Reengineering of a flavin-binding fluorescent protein using ProteinMPNN.

Area of Science:

• Protein engineering
• Computational biology
• Biochemistry

Background:

• Machine learning advances enable novel protein design and engineering methods.
• ProteinMPNN is a tool that predicts amino acid sequences for user-defined protein structures.
• Previous ProteinMPNN applications focused on standard proteins and binding proteins.

Purpose of the Study:

• To evaluate ProteinMPNN's efficacy in reengineering a non-proteinaceous ligand-binding protein, specifically the flavin-based fluorescent protein CagFbFP.
• To assess if ProteinMPNN can generate functional variants of fluorescent proteins while preserving ligand-binding capacity.

Main Methods:

• The native backbone of CagFbFP was maintained.
• 20 amino acids critical for flavin mononucleotide (FMN) binding were kept unchanged.
• ProteinMPNN predicted sequences for the remaining amino acids.

Main Results:

• ProteinMPNN suggested replacing 36-48 of 86 amino acids, resulting in 55%-66% sequence identity to the original CagFbFP.
• Three experimentally tested designs showed varying expression levels but all successfully bound FMN.
• The engineered proteins exhibited fluorescence, thermal stability, and other properties comparable to the native CagFbFP.

Conclusions:

• ProteinMPNN is effective for generating diverse, unnatural variants of fluorescent proteins.
• The tool can be broadly applied to reengineer proteins without compromising their ligand-binding functions.