The protein structures of GLUT-4, IR, Akt, and IRS-1 were obtained from the Protein Data Bank (PDB) using the PDB IDs: 7WSN, 4XLV, 1O6L, and 1K3A, respectively. These structures were visualized and analyzed using PyMOL (PyMOL molecular graphics system, version 2.0, Schrödinger, LLC) after removing water molecules. To prepare the target protein input file for docking simulation, polar hydrogen atoms were added to the PDB protein file, and ions, water molecules, subunits, and ligands were removed from the original structure file.

The 3D models of all proteins were acquired and assessed by the use of PROCHECK (Laskowski et al., 1993), which was obtainable from the SAVES server and also utilized by the ProSA server for validation (Wiederstein and Sippl, 2007). The Ramachandran (RC) plot, which compares the overall quality of the structure to a well-refined structure of comparable resolution, is provided by the PROCHECK server, which also assesses the stereochemical quality of the proteins (Carrascoza et al., 2014).

ProSA is a popular tool for examining possible mistakes in 3D protein models that come from theoretical models, protein engineering, or experimental elucidation. The software program provides an energy and Z-score map for the protein. Protein residues’ solvent exposure potential and distance-based methods are used to calculate the energy of the structure. The total model quality of the protein 3D structure is shown by the Z-score. In the ProSA energy plot, energy is displayed as a function of the amino acid sequence position and is utilized to determine the 3D model’s local quality. Negative values signify a stable and sound model, while positive models correspond to incorrect portions (Wiederstein and Sippl, 2007).

The plant-based glycosidic compound (stevioside) is retrieved from PubChem (https://pubchem.ncbi.nlm.nih.gov) using a “similar structure” search (Kim et al., 2016). The physiochemical properties of the selected ligands were calculated, and their structures were loaded onto a workstation for docking studies to assess their anti-diabetic activity.

The Computed Atlas of Surface Topography of proteins (CASTp) was used to characterize the active sites of the GLUT-4, Akt, IR, and IRS-1 proteins (available online: http://sts.bioe.uic.edu/castp/calculation.html accessed on 20 March 2020) and analyzed using Chimera version 1.12. The binding pocket of each protein was characterized by volume, surface area, and the presence of cavities in a solvent (Tian et al., 2018).

ADMET (absorption, distribution, metabolism, excretion, and toxicity) is a crucial aspect of in silico drug design as it helps assess the pharmacokinetic and pharmacodynamic properties of potential drug candidates. A popular computational method for ADMET prediction is pkCSM (http://structure.bioc.cam.ac.uk/pkcsm), which provides a full range of features to assess drug toxicity, distribution, metabolism, and absorption (Abdullahi et al., 2022).

To determine the molecular basis of stevioside specificity for these targets, the 3D structures of the IR, GLUT-4, Akt, and IRS-1 proteins (target) in association with stevioside (ligand) were evaluated using computational ligand–target docking. PyRx and HDOCK were used for molecular docking, with the AutoDock Vina option selected based on the scoring function (Yan et al., 2020; Pawar and Rohane, 2021). Using grid-based atomic affinity potentials, the interaction energy between stevioside and IR, GLUT-4, Akt, and IRS-1 proteins was determined at every step of the docking procedure (Vilar et al., 2008).

The RCSB PDB database was used to obtain the target proteins GLUT-4 (Supplementary Figure S1A), IR (Supplementary Figure S1B), Akt (Supplementary Figure S1C), and IRS-1 (Supplementary Figure S1D). X-ray crystallography was used to confirm the structures of all proteins. The A/B chains of each target protein were sequenced with chain lengths ranging from 328 to 520 amino acids. The samples were identified by their PDB IDs (7WSN, 4XLV, 1O6L, and 1K3A) and had the maximum (3,567) and the minimum number of atoms (2,548), respectively. The highest resolution was 3.31 and the lowest was 1.60 (Supplementary Table S1). In in silico protein, resolution refers to the clarity of atomic distances between amino acid residues when visualized in software. A higher resolution indicates a clearer molecular image (Qian et al., 2007). Since the target protein had a secondary structure with folds and consisted of one or more chains, its 3D structure was retrieved from a tertiary database (Rahman et al., 2020). The three protein structures were subjected to solvent sterilization, and the original ligands were modified using PyMol software to optimize the binding energy during molecular docking simulations (Baran et al., 2017; Supplementary Figure S1.

