Flexible Docking in Solution Using Metadynamics

Francesco Luigi Gervasio, Alessandro Laio, Michele ParrinelloJournal of the American Chemical Society 2005 · DOI

TL;DR

This paper introduces a novel application of metadynamics, a computational method that simulates molecular dynamics in a "biased" way, to model the complex process of drug-like molecules (ligands) binding to their protein targets (receptors) in a realistic water environment. The key innovation is its ability to not only predict the final bound structure and binding energy but also to map out the entire energy landscape, including the pathways and energy barriers involved in the binding process.

What Problem Does This Paper Address?

Predicting how a small molecule (ligand) will bind to a protein target is a cornerstone of drug discovery. This process, known as docking, is challenging because both the ligand and the protein can change their shapes (are flexible), and the binding occurs in a complex environment like water. Traditional docking methods often simplify these aspects, either by assuming rigid molecules or by using simplified models of the solvent. While these methods are fast and useful for initial screening, they struggle to accurately predict binding affinities and the detailed pathways of binding. More accurate methods, like molecular dynamics (MD), can capture flexibility and solvation but are computationally expensive, making it difficult to explore the vast conformational space and find the true binding minimum within realistic timescales. This paper addresses the need for a method that can accurately and efficiently explore the entire binding process, from a freely solvated ligand to its bound state within a flexible receptor, and quantify the energy landscape involved.

Methods

The core of this work is the application of metadynamics, a computational technique that enhances molecular dynamics simulations by adding a history-dependent bias potential. This bias discourages the system from revisiting configurations it has already explored, effectively pushing it to discover new low-energy states.

Here's how it works in this context:

  • Collective Variables (CVs): Instead of simulating all degrees of freedom, metadynamics focuses on a few key "collective variables" that describe the essential aspects of the binding process. For ligand-receptor docking, these typically include:
  • The angle between the ligand's principal axis of inertia and the line connecting the ligand and receptor centroids.
  • The distance between the ligand and a specific atom or the centroid of the receptor's binding site.
  • Bias Potential: As the simulation progresses, metadynamics adds Gaussian "hills" to the free energy surface (FES) along these CVs. These hills effectively raise the energy of already visited regions, encouraging the simulation to explore new areas and eventually discover the binding site and other relevant minima.
  • Free Energy Surface Reconstruction: The sum of these added Gaussians, when the simulation is run long enough, provides a negative image of the FES. This allows researchers to visualize the entire energy landscape, including binding pockets, intermediate states, and transition barriers.
  • Refinement: The authors also describe a two-stage approach: an initial coarse-grained exploration with larger Gaussians, followed by a refinement step with smaller Gaussians for higher accuracy.

This approach is novel in its ability to reconstruct the complete FES, including transition states and intermediate minima, which is a significant advantage over methods that only find the lowest energy state.

Datasets

The paper evaluates the method on four established ligand-protein complexes:

  1. β-trypsin/benzamidine: A classic, relatively rigid system.
  2. β-trypsin/chlorobenzamidine: Similar to the above, but with a slightly modified ligand to test sensitivity to binding energy changes.
  3. Immunoglobulin McPC-603/phosphocholine (PC): A system known for its protein flexibility playing a significant role.
  4. Cyclin-dependent kinase 2 (CDK2)/staurosporine: A more complex inhibitor to test the method's performance on larger molecules.

The crystal structures of these complexes were used as starting points for the simulations. Experimental binding energies were used for comparison.

Key Insights and Ideas

  • Beyond the Minimum: The most significant insight is that metadynamics doesn't just find the "best" binding pose; it maps the entire journey. This is crucial for understanding how a drug works and for designing better drugs by optimizing binding pathways.
  • Realistic Environment: By simulating in explicit water, the method captures the crucial role of solvation in the binding process, which is often simplified or ignored in other methods.
  • Flexibility is Key: The method successfully handles flexible ligands and receptors, demonstrating its applicability to more realistic biological systems. The difference in protein backbone RMSD between the trypsin and McPC-603 cases highlights the method's ability to capture protein conformational changes.
  • Efficiency: While computationally intensive, metadynamics offers a more efficient way to explore the FES compared to traditional methods like extensive umbrella sampling, achieving comparable accuracy in significantly less time.
  • Predicting Subtle Differences: The method can quantitatively predict small differences in binding energy arising from minor ligand modifications (e.g., benzamidine vs. chlorobenzamidine).

Key Results

  • Accurate Docked Geometries: The predicted docked geometries closely matched experimental crystallographic structures, with average RMSDs typically below 0.5 Å.
  • Quantitative Binding Energies: The calculated binding free energies (ΔGbinding) were in good agreement with experimental values, generally within 1 kcal/mol. For example, β-trypsin/benzamidine binding energy was predicted as -6 kcal/mol compared to the experimental -6.5 kcal/mol.
  • Complete FES Reconstruction: The method successfully reconstructed the 2D free energy surfaces for all four systems, revealing intermediate minima and transition barriers.
  • Efficiency Comparison: The metadynamics approach achieved comparable accuracy to 15 ns of umbrella sampling in a fraction of the time (around 3-7 ns of simulation).
  • Sensitivity to Ligand Modification: For β-trypsin/chlorobenzamidine, the method accurately predicted a ~1 kcal/mol decrease in binding affinity compared to benzamidine.

Limitations and Future Work

  • Choice of Collective Variables: The accuracy and efficiency of metadynamics heavily depend on the judicious choice of collective variables. For systems requiring significant, non-local protein rearrangements, these CVs would need to be explicitly included, potentially increasing computational cost.
  • Overfilling the FES: The authors caution against running simulations for too long, as "overfilling" the FES can lead to spurious minima and potentially protein unfolding.
  • Computational Cost: While more efficient than some alternatives, metadynamics simulations are still computationally demanding, requiring significant CPU hours.
  • Future Directions: The authors suggest that the detailed FES information provided by metadynamics can be a powerful tool for optimizing binding pathways and advancing rational drug design. Further exploration of more complex systems and protein conformational changes is implied.

Summary updated

Abstract

We apply our recently developed metadynamics method to the docking of ligands on flexible receptors in water solution. This method mimics the real dynamics of a ligand exiting or entering an enzyme and in so doing reconstructs the free energy surface. We apply it to four docking cases: beta-trypsin/benzamidine, beta-trypsin/chlorobenzamidine, immunoglobulin McPC-603/phosphocholine, and cyclin-dependent kinase 2/staurosporine. In every case studied, the method is able to predict the docked geometry and the free energy of docking. Its added value with respect to many other available methods is that it reconstructs the complete free energy surface, including all the relevant minima and the barriers between them.
BibTeX
@article{gervasio2005flexible,
  author = {Francesco Luigi Gervasio and Alessandro Laio and Michele Parrinello},
  title = {{Flexible Docking in Solution Using Metadynamics}},
  journal = {Journal of the American Chemical Society},
  year = {2005},
  doi = {10.1021/ja0445950},
}
Figure