Targeting the coronavirus SARS-CoV-2: computational insights into the mechanism of action of the protease inhibitors lopinavir, ritonavir and nelfinavir

SARS-CoV-2, an ssRNA betacoronavirus first identified in December 2019, rapidly spread worldwide causing COVID-19 with significant mortality and socio-economic impact. The urgency of delivering treatments conflicts with typical drug discovery timelines, motivating accelerated strategies including drug repurposing. Early genomic and structural data revealed high similarity to SARS-CoV (approximately 80% genome identity; Mpro shares 96.1% sequence identity), enabling transfer of knowledge and structure-based approaches. The SARS-CoV-2 main protease (Mpro/C30 endopeptidase) is essential for viral replication and maturation, making it a prime antiviral target alongside structural proteins involved in entry (e.g., ACE2-mediated). With the rapid release of Mpro’s crystal structure (PDB ID: 6LU7), structure-based drug design and repurposing became feasible. HIV protease inhibitors, notably lopinavir/ritonavir (undergoing randomized clinical trials in China) and nelfinavir (with prior in vitro activity against SARS-CoV protease), emerged as candidates. This study aims to elucidate, at atomic detail, the putative recognition and binding mechanisms of lopinavir, ritonavir, and nelfinavir to SARS-CoV-2 Mpro using supervised molecular dynamics (SuMD), thereby informing repurposing prospects and guiding inhibitor design.

Multiple computational efforts, including molecular docking-based virtual screening, rapidly evaluated approved drugs against SARS-CoV-2 Mpro. Convergence across studies frequently highlighted HIV protease inhibitors as potential Mpro binders, consistent with earlier repurposing evidence during SARS-CoV and MERS-CoV outbreaks. Prior literature documents promising experimental activity for several HIV protease inhibitors against related coronaviral proteases, with nelfinavir reported to inhibit SARS-CoV replication in vitro and show activity against SARS-CoV protease. Clinical interest led to at least three randomized trials in China evaluating lopinavir/ritonavir for COVID-19. The availability of the SARS-CoV-2 Mpro crystal structure (PDB 6LU7) with a covalent peptidomimetic (N3) reinforced the rationale for structure-guided assessment. SuMD, previously validated to reconstruct ligand–protein recognition pathways at nanosecond timescales, provides a dynamic alternative to docking by incorporating protein flexibility, water mediation, and the ability to reveal metastable states during binding.

Structural data and preparation: The Mpro (C30 endopeptidase) crystal structure in complex with covalent inhibitor N3 (PDB ID: 6LU7) was retrieved. Given the homodimeric symmetry and two equivalent catalytic sites, only one monomer (chain A) was used. The covalent ligand was removed; residue Y154 was restored; missing atoms built using AMBER14 topology. Hydrogen atoms and appropriate protonation states were assigned via MOE Protonate-3D. Coordinates for lopinavir, ritonavir, and nelfinavir were built in MOE and initially placed at least 30 Å from the binding site to avoid premature interactions (beyond the 9 Å electrostatic cutoff of the employed force field). System setup: Each protein–ligand system was explicitly solvated in a cubic TIP3P water box with at least 15 Å padding from solute. Systems were neutralized with Na+ and adjusted to 0.154 M ionic strength. Energy minimization: 500 steps of conjugate-gradient. Equilibration: 500,000 steps (≈1 ns) of NVT with 2 fs timestep, harmonic positional restraints on protein and ligand heavy atoms (1 kcal mol−1 Å−2) gradually reduced by a 0.1 scaling factor. Temperature 310 K (Langevin thermostat, 1 ps−1 damping), pressure 1 atm (Monte Carlo barostat). M-SHAKE constrained bonds to hydrogens. Electrostatics via particle-mesh Ewald (1 Å grid), Lennard-Jones cutoff 9.0 Å. SuMD protocol: SuMD supervises sequential short MD windows (nominally picoseconds-scale per window; supervision intervals of 60 ns reported for monitoring) by fitting the time series of the distance between the ligand and the protein binding site to a linear function; only windows with a negative slope (approach) are accepted; otherwise velocities are randomized and the step repeated. Supervision proceeds until the ligand–site distance falls below 5 Å, after which supervision is disabled and classical MD continues to refine the bound pose. For each ligand, up to 10 SuMD binding trajectories were generated; the best trajectory was selected for detailed analysis. Analyses: Ligand–protein interaction energies (electrostatic + van der Waals) computed with NAMD using AMBER ff14SB for post-processing; interaction energy landscapes plotted versus ligand–protein mass center distances. Per-residue interaction contributions were computed for residues within 4 Å dynamically during binding. Trajectory analysis employed an in-house Python tool leveraging ProDy; visualizations with VMD. General molecular modeling was performed in MOE (2018.0101). Hardware: Linux workstation with 8 CPUs (Intel Xeon E5-1620 3.50 GHz). Supplementary SuMD videos (V1–V3) synchronized molecular graphics with distance and energy timelines.

