Fluctuation Flooding Method (FFM) for Enhancing Conformational Sampling of Proteins

Abstract

A simple and powerful method for enhancing conformational transitions of proteins, referred as Fluctuation Flooding Method (FFM), is proposed. In FFM, biologically relevant anisotropic fluctuations of proteins are firstly characterized by the principal component analysis (PCA) from trajectories of molecular dynamics (MD) simulations. Then largely fluctuated snapshots along the anisotropic directions are assumed as candidates that tend to induce conformational transitions with high probabilities and selected as initial structures for Multiple Independent Molecular Dynamics (MIMD) simulations. In MIMD, a series of short time MD simulations are performed with re-generating initial velocities for selected candidates to enhance conformational transitions. Due to the re-organizations of the initial velocities, some of the candidates might make a conformational transition to another meta-stable states. The multiple trajectories from MIMD are characterized by the PCA to extract largely fluctuated snapshots as candidates for the next MIMD step. These procedures are repeated until distributions of the conformational sampling are well converged. In addition to FFM, Multiple Independent Umbrella Sampling (MIUS) using reference structures selected from MIMD can provide Free Energy Landscape (FEL). FEL calculations by the combination with MIUS enable us to quantitatively determine conformational transition pathways and estimate structural stabilities of newly found meta-stable states.To assess the reliability of the proposed method, we applied FFM to a toy model and confirmed that probability densities calculated from FFM showed a good agreement with the analytical solution. As an application to a real protein system, we applied FFM to T4-Lysozome in explicit water. Although 1-μs canonical MD failed to sample the closed state, FFM with 10-ns MIMD succeeded in finding conformational transitions from the open to closed states, where the minimum root-mean square deviation between the predicted closed and experimental X-ray structures was 0.76 A.