Table 1.
Overview of Drug Promiscuity Studies.
Figure 1.
Drug promiscuity: Ligand flexibility vs. binding site similarity.
(A) A flexible ligand, tretinoin (on the left), with two distinct conformations is able to bind to very different binding sites. (B) The drug BVDU (orange) binding to a viral thymidine kinase (green, 1osn) and a human heat shock protein (blue, homology model [32]). The two targets share a similar binding site, which allows the promiscuous binding of the drug in the same conformation.
Figure 2.
Promiscuity of drugs in the PDB.
30% of the drugs have three or more targets.
Table 2.
The 10 most promiscuous drugs.
Figure 3.
Visualization of possible correlations with degree of drug promiscuity (drug target count).
Green dots denote the mean. (A) Molecular weight. There is no correlation (). (B) Hydrophobicity. There is no correlation (
). (C) Bound Drug conformer clusters. There is a weak correlation (
). (D) The number of target Pfam families is correlated with the drug target count (
). (E) Global structural alignment. There is a correlation between the number of targets of a drug and the square root of the number of structurally similar proteins among its targets (
). (F) Similar binding sites. There is a correlation between the target count of a drug and the square root of the similar binding site count of its targets (
).
Table 3.
The most flexible drugs.
Figure 4.
Conformer count of promiscuous drugs in the PDB.
Figure 5.
Pipeline of the binding site similarity analysis.
Starting from 543 drugs, we identify 164 promiscuous drugs, each binding to three or more non-redundant targets (712 in total). The binding site alignment with SMAP is performed for all 2284 structures (i.e. the redundant targets). Subsequently, target pairs are clustered by 95% sequence identity – giving 712 non-redundant targets – and ranked with LigandRMSD.
Figure 6.
Comparison of the SMAP P-Value to LigandRMSD.
A P-Value of gives a significant binding site alignment. The LigandRMSD gives the conformational similarity between the bound ligands and is ≤3 Å for similar binding sites. The thresholds are displayed as solid lines in the plot. In total, 3948 non-redundant target pairs were compared.
Table 4.
The 10 drugs with the most similar binding sites.
Figure 7.
The heatmaps for the targets of methotrexate (A), acarbose (B) and quercetin (C) show that the target sequences are dissimilar (left), the global structural similarity (middle) is comparable to the sequence identity and the binding sites are overall more similar (right). (A) Two DHFR (1dg5, 3dl6) with a conserved 3D structure and similar binding sites for methotrexate are highlighted. (B) Although the proteins 4--glucanotransferase (1k1x) and glucoamylase (2f6d) have globally distinct sequences and structures, they bind acarbose in a very similar way. (C) The two protein kinases PI3KCG (3lj3) and PIM1 (3ma3) share a similar binding pocket for quercetin.
Figure 8.
Structural details of binding site similar targets.
The binding site alignments for the targets of (A) methotrexate, (B) acarbose and (C) quercetin (highlighted in blue in Figure 7) are visualized. Binding sites are highlighted in red and ligands are displayed in orange. PDB IDs are given below the structures. If the given ID is a representative of a cluster, the PDB ID of the underlying structures is given in parentheses.