Date Published: February 01, 2017
Publisher: International Union of Crystallography
Author(s): Robert A. Nicholls.
The process of ligand fitting with CCP4 is reviewed, including identifying ligand density in the map, ligand fitting, refinement and subsequent validation. Recent developments are discussed, and are illustrated using instructive examples demonstrating practical application.
Macromolecular crystallography is a useful technique for determining how ligands interact with proteins. Following structure determination, crystal structures of protein–ligand complexes are often used in structure-based drug design, calculation of interaction energies and protein-induced strain, and to make other biological inferences. In addition to being used by the original scientists who determined the structure, models of crystal structures deposited in the Protein Data Bank (PDB; Berman et al., 2002 ▸) are also used by other scientists who analyse the deposited models, as well as in more general studies involving PDB data mining and analysis. Conclusions made using modelled crystal structures of protein–ligand complexes can be highly sensitive to model quality and errors. Indeed, small changes in atomic positions may have a substantial impact on perceived chemical interactions, potentially leading to different results in subsequent analyses. Consequently, it is important that crystal structures are as accurate as possible and sufficiently reliable to give a definitive answer as to the binding mode of the ligand.
Typically, ligand fitting begins after all macromolecules have been built and refined. The first step is to identify any blobs of unmodelled density, which may correspond to ligand-binding sites. In some cases it may be clear that a blob corresponds to a particular ligand, but in other cases it may be less obvious (e.g. owing to problems encountered during co-purification; Fischer, Hopkins et al., 2015 ▸). It may be useful to compile a list of potential compounds (buffer components etc.) that could reasonably correspond to the blobs, and attempt to fit each of them before deciding which is the most likely.
The first question to be answered is: is there any electron density in the map that corresponds to the ligand of interest? The answer to this question is quite often no, in which case further crystallization experiments may need to be undertaken (Mueller, 2017 ▸; Hassell et al., 2007 ▸). Ligands are typically built after the rest of the model has been built and refined as well as reasonably possible (except perhaps for waters); this is usually assessed by considering the convergence of R factors, reduction of Rfree, satisfaction of geometry/chemical validation and inspection of electron-density maps. At this point, it is hoped that the ligands, if present, will be clearly visible in the difference density (mFo − DFc) map (the difference map can readily be searched in Coot using the ‘Difference Map Peak’ tool found in the ‘Validate’ menu). If there is convincing electron density in the map that could correspond to the ligand, then one can proceed with fitting. It should be noted that crystallization and data-collection conditions, such as temperature, can have an effect on such factors and thus on the potential ability to model a ligand (Fischer, Shoichet et al., 2015 ▸).
The CCP4/REFMAC monomer library (Vagin et al., 2004 ▸) comprises pre-computed descriptions for many common ligands (such as peptides). The restraints for these ligands are already distributed as part of the CCP4 suite, and are automatically used during fitting and refinement without any manual intervention required.
Following generation of the ligand description and initial coordinates, the next step is ligand fitting. This is typically performed in real space, and involves attempting to correctly position and orient the ligand as well as selecting the correct conformation. The following sections describe some of the tools available in Coot for real-space fitting, including tools for making post-translational modifications and carbohydrate fitting. Details of the tools implemented in Coot for dealing with ligands have been described by Debreczeni & Emsley (2012 ▸).
Following successful ligand building and fitting, full-model refinement can then be performed. This allows the ligand and protein to co-refine, synergistically optimizing the agreement between model and experimental data. Since the ligand contributes to the model phases, subsequent density interpretation must be performed with care owing to the potential for model bias. Careful inspection of the difference density and OMIT maps is often necessary in order to ensure that the ligand is actually present and in the modelled state.
In this paper, some of the features available in various software tools designed to ease ligand building, fitting, refinement and validation have been described, and a number of scenarios exemplifying practical usage have been considered. These tools, distributed as part of the CCP4 suite, include AceDRG for the generation of ligand restraint dictionaries and conformers, Coot for map interpretation, ligand finding, fitting, conformer generation, real-space refinement, Jiggle Fit and carbohydrate building (LO/Carb), JLigand for manually creating and editing ligands, restraints and link records in three dimensions, Lidia for manually creating and editing ligands in two dimensions, as well as visualization and analysis of the chemical environments of molecules (FLEV), and REFMAC5 for macromolecular refinement. Along with the CCP4i2 GUI, these tools aim to ease the ligand-fitting process, facilitating the ability for ligands to be modelled quickly and reliably whilst providing diagnostics to help tackle the varied problems that may be encountered during the procedure. Substantial advancements have recently been made, designed to make life easier for the computational crystallographer. However, ultimately, there is no substitute for manual inspection and due diligence.