The overlap of small molecule and protein binding sites within families of protein structures

Abstract

Protein-protein interactions are challenging targets for modulation by small molecules. Here, we propose an approach that harnesses the increasing structural coverage of protein complexes to identify small molecules that may target protein interactions. Specifically, we identify ligand and protein binding sites that overlap upon alignment of homologous proteins. Of the 2,619 protein structure families observed to bind proteins, 1,028 also bind small molecules (250-1000 Da), and 197 exhibit a statistically significant (p<0.01) overlap between ligand and protein binding positions. These "bi-functional positions", which bind both ligands and proteins, are particularly enriched in tyrosine and tryptophan residues, similar to "energetic hotspots" described previously, and are significantly less conserved than mono-functional and solvent exposed positions. Homology transfer identifies ligands whose binding sites overlap at least 20% of the protein interface for 35% of domain-domain and 45% of domain-peptide mediated interactions. The analysis recovered known small-molecule modulators of protein interactions as well as predicted new interaction targets based on the sequence similarity of ligand binding sites. We illustrate the predictive utility of the method by suggesting structural mechanisms for the effects of sanglifehrin A on HIV virion production, bepridil on the cellular entry of anthrax edema factor, and fusicoccin on vertebrate developmental pathways. The results, available at http://pibase.janelia.org, represent a comprehensive collection of structurally characterized modulators of protein interactions, and suggest that homologous structures are a useful resource for the rational design of interaction modulators.

Figure 1. The overlap of ligand and protein binding sites within protein families.

(A) The distribution of overlap scores (Eqn 1) is shown for all families that bind both ligands and proteins (grey; n = 1,028), and the subsets of families with a statistically significant overlap (p0.01; solid; n = 197) or non-overlap (p0.01; dashed; n = 113). The highest overlap score observed is 10.83 (not shown). (B) The residue type propensity (Eqn 3) and (C) conservation (Eqn 4) at alignment positions that bind both ligands and proteins (black; n = 102,436), bind ligands (cyan; n = 46,610), bind proteins (orange; n = 491,723) in comparison to all solvent-exposed residues (grey; n = 1,147,882). The statistical significance of the residue propensities was estimated by a bootstrap resampling procedure (Table S5).

Figure 2. Ligand-protein binding site overlap observed at protein–protein interactions as a function of sequence identity.

The maximum observed ligand binding site overlap (y-axis) for (A) inter-molecular domain–domain and (B) domain–peptide interactions, as a function of the ligand binding site sequence identity (x-axis). The densities in these plots are represented by colors that range from yellow (no density) to blue (maximum density). The (C) maximal and (D) cumulative overlap profile is shown at a minimum ligand binding site identity threshold of 30% for inter-molecular (black), intra-molecular (orange) domain–domain, and domain–peptide (cyan) interactions. Tick marks indicate interfaces that exhibit a particular level of interface coverage. The overlap score refers to the fraction of interface residues aligned to ligand binding site residues (Eqn 2).

Figure 3. Small molecule binding sites overlapped with four broad classes of protein–protein interfaces.

(A) Enzyme – protein inhibitors: eg, 3′-phosphothymidine (3′–5′)-pyrophosphate adenosine 3′-phosphate (PDB 1U1B:PAX) overlapped with the ribonuclease (orange, 2Q4G)–inhibitor (purple, 2Q4G) interface. (B) Enzyme–protein substrate: eg, Kni-577 (cyan, 1MRW:K47) bound to the HIV-protease dimer (grey backbone, 1MRW:A,B; orange, 1A8K:A,B) at the same positions as its peptide substrate (purple, 1A8K:C). (C) Structural or regulatory interfaces: eg, kabiramide-C (cyan, 1QZ5:KAB) bound to Actin (grey backbone, 1QZ5:A; orange, 1H1V:A) at the same position as Gelsolin (purple, 1H1V:G). (D) Several ligands complemented protein interfaces: eg, bepridil (cyan, 1lxf:BEP) bound at the interface between troponin C (orange, 1LXF:C) and troponin I (purple, 1LXF:I). Figure produced by PyMOL (http://pymol.org).

Figure 4. Overlapping binding sites suggest structural mechanisms for observed small molecule effects.

(A) Sanglifehrin (cyan, PDB 1NMK:SFM) binds to cyclophilin A (grey, 1NMK; orange, 1AK4:A) at the same position that binds the HIV Gag capsid protein (purple, 1AK4:D). (B) Fusicoccin (cyan, 1O9E:FSC) binds to a region of the plant 14-3-3 protein (grey, 1O9E) that is homologous to the 14-3-3- (orange, 1A38:A) binding site for phosphopeptides (purple, 1A38:P). (C) Bepridil (cyan and blue, 1LXF:BEP) binds to Troponin C (grey, 1LXF:C) at positions that are homologous to the calmodulin (orange, 1K93:D) interface for anthrax edema factor (purple, 1K93:A). Troponin C aligns to both EF-hand motifs in calmodulin: The binding site aligned with EF-motif 2 (cyan) exhibits greater overlap with the anthrax edema factor interface than EF-motif 1 (blue).