Structural and Binding Analysis of SARS-CoV-2 Spike Protein

Samples of Arbidol 1 and derivatives were purchased from ChemDiv, including 1 (1635-0087), 1b (8015-5742) and 1c (H027-0218C) and 1d (H027-0205C). Other compounds were purchased from Selleck Chemicals, including Clofazimine 2 (S4107), Toremifene 3 (S1776), Ecliptasaponin A 4 (S9403) and Ivermectin (S1351). While Toremifene Citrate (S1776) and Ivermectin (S1351) came as a stock solution in 10 mM DMSO, all other purchased compounds were prepared as a standard 10 mM DMSO stock solution from an exact known weight (mg) of each compound. Compound DMSO stock solutions were prepared from the same source of DMSO used in the SPR experiments to minimize observed bulk responses in compound injections.
All SPR experiments were performed at Reaction Biology Corporation (Malvern, PA, USA) using a Biacore 8K+ (Cytiva) instrument with high sensitivity [32 (link)]. For immobilizations, a “series S” and “SA” sensor chips (Cytiva) were used to capture Avi-tag biotinylated protein samples of the SARS-CoV-2 Spike protein. Two biotinylated recombinant samples of the Spike protein were purchased from Acro Biosystems: a full-length trimeric SARS-CoV-2 Spike construct with the D614G mutation (Biotinylated SARS-CoV-2 S protein, catalog number SPN-C82E3) and a trimeric construct of the SARS-CoV-2 S2 segment (Biotinylated SARS-CoV-2 S2 protein, catalog number S2N-C52E8). The experimental design aims to compare a full-length Spike trimer to an equivalent trimeric S2 construct, and the two constructs utilized were the best available equivalents that were biotinylated and resulted in successful immobilizations and subsequent SPR binding experiments. Both constructs incorporate a series of substitutions that stabilize the folded trimeric prefusion conformation. The full-length construct with the D614G mutation (catalog number SPN-C82E3) contains Proline substitutions (F817P, A892P, A899P, A942P, K986P, V987P) to stabilize the trimeric prefusion state and Alanine substitutions (R683A and R685A) to remove the furin cleavage site. The S2 construct (catalog number S2N-C52E8) contains the same series of Proline substitutions (F817P, A892P, A899P, A942P, K986P, V987P) to stabilize the folded trimeric conformation and minimize the formation of misfolded aggregates during protein production. The S2 construct is shown by the vendor to reliably bind a Spike S2 subunit antibody (Human IgG1) by Avitag ELISA assays (Acro Biosystems, 1 Innovation Way, Newark, DE 19711, USA), where the full-length construct (catalog number SPN-C82E3) is shown by the vendor to reliably bind ACE2 and an Anti-Spike RBD neutralizing antibody (Human IgG1) (Cat. No. SAD-S35) by Avitag ELISA. The series of substitutions in both constructs act to stabilize the full-length folded trimer, as well as to stabilize the folded S2 trimer that reliably binds to a Spike S2 subunit antibody. Under our best conditions after immobilization, both constructs were able to demonstrate binding of small molecules, which diminished over time (after two hours) as is often observed for some folded proteins that may become unfolded over time once immobilized on the biosensor surface. Thus, while we do not know for sure if the exact folded structure of the two constructs is the same, SPR evidence from small-molecule binding experiments suggest that the immobilized samples are folded trimers rather than an unfolded immobilized peptide, which may aggregate on a biosensor surface.
Numerous attempts were made to prepare the biosensor surface and optimize the assay conditions to improve the quality of the binding data (See Section 3). The protein samples were immobilized on the SA chips using a 5 (mL/min) flow rate with a running buffer composed of PBS with 0.05% Tween 20, resulting in ranges of 3000–5000 RU. Following immobilization, both the sample surface and the reference channel surface were blocked with biotin to attempt to minimize non-specific binding. In the first round of assay development and initial data collection on the surfaces described above, small-molecule analytes were analyzed using running buffer composed of PBS with 0.05% Tween 20 and either a 1% DMSO or 2% DMSO solutions. Titrations of each analyte were performed using multi-cycle kinetics mode, with 200 mM as the highest concentration for a 2-fold serial dilution of 10 concentrations. In this first round of data collection, serial dilutions were performed on a plate, where the 200 mM concentration was prepared by mixing 10 mM DMSO stock solution with 0% DMSO running buffer to achieve either a 1% or 2% DMSO solution of analyte to match the running buffer.
