Computational design of cysteine proteases.

Area of Science:

• Biochemistry
• Enzyme Engineering
• Computational Biology

Background:

• Enzyme design has advanced, but challenges remain for reactions with high energy barriers, like polypeptide hydrolysis.
• Amide bonds in proteins are highly stable due to a significant energy barrier to hydrolysis, persisting for centuries in aqueous solutions.

Purpose of the Study:

• To de novo design enzymes capable of hydrolyzing the polypeptide backbone with high efficiency and sequence specificity.
• To leverage a novel deep learning method for creating enzymes with unique structures distinct from natural proteases.

Main Methods:

• Utilized a new deep learning method, RFD2-MI, for de novo enzyme design.
• Engineered enzymes employing an activated cysteine nucleophile to catalyze polypeptide backbone hydrolysis.
• Validated enzyme designs through structural analysis, comparing crystal structures to computational models.

Main Results:

• Achieved rate enhancements (kcat/kuncat) of up to 3 × 10^7 for polypeptide hydrolysis.
• Designed enzymes with novel folds dissimilar to natural proteases (TM-score < 0.50).
• Demonstrated high accuracy in design methodology with crystal structures closely matching designed models (Cα RMSDs < 1.2 Å).

Conclusions:

• The deep learning approach successfully generated novel enzymes with significant catalytic activity for polypeptide hydrolysis.
• The designed enzymes possess unique structural folds, distinct from those found in nature.
• This methodology holds broad potential for developing new proteases for biotechnological and medical applications.