qFit-ligand Reveals Widespread Conformational Heterogeneity of Drug-Like Molecules in X-Ray Electron Density Maps

Abstract

Proteins and ligands sample a conformational ensemble that governs molecular recognition, activity, and dissociation. In structure-based drug design, access to this conformational ensemble is critical to understand the balance between entropy and enthalpy in lead optimization. However, ligand conformational heterogeneity is currently severely underreported in crystal structures in the Protein Data Bank, owing in part to a lack of automated and unbiased procedures to model an ensemble of protein-ligand states into X-ray data. Here, we designed a computational method, qFit-ligand, to automatically resolve conformationally averaged ligand heterogeneity in crystal structures, and applied it to a large set of protein receptor-ligand complexes. In an analysis of the cancer related BRD4 domain, we found that up to 29% of protein crystal structures bound with drug-like molecules present evidence of unmodeled, averaged, relatively isoenergetic conformations in ligand-receptor interactions. In many retrospective cases, these alternate conformations were adventitiously exploited to guide compound design, resulting in improved potency or selectivity. Combining qFit-ligand with high-throughput screening or multitemperature crystallography could therefore augment the structure-based drug design toolbox.

Figure 1.

Ligand structural dynamics and minor changes during fragment optimization lead to new binding modes and drive drug design. A) Subtle changes in chemical structure of ligands can impose new binding modes. B) Near isoenergetic receptor-ligand conformations exchange in dynamic equilibrium in crystal structures. These conformations can inform the design of a ligand with higher affinity. C) Hsp90 inhibitors in gray (PDB ID 4CWO) and gold (PDB ID 4CWN). D) Subtle changes in chemical structures lead to changes in binding pose for Lp-PLA2, changing the course of design (PDB IDs 5JAL, 5JAO). E) New Lp-PLA2 inhibitor designed as a result of observed alternate binding poses of fragments (PDB ID 5JAP). F) Evidence of difference density in x-ray crystal structure of Lp-PLA2 fragment shows alternate binding poses pre-exist at low occupancy (PDB ID 5JAL). G) Alternate conformations exploited in the design of EphB4-binding ligands (PDB ID 2VWX). Electron densities are shown at 1.5σ. Positive (green) and negative (red) difference densities in panel F are shown at +3.0σ and −3.0σ, respectively.

Figure 2.

Benchmark statistics. A) Categories of alternate conformations present in the benchmark. B) Conformer occupancies pre- (blue) and post- (orange) re-refinement. C) Ligand A to B conformer RMSD, pre (blue) and post (orange) re-refinement. D) Occupancy shift versus mean B-factor difference after re-refinement.

Figure 3.

qFit-ligand workflow and statistics on synthetic data. A) Rigid clusters are defined as rings or terminal ends, and rotatable bonds as any bond that is not part of a ring. The local search finds possible positions and orientations of each cluster in the binding site, avoiding steric clashes with the protein. As clusters are joined, torsions and degrees of freedom are sampled. Up to 5 ligand conformations that best match the ligand density are selected and combined with the protein model to give a final ligand multiconformer model. B) Violin plot of categories at 1.60Å resolution and occupancy 0.50 across 90 test cases. The white dot represents the median, the bold center line represent the interquartile range (IQR), and the thin center line represents the percentile range 25^th-1.5 IQR to 75^th+1.5 IQR. Lower nRMSD is better. C) Heatmap of category vs. resolution at 0.50 occupancy. Lower nRMSD is better. D) Violin plot for representative resolution 1.60Å and 0.50 / 0.50 occupancy at optimal parameters. E) Heatmap of normalized RMSD at optimal parameters (resolution vs. occupancy).

Figure 4.. Electron density support measures and energetics of qFit-ligand multi-conformers are similar to those of single-conformer models for the benchmark set, while reporting on conformational heterogeneity.

A) nRMSD distributions by category of conformational heterogeneity. The nRMSD of terminal end flips was determined from only atoms affected by the dihedral changes. Lower nRMSD is better. B) Histogram of R_work and R_free differences between refined qFit-ligand models and single conformer structures. Negative values indicate a lower R for refined qFit-ligand models. C) The distribution of occupancy-weighted, ligand internal energies of qFit-ligand multiconformer ligands relative to single ‘A’ conformation. Negative values indicate that the multiconformer ensemble has lower internal energy than the single benchmark ‘A’ conformation. Positive values indicate the multiconformer ensemble has higher internal ligand energies. D) Ligand EDIAm scores for re-refined single conformer models against automatically refined qFit-ligand multiconformer models. Higher EDIAm scores are better, with scores greater than 0.8 indicating that the model is well-supported by the density (shaded area for single conformer models). E) Examples where the ligand EDIAm was improved by modeling alternate conformations with qFit-ligand. The 2mFo-DFc maps are shown in blue at a contour level of 1σ.

Figure 5.

Prospective discovery of additional conformations and recovered conformations from the D3R and Twilight dataset. A) The deposited multiconformer model, qFit-ligand multiconformer model, and manually edited multiconformer model of ERK2 (PDB ID 4FV3; new PDB ID 6DMG). B) Single conformer and qFit-ligand multiconformer models of serratia fonticola carbapenemase E166A mutant with the acylenzyme intermediate of meropenem (PDB ID 4EV4; new PDB ID 6DMH). C) Prospective application of qFit-ligand to inhibitor 5T5 of BACE-1 (PDB ID 5EZX; new PDB ID 6DMI). Single conformer and qFit-ligand multiconformer models shown. D) Overlay of multiconformer model of inhibitor 5T5 (green) with inhibitor 5T6 (magenta) (PDB ID 5EZZ). Electron densities are shown at 1.5σ (blue) and 0.3σ (purple). Positive (green) and negative (red) difference densities are shown at +3.0σ and −3.0σ, respectively. All structures shown have been refined using Phenix.

Figure 6.

Prospective application of qFit-ligand to BRD compounds A) Single conformer crystal structure, qFit-ligand multiconformer, and manually edited multiconformer models including alternate conformations of Asp145 and compound BDOIA383 bound to BRD4 (PDB ID 5CFW; new PDB ID 6DMJ). B) Protein-ligand interactions of the major, crystal-contact stabilized ‘A’ and minor ‘B’ BDOIA383 conformations of the final qFit-ligand model (panel 6A). C) Single conformer crystal structure, qFit-ligand multiconformer, and manually edited multiconformer models of a isoxazolyl-benzimidazole ligand bound to CBP BRD (PDB ID 4NR5, new PDB ID 6DMK). D) Single conformer and qFit-ligand multiconformer models of ligand 9BM bound to BRD4 (PDB ID 4BW3; new PDB ID 6DML). Viewing orientation differs from panel A,C to clearly show ligand alternate conformations. E) Ligand S5B’s single binding conformation in BRD2 (PDB ID 4AKN). Electron densities are shown at 1.5σ (blue) and 0.3σ (purple). Positive (green) and negative (red) difference densities are shown at ±3.0σ. Distances in Ångstroms.