PREreview of Exploring ensemble structures of Alzheimer’s amyloid β (1-42) monomer using linear regression for the MD simulation and NMR chemical shift

Molecular dynamics simulations are often used to examine protein folding and dynamics at a detailed molecular level. In order to accurately reproduce the relevant conditions in simulations, models and methods are being continuously improved by various groups all over the world. In this paper, the authors propose an altogether different approach of altering the simulation results to match with experimental data, and they use this method to study the misfolding of an amyloid peptide (Aβ42).

First, they experimentally characterized Aβ42 using dynamic light scattering (DLS), circular dichroism (CD) and nuclear magnetic resonance (NMR) spectroscopy and compared the observations with replica-exchange molecular dynamics (REMD) simulations of Aβ42, using both implicit and explicit solvent models. The computational data did not agree well with the experiments – the predicted hydrodynamic radius from explicit and implicit solvent simulations were respectively lower and higher than that obtained from DLS experiments. Moreover, the proportion of secondary structures in simulations did not match with experiments and the chemical shifts obtained from NMR experiments were also significantly different from that predicted in simulations.

To bridge the experimental and the computational results, they developed a multiple linear regression analysis to find the most realistic ensemble from all the structures sampled during the MD simulation. They compared the predicted chemical shift of the sampled ensemble to the actual chemical shift obtained from NMR experiments and assigned a probability score for each conformation. This produced the “most compelling ensemble”, consisting of various monomeric conformations and their populations. The hydrodynamic radius calculated from this ensemble was much closer to that observed in DLS experiments, and the proportion of secondary structures were also similar to those predicted using CD spectra.

To assess the generalizability of their approach, the authors carried out REMD simulations of two other well-studied folded proteins. The ensembles obtained after multiple linear regression were quite similar to previously published solution NMR structures. Although the authors claim that their approach would be broadly useful for obtaining and studying protein ensemble structures, in my opinion –

• The linear regression step filters the simulation trajectory to best fit the experimental chemical shifts, so the scope of this approach to provide new insights (beyond what is already known from experiments) is not clear.

• Since proteins can behave very differently even in slightly different environments, the linear regression results would be limited to the specific experimental conditions as well as the quality of experimental data.

• The methods described here might be useful mostly for supplementing/visualizing NMR chemical shift data by generating trajectories which match them.

• An alternative approach could have been to use the NMR chemical shift data to improve the model itself – allowing us to generate new hypothesis and make specific predictions, which could then be experimentally validated.