Modular protein scaffold architecture and AI-guided sequence optimization facilitate de novo metalloenzyme engineering

Incorporating metal cofactors into computationally designed protein scaffolds provides a versatile route to novel protein functions, including the potential for new-to-nature enzyme catalysis. However, a major challenge in protein design is to understand how the scaffold architecture influences conformational dynamics. Here, we characterized structure and dynamics of a modular de novo scaffold with flexible inter-domain linkers. Three rationally engineered variants with different metal specificity were studied by combining X-ray crystallography, NMR spectroscopy, and molecular dynamics simulations. The lanthanide-binding variant was initially trapped in an inactive conformational state, which impaired efficient metal coordination and cerium-dependent photocatalytic activity. Stabilization of the active conformation by AI-guided sequence optimization using ProteinMPNN led to accelerated lanthanide binding and a 10-fold increase in k_cat/K_m for a photoenzymatic model reaction. Our results suggest that modular scaffold architectures provide an attractive starting point for de novo metalloenzyme engineering and that ProteinMPNN-based sequence redesign can stabilize desired conformational states.