Assessing Substrate Scope of the Cyclodehydratase LynD by mRNA Display-Enabled Machine Learning Models.

Area of Science:

• Biochemistry
• Enzymology
• Natural Product Synthesis

Background:

• YcaO enzymes install azoline heterocycles into peptides, enhancing bioactivity.
• Azoline moieties contribute to the structural rigidity and potent bioactivities of natural products.
• Enzyme substrate promiscuity offers potential for synthesizing novel peptide-based inhibitors.

Purpose of the Study:

• To investigate the substrate promiscuity of the YcaO cyclodehydratase, LynD, using mRNA display.
• To identify trends in LynD's substrate processing and tolerance.
• To develop a predictive model for LynD activity and understand enzyme evolution.

Main Methods:

• Utilized mRNA display, a high-throughput peptide library technology.
• Assayed a large library of potential LynD substrates to determine activity trends.
• Employed deep learning to model substrate processing and epistatic interactions.

Main Results:

• Identified specific amino acid preferences and disfavored residues for LynD modification.
• Found that charged residues and multiple adjacent heterocyclizations are disfavored.
• Developed a deep learning model accurately predicting LynD substrate processing.

Conclusions:

• LynD exhibits specific substrate preferences, disfavoring charged residues and multiple heterocyclizations.
• A deep learning model can predict and explain LynD's peptide modification.
• Understanding YcaO substrate scope aids in designing novel peptide therapeutics and probes.