46 Encounter complexes and dimensionality reduction in protein-protein association

Dima Kozakov, Pirooz Vakili, Ioannis Ch. Paschalidis, Sándor VajdaJournal of Biomolecular Structure and Dynamics 2015 · DOI

TL;DR

Proteins rarely act alone, forming complexes through transient "encounter complexes." This study uses advanced computational docking and experimental data to reveal that protein association follows surprisingly constrained, "canyon-like" pathways, similar to how proteins navigate DNA. This dimensionality reduction offers a new perspective on protein binding and could improve computational prediction methods.

What Problem Does This Paper Address?

Understanding how proteins find and bind to their partners is a fundamental challenge in biology. This process isn't a direct lock-and-key fit but involves a dynamic search through various intermediate states, known as "encounter complexes." These states are fleeting and difficult to observe directly, making their structures and the pathways leading to the final complex poorly understood. Existing computational docking methods often generate many low-energy structures, but it's been unclear which of these represent biologically relevant intermediate states. This paper aims to bridge this gap by using experimental data to validate computational models of these encounter complexes and to characterize the energy landscape governing protein association.

Methods

The researchers employed a multi-pronged approach:

  • Physics-Based Docking (PIPER): They used a Fast Fourier Transform (FFT)-based docking program called PIPER to exhaustively sample the conformational space of two interacting proteins. This method systematically explores all possible relative orientations and positions, generating a large ensemble of potential complexes. The energy function used in PIPER includes van der Waals, electrostatic, and solvation terms.
  • Paramagnetic Relaxation Enhancement (PRE): This experimental technique was crucial for validating the computational models. By attaching paramagnetic labels (EDTA-Mn²⁺) to one protein and measuring their effect on NMR signals of the other, researchers can infer distances between specific atoms. This provides experimental constraints on the proximity of different parts of the interacting proteins.
  • Ensemble Averaging: Instead of comparing experimental PRE data to a single docked structure, the authors averaged the theoretical PRE profiles calculated from their entire ensemble of low-energy docked structures. This approach accounts for the dynamic nature of encounter complexes.
  • Principal Component Analysis (PCA) on Energy Landscapes: To understand the "shape" of the energy landscape governing protein association, they transformed the 6D rotational/translational space into a more manageable Euclidean space using exponential maps. PCA was then applied to identify the principal directions of motion and energy variation in the vicinity of the native complex. This revealed "permissive" (low energy barriers) and "restrictive" (high energy barriers) subspaces.

Datasets

The study utilized existing protein structures from the Protein Data Bank (PDB) for docking calculations. Specifically, they used unbound structures of the N-terminal domain of Enzyme I (EIN) and the histidine-containing phosphocarrier protein (HPr) (PDB entry 1ZYM and 2JEL, respectively) for their primary example. They also analyzed a diverse set of 42 protein pairs from the protein docking benchmark (Chen et al., 2003) to generalize their findings. The experimental PRE data used for validation was from previously published studies on the EIN-HPr system.

Key Insights and Ideas

  • Encounter Complexes are Biologically Relevant: The study provides strong evidence that many of the "false positive" low-energy structures generated by docking programs are not artifacts but represent actual encounter complexes that proteins pass through during association.
  • Dimensionality Reduction in Protein Binding: The most striking finding is that protein-protein association is not a free exploration of all possible orientations. Instead, it's highly constrained, with proteins tending to move along specific, low-energy pathways. This "dimensionality reduction" means that a large fraction of the possible motions is confined to a low-dimensional subspace.
  • "Canyon-Like" Energy Landscapes: The energy landscape of protein association is visualized as a canyon, with a smooth, funnel-like bottom representing the permissive subspace where proteins can move relatively freely. This is contrasted with higher energy regions that act as walls, restricting movement.
  • Analogy to DNA Binding: This constrained, pathway-driven association is remarkably similar to how proteins locate target sites on DNA, suggesting a common principle in molecular recognition.
  • Validation of Docking Ensembles: The agreement between experimental PRE data and the ensemble-averaged theoretical PRE profiles strongly validates the use of large ensembles of docked structures for studying transient states.

Key Results

  • Improved PRE Prediction: Accounting for encounter complexes generated by docking significantly improved the correlation between theoretical and experimental PRE profiles compared to using only the native complex structure. For the EIN-HPr system with a distant paramagnetic label, the correlation coefficient increased from 0.47 to 0.705.
  • Dominant Permissive Subspace: Principal Component Analysis revealed that for most protein pairs studied, the first two principal components (eigenvectors) accounted for over 75% of the total variance in the low-energy conformations. This indicates a highly restricted, low-dimensional subspace for association.
  • Consistent Restrictive Subspaces: The restrictive subspaces identified by PCA were remarkably similar between two different docking programs (PIPER and RosettaDock), suggesting that this dimensionality reduction is an inherent property of protein-protein association, not an artifact of a specific computational method.
  • Generalizability: The findings were consistent across 42 different protein complexes, reinforcing the universality of the observed dimensionality reduction.

Limitations and Future Work

  • Rigid Body Approximation: While local adjustments were allowed, the primary focus was on rigid-body motions. The study acknowledges that significant conformational changes upon binding might require different approaches.
  • Computational Cost: While PCA helps understand the landscape, generating the initial large ensembles of docked structures is computationally intensive.
  • Future Work: The authors suggest that understanding these reduced dimensionality subspaces can lead to more efficient docking algorithms by focusing the search on these permissive directions. Further exploration of how these pathways are influenced by factors like protein size, charge, and specific interaction types would be valuable. The study also opens avenues for better understanding the dynamics of protein complex formation and evolution.

Summary updated

BibTeX
@article{kozakov2015encounter,
  author = {Dima Kozakov and Pirooz Vakili and Ioannis Ch. Paschalidis and Sándor Vajda},
  title = {{46 Encounter complexes and dimensionality reduction in protein-protein association}},
  journal = {Journal of Biomolecular Structure and Dynamics},
  year = {2015},
  doi = {10.1080/07391102.2015.1032595},
}
Figure