Chapter 40Structure-Based Protein-Ligand Binding
This chapter discusses uncertainty and dynamics in the protein-ligand docking, and then reviews some methods for flexible protein-ligand docking.
Uncertainty in Experimentally Derived Structures
Uncertainty in X-Ray Structures
We now discuss possible uncertainties in crystal structures from X-ray diffraction [1].
Interpretation of X-ray Diffraction Data The raw diffraction data are generally converted to an electron density map by inverting the Fourier transformation after determining the phases. More precisely, the electron density is written by
where is the volume of the unit cell, is the structure factor, and is the phase angle.
Then a solid atomic model is fit to interpret the density map by iteratively adjusting the model parameters, such as the atomic coordinates. During the interpretation of the density map data, we must consider the following issues:
- Missing or Uninterpretable Densities During the interpretation of the density map data, the missing or uninterpretable densities need to be corrected according to the crystallographers’ experience.
- Resolution of the Data The resolution of the data is expressed in Å. The resolution of a diffraction pattern, , can be obtained by Bragg’s law:
where denotes the wavelength of the radiation, and denotes the diffraction angle where the diffraction pattern fades away because of the disorder in the crystal. A higher resolution means more reliability of experimental data in terms of accuracy and precision.
- Temperature Factor (or B-factor) The B-factor is used to model the effects of static and dynamic disorder in the crystal, and provide a relative indication of the reliability of different parts of the model. The dynamics disorder factor is defined by
where is the mean square displacement of the atomic vibration. If -factors are high, this means that little or no electron density is available for the atoms in the corresponding side-chain, and thus the coordinates are less reliable.
How well the model predicts the experimental data is usually assessed by the R-free value. In addition to the interpretation of the raw X-ray data, we must consider the intrinsic uncertainties in the protein crystal, such as uncertainties in crystal packing, difference in crystalline and solution states, internal motions, and so on. Also, we need to consider the different levels of uncertainties in X-ray structures, such as uncertainties in backbone and side-chains, heteroatoms (water, ligand, and metal), and the ionization state of the protein or ligand.
Uncertainty in NMR Structures
The accuracy of the structures determined by the NMR technique is related to
- the number and quality of restraints, and
- the approximate solution generated by computational algorithms.
Generally, we can represent the uncertainties in NMR structure by an ensemble of structures. These issues are discussed in chapters 35 and 42.
Protein Dynamics
Protein dynamics can be represented as frequencies (timescales) and magnitudes. Some NMR experiments can measure the protein motions within micro- () to milliseconds (), whereas dynamics within pico- () to nanoseconds () are measured by NMR relaxation measurements. So far, there is no uniform mathematical framework for representing various dynamics across all timescales, although this is an active area of research.
The following dynamics, for example, can be measured by current NMR techniques:
- Backbone dynamics from the NH vector.
- Backbone dynamics for other backbone vectors.
- Backbone dynamics from RDCs.
- Side-chain dynamics.
- Dynamics from chemical (or conformational) exchanges: slow conformation exchange.
- Dynamics from H-D exchange.
An engineer or mathematical physicist would be surprised by this use of the term “dynamics.” The phenomena measured by NMR are surely caused by dynamics. But the readout is more like stiffness. For example, local order parameters () for an NH bond vector are similar to the stiffness of that vector’s orientation. See chapter 24.
Probabilistic Representations of Uncertainty and Dynamics
Because of the uncertainty in measured experimental data and dynamics of protein motions, it may sometimes be difficult to know the precise position of each atom in the protein. A natural representation of atom coordinates is a probabilistic framework, that is, we may in principle assign a probability for each constraint. This would require more complicated algorithms for determining the protein structures. Examples of such algorithms are given in chapter 7.2, and by Nilges and coworkers [4, 5].
Representation of Protein Flexibility: Ensemble Docking
In computational docking algorithms, the flexibility of protein-ligand docking is considered at three levels, according to their degree of approximation [2]: (1) rigid body docking, in which both the protein and ligand are regarded as rigid solid bodies; (2) semiflexible docking, in which the protein is regarded as rigid, while the ligand is considered flexible; and (3) flexible docking, in which both protein and ligand are considered flexible.
In Knegtel et al. [3], an ensemble of experimental receptor structures is used to model flexible protein-ligand docking. Two related docking schemes are proposed to use the information from conformational variability: energy-weighted average and geometry-weighted average methods. In the energy-weighted average approach, an average ligand-protein interaction energy for each atom is calculated and considered during docking. In the geometry-weighted average approach, the following steps are executed: (1) Ensembles of protein structures are superimposed on residues near the binding sites; (2) mean and standard deviations for all atom coordinates are calculated; (3) each single structure is divided into ordered and disordered regions; (4) an ensemble of structures is stored for disordered regions, while one average structure is computed for ordered regions; and (5) the scoring function is computed based on the individual ordered regions and ensemble-average disordered regions.
FDS: Flexible Ligand and Receptor Docking with a Continuum Solvent Model and Soft-Core Energy Function
The following steps are applied in the FDS protocol [6] for protein-ligand docking with flexible ligand and flexible side-chains of receptor:
- Determine the hydrogen bond motifs or cliques. Each clique is a set of hydrogen-bonded atom pairs from the protein and ligand that simultaneously satisfy the given distance constraints.
- Generate the structures that satisfy all cliques by using a distance-geometry embedding algorithm, such as DGEOM95.
- Cluster the cliques using a basic root-mean-square distance (RMSD) to produce representative structures that sample important regions of the active site.
- The preceding structures are submitted to a Monte Carlo algorithm to find the final binding modes. During the simulation, both the dihedral angles in ligands and side-chains in proteins are sampled, either randomly or according to a rotamer library. A soft-core function for vdW repulsion and a molecular mechanics force field with the generalized born/surface area (GB/SA) continuum model are used.
Exercise Compare this approach to the algorithm (chapter 12), when side-chain and ligand rotamers are used to model flexibility, but no mutations are allowed. How does allow the pose of the ligand to minimize? Contrast the algorithms, and discuss their similarities and differences.
References
- [1] A. M. Davis, S. J. Teague, and G. J. Kleywegt. Application and limitations of X-ray crystallographic data in structure-based ligand and drug design. Angewandte Chemie International Edition 42;24 (2003 Jun 23):2718–2736.
- [2] I. Halperin, B. Ma, H. Wolfson, and R. Nussinov. Principles of docking: An overview of search algorithms and a guide to scoring functions. Proteins 47;4 (2002 Jun 1):409–443.
- [3] R. M. Knegtel, I. D. Kuntz, and C. M. Oshiro. Molecular docking to ensembles of protein structures. Journal of Molecular Biology (1997 Feb 21)266(2):424–440.
- [4] W. Rieping, M. Habeck, and M. Nilges. Inferential structure determination. Science 309;5732 (2005 Jul 8): 303–306.
- [5] W. Rieping, M. Nilges, and M. Habeck. ISD: A software package for Bayesian NMR structure calculation. Bioinformatics 24;8 (2008 Apr 15): 1104–1105. Epub 2008 Feb 28.
- [6] R. D. Taylor, P. J. Jewsbury, and J. W. Essex. FDS: flexible ligand and receptor docking with a continuum solvent model and soft-core energy function. Journal of Computational Chemistry 24;13 (2003 Oct):1637–1656.