A Nature study published July 22 combined ProteinMPNN redesign with laboratory evolution to produce botulinum neurotoxin proteases that were more stable and better able to acquire new functions than natural starting enzymes. One redesigned lineage reached more than 79-fold greater selected specificity for the neurodegeneration-linked protein ataxin-2, but the work remains preclinical protein engineering rather than a demonstrated treatment.
Researchers at the Broad Institute, Harvard University and collaborating institutions have shown that AI-stabilized proteins can give laboratory evolution a stronger starting point. Their study, published in Nature on July 22, used ProteinMPNN to redesign three botulinum neurotoxin proteases before subjecting them to directed-evolution campaigns.
The result is a protein-engineering workflow, not a new version of commercial Botox. Botulinum proteases were used as programmable research enzymes because their cutting activity can be redirected toward different protein targets.
Why redesign the starting enzyme
Directed evolution repeatedly mutates and selects proteins for a desired function, but useful mutations can also reduce stability or expression. The team used ProteinMPNN to propose amino-acid sequences that preserved the enzymes' overall structures while improving their stability and retaining or increasing catalytic performance.
In side-by-side evolution campaigns, the redesigned starting points consistently reached higher-activity solutions than the corresponding natural enzymes. The researchers found that some beneficial mutations worked only in the redesigned backgrounds, supporting their explanation that the extra stability opened access to functional sequence combinations that natural starting proteins could not tolerate.
The experiments covered three distinct botulinum proteases and multiple target substrates. The paper reports that 58 of 74 initial ProteinMPNN designs were functional, while 33 matched or exceeded the natural BoNT/E protease's cleavage rate. Selected designs also showed higher soluble expression and thermal stability.
The ataxin-2 result and its limits
The team then evolved a redesigned BoNT/E protease to cleave ataxin-2, a protein implicated in neurodegenerative disease. The best redesigned lineage achieved more than 79-fold greater selected specificity for ataxin-2 than the best lineage evolved from the natural enzyme. In the reported assays, the redesigned proteases also showed no detected cleavage of the native SNAP25 target at the tested concentration.
Those findings establish a research method, not therapeutic efficacy. The study demonstrates protein redesign and biochemical selection; it does not show that an ataxin-2-targeting protease is safe or effective in patients. The authors also disclose patent applications related to the work, and senior author David Liu reports several biotechnology-company relationships.
For protein engineers, the practical result is that model-guided stabilization can be more than a final polishing step. It can expand the set of mutations that directed evolution can explore, potentially improving the odds of reaching a stable, specific enzyme with a new function.
Key Points #
- 1ProteinMPNN redesigned three botulinum neurotoxin proteases before laboratory evolution, improving the stability of the starting enzymes while retaining catalytic function.
- 2A redesigned BoNT/E lineage reached more than 79-fold greater selected specificity for ataxin-2 than the best lineage evolved from the natural enzyme.
- 3The work is a preclinical protein-engineering study and does not establish a safe or effective treatment or modify commercial Botox.
Scoring Rationale #
The study presents a reproducible AI-plus-directed-evolution workflow with broad protein-engineering relevance and a quantified ataxin-2 result, while remaining preclinical and therefore below the highest immediate-impact tier.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.