Marine Compounds Identified as Potential GLP-1 Receptor Agonists

Researchers have identified three marine-derived compounds that may act as small-molecule agonists for the glucagon-like peptide-1 receptor (GLP-1R), a validated therapeutic target for type 2 diabetes and obesity. The primary objective was to find compounds with predicted oral bioavailability and favorable safety profiles that could serve as functional food ingredients or dietary supplements for metabolic health management.

The investigation began with a large-scale virtual screening of 51,199 prepared three-dimensional ligand states from the CMNPD. Using Glide Extra Precision (XP) docking, the team identified 50 unique hits that produced docking scores more favorable than the reference threshold of −10.608 kcal/mol. This threshold was established by re-docking the co-crystallized reference agonist Danuglipron, which served as a structural benchmark. The docking protocol was validated by reproducing the crystallographic binding orientation of Danuglipron with a heavy-atom root-mean-square deviation (RMSD) of 0.97 Å, well within the predefined acceptance criterion of 2.0 Å.

To manage chemical diversity, the 50 unique hits were clustered using average-linkage hierarchical clustering based on pairwise Tanimoto similarities. The optimal number of clusters was determined to be 13, with the four most populated clusters accounting for 72% of the total hits. This clustering step was designed to reduce redundancy among recurrent structural families while preserving structurally distinct chemotypes. Representatives from the four most populated clusters were selected for detailed structural interaction analysis to illustrate the binding-mode diversity of the screened compounds.

Binding free energy refinement was performed using Prime MM/GBSA calculations to prioritize compounds based on both overall predicted binding energetics and engagement of the key vestibular residue Trp33. Two dataset-derived quartile thresholds were applied: a binding free energy (ΔGbind) of ≤ −64 kcal/mol and a Trp33 interaction energy of ≤ −7.5 kcal/mol. Five compounds met these dual-criterion requirements: CMNPD4578, CMNPD24293, CMNPD26068, CMNPD17793, and CMNPD3660. These candidates were further evaluated using a multiparameter physicochemical and ADMET assessment to guide the selection of candidates for molecular dynamics (MD) simulations.

ADMET profiling revealed distinct characteristics among the five priority hits. CMNPD4578 exhibited the most favorable predicted MM/GBSA ΔGbind value of −91.927 kcal/mol and the strongest Trp33 interaction energy of −10.978 kcal/mol. However, it showed very low predicted Caco-2 permeability and a low human oral absorption class, representing substantial exposure-related limitations. In contrast, CMNPD26068, known as Varioloid A, displayed the most balanced profile, including favorable binding energetics, high predicted Caco-2 permeability, and the lowest predicted number of likely metabolic reactions. CMNPD17793 combined favorable binding energetics with a high qualitative human oral absorption classification.

Based on this multiparameter prioritization, CMNPD4578, CMNPD17793, and CMNPD26068 were selected for 500 ns MD simulations to assess the dynamic stability of the ligand-receptor complexes. The simulations were performed using GROMACS with the CHARMM36m force field for the protein and the CHARMM General Force Field for the ligands. The systems were embedded in a POPC bilayer and solvated with TIP3P water, with counterions added to neutralize the charge and reach a 0.15 M salt concentration. Production NPT dynamics were run at 303.15 K and 1 bar for a total of 500 ns.

The MD simulations revealed significant differences in the dynamic stability of the three candidates. CMNPD17793 and CMNPD26068 exhibited stable behavior, with protein backbone RMSD values remaining in the range of 0.2–0.4 nm throughout the simulation. In contrast, CMNPD4578 developed a pronounced drift in the final portion of the trajectory, with protein RMSD approaching 0.9 nm and ligand RMSD reaching approximately 3 nm. This substantial increase in ligand RMSD was consistent with near-complete escape from the original binding pocket, indicating that CMNPD4578 was dynamically unstable despite its favorable static docking scores.

Hydrogen-bond dynamics analysis provided further insight into the interaction patterns. CMNPD4578 formed the highest average number of hydrogen bonds, with 2.18 ± 1.34 per frame, but this count declined during the later simulation period as the ligand displaced. CMNPD17793 and the reference agonist Danuglipron exhibited comparable hydrogen-bond counts of 1.21 ± 0.92 and 1.20 ± 0.86 per frame, respectively. CMNPD26068 formed fewer hydrogen bonds, averaging 0.43 ± 0.65 per frame, suggesting that its retention within the binding pocket was primarily supported by nonpolar and van der Waals interactions rather than an extensive hydrogen-bond network.

Toxicity and structural-alert profiling were conducted using the Deep-PK web server and SwissADME. None of the investigated compounds was free of predicted toxicity alerts across all evaluated endpoints. CMNPD17793 and CMNPD26068 were classified as potentially mutagenic in the Ames endpoint, with probabilities of 0.984 and 0.988, respectively. CMNPD26068 additionally showed a high-confidence hERG inhibition prediction with a probability of 0.956. Structural-alert screening revealed that CMNPD26068 was free of both PAINS and Brenk alerts, whereas CMNPD17793 contained quinone- and Michael-acceptor-related alerts, which may be associated with chemical reactivity or assay interference.

