Stereoelectronic effects in stabilizing protein-N-glycan interactions revealed by experiment and machine learning.

The energetics of protein-carbohydrate interactions, central to many life processes, cannot yet be manipulated predictably. This is mostly due to an incomplete quantitative understanding of the enthalpic and entropic basis of these interactions in aqueous solution. Here, we show that stereoelectronic effects contribute to stabilizing protein-N-glycan interactions in the context of a cooperatively folding protein. Double-mutant cycle analyses of the folding data from 52 electronically varied N-glycoproteins demonstrate an enthalpy-entropy compensation depending on the electronics of the interacting side chains. Linear and nonlinear models obtained using quantum mechanical calculations and machine learning explain up to 79% and 97% of the experimental interaction energy variability, as inferred from the R2 value of the respective models. Notably, the protein-carbohydrate interaction energies strongly correlate with the molecular orbital energy gaps of the interacting substructures. This suggests that stereoelectronic effects must be given a greater weight than previously thought for accurately modelling the short-range dispersive van der Waals interactions between the N-glycan and the protein.