Biosensor and machine learning-aided engineering of an amaryllidaceae enzyme

Area of Science:

• Biotechnology
• Enzyme Engineering
• Machine Learning

Background:

• Therapeutic alkaloid production faces challenges due to slow biocatalyst engineering.
• Amaryllidaceae alkaloids, like galantamine, are valuable but difficult to synthesize, necessitating plant extraction.
• Current methods rely on low-yielding plant sources, limiting industrial-scale biomanufacturing.

Purpose of the Study:

• To develop an efficient biosensor-machine learning technology stack for rapid biocatalyst development.
• To engineer an Amaryllidaceae enzyme in Escherichia coli for improved therapeutic alkaloid synthesis.
• To overcome limitations in current biomanufacturing processes for complex plant secondary metabolites.

Main Methods:

• Directed evolution was employed to create a sensitive biosensor for 4'-O-methylnorbelladine.
• A structure-based residual neural network (MutComputeX) was developed for enzyme variant generation.
• Engineered enzyme variants were rapidly screened using the developed biosensor.

Main Results:

• A highly sensitive and specific biosensor for a key Amaryllidaceae alkaloid intermediate was developed.
• MutComputeX identified enzyme variants with a 60% increase in product titer.
• Engineered variants exhibited 2-fold higher catalytic activity and 3-fold reduced off-product formation.

Conclusions:

• The biosensor-machine learning technology stack significantly accelerates biocatalyst engineering.
• This approach enables efficient industrial-scale biomanufacturing of therapeutic alkaloids.
• Structural insights elucidated mechanisms for beneficial mutations, guiding future enzyme optimization.