In a final round of compound characterization using optimized conditions, a new SA chip was prepared, with the aim of facilitating collection of duplicate sensorgrams from two surfaces. Given the high lipophilicity of some of the compounds, rather than performing the dilution on the plate, the serial dilution was performed in DMSO first to avoid solubility issues. Ten concentrations were prepared in DMSO at 50× the concentration used in the assay and then transferred to the plate and mixed with the DMSO-free running buffer to achieve a 2% DMSO solution of analyte to match the 2% DMSO running buffer. Independent duplicates were compared for each compound and two separate concentration series were performed, with starting concentrations of 100 mM and 50 mM, respectively. Titrations of each analyte were performed using multi-cycle kinetics mode. All SPR data were appropriately solvent-corrected [33 (link)], reference-subtracted and analyzed while fitted to a steady-state affinity model using Biacore Insight Evaluation Software.
All molecular docking and free-energy calculations were performed using CHARMM [34 (link)] and the previously described CHARMM-based computational methods established by our laboratory [35 (link),36 (link)]. Molecular docking utilized the LPDB CHARMm force field to model small-molecule potential functions and the resulting protein–ligand interactions [37 (link),38 (link)]. As previously described, a two-step scoring approach was utilized to rank the final TOP5 poses from any docking attempt. For the final TOP5 docking poses, a final energy minimization of the protein–ligand complex was performed using the Generalized Born using Molecular Volume (GBMV) implicit solvent method [39 (link),40 (link)]. Starting from the minimized complex, minimizations of the bound and free state were performed where potential energy components (VDW), (ELEC) and solvation (SOLV) were calculated in order to approximate the free energy of binding (∆G_bind) using a linear interaction energy scoring approach with previously determined empirical generalized parameters [35 (link)]. Results using the predicted (∆G_bind) values for the TOP5 poses of each individual docking “trials” were pooled and sorted by (∆G_bind), and the top-ranked members of a geometric cluster (RMSD < 2.0 Å) were identified. Statistics for (∆G_bind) were calculated from the average and standard deviation from the three top-ranked members of a geometric cluster (RMSD < 2.0 Å) or a triplicate representing the geometric cluster. For all work performed in this study, independent docking “trials” were initiated from 20 generated conformations of a given small-molecule ligand, such that the initial geometry was entirely independent of any CHARMM-based procedure. MarvinSketch version 15.8.31 is a publicly available 3D conformation generator that was used to generate non-identical low-energy conformations [41 ].
Our laboratory had previously used a pharmacophore procedure to identify the most favorable TOP50 binding sites on the SARS-CoV-2 Spike protein S2 segment [19 (link)]. This structural analysis was performed using the 3.2 Å CryoEM structure (6vyb.pdb) of the full-length Spike protein where the ectodomain was in the “closed” state (6vxx.pdb) [42 (link)]. From this model (6vxx.pdb), molecular docking was performed using a hierarchical approach, such that 10 conformations of Clofazimine were initially docked to all 50 sites on the Spike S2 segment. Then, after identification of the TOP5 most favorable sites from this first step, more extensive sampling was used to refine the ranking of the TOP5 sites, and 20 conformations of Clofazimine were docked. Using this model (6vxx.pdb), the consensus binding mode for Cofazimine binding to the SARS-CoV-2 Spike S2 segment (Figure 3B) was used to dock 18 derivatives of Clofazimine [23 (link)]. These derivatives were modeled at five binding sites: Site 1 and Site 2 on the S2 segment as shown in (Figure 3), as well as for the lowest-energy binding sites [19 (link)] in the Nsp5 Main Protease (6w63.pdb) [43 ], Nsp13 Helicase (6jyt.pdb) [44 (link)], and the Nsp16 2′-O methyltransferase (6wkq.pdb) [45 ]. The results for docking the series into the Nsp5 Main Protease and Nsp16 are expected to represent negative controls, where Clofazimine series SAR data would not be expected to show well-modeled binding to these two sites. As our previous work has highlighted, the Nsp5 Main Protease and Nsp16 binding sites in particular [19 (link)] are thermodynamically favorable for the binding of a variety of small-molecule fragment ligands. This makes them more challenging negative control “decoy” binding sites, particularly compared to most of the possible binding sites on the S2 segment, which are less thermodynamically favorable “decoy” binding sites. The physical basis for this is that the specific molecular shape and the hydrophobicity of the Nsp5 and Nsp16 binding sites are favorable for binding hydrophobic small molecules and result in more thermodynamically favorable and more negative (∆G_bind) values when performing virtual screening of a library of compounds. All molecular graphics images of protein structures and molecular interactions were generated with UCSF Chimera [46 (link)].