- Lopinavir: A complete recognition trajectory was sampled within ~20 ns. First contacts occurred at ~5 Å from the catalytic site. The predicted bound state featured a persistent double hydrogen bond to the Glu166 backbone, a key anchoring interaction observed in SARS-CoV and in the 6LU7 covalent complex. The cyclic urea moiety formed an additional hydrogen bond with Gln189. Interaction energy analysis indicated a stable binding mode consistent with these contacts. - Ritonavir: A binding trajectory was also obtained within ~20 ns. Final poses displayed hydrogen bonding to Glu166 and Gln189; however, interaction energy landscape comparisons suggested lower energetic stability versus lopinavir. The urea moiety remained solvent-exposed, indicating suboptimal accommodation and weaker stabilization relative to lopinavir. - Nelfinavir: Required a slightly longer SuMD to traverse an initial metastable vestibular region near the catalytic site, where it engaged several polar residues for ~20 ns. Subsequent conformational adjustments over the last ~10 ns led to an optimally accommodated pose within the binding cleft, stabilized by a dense hydrogen-bond network. Key residues mediating direct and water-bridged interactions included His166, Glu166, Gln189, Thr190, and Gln216. In the final nanoseconds, a stabilizing salt bridge formed between Glu166 side chain and the positively charged octahydro-1H-isoquinoline moiety of nelfinavir. The predicted binding mode showed strong analogies to the N3-bound crystallographic pose, with interaction energies indicating a highly stabilized final state. - Across ligands, Glu166 emerged as a central determinant of binding stability, consistent with literature reports that its mutation disrupts protease dimerization and activity in SARS-CoV and MERS-CoV.

This preliminary computational investigation leveraged the SARS-CoV-2 Mpro structure and SuMD to elucidate dynamic recognition pathways of three repurposing candidates. Findings support that lopinavir can achieve a stable bound state through key interactions (notably Glu166 backbone and Gln189), while ritonavir, although capable of engaging the site, adopts less energetically favorable accommodations due to incomplete insertion of its urea motif. Nelfinavir exhibited the most favorable dynamic stabilization among the three, transitioning from a metastable vestibular site to a densely hydrogen-bonded final pose that mirrors critical features of the reference covalent inhibitor and forms a salt bridge to Glu166. The centrality of Glu166 across all trajectories underscores its importance in ligand anchoring and aligns with mutational data linking Glu166 to protease dimerization and function. These insights refine mechanistic hypotheses for how HIV protease inhibitors might interact with SARS-CoV-2 Mpro, informing rational optimization and prioritization for experimental validation within repurposing pipelines.

SuMD simulations provided atomistic pathways and plausible bound poses for lopinavir, ritonavir, and nelfinavir at SARS-CoV-2 Mpro. Lopinavir formed stable, literature-consistent interactions (notably with Glu166 and Gln189), ritonavir showed comparatively weaker stabilization due to suboptimal binding-site engagement, and nelfinavir displayed a particularly robust binding mode with extensive hydrogen bonding and a Glu166 salt bridge resembling features of the crystallographic reference inhibitor. The recurrent involvement of Glu166 highlights a conserved anchoring hotspot relevant to inhibitor design. These computational insights support further experimental evaluations (biochemical inhibition, structural studies, and cellular assays) of nelfinavir and lopinavir against Mpro and suggest structure-guided optimization around interactions with Glu166/Gln189. Future work may include free energy calculations, explicit consideration of dimeric states, assessment of water networks, and mutational analyses to validate predicted interaction determinants.

The study is computational and preliminary, lacking direct experimental validation of the predicted binding modes and affinities. Only one monomer of the homodimeric Mpro was simulated, potentially underrepresenting dimerization effects known to influence activity. SuMD selected and analyzed the best trajectory among multiple runs, which may introduce selection bias. The force-field and solvent models (AMBER14/ff14SB, TIP3P) and simulation timescales (nanoseconds) impose approximations, and long-timescale conformational dynamics or alternative binding pathways may not be fully captured.