The Ramachandran plot of GLUT-4, Akt, IR, and IRS-1 proteins was obtained from the SAVES server (Supplementary Figure S2), while their plot statistics are presented in Table 1. The plot statistics of GLUT-4 indicated that 91.2% of amino acid residues were present in the most preferred region and 8.8% occurred in the extra allowed region according to the plot, while none of the amino acid residues was found in the regions that were liberally allowed or banned (Supplementary Figure S1A). The plot statistics of Akt protein indicated that 90.0% of the residues of amino acids were found in the most favored region, 7.6% were in the extra allowed region, 1.0% was in the generously allowed zone, and 1.4% was in the prohibited region (Supplementary Figure S2B). The plot data of IR protein revealed that 91.2% of amino acid residues were found in the most favorite region, 8.8% were in the extra permitted zone, and none were found in the regions that either liberally permitted or outlawed (Supplementary Figure S2C). Analysis of IRS-1 amino acid distribution using the Ramachandran plot revealed that 93.5% residues resided in the most favored regions while 6.2% were found in the additionally allowed regions. Only 0.4% of the residues occupied the generously allowed regions, and none were located in the disallowed regions (Supplementary Figure S2D). Hence, the created 3D models are good in terms of stereochemical quality.

The GLUT-4, Akt, IR, and IRS-1 proteins were submitted to the ProSA web server, and a Z-score of GLUT-4 protein was obtained at −6.98 (Supplementary Figure S3A), Z-score of Akt protein was obtained at −7.98 (Supplementary Figure S3B), the Z-score of IR protein was obtained at −8.74 (Supplementary Figure S3C), and the Z-score of IRS-1 protein obtained at −8.46 (Supplementary Figure S3D). The plot of the Z-scores of all protein chains in PDB as calculated via NMR spectroscopy (dark blue region) and X-ray crystallography (light blue region) is presented in Supplementary Figure S3. The energy profiles of GLUT-4 (Supplementary Figure S4A), Akt (Supplementary Figure S4B, IR (Supplementary Figure S4C), and IRS-1 (Supplementary Figure S4D) proteins obtained using the ProSA web server are shown in Supplementary Figure S4. The ProSA energy map derived for each model demonstrates that the maximum residues were located within the negative energy zone, indicating that each model is stable.

The ligand (stevioside) obtained from PubChem consisted of PubChem CID 442089. The molecular weight of stevioside was 804.9 g/mol, the molecular formula of stevioside was C₃₈H₆₀O₁₈, and the 2D and 3D structures of stevioside are shown in Figure 1.

The CASTp server was used to identify the amino acid residues that lined the binding site positions of GLUT-4, IR, Akt, and IRS-1 proteins. Additionally, three major binding pockets for each protein were predicted using CASTp. These pockets were subsequently analyzed using Chimera 1.12, respectively. Furthermore, binding pockets with a small surface area and volume were excluded from consideration as potential ligand-binding sites for virtual screening.

According to the CASTp server predictions, the GLUT-4 active site positions residues within pocket 1, the largest and most surface-exposed pocket on the GLUT-4 protein. This pocket was also lined with several residues that were known to be crucial for glucose transport, including Thr21, Leu24, Phe38, Ile42, Phe88, and Ser95. The GLUT-4 pocket 1 comprises 104 amino acid residues, including Thr21, Leu24, Ser35, Phe38, Ile42, Phe88, Ser95, Ser96, Phe97, Ile99, Gly100, Ile101, Ser103, Gln104, Arg108, Gly150, Thr152, Ser153, Gly154, Val156, Pro157, Met158, Val160, Gly161, Glul62, Ile163, Ala164, Pro165, Thr166, His167, Leu168, Arg169, Gly170, Leul72, Gly173, Thr174, Asn176, Gln177, Ile180, Val181, Ile184, Pro227, Arg228, Ile233, Leu244, Leu247, Thr248, Trp250, Val253, Val256, Glu259, Leu260 Asp262, Glu263, Lys266, Leu267, Thr337, Ser340, Val341, Val344, Glu345, Arg349, Phe395, Glu396, Gly400, Pro401, Trp404, Phe405, Asn427, Trp428, Asn431, Thr471, Thr471, Arg472, Gly473, Arg474, Thr475, Phe476, Asp477, and Ile479. Using CASTp software with default parameters, 50 active site positions were identified and analyzed within the GLUT-4 protein structure. All pockets were described to determine the residues surrounding them within a probe radius of 1.4 Ǻ. Of these, the biggest active site has a zone of 2158.359 Ǻ² and a volume of 2765.094 Ǻ³ during analysis. The position of the largest active site in the protein, which ranges from amino acids 21 to 479, is shown in Supplementary Figure S5 and is highlighted in green.

