Predicting where small molecules bind at protein-protein interfaces

Abstract

Small molecules that bind at protein-protein interfaces may either block or stabilize protein-protein interactions in cells. Thus, some of these binding interfaces may turn into prospective targets for drug design. Here, we collected 175 pairs of protein-protein (PP) complexes and protein-ligand (PL) complexes with known three-dimensional structures for which (1) one protein from the PP complex shares at least 40% sequence identity with the protein from the PL complex, and (2) the interface regions of these proteins overlap at least partially with each other. We found that those residues of the interfaces that may bind the other protein as well as the small molecule are evolutionary more conserved on average, have a higher tendency of being located in pockets and expose a smaller fraction of their surface area to the solvent than the remaining protein-protein interface region. Based on these findings we derived a statistical classifier that predicts patches at binding interfaces that have a higher tendency to bind small molecules. We applied this new prediction method to more than 10,000 interfaces from the protein data bank. For several complexes related to apoptosis the predicted binding patches were in direct contact to co-crystallized small molecules.

Figure 1. Schematic representation of a PP complex (left) and a PL complex (right).

PP and PL share an identical or similar ‘reference protein’. The interface areas shown in light grey were determined using a distance criterion as described in the methods section. Overlap residues form contacts in the PP complex as well as in the PL complex whereas non-overlapping residues form contacts only in one of the complexes.

Figure 2. Identification of competitive PP:PL pairs using structural superposition.

Figure 3. Construction of a surface patch.

(a) Shown is a pre-patch of size n = 5, consisting of a central surface residue (dark grey) and n-1 = 4 nearest neighbors residues (light grey). (b) Using the coordinates of the C_α-atoms of the residues, the center of mass (COM) for this pre-patch is calculated. For any surface residue including the central residue, a solvent vector is defined using the coordinates of its C_α-atom and the COM. (c) The final surface patch contains the central residue and the closest n-1 surface residues (grey) for which the angle α_i between their solvent vector and the solvent vector for the central residue is between 0° and 110°.

Figure 4. Pairs of protein-protein complexes and the related protein-ligand complexes.

(a) and (b) show trypsin bound to the native protein inhibitor Bovine Pancreatic Trypsin Inhibitor (BPTI) (a, PDB code 1BZX E:I) or to the small-molecule ligand benzamidine (b, 1MBQ A:BEN). Here, the identifiers denote the chains and ligands used. The chain identifiers of the reference proteins are marked in bold. (c) and (d) show insulin growth factor protein bound to an IGF-binding protein (c, 2DSQ I:G) or to a detergent molecule (d, 1GZR B:C15). (e) and (f) show barnase bound to the protein inhibitor barstar (e, 1X1U A:D) or to its natural ligand RNA, in this case the tri-nucleotide GMP (f, 1GOY A:3GP). (g) and (h) display a sugar binding protein and its natural ligand N-acetylglucosamine (h, 1DBN A:NAG) or when it forms a quaternary homomeric structure (g, 2DVG C:B).

Figure 5. PP:PL pair 1BZX E:I (left) and 1MBQ A:BEN (right).

The reference proteins are marked in dark grey, the ligand and the binding protein partner in light grey. Additionally, the overlapping residues on the surface of the reference proteins are marked as bright spheres. The corresponding regions are circled.

Figure 6. Interface statistics.

Shown is the total number of interface residues against (a) the number of overlap residues for PP interfaces and (b) the number of overlap residues for PL interfaces.

Figure 7. Distribution of conservation ranges obtained from the Consurf webserver.

Conservation scores are shown separately for overlap (black line) and non-overlap (grey line) residues for PP interfaces. As a reference, the conservation of all residues on the protein surface (dotted line) is plotted. Negative values indicate residues that are more conserved. Using the Welch t-test the two classes showed a statistically significant difference (p-value<2.2e-16).

Figure 8. Binned distribution of the protrusion index for overlap and non-overlap residues of PP interactions.

Values close to zero indicate buried residues. Using the Welch t-test the two classes showed a statistically significant difference (p-value < 2.2e-16).

Figure 9. Distribution of surface fractions for overlap and non-overlap residues for PP interfaces.

Using the Welch t-test the two classes showed a statistically significant difference (p-value = 6.373e-11).

Figure 10. Significance of interface features for predicting overlap residues.

MeanDecreaseAccuracy reflects the suitability of a feature as a reliable predictor. Precisely, it reflects how much the average prediction accuracy decreased when randomly shuffled values were used for a particular feature in the testing phase. In the diagram, this quality measure decreases from top to bottom. Cons#n refers to the conservation score of the n-th nearest surface residue starting from the central residue (n = 1∶ central residue itself). Analogously, protru#n refers to the protrusion value of the n-th nearest residue starting from the central residue (n = 1∶ central residue itself). Density and surfaceFraction describe the contact density and the surface fraction of the central residue. HsfPatch8 denotes the relative frequency of predicted hot-spots in a surface patch of size 8.

Figure 11. Surface-patch based predictions.

The dark grey bars indicate the fraction of positive classifications for 7-residues-patches with increasing proportion of predicted overlap residues (1 out of 7 to 7 out of 7 predicted overlap residues). The light grey line represents the absolute frequency for every patch, e.g. there are about 300 patches for which 5 out of 7 residues in the patch were predicted as ‘overlap’.

Figure 12. Distribution of the number of 7-residue surface patches per protein-protein binding interface.

The central residue must be predicted as ‘overlap’ and at least 4 of the 6 remaining residues must be predicted as ‘overlap’ as well.

Figure 13. Visualization of complexes related to apoptosis.

Predicted overlap residues are colored in dark grey whereas all other surface residues appear in light grey. (a) 1PQ1A, (b) 1RE1B, (c) 1OLGA, (d) 2C2ZB, (e) 2TNFB, (f) 1DU3D, (g) 1BH5B, (h) 1PYOD.