Large-scale computational drug repositioning to find treatments for rare diseases

Abstract

Rare, or orphan, diseases are conditions afflicting a small subset of people in a population. Although these disorders collectively pose significant health care problems, drug companies require government incentives to develop drugs for rare diseases due to extremely limited individual markets. Computer-aided drug repositioning, i.e., finding new indications for existing drugs, is a cheaper and faster alternative to traditional drug discovery offering a promising venue for orphan drug research. Structure-based matching of drug-binding pockets is among the most promising computational techniques to inform drug repositioning. In order to find new targets for known drugs ultimately leading to drug repositioning, we recently developed eMatchSite, a new computer program to compare drug-binding sites. In this study, eMatchSite is combined with virtual screening to systematically explore opportunities to reposition known drugs to proteins associated with rare diseases. The effectiveness of this integrated approach is demonstrated for a kinase inhibitor, which is a confirmed candidate for repositioning to synapsin Ia. The resulting dataset comprises 31,142 putative drug-target complexes linked to 980 orphan diseases. The modeling accuracy is evaluated against the structural data recently released for tyrosine-protein kinase HCK. To illustrate how potential therapeutics for rare diseases can be identified, we discuss a possibility to repurpose a steroidal aromatase inhibitor to treat Niemann-Pick disease type C. Overall, the exhaustive exploration of the drug repositioning space exposes new opportunities to combat orphan diseases with existing drugs. DrugBank/Orphanet repositioning data are freely available to research community at https://osf.io/qdjup/.

Fig. 1

Performance assessment for the recognition of pockets binding similar ligands. The performance of eMatchSite alone and including virtual screening, labeled as eMatchSite/VS, is evaluated with BEDROC against the Huang dataset and compared to that of a random classifier. The Huang dataset contains ligand-bound and unbound structures. Boxes end at the quartiles Q₁ and Q₃, the horizontal line in a box is the median, and whiskers point at the farthest points that are within 3/2 times the interquartile range

Fig. 2

Example of a successful drug repositioning with eMatchSite. Staurosporine is repositioned from serine/threonine-protein kinase Pim-1 to synapsin Ia. a The local superposition of binding pockets according to the eMatchSite alignment. The computer-generated model of the primary target, Pim-1, is represented by transparent purple ribbons, whereas the model of the off-target, synapsin Ia, is represented by solid gold ribbons. Binding residues predicted by eFindSite are shown as spheres. The repositioned drug is presented as solid sticks colored by atom type (gold–carbon, red–oxygen, blue–nitrogen). ATP bound to synapsin Ia is shown as transparent sticks colored by atom type (teal–carbon, red–oxygen, blue–nitrogen, yellow–sulfur, peach–phosphorus). b A scatter plot for the correlation of ranks from virtual screening conducted by Vina. Each dot represents one library compound, whose ranks against the primary target and off-target are displayed on x and y axes, respectively. A dashed black line is the diagonal corresponding to a perfect correlation

Fig. 3

Flowchart of the drug repositioning procedure devised in this study. For protein sequences from DrugBank and Orphanet (a), template-based structure modeling is conducted with eThread to construct 3D models (b). Protein models are subsequently annotated by eFindSite with drug-binding sites (c). A similarity-based ligand docking is performed for DrugBank drug-protein pairs, i.e., a globally similar template is aligned onto the target structure with Fr-TM-align and then the drug is aligned onto the template-bound ligand with KCOMBU (d). The modeling procedure produces drug-bound structures for DrugBank and unbound structures for Orphanet proteins (e). Next, all-against-all matching of drug-binding pockets in DrugBank and Orphanet proteins is conducted with eMatchSite (f). The DrugBank compound is transferred to the Orphanet model when the similarity of binding pockets is sufficiently high and the resulting complex is refined (g). Finally, the quality of final Orphanet complex models is assessed with DFIRE and virtual screening (h)