During the study, pockets 2 and 3 were accurately predicted by CASTp as potential binding sites, but their likelihood of being active sites was lower than that of pocket 1. Pocket 2 was smaller and less exposed to the surface of the protein than pocket 1. It consists of 17 amino acid residues, namely, Gln49, Pro74, Thr78, Tyr309, Ser312, Ile313, Thr316, Leu371, Phe438, Gln439, Tyr440, Ala442, Glu443, Gly446, Pro447, Val449, and Phe450. The predicted active site of pocket 2 has a surface area of 152.196 Ų and a volume of 73.774 Ǻ³.

Among the pockets on the GLUT-4 protein, pocket 3 was the smallest and least exposed to the surface. The residues that consistently form pocket 3 were Thr283, His284, Pro287, Leu410, Phe411, Ser412, Pro415, Phe476, Ile479, Ser480, and Phe483. Pocket 3 has an active site with an area of 83.119 Ų and a volume of 32.407 Ǻ³. The three pockets predicted by the CASTp tool for the GLUT-4 protein are shown in Figure 2 and Table 2. Pocket 1 is represented in red, pocket 2 in orange, and pocket 3 in yellow.

Based on the predictions made by the CASTp server for the Akt protein, pocket 1 of the Akt protein stands out as the largest and most surface-exposed pocket, and it has been identified as the active site. Akt protein pocket 1 includes 49 amino acid residues, namely, Leu158, Gly159, Lys160, Gly161, Thr162, Phe163, Gly164, Lys165, Val166, Ala179, Lys181, Leu183, Arg184, Val187, Ile188, Lys191, Glu193, His196, Thr197, Glu200, Leu204, Thr213, Phe227, Met229, Glu230, Tyr231, Ala232, Glu236, Phe239, Asp275, Lys277, Glu279, Asn280, Met282, Thr292, Asp293, Phe294, Gly295, Leu296, Tyr438, Phe439, Asp440, Phe443, Arg6, Thr7, Thr8, Ser9, Phe10, and Ala11. A total of 22 active sites were evaluated in the structure through CASTp software with ideal parameters. The biggest active site has an area of 579.259 Ų and a volume of 355.567 Ǻ³. All pockets were characterized to determine their residues around the probe radius of 1.4 Ǻ. The protein’s biggest active site is located between amino acids 158 and 443, as shown in Supplementary Figure S6, in green color.

The CASTp server also predicted pockets 2 and 3 comparatively lesser than pocket 1 and also predicted that pocket 1 is more exposed than pocket 2 and consists of 16 amino acid residues such as Asp303, Gly304, Ala305, Thr306, Asp326, Tyr327, Gly328, Arg329, Ala330, Pro389, Lys390, Gly394, Gly395, Gly396, Pro397, and Asp399. The biggest active site has an area of 174.712 Ų and a volume of 174.712 Ǻ³.

Pocket 3 is the smallest and least surface-exposed pocket on the Akt protein; pocket 3 of the Akt protein consists of 13 residues such as Arg176, Tyr177, Tyr178, Thr213, Glu230, Tyr231, Ala232, Asn233, Asp284, Lys285, Lys290, Glu433, and Trp479. The biggest active site has an area of 121.624 Ų and a volume of 81.980 Ǻ³. The three binding pockets were predicted for the Akt protein using the CASTp tool (Table 3; Figure 3). Furthermore, pocket 1 is highlighted in red, pocket 2 in orange, and pocket 3 in yellow.

