A repurposed drug screen identifies compounds that inhibit the binding of the COVID-19 spike protein to ACE2

Abstract

Repurposed drugs that block the interaction between the SARS-CoV-2 spike protein and its receptor ACE2 could offer a rapid route to novel COVID-19 treatments or prophylactics. Here, we screened 2701 compounds from a commercial library of drugs approved by international regulatory agencies for their ability to inhibit the binding of recombinant, trimeric SARS-CoV-2 spike protein to recombinant human ACE2. We identified 56 compounds that inhibited binding by <90%, measured the EC₅₀ of binding inhibition, and computationally modeled the docking of the best inhibitors to both Spike and ACE2. These results highlight an effective screening approach to identify compounds capable of disrupting the Spike-ACE2 interaction as well as identifying several potential inhibitors that could serve as templates for future drug discovery efforts.

Introduction

COVID-19 is currently a global pandemic, causing extensive mortality and economic impact. While the success of rapidly developed vaccines offers hope to control the virus¹, treatments that improve disease outcomes are also critically needed. Remdesivir, a nucleoside inhibitor of viral RNA polymerase, is the only drug approved by the FDA to treat COVID-19², though its efficacy is disputed³. Monoclonal antibodies⁴ or recombinant Ace2⁵, which block the interaction between the SARS2 spike protein and its obligatory receptor ACE2, have also shown promise, but are expensive and suffer production limitations. Repurposing already-approved small molecule drugs could allow for rapid deployment of low-cost and widely available therapeutics⁶, but candidates studied in robust clinical trials have thus far failed^3,7,8. Here, we took an unbiased approach to screen 2701 drugs approved by global regulatory agencies for the ability to block the interaction between recombinant, trimeric SARS2 spike protein⁹ and latex-bead-conjugated recombinant human ACE2.

Methods

Drug Screening

The “FDA-approved drug screening library” (Cat # L1300) was purchased from Selleck Chemicals. In a 96-well plate format, we briefly incubated recombinant, biotinylated, trimeric spike protein⁹ with either 200uM or 1mM of each drug, in duplicate, then added 5-micron flow cytometry beads (Luminex) coated with recombinant ACE2 (for detailed methods, see¹⁰). Three replicates per plate of positive (vehicle) controls and negative no-spike-protein controls were included, 31 plates in total. After washing and the addition of streptavidin-PE to bind spike attached to ACE2, plates were washed again on a magnetic plate washer and read on an Acea Novocyte flow cytometer. Data were expressed as the median PE fluorescence intensity (MFI), and converted to % inhibition using the formula 1-(MFI_drug/MFI_{positive control}). For EC₅₀ studies, serial dilutions of drugs were performed, in duplicate, and run as above.

In Silico Modeling

The S1 subunit of SARS-CoV-2 spike protein comprising the receptor binding domain (RBD) was modelled using the SWISS-MODEL server¹¹, the FASTA sequence (residues 316-530) was retrieved from UniProtKB - P0DTC2 and used as a query sequence, PDB ID: 6VSB with 100% sequence identity was used as a template^12,13. The structure quality of the modelled protein was validated on PROCHECK, which showed 89.2% residues in the core regions, the quality factor of 89.77% was obtained from Verify 3D on SAVES server and ProSA web gave a Z-score of -6.15 ^14–16. The crystal structure of ACE2 protein was retrieved from PDB 1D: 2AJF and used for docking into the RBD interface. The SDF structures of the selected FDA-approved drugs were downloaded from Selleck chemicals and PubChem. Ligand and protein preparation was performed using the Ligprep and protein preparation wizard tool on Schrödinger Maestro version 12.2. Structural-based docking to the RBD interface of the Spike and ACE2 was also performed using Schrödinger Maestro and BIOVIA Discovery Studio (Dassault Systems) for docking analysis and visualization.

