The design of artificial photoenzymes by incorporating synthetic chromophores into proteins represents a promising strategy to achieve non-natural biocatalytic transformations with high levels of stereocontrol. Selecting an appropriate protein scaffold is a crucial step in this approach, which so far has been limited to naturally occurring proteins. Here, we tested the suitability of computationally designed scaffolds for this purpose. We chose a de novo helical bundle protein that has a central cavity for small molecule binding but no inherent catalytic activity. To generate a starting point for photoenzyme engineering, we installed a thioxanthone-based triplet sensitizer via cysteine bioconjugation. Guided by computational modeling and molecular dynamics (MD) simulations, three rounds of directed evolution toward the 2 + 2 photocycloaddition of a 3-alkenyloxy-substituted quinolone resulted in enzyme variants with catalytic efficiencies of kcat/Km > 1000 M–1 s–1 and opposite enantioselectivity. Upon visible-light irradiation, both product enantiomers were accessible with quantitative yield and >90:10 enantiomeric ratio. Furthermore, we obtained high-resolution crystal structures of the evolved designer enzymes. When exposing crystals of substrate-bound protein to blue light, we observed product formation in crystallo and could rationalize the enantioselectivity. Our work highlights the potential of de novo designed protein scaffolds to efficiently generate and evolve stereoselective artificial photoenzymes.
Boesen et al. (2026) studied this question.