The CASTp server predicted the IR protein active sites (Table 4); pocket 1 was the largest and most surface-exposed pocket on the IR protein. The IR protein of pocket 1 consists of 53 amino acid residues, namely, Arg1000, Glu1001, Leu1002, Gly1003, Gln1004, Gly1005, Ser1006, Phe1007, Gly1008, Val1010, Glu1012, Ala1028, Lys1030, Thr1031, Val1032, Arg1039, Glu1040, Glu1043, Phe1044, Glu1047, Met1051, Val1060, Met1076, Glu1077, Leu1078, Met1079, Ala1080, His1081, Gly1082, Asp1083, Ser1086, Tyr1087, Ser1090, Asn1097, Pro1099, Arg1101, Arg1131, Asp1132, Arg1136, Asn1137, Met1139, Phe1144, Gly1149, Asp1150, Phe1151, Gly1152, Met1153, Arg1155, Lys1165, Gly1169, Leu1170, Leu1171, and Pro1172. Using the optimal settings provided by CASTp software, a total of 24 active sites within the structure were assessed. All the pockets were defined to determine their residues approximately at the probe radius of 1.4 Ǻ, and of them, the largest active site has a volume of 686.806 Ǻ³ and an area of 762.651 Ų, respectively. The biggest active site location in the protein was between 1000 and 1172 amino acids and highlighted in green (Supplementary Figure S7).

The prediction of the CASTp server proved that pocket 2 and 3 lesser extent than pocket 1. However, it has been observed that pocket 2 surface exposure was comparatively lower than pocket 1 and contains 17 amino acid residues such as Arg1131, Lys1165, Gly1167, Lys1168, Gly1169, Leu1170, Leu1171, Pro1172, Val1173, Met1176, Ser1180, Leu1181, Gly1184, Phe1186, Asn1215, Glu1216, and Leu1219. The biggest active site has an area of 215.893 Ų and a volume of 254.117 Ǻ³.

Pocket 3 is the least surface-exposed pocket on the IR protein; pocket 3 of the IR protein consists of 18 residues, namely, Ile1019, Glu1022, Thr1025, Arg1026, Cys1056, His1057, Val1059, Arg1061, Leu1063, Glu1077, Leu1078, Met1079, Ala1080, Ala1141, His1142, Asp1143, Thr1145, and Lys1147. The biggest active site has an area of 193.749 Ų and a volume of 198.010 Ǻ³. All three pockets were predicted by the CASTp tool of the IR protein (Figure 4); the red color represented pocket 1, the orange color represented pocket 2, and the yellow color represented pocket 3.

The CASTp server predicts the IRS-1 protein active sites (Table 5); pocket 1 was the largest and most surface-exposed pocket on the IRS-1 protein. The IRS-1 protein of pocket 1 consists of 59 amino acid residues, namely, Arg973, Leu975, Gly976, Gln977, Gly978, Ser979, Phe980, Gly981, Met982, Val983, Glu985, Arg999, Ala1001, Lys1003, Thr1004, Glu1013, Glu1016, Phe1017, Asn1019, Glu1020, Ala1021, Val1023, Met1024, Val1033, Leu1035, Val1047, Met1049, Glu1050, Leu1051, Met1052, Thr1053, Gly1055, Asp1056, Val1102, His1103, Arg1104, Asp1105, Arg1109, Asn1110, Cys1111, Met1112, Gly1122, Asp1123, Phe1124, Gly1125, Met1126, Thr1127, Arg1128, Ile1130, Thr1133, Asp1134, Arg1137, Lys1138, Gly1142, Leu1143, Leu1144, Val1158, Phe1159, and Thr1160. Using the optimal settings provided by CASTp software, a total of 35 active sites within the structure were assessed. All the pockets were defined to determine their residues approximately at the probe radius of 1.4 Ǻ, and of them, the largest active site has a volume of 1073.500 Ǻ³ and an area of 1047.925 Ų, respectively. The biggest active site location in the protein was between 973 and 1160 amino acids and highlighted in green (Supplementary Figure S8).

Although the CASTp server identifies pockets 2 and 3, these are predicted to be of lesser significance compared to pocket 1; pocket 2 consists of 20 amino acid residues, namely, His1030, His1031, Leu1057, Tyr1060, Leu1061, Leu1064, Pro1077, Lys1081, Gln1084, Met1085, Glu1088, Val1113, Asp1116, Phe1117, Thr1118, Val1119, Glu1241, Gly1243, Phe1244, and Val1247. The biggest active site has an area of 203.855 Ų and a volume of 136.855 Ǻ³.

