Research Article: Proper modelling of ligand binding requires an ensemble of bound and unbound states

Date Published: March 01, 2017

Publisher: International Union of Crystallography

Author(s): Nicholas M. Pearce, Tobias Krojer, Frank von Delft.


The importance of modelling the superpositions of ligand-bound and unbound states that commonly occur in crystallographic data sets is emphasized and demonstrated. The generation of an ensemble that models not only the state of interest is important for the high-quality refinement of low-occupancy ligands, as well as to explain the observed density more completely.

Partial Text

Crystallographic diffraction experiments reveal the atomic composition of protein crystals, but when the crystal is composed of objects in multiple states the resulting diffraction pattern is a weighted average of these states. Ligands will often bind at sub-unitary occupancy in the crystal, as shown by examples where extensive experimental optimization is required to obtain interpretable electron density (McNae et al., 2005 ▸; Müller, 2017 ▸): not only ligand affinity but also solubility and the crystal lattice play a part in determining the occupancy of a ligand (Danley, 2006 ▸). Noncovalent ligands are always subject to binding equilibrium, so in general, even crystal forms that form only by the co-crystallization of ligand and protein cannot be assumed to have the ligand bound at full occupancy, as even high-affinity ligands may be partially displaced by unpredictable experimental artefacts.

We present four contrasting examples where the inclusion of a complementary solvent model leads to a better description of the crystal, and thereby to a higher quality ligand model. The ligands here were all identified and modelled using the PanDDA method (Pearce et al., 2016 ▸). The model of the ligand was in each case derived from PanDDA event maps, and we investigate here only the effect that the inclusion/absence of the superposed solvent model has on the interpretation of the data. Models are generated and refined as described in §2.2. Validation metrics are calculated for only the ligand residue in each of the models. Crystallographic model parameters, including ligand validation scores, may be found in Supplementary Tables S1–S4. The chemical structures of the modelled compounds are shown in Supplementary Fig. S2.

The examples presented here provide consistent evidence for ground-state molecules co-existing with ligand-bound molecules in crystals across a range of non-unitary occupancies. Moreover, the inclusion of a superposed ground-state model, obtained from a reference data set, improves the quality of the obtained ligand models in all cases. In the case of some weak ligands, the ground state model is crucial for the refinement of the protein–ligand complex (§3.1); in other cases it acts simply to remove ‘extraneous’ difference density that could be interpreted by an overzealous modeller as being caused by a ligand in multiple conformations (§3.2). The modelling approach can affect the interpretation of intermolecular interactions (§3.3), and in the case of high occupancy a superposed ground state can still marginally improve the ligand model, alongside providing a more complete model of the crystal (§3.4).

All crystallographic data for the various versions of each model have been uploaded to Zenodo ( The KDM4D structures are labelled as JMJD2D for consistency with the original PanDDA manuscript. Interactive HTML summaries for all of the fragment-screening data sets can also be found at (for JMJD2D), (for BAZ2B) and (for BRD1).

The following references are cited in the Supporting Information for this article: Lang et al. (2014 ▸).




Leave a Reply

Your email address will not be published.