In-silico and in-vitro hybrid approach to identify glucagon-like peptide-1 receptor agonists from anti-diabetic natural products

Abstract Natural products have contributed immensely towards the treatment of various diseases including diabetes. Here, a database of small molecules from nature possessing antidiabetic properties was analysed and shortlisted according to their structural diversity. Later, those structures were screened by in-silico docking studies to understand their affinity towards glucagon-like peptide-1 (GLP-1) receptor. The selected molecules were isolated and investigated further by integrated in-vitro and in-silico approaches. Alpha-mangostin was found to be suitable due to its excellent docking score and isolation yield. A pancreatic beta cell line was used to test the activity of alpha-mangostin and observed a 3-fold increase in insulin secretion compared to 15 mM glucose control. Further, in-silico molecular dynamics simulations studies have validated its target by showing conformational changes at the functionally active part of the GLP-1 receptor. This screening strategy can be applied to identify pertinent natural products rapidly for various therapeutic targets. Graphical Abstract


Introduction
Peptide based glucagon-like peptide-1 receptor agonists (GLP-1 RA) emerged as the most promising drugs for the treatment of type-2 diabetes mellitus (T2DM) (Meier 2012;Edmonds and Price 2013). However, high cost of these drugs restricts its wide acceptance which can be resolved by identifying small molecule GLP-1 RA (Hansen et al. 2009). Multiple receptor conformations of G protein-coupled receptors (GPCRs) confirms that different ligands can stabilize the different part of the receptor and can bring out the desired therapeutic response via biased agonism. The promising results of GLP-1 RA and biased agonism of the GLP-1 R leads to the efforts in the direction of finding out orally active small GLP-1 R nonpeptide agonists which can bind and activate the receptor (Wootten et al. 2016). In recent years, computer aided drug design (CADD) has helped to identify suitable molecules for different molecular targets (Myrianthopoulos et al. 2018). Molecular Docking and Molecular Dynamics (MD) simulations are extensively employed to explain the molecular mechanism of protein-ligand binding (Ol gac¸et al. 2017). Natural products remain a well-established source of therapeutics and are uniquely positioned in drug lead development (Cragg et al. 1997;Butler 2004;Newman and Cragg 2020). Dictionary of Natural Products (DNP), a subset of the CRC Chemical Database includes detailed information about 281000 natural products (CRC Press 2021). Here, we identified antidiabetic molecules from DNP which were segregated and selected based on their structural class of natural products. We screened these compounds by an in-silico and in-vitro hybrid approach to find a compound having the ability to secrete insulin through GLP-1R.

