Back
Science

AI-Redesigned Protease Starting Points Enhance Directed Evolution Outcomes

View source

AI-Driven Redesign Supercharges Enzyme Evolution, Achieving 79-Fold Gain in Specificity

A new study in Nature demonstrates how combining AI-based protein design with automated evolution dramatically improves the engineering of botulinum neurotoxin (BoNT) proteases.

The research team integrated the deep-learning model ProteinMPNN with the stabilization method PROSS to create stabilized variants of BoNT proteases. These redesigned enzymes were then subjected to phage-assisted continuous evolution (PACE).

Key Findings

  • Redesigned enzymes outperformed wild-type starting points in matched evolution campaigns, adapting faster and accessing mutational spaces that were non-functional in natural backgrounds.
  • When evolved to cleave human ataxin-2, an AI-redesigned protease variant achieved over 79-fold greater specificity than the top enzyme evolved from the wild-type sequence.
  • The top redesigned variant showed 16% sequence divergence from the natural protein.

Methods

  • 74 ProteinMPNN designs were tested; 78% retained catalytic activity.
  • Top variants showed catalytic efficiencies 1.7–2.8 times that of wild-type and increased melting temperatures (up to 59.5°C).
  • Parallel continuous evolution campaigns with 44 lines were run on an automated eVOLVER platform.
  • Substrates of varying difficulty (SNAP25 variants and ataxin-2) were used to challenge the enzymes.

Implications

The study demonstrates that AI redesign can mitigate the stability-activity trade-off in enzyme engineering, expanding accessible mutational space and enabling evolution of highly specific non-native catalytic activities.

Further research is needed to verify these advantages across other enzyme families and to establish delivery, efficacy, and safety in disease models.