The SARS‑CoV‑2 main protease (Mpro) is highly conserved across coronaviruses and performs an essential role in viral polyprotein processing and replication. Because of these properties, Mpro remains a key target for antiviral drug discovery against COVID‑19. The present study used structural information from existing Mpro–inhibitor complexes to guide an in‑silico search for novel non‑covalent inhibitors within a curated Thai natural product database.
The investigators collected 148 crystal structures of SARS‑CoV‑2 Mpro bound to non‑covalent inhibitors from the Protein Data Bank. From this ensemble of complexes, they derived protein–ligand interaction fingerprints (IFPs) to characterize recurrent interaction patterns across ligands and to map the inhibitor binding site. The IFP analysis identified recurring contacts and interaction motifs that define the preferred binding geometry for non‑covalent inhibitors.
Key output from the mapping was the observation that the S2 binding site functions as a major anchoring region for most non‑covalent inhibitors present in the crystallographic set. The S2 pocket therefore represented a primary structural feature to target during virtual screening, with specific residue interactions and contact types informing what a desirable ligand pose should reproduce.
Using the interaction fingerprint map as a guide, the team virtually screened compounds from an in‑house Thai natural product database. The screening workflow prioritized not only docking scores but also agreement between predicted ligand–protein contacts and the experimentally derived IFPs, thereby favoring compounds predicted to reproduce key anchoring interactions at subsites such as S2.
Docking procedures evaluated pose quality and energetic criteria alongside the interaction profile. The combined use of structural IFP information and conventional docking metrics aimed to enrich for compounds likely to recapitulate the interaction pattern observed among known non‑covalent Mpro inhibitors.
From the virtual screen, seven hit compounds were selected for further evaluation. Selection criteria explicitly included docking pose quality, the extent to which predicted interactions matched the IFP‑derived map, and favorable energetic metrics from docking. The abstract lists three compounds that stood out during follow‑up analysis (see next sections); full chemical identities and the other four hits were not detailed in the abstract.
To assess the dynamical stability of predicted Mpro–hit complexes and the persistence of key interactions, molecular dynamics (MD) simulations were performed on the selected complexes. MD analysis confirmed the stability of protein–ligand complexes for the hits examined and provided time‑resolved evidence that specific interactions with residues in the S2 site were maintained.
The simulations were used to examine interaction types—such as π–π stacking, hydrogen bonds, and H–π contacts—and the consistency of these interactions over the simulation trajectories. This dynamic validation step served to filter docking artifacts and prioritize compounds with stable, reproducible contact patterns in a simulated aqueous and protein environment.
Three compounds emerged as particularly interesting virtual hits based on docking and MD results:
Panduratin A (CPD3): Predicted to form consistent strong interactions at the S2 binding site, including aromatic stacking and hydrogen‑type contacts that help anchor the ligand.
Deacetylmammea E/BA cyclo D (CPD4): Also showed persistent interactions with key S2 residues, suggesting a stable non‑covalent binding mode.
N‑pentyl beta‑carboline‑1‑propionate (CPD5): Demonstrated sustained π–π and H–π interactions with residues at the S2 pocket during MD simulations.
The abstract highlights that these three compounds consistently formed strong interactions at S2, mediated by π–π, H–π, and hydrogen bonds, across the simulated trajectories.
Implications:
Limitations noted by source content:
Recommended next steps (based on the nature of the reported work):
The authors declared no conflict of interest. Citation and indexing details reported in the source: J Comput Aided Mol Des. 2026;40(1):234. PMID: 42758356. DOI: 10.1007/s10822-026-00941-z.
Note: The abstract summarizes the computational strategy and highlights candidate compounds; specific experimental validation data were not reported in the abstract.