Dengue virus remains one of the world’s most burdensome mosquito-borne pathogens, and there is still no specific antiviral drug for the disease it causes. According to 2024 surveillance data from the World Health Organization cited in the study, more than 14 million dengue cases and over 11,000 dengue-related deaths were reported worldwide that year, the highest annual global burden recorded to date. A new computational study, published as open-access research in the journal Discover Viruses, reports the identification of four natural product candidates that, in silico, bind more strongly to a key viral enzyme than a known reference inhibitor. The work is entirely computational: no compound has yet been tested in the laboratory against dengue virus.
The enzyme in the crosshairs is the RNA-dependent RNA polymerase (RdRp), the C-terminal domain of the viral NS5 protein, which copies the viral RNA genome during replication. The structure used in the study was the DENV RdRp crystal structure deposited in the Protein Data Bank under accession 5K5M, resolved at 2.01 angstroms and co-crystallized with a known inhibitor abbreviated 68T. That reference compound served two purposes: it defined the binding pocket and it provided a benchmark docking score against which the screened natural products could be compared.
The researchers, led by Aishwarya Khare and colleagues, drew their chemical starting material from COCONUT, the Collection of Open Natural prodUcTs, a freely accessible database of structurally diverse natural products. A library of 407,270 compounds was passed through a three-tier structure-based virtual screening pipeline built on the Glide docking engine within the Maestro-Schrödinger suite. The first stage, high-throughput virtual screening, reduced the pool; the second, standard-precision docking, refined it; and the third, extra-precision docking, scrutinized the survivors. Only about 10 percent of ligands advanced at each stage. Every compound was prepared with tautomeric states generated at pH 7.0, and a QikProp filter removed molecules with unfavourable predicted pharmacokinetic and physicochemical properties before docking began.
From this funnel, 1,420 compounds emerged with docking scores ranging from −13.235 to −9.029 kcal/mol. To guard against false positives, the team then rescored the docked poses with the Molecular Mechanics and Generalized Born Surface Area method, MM/GBSA, which estimates binding free energy and yielded values from −81.64 to 21.31 kcal/mol across the screened set. Four compounds rose to the top: CNP0243663, CNP0306108, CNP0384605 and CNP0226892, with MM/GBSA binding free energies of −81.64, −79.97, −77.64 and −74.97 kcal/mol respectively. Their docking scores were −11.588, −9.639, −11.807 and −10.284 kcal/mol, all better than the −9.368 kcal/mol recorded when the reference inhibitor 68T was re-docked into the same pocket.
The interaction analysis showed why the compounds scored well. CNP0243663 formed seven hydrogen bonds with residues including Lys460, Arg729, Arg737, Arg792, Thr794, Trp795 and His798, alongside hydrophobic contacts with Leu512, Leu734, Tyr758, Met761, Trp795 and Trp803, salt bridges with Lys460 and Arg729, and π-π stacking with Trp795. CNP0384605 made three hydrogen bonds with Arg737, Thr794 and Trp795 plus three π-π stacking interactions with Trp795, while CNP0306108 and CNP0226892 formed two and three hydrogen bonds respectively. The re-docked 68T reference formed five hydrogen bonds with Arg729, Arg737, Thr794, Lys800 and Glu802. The overlap in contact residues indicated that the natural compounds occupy the same binding region as the reference ligand.
Stability over time was then tested with 200-nanosecond molecular dynamics simulations run under the NPT ensemble at 300 K and 1.01325 bar, using the OPLS-2005 force field and explicit TIP4P water with neutralizing counterions. Protein backbone root-mean-square deviation stayed below 3.5 angstroms in all five complexes, indicating that the enzyme itself did not undergo major conformational changes. Ligand behaviour varied more. CNP0243663 was the most stable ligand, equilibrating at roughly 3.5 to 4.0 angstroms and remaining consistently below the reference ligand, which deviated by up to 5.4 angstroms. CNP0226892 showed the largest excursion, peaking at 6.5 angstroms before settling, and CNP0384605 fluctuated between roughly 3.0 and 4.5 angstroms with occasional peaks near 5.5 angstroms, movements the authors interpret as repositioning within, rather than escape from, the pocket.
