Drug repurposing is a long-established strategy for identifying new applications for approved or investigational therapeutics that are beyond the scope of the original medical indication.¹ Not since the serendipitous repositioning of a drug designed for hypertension and angina to become Pfizer’s infamous “little blue pill,” commonly referred to as Viagra, has drug repurposing been so prominent in the public eye. The global urgency for therapeutics to treat COVID-19 has brought this drug strategy once again to mass attention.

3D protein modeling includes several molecular modeling and simulation methods for drug repurposing. A previous post in this series explored molecular docking as a method for identifying possible inhibitors binding to the SARS-CoV-2 main protease. Can an alternative in silico method such as pharmacophore modeling achieve comparable results and provide additional evidence supporting the selection of one candidate over another?

Taking an Alternative Route

Pharmacophore modeling provides an abstraction of the molecular features that are necessary for the recognition of a ligand by a protein target. Its representation of molecular interactions and binding provides a contrasting perspective to classical simulation methods.

We obtained an initial dataset of several SARS-CoV-2 main protease protein structures from Diamond Light Source.² We aligned these structures and then used the ‘Interaction Pharmacophore Generation’ protocol available in BIOVIA Discovery Studio to generate pharmacophores representing the non-bond interactions of each receptor-ligand complex. The total number of features in the pharmacophores ranged from two features for the 5R80 and 5R7Y crystal structures to a nine-feature model for 6LU7. We merged the pharmacophores from the individual complexes into a single model and edited closely clustered features. The final model included 14 features that represent intermolecular contacts between the protease and possible small molecule binders.

We subsequently performed in silico alanine-scanning mutagenesis on the active site residues of all complexes to identify which residues reduced the binding affinity (hotspot) of that protein-ligand complex when mutated. From all of the complexes, we identified eight residues (HIS41, MET49, ASN142, HIS163, MET165, PRO168, GLN189 and GLN192) as hotspots in at least one complex and three residues that were hotspots in at least four complexes. In the 14-feature pharmacophore model, six features corresponded with interactions with one of the eight hotspot residues. A second dataset of non-covalent M^PRO ligand complexes released by Diamond Light Source ² revealed a number of different binding modes, leading us to separate the six key pharmacophore features into two groups for use in the next step of the virtual screening.

We now used this pharmacophore model to perform virtual screening of a selected library containing 2,650 FDA-approved drugs that have multiple ligand confirmations pre-generated for fast searching. We screened these ligands against the 14-feature pharmacophore, with a requirement that at least one feature from each of the two required groups needed to match, as well as at least two features from the remaining eight features, to identify a hit. This is to allow diverse exploration of different binding modes of the ligands, while also ensuring that only poses that can form at least two interactions with significant residues identified in the alanine-scanning mutagenesis are retained. The algorithm essentially explores over 12,000 possible pharmacophore combinations. We then minimized the best fitting hits for each pharmacophore combination within the protein. Finally, we determined the binding energies of the optimized pharmacophore-derived poses using CHARMm with a Generalized Born using Molecular Volume (GBMV) implicit solvent model. This method is effectively a molecular docking calculation (as in our previous blog), but here we have used our optimized pharmacophore model to restrain and filter the docked results.

We calculated the non-bond interactions of each pose and filtered to focus on only those poses with interactions to four key residues – HIS41, MET49, CYS145 and MET165. HIS41, MET49 and MET165 were the three residues common to at least four complexes identified earlier, and CYS145 is the second residue in the significant HIS41/CYS145 catalytic dyad found in each subunit of the main protease homodimer. We sorted the hits with the calculated binding free energy to identify the top ten plausible candidates for further exploration. Ritonavir was the only common ligand identified here and in the previous blog. Ritonavir had the seventh best binding free energy and was also previously ranked seventh in the docking study. Ritonavir is currently undergoing a number of clinical trials for COVID-19.³

Video 1: Best scored pose for Ritonavir interacting with HIS41, MET49, SER144, CYS145, MET165, GLU166 and GLN189.

The best scoring ligand in the method presented here was Montelukast, a cysteinyl leukotriene receptor antagonist. A recent published paper has hypothesized its use to limit progression of the disease on COVID-19 infection.⁴

 

Three additional top-scoring compounds included the drugs Telmisartan, Moexipril and hydroxycloroquine. The potential of all these compounds as repurposed drugs for COVID-19 has recently come under scrutiny,⁵ with hydroxychloroquine garnering the most controversy. One additional difference between the top ten hit lists of the two methods is the diversity of the drug targets included. The top ten docking results include seven HCV or HIV protease inhibitors. The pharmacophore-prioritized hit list includes drugs with seven different target classes.

Our pharmacophore-derived virtual screening produced a prioritized hit list including several new potential COVID-19 drugs not identified in our previous docking work. With this approach, we demonstrate the utilization of in silico alanine-scanning mutagenesis as a useful technique for refining the pharmacophore model. We also imposed the requirement that certain features be present for a fit during the screening process. The results of these two studies are not directly comparable—not so much due to the differences in the basic algorithms, but rather due to the implementation of different supporting strategies.

The brief comparison of results between the two methods showed little overlap in leads to focus on. One could say that the commonality of Ritonavir identified by both methods provides evidence to support the selection of this candidate for further study. Previous studies have shown that using a consensus of protein-ligand docking scoring functions improves the identification of putative drug candidates.⁶ For this reason, researchers could use the unique best-scoring compounds from each method to prioritize compounds for experimental testing.

In conclusion, pharmacophore-derived virtual screening provides a supplementary and complementary method for docking, contributing to a consensus and greater confidence in candidate selection for experimental validation. Beyond the urgency of providing a drug candidate for repurposing, this pharmacophore-driven method may also identify ligands that are more diverse for subsequent lead optimization.

References:

1. Pushpakom et al. Drug repurposing: progress, challenges and recommendations. Nat Rev Drug Discov. 2019, 18(1):41-58.
2. https://www.diamond.ac.uk/covid-19/for-scientists/Main-protease-structure-and-XChem.html
3. https://www.bioworld.com/COVID19products
4. Fidan and A. Aydoğdu. As a potential treatment of COVID-19: Montelukast. Med Hypotheses. 2020, May.
5. Vaduganathan et al. Renin–Angiotensin–Aldosterone System Inhibitors in Patients with Covid-19. N Engl J Med. 2020; 382:1653-1659.
6. J-M. Yang et al. Consensus Scoring Criteria for Improving Enrichment in Virtual Screening. J Chem Inf Model. 2005, 45(4):1134-46.