Fig. 4

Multiple drugs repositioned through a single pocket alignment. Schematic of (a) a single DrugBank target binding two drugs (teal and yellow), (b) an Orphanet target (green), and (c) two modeled complexes of DrugBank drugs and an Orphanet protein (teal-green and yellow-green). d A real example of tolcapone (teal) and entacapone (yellow) repositioned to guanine nucleotide-binding protein subunit alpha-11 (green) based on its local alignment with catechol O-methyltransferase. Non-carbon ligand atoms in panel (d) are colored by atom type (blue–nitrogen, red–oxygen)

Fig. 5

Multiple models of a single drug-target complex constructed based on multiple pocket alignments. Schematic of (a) three DrugBank targets binding the same drug (teal, orange, and yellow), (b) an Orphanet target (green), and (c) three poses of a DrugBank drug within the binding site of an Orphanet protein (teal/orange/yellow-green) modeled from different pocket alignments. (d) A real example of ponatinib repositioned to Ras-related protein Rab-23 (green) based on its local alignment with Lck/Yes-related novel protein tyrosine kinase (ponatinib is teal), lymphocyte cell-specific protein-tyrosine kinase (ponatinib is orange), and proto-oncogene tyrosine-protein kinase Src (ponatinib is yellow). Non-carbon ligand atoms in panel (d) are colored by atom type (blue–nitrogen, red–oxygen)

Fig. 6

Correlation between interaction energies calculated for DrugBank and Orphanet complex models. Each gray dot represents a drug-target pair from the DrugBank database, whose DFIRE score is displayed on the x-axis. Since a drug can be repositioned to multiple Orphanet proteins, the mean DFIRE score ± standard error is displayed on the y-axis. Linear regression is shown as a solid line

Fig. 7

Example of a recently determined structure corroborating repositioning prediction by eMatchSite. Ibrutinib repositioned from tyrosine-protein kinase BTK to tyrosine-protein kinase Blk is compared to the X-ray structure of tyrosine-protein kinase HCK complexed with ligand OOS. a Chemical structures of the repositioned drug, ibrutinib, and the co-crystallized ligand, OOS. b The modeled structure of the ibrutinib-Blk complex, colored in purple, is globally superposed onto the experimental OOS-HCK structure, colored in gold. Proteins are shown as ribbons, ligands as sticks, and binding residues predicted by eFindSite in Orphanet models as spheres. Non-carbon atoms in ligands are colored by atom type (red–oxygen, blue–nitrogen, yellow–sulfur)

Fig. 8

Repositioning of exemestane from cytochrome P450 aromatase (CYP19A1) to Niemann-Pick disease type C2 protein (NPC2). CYP19A1 and NPC2 proteins are colored in purple and gold, respectively, whereas ligands are colored by atom type (green/teal–carbon, red–oxygen, yellow–sulfur). a Global superposition of the modeled complex between CYP19A1 (purple ribbons) and exemestane (thick sticks), and two experimental structures of CYP19A1 (teal ribbons) bound to androstenedione and exemestane (thin sticks). Binding residues are shown as spheres. b Global superposition of the NPC2 model (gold ribbons) and two experimental structures of NPC2, human and bovine (teal ribbons), bound to cholesterol sulfate (thin sticks). Binding residues are shown as spheres. In addition, the steroid-binding pocket predicted by eFindSite is represented by a cluster of template-bound ligands (transparent sticks) extracted from the following template proteins superposed onto the NPC2 model (template-proteins are not shown): GM2A (PDB-IDs: 2ag2, 1tjj, 2agc), LY96 (PDB-IDs: 2e59, 2e56, 4g8a, 3fxi, 2z65, 3mu3, 3rg1, 5ijd, 3vq2, 3m7o), DERF2 (PDB-ID: 1xwv), and NPC2 (PDB-IDs: 5kwy,2hka, 3web). c Cross section of the internal cavity in the NPC2 structure exposing the repositioned exemestane (thick sticks). CYP19A1 (purple ribbons) and NPC2 (gold surface) are locally superposed according to the sequence order-independent pocket alignment by eMatchSite. Annotated binding residues in NPC2 are solid, whereas the remaining surface is transparent