Contact mapping across the trajectories highlighted which interactions persisted. In the CNP0243663 complex, Arg729 maintained a hydrogen bond with the ligand for the entire 200 nanoseconds, with water-mediated contacts through Glu459 and Glu733 also recurring. In the CNP0306108 complex, Glu464, Tyr758 and Thr793 each held hydrogen bonds for 100 percent of the simulation. For CNP0384605, Arg729 and Arg737 maintained hydrogen bonds for 100 and 80 percent of the time respectively. Root-mean-square fluctuation analysis showed maximum protein fluctuations of 4.0 to 4.5 angstroms for the natural compound complexes versus 4.8 angstroms for the reference, and solvent-accessible surface area analysis suggested CNP0306108 adopted the most buried binding mode.
Post-simulation MM/GBSA rescoring over the trajectories produced a somewhat different ranking than the docking stage. CNP0306108 came out strongest, with a mean binding free energy of −85.18 ± 4.43 kcal/mol, and CNP0384605 scored similarly to the reference at roughly −77 kcal/mol. CNP0226892 yielded −69.70 ± 9.15 kcal/mol, while CNP0243663, despite its strong docking performance, showed the weakest post-simulation value at −59.51 ± 6.47 kcal/mol. Energy decomposition attributed the binding largely to van der Waals and solvation terms rather than covalent or hydrogen-bond contributions. Principal component analysis of the trajectories found that the first three principal components captured between 38.4 and 40.3 percent of conformational variance across the four natural compound complexes, close to the 39.9 percent for the reference complex, indicating broadly similar essential protein motions upon ligand binding.
Electronic and drug-like properties were assessed separately. Density functional theory calculations at the B3LYP/6-31G** level characterized the HOMO-LUMO gaps, dipole moments, hardness, softness, electronegativity and electrophilicity of all four candidates and the reference compound, with band gaps ranging from 0.14925 Hartrees for 68T to 0.19648 Hartrees for CNP0243663. SwissADME profiling predicted low gastrointestinal absorption, no blood-brain barrier permeability and compliance with Lipinski’s rule of five for all compounds, properties the authors note are not disqualifying for a dengue antiviral because the RdRp functions in the cytoplasm. Three of the four compounds were predicted not to inhibit CYP2D6. ProTox-III toxicity predictions flagged CNP0306108, CNP0384605 and CNP0226892 as potentially hepatotoxic, neurotoxic and immunotoxic, whereas CNP0243663 was predicted to be non-hepatotoxic, non-neurotoxic, non-cardiotoxic, non-mutagenic, non-immunotoxic and non-cytotoxic, giving it the most favourable safety profile in the set.
The authors are explicit that these are computational predictions requiring experimental confirmation. Enzymatic assays against the purified polymerase and cell-based dengue infection models are the necessary next steps before any of the four candidates can be considered genuine antiviral leads. What the study does deliver is a prioritized, structurally diverse shortlist of natural products, together with a documented screening workflow, that future laboratory work can test directly against the dengue virus polymerase.
Subject of Research: Computational identification of natural product inhibitors of the dengue virus RNA-dependent RNA polymerase
Article Title: Natural product inhibitors targeting dengue virus RNA dependent RNA polymerase using molecular dynamics and density functional theory
Article References: Khare, A., Tripathi, V., Dwivedi, V. D., Pathak, N., & Srivastava, S. (2026). Natural product inhibitors targeting dengue virus RNA dependent RNA polymerase using molecular dynamics and density functional theory. Discover Viruses, 3(1), Article 8. https://doi.org/10.1007/s44370-026-00043-2
Image Credits: AI Generated
DOI: 10.1007/s44370-026-00043-2
Keywords: dengue virus, RNA-dependent RNA polymerase, natural products, virtual screening, molecular docking, molecular dynamics simulation, MM/GBSA, density functional theory, COCONUT database, drug discovery, antiviral candidates, in silico
News Source: Kristina Jarvis. (October 29, 2026). Computational screen of 407,270 natural compounds yields four predicted dengue polymerase inhibitors. Scienmag.
