In silico molecular dynamics simulation of the SARS-CoV-2 S/ACE2 complex and virtual screening.

A model was prepared considering the spike protein S1 of SARS-CoV-2 complexed with human ACE2 starting from the X-ray structure 6M0J (resolution, 2.45 Å) (7). This structure contains the receptor binding domain (RBD) of S1 and includes all the amino acid residues of the RBD, with well-defined density between the residues on S1 RDB and ACE2. The model was treated with PropKa (12) to assign protonation states and subjected to molecular dynamics simulations for 400 ns based on the AMBER ff14SB force field as described in detail previously (6). Final trajectories were analyzed in terms of backbone root mean square deviation (RMSD), hydrogen bonds formed, and cluster analysis. From the cluster analysis performed on the ensemble of poses generated from the MD simulations, one representative structure from each of the 8 dominant clusters was selected and analyzed with FPocket software (13). This approached enabled the identification of possible druggable pockets in the interfacial S1-ACE2 region in each of the selected structures and in the initial model prepared from the X-ray structure. These were described in detail previously (6).

Structure-based virtual screening was performed using the following four molecular libraries: (i) the French National Chemical Library (Chimiothèque Nationale) (70,000 molecules; https://chembiofrance.cn.cnrs.fr/fr/composante/chimiotheque), (ii) the Mu.Ta.Lig. Virtual Chemotheca (ca. 60,000 molecules; http://chemotheca.unicz.it), (iii) the Inhibitors of Protein-Protein Interactions Database (1,956 molecules; https://ippidb.pasteur.fr), and (iv) the ZINC FDA Approved and ZINC In-Trials Database (1,379 and 5,811 molecules for evaluation for drug repurposing; https://zinc.docking.org). These libraries were screened against the SARS-CoV-2 S/ACE2 complex (model 1) (Fig. 1C) and against SARS-CoV-2 S alone (model 2) (Fig. 1D), with two independent docking software programs and scoring functions: AutoDock Vina (14) and GOLD (PLP scoring function) (15). It should be noted that the different docking program results have different values and scales. The score for PLP is dimensionless, and the higher the score, the better the binding affinity. On the other hand, AutoDock Vina uses a metric that resembles binding free energy (in kcal/mol), so a more negative value suggests a stronger affinity.

Compounds were sorted based on their Vina and PLP scores against model 1 and model 2. A selection of 20 compounds based on the most appealing compounds from each library was then made. This selection took into consideration not only the score against both models with Vina and/or PLP but also the chemical diversity among the molecules selected and the physicochemical properties of the molecules (Table S1 and Fig. S2).