Stable and Functionally Diverse Versatile Peroxidases Designed Directly from Sequences.

Area of Science:

• Biochemistry
• Enzyme Engineering
• Computational Biology

Background:

• White-rot fungi utilize oxidoreductases, particularly versatile peroxidases (VPs), for efficient lignin decomposition.
• Recombinant production of VPs is challenging, limiting their research and industrial applications.
• Accurate enzyme structures are crucial for computational optimization, but experimental structures are scarce for many enzymes.

Purpose of the Study:

• To assess the reliability of deep-learning-based *ab initio* structure prediction for computational enzyme design.
• To engineer and functionally express novel versatile peroxidases (VPs) with improved properties.
• To demonstrate the utility of AI-driven methods for exploring natural enzyme diversity.

Main Methods:

• Utilized deep-learning *ab initio* structure prediction to generate models for VP optimization.
• Employed PROSS computational design for one-shot stability and functional enhancement of VPs.
• Expressed designed VP variants in yeast and characterized their activity, stability, and reactivity profiles.

Main Results:

• AI-predicted structures served as reliable starting points for computational enzyme design.
• Four designed VPs with up to 43 mutations were successfully expressed in yeast, unlike their wildtype counterparts.
• Three designs displayed significant diversity in reactivity and environmental tolerance.

Conclusions:

• Deep-learning structure prediction combined with computational design enables efficient optimization of enzymes with challenging expression.
• This approach expands the scope of computational enzyme engineering, facilitating the discovery of novel biocatalysts.
• The methodology allows for direct exploitation of functional diversity within natural enzyme families from genomic data.