Reaction-Coordinate-Dependent Non-Markovian Friction Governs Protein-Folding Dynamics

It is common to project the full atomic-resolution representation of a protein onto a one-dimensionalreaction coordinate (RC) to capture the protein-folding kinetics.

As a direct consequence ofthis dimensionality reduction, non-Markovian friction emerges in the framework of the general-ized Langevin equation (GLE).

All previous applications of GLEs to protein folding employed anRC-independent friction memory function and therefore did not account for the different frictionin the folded and unfolded states.

Using a recently derived GLE with RC-dependent mass andfriction memory function, we introduce a novel method to extract memory functions from timeseries data via a conditional Volterra equation.

When applied to molecular dynamics (MD) data ofsix fast-folding proteins, we find strongly RC-dependent memory friction in line with the intuitiveexpectation that friction is higher in the folded than in the unfolded state due to internal proteinfriction.

Our numerically efficient method to simulate the GLE confirms the accuracy of the GLEparameter extraction by comparison with the MD data.

We show that RC-dependent memoryfriction not only adds physical insight into the folding process but also significantly improves thedescription of protein folding kinetics using low-dimensional RCs.