Results

We took an unbiased approach to screen 2701 drugs approved by global regulatory agencies for the ability to block the interaction between recombinant, trimeric SARS2 spike protein⁹ and latex-bead-conjugated recombinant human ACE2 (Figure 1A). The use of a cell-free system prevented potential cytotoxic effects of drugs inherent to cell-culture based live-virus assays, and allowed us to focus solely on inhibition of the Spike-ACE2 interaction. Using a previously validated assay¹⁰, we quantified the amount of Spike-ACE2 co-association in the presence of high concentrations (200uM-1mM) of each drug, in duplicate, and calculated the percent inhibition using six replicates of vehicle control per plate (31 plates total). In this first-round screen, 114 drugs that exhibited 90% or greater inhibition were identified (Figure 1B and Table S1).

Figure 1: Inhibition of Spike-ACE2 binding by repurposed drugs.

A) In vitro assay design showing inhibition of ACE2-Spike binding by “effective” drugs. B) Example histogram from primary screen showing >90% inhibition of ACE2-Spike binding following drug addition. C, D) EC₅₀ data for the top candidate, Thiostrepton (structure inset) expressed as PE fluorescence (C) or percent inhibition (D). E) Three-dimensional and F) two-dimensional computational rendering of Thiostrepton binding to the SARS-CoV-2 spike protein.

We next performed serial dilutions of these 114 drugs, in duplicate, to measure EC₅₀s of the ACE2 – spike interaction (Figure 1C, D). Fifty-eight of the drugs were revealed to be either false positive hits (they showed no inhibition upon re-screening), or showed inhibition only at the highest concentration tested, and were eliminated. The drug with the highest EC₅₀ was Thiostrepton (EC₅₀ = 3.95 x 10⁻⁶ M), a cyclic oligopeptide used as a topical antibiotic in animals, and that interacts with the transcription factor FOXM1 to inhibit the growth of breast cancer cells in vitro¹⁷. Next was Oxytocin (EC₅₀ = 4.17 ⨯ 10⁻⁶ M), a peptide hormone that is administered to induce childbirth, and that may increase social cognition when administered intranasally¹⁸. The next four best candidates were actually two closely related pairs of drugs, which demonstrates the robustness of our screen in identifying each compound twice. Nilotinib (EC₅₀ = 4.20 ⨯ 10⁻⁶ M), which was identified as both a free base and an HCl salt, is a selective tyrosine kinase inhibitor used to treat chronic myelogenous leukemia¹⁹.

Hydroxycamptothecine (EC₅₀ = 7.25 ⨯ 10⁻⁶ M) and its stereoisomer S-10-Hydroxycamptothecine (EC₅₀ = 7.07 × 10⁻⁶ M) are DNA topoisomerase I inhibitors with anti-cancer activity²⁰. Interestingly, three derivatives that have also been approved for cancer therapy, Topotecan, Irinotecan, and Belotecan were included in the screening panel, but did not inhibit spike-ACE2 binding. The EC₅₀s of all 56 compounds are listed in Table S2; given the decreased severity of COVID-19 in females²¹, it is notable that Estradiol Benzoate (a synthetic estrogen) inhibited the interaction (EC₅₀ = 1.75 ⨯ 10⁻⁵ M).

We next performed molecular docking studies with both spike and ACE2 to unravel the binding and crucial molecular interactions of the top 12 candidates, focusing on the Spike-Ace2 binding interface. The compounds with the lowest EC₅₀s also gave the lowest docking (Glide²²) scores (Table 1), cross-validating our results. The two top hits, Thiostrepton and Oxytocin, bound more favorably to Spike than Ace2, and interacted with several key residues that mediate Spike-ACE2 binding²³ (Figure 1, Figure S1-S2, and Table S3). Thiostrepton in particular bound extensively to Spike residues, and an OH-group in Thiostrepton bound simultaneously with Lys417 of Spike and Asp30 of ACE2. Simultaneous binding of these two critical interface resides would likely disrupt the Spike-ACE2 interaction, resulting in the low observed EC50 value. Nilotinib and Hydroxycamptothecine exhibited glide scores that were slightly more favorable for ACE2, but also showed moderate binding with Spike. Interestingly, both interacted with Arg393 on ACE2, a critical spike-binding residue¹⁶, and both also interacted with the Spike receptor binding interface (Figure S3-S6 and Table S3). The remaining compounds generally gave higher glide scores, consistent with their lower affinity, but still docked appreciable to ACE2 and/or Spike (see Figure S7-S12, Table S3 and Supplementary Discussion).

