain includes 10 RNA recognition units, each designed for the specific detection of a unique RNA base type. Bottom: Illustration of the precise engagement of CU-REWIRE5 with the target RNA, showcasing the accurately binding and editing C₄₅₉ of EGFP mRNA. g Illustrations depicting the components of CU-REWIRE5, ProAPOBECs, and their deaminase donors. h Various editing rates on C₄₅₉ of EGFP mRNA of CU-REWIRE5s, assembled with ProAPOBECs in (i) and Supplementary Fig. 2d. Editing rates were measured by Sanger-sequencing with triplicates (refer to methods), values represent mean ± SEM. i Protein expression levels of CU-REWIRE5s in HEK293T cells. The CU-REWIREs were Flag-tagged, anti-Flag antibody was used to examine CU-REWIREs, and GAPDH was used as a loading control.