Results and discussion
Nature has produced numerous scaffolds possessing pharmacological properties, however, a fraction of them are converted into drugs or drug candidates owing to several reasons. This attrition is probably due to the limited understanding of the mechanism of action of these molecules ( (Jung et al., 2006); Hung et al. 2012). During our literature study, it was observed that many molecules from DNP have anti-diabetic properties, however, the mode of action for them is mostly unknown. A list of 251 secondary metabolites possessing hypoglycemic activity from DNP was sorted according to structural diversity such as molecules belonging to flavonoids (flavone, isoflavone, flavanone, flavonolignan, benzofuran), chalcones, diterpenoids (labdane), alkaloids (pyrrolizidine type, berberine type, vinca type), sesquiterpene (eudesmane-type), iridoid glycosides and their aglycons, glycosides (monoterpene, triterpene) tannins (gallotannins), catechols, stilbenoid and xanthones were selected in this study (Table S1). The isolation yield has been given a priority as sustainable production of natural products is a major bottleneck for their development (Khan 2018). The shortlisted 28 molecules were docked with the extracellular domain (ECD) of GLP-1R using Autodock software to check their binding affinity (Table S2). Cochinchinenin C, a non-peptide GLP-1R agonist was used as a positive control (Sha et al. 2017). The binding energy of the control (-5.66 kcal/mole) was considered as a reference value and the selected eight molecules were having scores better than the control. Alpha-mangostin owing xanthone scaffold (Figure 1) was selected for further evaluation due to its desirable binding score and isolation yield (3.32% w/w) (Table S3).
Molecular docking studies of alpha-mangostin and cochinchinenin C revealed that both the compounds showed a similar binding pattern with amino acid residues in the binding pockets. Additionally, alpha-mangostin has an edge due to its delicate balance of both hydrophobic and electrostatic interactions. It also showed polar interactions with the side chains of ASP67 and ARG121 residues at the binding pocket of GLP-1 R which were largely absent in the control ( Figure S1). To confirm insulin secretion capability of alpha-mangostin, the compound was isolated from the powdered pericarp of the Garcinia mangostana Linn. using a series of column chromatography and the structure was confirmed by 1 H, 13 C NMR and MS data ( Ee et al., 2006;Al-Massarani et al. 2013). Rat insulinoma cell line (INS-1E) was used for in-vitro assay and low glucose and high glucose medium were used as a negative control (Merglen et al. 2004). Alpha-mangostin has shown improved insulin secretion in a dose-dependent manner and its highest tested concentration showed a 3-fold increase in insulin secretion as compared to high glucose condition. It was observed that INS-1E cells secrete 4 ng of insulin per 10 5 cells at 0.5 mM of alpha-mangostin while it was 8 ng per 10 5 cells at 10 mM concentration of the tested compound ( Figure S2).
Comparative atomistic simulations of apo and complex structures were performed to affirm the mode of actions of alpha-mangostin in the dynamic environment. Overall, cochinchinenin C and alpha-mangostin behave similarly and the observed root mean square deviation (RMSD) was in the range of 0.2-0.25 nm. Importantly, they stabilized the ECD of GLP-1 R compared with the apo form of the protein. The root mean square fluctuation (RMSF) profile of both complexes showed an additional sharp peak between residue 52-60 as compared to the apo form of the receptor indicating conformational changes. Profile of alpha-mangostin/protein complex displays an additional peak at residue 42-48 ( Figure S3). The results propose the involvement of the first a-helix (31-53) and loop (54-64) in receptor activation mechanism which is supported by the literature (Santiago et al. 2018). Furthermore, the binding patterns of ligand-receptor complexes indicated the presence of 1-2 H-bonds between alpha-mangostin and ECD of GLP-1 R comparable to the standard molecule showing 2-3 H-bonds ( Figure S4). Some of these H-bonds were formed during the conformational changes that occur in the receptor.
We have studied and compared the environment of ligands in terms of residues present within 3.5 Å distance from ligands for both the protein complexes. It was observed that during simulation p-p and hydrophobic interactions were stabilizing these complexes. Moreover, polar interactions such as H-bonds have provided the specificity of the ligand-protein binding ( Figure S5). Loop residues (54-64) were showing conformational changes in the RMSF profile and also these residues are near to the part of the transmembrane helices which is shown to be involved in receptor activation (Santiago et al. 2018). So, it can be speculated that alpha-mangostin would be able to activate the GLP-1 R by providing the above-mentioned conformational changes that possibly trigger the interactions of ECD with transmembrane helices. However, it should be validated by simulations of full-length protein including the transmembrane domain. A published report indicated that alpha-mangostin increases insulin secretion by activating the insulin receptor (IR) and pancreatic duodenal homeobox1 (Pdx1). It causes phosphorylation of PI3K, Akt and ERK cascades and inhibits the phosphorylation of IRS-1 responsible for insulin resistance and pancreas dysfunction (Lee et al. 2018). A recent review mentioned the potential of alphamangostin to treat diabetes mellitus through glucokinase activation (Sharma et al. 2021). However, the anti-diabetic activity of alpha-mangostin is likely to be achieved through targeting multiple pathways and one of which could be the GLP-1 receptor.

Conclusion
An integrated experimental and computational approach was applied to find natural products exhibiting anti-diabetic properties via GLP-1R. The data revealed alpha-mangostin, kraussianone 1, and laserpitin can be potential GLP-1 receptor agonists. Alphamangostin showed 3-fold improvement in insulin secretion as compared to the control in INS-1E cell line. The binding of alpha-mangostin with GLP-1R was justified by the observed conformation changes at the functionally important part of the protein in simulation study. In all, this strategy can be applied to fit natural products in drug discovery pipelines by finding their biological targets of various diseases.