Application of advanced sampling and analysis methods to predict the structure of adsorbed protein on a material surface

Abstract

The use of standard molecular dynamics simulation methods to predict the interactions of a protein with a material surface have the inherent limitations of lacking the ability to determine the most likely conformations and orientations of the adsorbed protein on the surface and to determine the level of convergence attained by the simulation. In addition, standard mixing rules are typically applied to combine the nonbonded force field parameters of the solution and solid phases the system to represent interfacial behavior without validation. As a means to circumvent these problems, the authors demonstrate the application of an efficient advanced sampling method (TIGER2A) for the simulation of the adsorption of hen egg-white lysozyme on a crystalline (110) high-density polyethylene surface plane. Simulations are conducted to generate a Boltzmann-weighted ensemble of sampled states using force field parameters that were validated to represent interfacial behavior for this system. The resulting ensembles of sampled states were then analyzed using an in-house-developed cluster analysis method to predict the most probable orientations and conformations of the protein on the surface based on the amount of sampling performed, from which free energy differences between the adsorbed states were able to be calculated. In addition, by conducting two independent sets of TIGER2A simulations combined with cluster analyses, the authors demonstrate a method to estimate the degree of convergence achieved for a given amount of sampling. The results from these simulations demonstrate that these methods enable the most probable orientations and conformations of an adsorbed protein to be predicted and that the use of our validated interfacial force field parameter set provides closer agreement to available experimental results compared to using standard CHARMM force field parameterization to represent molecular behavior at the interface.

Fig. 1.

Representative model system for simulations with HEWL on the HDPE surface. The protein is displayed in cartoon representation, fixed water layer bellow the HDPE surface by ball-and-stick representation, and the HDPE surface and Cl⁻ counterions by vdW surface representation. The mobile water layer is shown in points for clarity. The graphical representation was generated using VMD (Ref. 89).

Fig. 2.

Graphical description of the angle of the adsorbed protein relative to the material surface (Θ°). The angle was defined as the angle between the protein vector (vector N × C) and the surface normal (vector n). The protein vector was calculated as the cross-product of two vectors, both of which start from center of mass of the protein and pass through N-terminus (vector N) or C-terminus (vector C).

Fig. 3.

Representative states from the top five most populated clusters obtained in 100 ns of sampling (for each simulation) of HEWL on HDPE (a) using standard MD starting from two distinct orientations of the protein on the surface, and (b) using TIGER2A advanced sampling using eight replicas, each starting with a different orientation of the protein on the surface.

Fig. 4.

(a) Ensemble average SASA profiles for the entire sequence of the protein are shown. The SASA profiles of the active site residues, E35 and D52, are also highlighted. The correlation between simulation and experimental SASA profiles of selected amino-acid residues is displayed on the right-hand side. (b) The active site residues in the native state and in the adsorbed state of the protein are shown. The adsorbed states are representative conformations from the top cluster in the TIGER2A simulations for the CHARMM FF and the tuned IFF for interfacial interactions.

Fig. 5.

Number of sampled states in each cluster (dark gray columns and above numbers, left-hand axis) and the relative free energies of each cluster (light blue columns, right-hand axis) of the adsorbed protein states obtained in TIGER2A enhanced sampling simulations using the CHARMM FF and the tuned IFF for the interfacial interactions.

Fig. 6.

Comparison of the predicted adsorption behavior of HEWL protein on HDPE in TIGER2A simulations using the CHARMM FF and the tuned IFF (seed 1 and seed 2) for the interfacial force field. Representative states from the two most populated clusters are displayed. On the top of each figure the overall orientation of the protein is shown; followed with the images framed in black, which highlight the residues found within 5 Å from the surface (i.e., residues predicted to directly interact with the surface). The bottom images show the residues interacting with the surface, as viewed from below the surface.

Fig. 7.

Assessment of the convergence between seed 1 and seed 2 in TIGER2A simulations of HEWL on HDPE using tuned IFF for the interfacial interactions. The simulation seeds were merged and cluster analysis was run on this merged ensemble of states, which identified eight clusters. The percent of states from each seed was then calculated for each cluster. The convergence percentages from cluster 1 to cluster 8 are 50.0%, 18.2%, 74.6%, 15.4%, 9.5%, 25.0%, 0%, and 0%, respectively.

Fig. 8.

Probability densities of the adsorbed protein's C_α RMSD from the crystal structure characterizing the adsorption-induced change in conformation (top plots), and angle between the protein's position and the normal vector from the surface characterizing adsorbed orientation (bottom plot) in regular MD simulations and TIGER2A simulations, both using tuned IFF for the interfacial interactions. Regular MD simulations started from two distinct orientations of the protein on the surface, and TIGER2A simulation started from two seeds, and each seed had eight different orientations of the protein on the surface.

Fig. 9.

Conformational assessment of the protein. Secondary structure content and C_α RMSD from the crystal structure of HEWL obtained over 100 ns of TIGER2A sampling are shown. The secondary structure contents were calculated for every 5 cycles, which is equivalent to 5 ns of TIGER2A.