Table 1: Summary of the top 12 drug candidates.

Computationally modeled glide scores for ACE2 and spike binding and EC₅₀ values measured with the recombinant Spike-ACE2 binding inhibition assay are displayed.

Discussion

Overall, this study identified 56 approved drugs that show some efficacy in blocking the interaction between the COVID spike protein and its receptor, ACE2. Many of the identified drugs are already approved for clinical use in humans (see Table 2). However, aiming towards clinical translation, there are several obvious issues. First, many of the drugs are chemotherapy agents, and have toxic side-effects that would not be tolerable in COVID-19 patients. For example, Nilotinib inhibits a kinase important in B cell signaling^19,24, and may prevent normal immune function, while Picropodophyllin²⁵ and Docetaxel²⁶ both produce moderate to severe side effects when used in the context of chemotherapy. Secondly, several of the drugs have peak plasma concentrations that are orders of magnitude lower than the concentration required to inhibit spike-Ace2 binding in the in vitro assay, for example Oxytocin²⁷ and Doramectin²⁸ (0.005 vs 4.2 uM and 0.014 vs 12.8 uM, respectively). Finally, Oxytocin’s clearance rate would necessitate continuous infusion, which seems impractical.

Table 2: Reported pharmacokinetic properties of top hits^24–33.

Drugs in grey indicate dosing studies were performed on non-human animals.

In light of known side-effects and pharmacokinetic data of many of our high-ranking drug candidates, Selamectin may be the top candidate for further study. Selamectin is an anti-parasitic used in dogs and cats to prevent infestation with nematode and arthropod species. Oral Selamectin is well tolerated in dogs and can achieve a peak plasma concentrations (9.9uM) comparable to the measured EC₅₀ (8.5 uM)²⁹. However, we were not able to identify any studies using Selamectin in humans, since Ivermectin is the standard alternative. Thiostrepton would also be a top candidate, but pharmacokinetic data are lacking.

An advantage of our recombinant approach, using a cell-free system, is that the effects of drug toxicity on cell growth do not confound the readout of Spike-Ace2 binding. However, our results would need to be replicated in a cell or animal model using live virus to ensure that anti-SARS effects appear at drug concentrations low enough to prevent toxicity. The relatively weak (micromolar) binding kinetics of the drugs identified here, as well as their known toxicities, bioactivities and/or high clearance rates, suggest that many would currently be unlikely to be viable for treating acute disease or for prophylactic use. However, they could serve as starting points for future medicinal chemistry optimization efforts to rationally design derivatives that are both less toxic and bind to the COVID spike with higher affinity.

Author Contributions

SEPS conceived the study. KT, CA, EPG, JW and SEPS planned experiments. NS provided recombinant COVID proteins for the inhibition assay. KT and EPG performed microsphere-based experiments, CA performed modeling studies under the supervision of JW. KT, CA, JW and SEPS, analyzed the data. KT, CA, JW and SEPS wrote the manuscript, and all authors read and approved the manuscript.

Acknowledgements

The authors thank Jason McLellan and Barney Graham for their generous gift of the SARS-CoV-2 recombinant trimer and hACE2 expression constructs; This study was funded by a COVID-19 research award from the University of Washington Institute for Translational Heath Sciences (via grant UL1 TR002319) and by a COVID19 award from the Research Integration Hub at Seattle Children’s Research Institute (to SEPS), and by R01 AI140951 (to DNS). The authors declare no conflicts of interest.