The source organisms of the stable candidates provide additional context for their potential nutraceutical applications. CMNPD17793 is derived from the brown seaweed Sargassum sagamianum, which is consumed as a traditional food in several East Asian countries and has been scientifically validated for its antidiabetic properties. CMNPD26068 originates from the marine-derived fungus Paecilomyces variotii, a known producer of bioactive secondary metabolites with anti-inflammatory and antioxidant activities. These external biological precedents support the rationale for investigating these compounds in the context of metabolic health management.

The study highlights the importance of dynamic simulation in de-risking candidate selection. While CMNPD4578 appeared strongly bound based on static scoring, its dynamic instability demonstrated that favorable docking scores do not guarantee stable binding. The stable behavior of CMNPD17793 and CMNPD26068, combined with their engagement of the critical Trp33/Trp203 clamp, suggests that they occupy a binding region associated with structurally characterized non-peptide agonists. This convergence supports the structural plausibility of their predicted binding modes and warrants further experimental evaluation.

However, the authors emphasize that in silico prediction of binding affinity does not guarantee functional agonism. Stable binding to the orthosteric site could theoretically correspond to antagonistic activity or receptor stabilization without activation. Therefore, the identified compounds should be regarded as putative agonists or high-priority candidates. Experimental validation, including cAMP accumulation assays and β-arrestin recruitment assays, is essential to definitively confirm their agonistic efficacy and potency in vitro. The study provides a robust foundation for translating these in silico discoveries into tangible, next-generation functional ingredients for diabetes management.

Beyond static docking orientations, short-range interaction energy analyses revealed that dynamic stability was governed predominantly by nonbonded van der Waals forces. Both CMNPD17793 and CMNPD26068 sustained consistently favorable Lennard-Jones interaction terms averaging approximately −210 kJ/mol and −190 kJ/mol, respectively. In contrast, CMNPD4578 initially exhibited strong electrostatic attraction, but its van der Waals contributions deteriorated sharply after 300 nanoseconds as the molecule drifted from the binding site. Hydrogen-bond tracking over the 5,001 simulation frames identified specific anchoring residues. CMNPD17793 formed persistent interactions linking lysine 197 to ligand atom O3 with an occupancy of 43.17 percent, and glutamine 221 to atom O6 with 33.09 percent occupancy, supplemented by contacts with tryptophan 297 and glutamine 37. Conversely, CMNPD26068 maintained pocket retention despite fewer polar contacts, forming its most frequent hydrogen bonds with serine 206 at 15.98 percent occupancy and tryptophan 33 at 4.70 percent. For danuglipron, the reference agonist, the most persistent hydrogen bonds engaged glutamine 234 and glutamine 37, displaying trajectory occupancies of 40.65 percent and 32.99 percent, respectively.

The computational workflow also detailed clear physiological rationales for excluding the remaining dual-criterion hits from dynamic simulation. Although CMNPD24293 demonstrated a favorable MM/GBSA binding free energy of −89.045 kcal/mol and a tryptophan 33 interaction energy of −8.878 kcal/mol, it was eliminated due to a low human oral absorption ranking, a modest Caco-2 permeability of 162.565 nm/s, a high predicted metabolic burden of 13 reactions, and the most unfavorable hERG liability score among the candidates at −6.332. Similarly, CMNPD3660 achieved a binding free energy of −82.238 kcal/mol but was disqualified because of severe aqueous insolubility at −8.218 log mol/L, excessive serum albumin binding indicated by a QPlogKhsa of 1.313, 12 predicted metabolic reactions, and two Brenk structural alerts flagged for an isolated alkene and a Michael acceptor motif.

Quantitative toxicity evaluations using Deep-PK provided additional exposure benchmarks across the prioritized chemotypes. CMNPD26068 achieved a maximum recommended tolerated dose of 0.770 log mg/kg/day, classifying it in the high tolerability tier alongside the synthetic reference danuglipron at 0.700 log mg/kg/day. In comparison, CMNPD17793 and CMNPD4578 were classified in the low tolerated dose category at 0.280 and −0.690 log mg/kg/day, with predicted rat acute median lethal doses of 2.770 and 2.560 mol/kg, respectively. Furthermore, predicted rat chronic oral lowest-observed-adverse-effect levels ranged from 1.590 log mg/kg body weight per day for CMNPD4578 to 3.400 for CMNPD24293, compared to 1.160 for danuglipron. The investigators emphasized that while dietary consumption of organisms like Sargassum sagamianum offers preliminary rationale, natural consumption history does not confirm that isolated compounds are safe, bioaccessible, or pharmacologically active. Because marine natural products frequently occupy beyond-rule-of-five chemical space, experimental functional profiling remains necessary to differentiate therapeutic activation from benign binding or antagonism.

Subject of Research: Computational screening of marine natural products for GLP-1 receptor agonist activity

Article Title: Computational identification of potential GLP-1 receptor agonists from marine natural products for nutraceutical applications in diabetes and obesity management

Article References: Hajipour, F., Hamedani, O. T., Foroumadi, A., Gomez, N. G., Gonzalez, O. L., & Firoozpour, L. (2026). Computational identification of potential GLP-1 receptor agonists from marine natural products for nutraceutical applications in diabetes and obesity management. Results in Chemistry, 32, Article 104042. https://doi.org/10.1016/j.rechem.2026.104042

Image Credits: AI Generated

DOI: 10.1016/j.rechem.2026.104042

Keywords: GLP-1 receptor, marine natural products, nutraceuticals, molecular docking, molecular dynamics, diabetes, obesity, virtual screening, ADMET profiling, small-molecule agonists, Computational, identification