Molecular Dynamics Simulation of Protein-Protein Complex Formation in a Crowded Environment

Principal Investigator:
Martin Zacharias

Physik-Department T38, Technische Universität München

Local Project ID:

HPC Platform used:
SuperMUC of LRZ

Date published:

The process of protein-protein complex formation is of fundamental importance for a better understanding of a variety of biological processes. In a cellular environment the high concentration of surrounding proteins can influence the association process between proteins. Aim of the research project was to simulate the formation of specific and non-specific protein-protein complexes and to investigate the effect of additional protein molecules (crowding) on complex formation in atomic detail. 

The project also involved methodological advancements to accurately calculate the free energy profiles of biomolecular interactions. So far protein association has been studied mostly using simplified coarse-grained protein models without an explicit inclusion of solvent molecules. Alternatively, Brownian Dynamics simulations using rigid partner proteins can also be used to study protein-protein association and encounter [1]. However, such techniques largely neglect the conformational flexibility of proteins and often approximate the aqueous environment poorly. Both conformational adaptation of protein partners and hydration effects are of critical importance for protein-protein interactions and need to be accounted for accurately in order to realistically simulate and understand complex formation events. In the present study atomistic Molecular Dynamics (MD) simulations of proteins in the presence of explicit solvent and ions were used.

As a model system the complex formation of two well known small proteins, Colicin E9 (E9) and the Immunity protein 9 (Im9), known to form very specific high-affinity complexes [2, see Fig. 1]. The E9 protein is a DNA hydrolyzing enzyme and Im9 is a high-affinity protein inhibitor that blocks the E9 activity upon binding. Both proteins consist of ~100 residues and structures of the complex and the isolated partners are known [2]. Specific binding of protein partners may occur directly from the separate fully solvated unbound states of the partner molecules. However, it is also possible that the process involves several intermediate states (encounter complexes) and initial non-specific binding followed by diffusion on the protein surfaces towards the specific complex structure. Since in MD simulations each protein molecules is flexible it will be possible to also directly investigate conformational changes associated with binding events at an unprecedented resolution in time and space. In addition, the role of solvent and dehydration (de-wetting) during binding can be studied at a high level of detail during multiple binding events. The simulation model system contained multiple proteins and allowed studying whether the crowding environment in a cell (high protein density) has a direct influence on the association of protein molecules.


The simulations can help to answer several fundamentally important questions on the mechanism of protein-protein complex formation. How do transient non-specific encounter complexes look like? Do they still have a wet (solvated) or partially dry interface structure? This question is also related to the possibility of anomalous diffusion that may play a role for protein-protein binding in an in vivo situation.


Are there intermediates on the path towards a specific protein-protein complex? Does the complex formation process involve the diffusion of proteins on the surface of the partner (initially bound at a nonspecific site) before reaching the native binding site? Are there trapped incorrectly bound states competing with specific binding? These questions cannot be answered accurately using simpler coarse-grained protein models or rigid protein partners and a rough implicit description of the surrounding solvent.

MD simulations of multiple copies of the colicin E9 and Im9 protein were started from random initial placements in simulation box which contained up to 27 copies of each protein. The simulation system also contained surrounding explicit water molecules and ions (200 mM) resulting in a total number of atoms of 1.2·106. The mean distance of ~3-5 nm between proteins corresponded to a dense protein solution similar to the crowded environment of a cell. After equilibration of the simulation system several production runs using the GROMACS software [3] were performed resulting in an aggregate simulation time of almost 1 us (990 ns). This was possible due to the excellent scaling of the program for the large simulation system on multiple cores. Typically, between 512-2048 cores were used.

During the simulations multiple association and dissociation events were observed. This included formation of specific contacts (contacts that are observed in the native protein-protein complex) and non-specific contacts. During the aggregate simulation time ~25% of the E9 and Im9 pairs formed transient complexes. The great majority of association events corresponded to non-specific binding events with a mean number of contacts that corresponded to ~30% of the number of contacts observed in the native specific complex. A contact is here defined as a minimum distance between two residues in two proteins of < 0.625 nm. This cutoff includes also transient association events that contain interfacial hydration layers. Around 10% of the transient complexes included specific contacts (that also occur in the native protein complex). The analysis of the transiently formed complexes is still ongoing and requires the design of new tools to systematically investigate the large number of possible protein-protein encounters.

Besides of the study of protein-protein interactions using unrestrained MD simulations new methods to extract the change in free energy upon biomolecular association were developed employing the parallel computing capabilities of SuperMUC. The method is based on the well-established Umbrella sampling approach to induce the dissociation or association of two binding partners along a distance coordinate. In such simulations accurate convergence of calculated free energy changes is difficult to achieve and one obtains usually different numbers for the association and dissociation processes. By allowing exchanges between neighboring umbrella sampling windows and by adding a potential that offsets the free energy change along the reaction coordinate in an iterative process we were able to achieve very rapid convergence of calculated free energy changes along the separation coordinate [4]. This method is well suited to systematically study non-specific versus specific binding for pairs of proteins.

On-going Research / Outlook

In our ongoing research we currently design tools to systematically analyze the transient association events observed during the MD-simulations of the colicin E9 and Im9 proteins. The main goal is to characterize the types of transient contacts. Are these contacts mainly electro-statically driven and involve interfacial waters or do they involve mostly nonpolar hydrophobic protein-protein contacts? Another interesting issue is the role of conformational adaptation during transient complex formation. Of particular interest is also the diffusion of proteins on the surface of the protein partners. The ongoing and future analysis is also focused on the question of the time range for the transient interactions which likely requires longer simulation times. It is planned in the future to perform more extensive simulations possibly on even larger systems. The binding affinity of long-lived transient complexes will be investigated using umbrella sampling simulations (indicated above) and will be compared to the native binding mode. This will help to relate the strength of non-specific association to the affinity of protein-protein binding in a native complex.

Principal investigator: Prof. D. Martin Zacharias

Researchers: Aliaksei Krukau, Rainer Bomblies, Alexander Knips, Florian Kandzia, Manuel Luitz, Guiseppe La Rosa, Katja Ostermeir, Fabian Zeller

References and Links

[1] Gabdoulline RR, Wade RC. Brownian dynamics simulation of protein-protein diffusional encounter. Nature Methods.14 (1998) 329-41.
[2] Papadakos G, Wojdyla JA, Kleanthous C. Nuclease colicins and their immunity proteins. Quart. Rev. Biophys.114 (2011) 1-47.
[4] Zeller F, Zacharias M Adaptive Biasing combined with Hamiltonian Replica Exchange to improve umbrella sampling free energy simulations. J ChemTheo Comput 10 (2014) 703-710.

Scientific Contact:

Prof. Dr. Martin Zacharias
Physik-Department T38
Technische Universität München
James-Franck-Str. 1, D-85747 Garching/Germany

Tags: LRZ Life Sciences Health and Medicine