Bayesian Optimization-Assisted Engineering of Formate Dehydrogenase Encapsulation in Multivariate Zeolitic Imidazolate Framework

Abstract

Engineering of multivariate zeolitic imidazolate frameworks (mZIFs) offers substantial potential for optimizing enzyme encapsulation by enhancing encapsulation efficiency (EE), enzyme loading capacity (P_loading), retained enzymatic activity (REA), and protection. However, this area remains underexplored. In this study, we rationally employed three imidazole-based ligands with distinct functionalities─HeIM (2-ethylimidazole), HTz (1,2,4-triazole), and HIM (1-(2-hydroxyethyl)imidazole)─to fine-tune hydrophobicity and defect simultaneously within FDH@mZIF (FDH = formate dehydrogenase). Leveraging an iterative Bayesian optimization-assisted training-design-synthesis-measurement workflow, we efficiently identified F190 as the best FDH@mZIF, achieving EE = 89.3%, REA = 14.9%, and P_loading = 30.3 wt%. This establishes F190 as the leading FDH-based biocatalyst in the literature. The optimal FDH-mZIF interactions in F190 were reflected by minimal structural perturbation of encapsulated FDH, as evidenced by the ATR-FTIR and fluorescence studies. Additionally, F190 can effectively safeguard the encapsulated FDH against thermal and proteolytic degradation and catalyze CO₂-to-formate conversion while maintaining activity for at least five cycles without significant activity loss.