Pocket 3 is the least surface-exposed pocket on the IRS-1 protein; the pocket 3 of the IRS-1 protein consists of 19 residues, namely, Trp962, Val992, Glu995, Thr998, Arg999, Met1024, Lys1025, Phe1027, Asn1028, Cys1029, His1030, Val1032, Val1033, Arg1034, Leu1035, Leu1036, Glu1050, Lys1120, and Phe1124. The biggest active site has an area of 152.167 Ų and a volume of 116.874 Ǻ³. All three pockets were predicted by the CASTp tool of the IRS-1 protein (Figure 5); the red color represented pocket 1, the orange color represented pocket 2, and the yellow color represented pocket 3.

The ADMET (absorption, distribution, metabolism, excretion, and toxicity) values for the compound under investigation have been computed using the web server pkCSM. The pharmacokinetic properties of the molecule should be taken into consideration as important variables throughout the virtual screening process rather than just focusing on enhancing the binding affinity and increasing specificity. Analysis has been done on the current status of theoretical models for predicting characteristics of drug absorption, such as Caco-2 permeability, intestinal absorption, and blood–brain partitioning. The importance of predictive and physical and chemical characteristic in evaluating passive drug absorption was emphasized.

The obtaining result of water solubility was found at −2.733, which indicates they are highly soluble in liquid medium, and the value of Caco-2 permeability was obtained at −0.366. The intestinal absorption (human) value of stevioside is 0, which means that it was not absorbed to blood through the intestine. The VDss (human) level range was −0.305 log L/kg, fraction unbound (human) was 0.379 Fu, BBB permeability score was −2.003 log BB, and CNS permeability value was −6.742 log PS. Drugs do not display inhibition or substrate against the CYP3A4 substrate, CYP2D6 substrate, CYP2C19 inhibitor, CYP1A2 inhibitor, CYP2C9 inhibitor, CYP3A4 inhibitor, and CYP2D6 inhibitor, and these enzymes were mostly present in the liver. Total clearance of stevioside was found at 0.746, and no renal OCT2 substrate was present, respectively (Supplementary Table S2).

The study of medication toxicity to aquatic and non-aquatic species is characterized by its effects on these two groups of animals. The nature of drugs is chemical. Therefore, during production at a pharmaceutical or drug manufacturer’s factory, they may degrade and combine, which could be hazardous to the ecosystem. In addition, after being administered, they might have harmful or cancerous effects on human body, and this makes the current study very significant. The aforementioned medications do not cause hepatotoxicity, AMES toxicity, or skin sensitivity; nevertheless, ordinary metformin hydrochloride may cause these side effects. The highest level of oral rat acute toxicity was 2.591 mol/kg/day, the maximum level of oral rat chronic toxicity was 5.552 mg/kg/day, and the maximum tolerable dose was −0914 mg/kg/day. All of them suggest that these medications have improved pharmacokinetic and physiochemical characteristics (Supplementary Table S3).

The docking of the GLUT-4 protein with stevioside is predicted by the HDOCK server with a docking score of −327.28, with a confidence score of 0.9720 and ligand RMSD of 175.53 (Å) (Figure 6B). Here, the GLUT-4 docked position with stevioside reliably shows the stevioside binding position to the GLUT-4 protein (Figure 6A). Stevioside, in combination with GLUT-4, revealed a greater binding energy of −9.8 kcal/mol (Table 6) using PyRx software, with four salt bridges, and developed eight hydrogen bond interactions with amino acids, namely, TYR-58, ILE-59, ILE-61, GLU-86, LYS-87, VAL-88, ASN-92, and ARG-111, possessing hydrogen bond distances of 3.26 Å, 1.91 Å, 1.87 Å, 2.03 Å, 2.81 Å, 2.09 Å, 1.92 Å, and 1.88 Å, respectively.

The docking of the Akt protein with stevioside by the HDOCK server with the docking score was recorded as −212.94, along with a confidence score of 0.7788 and ligand RMSD of 130.23 (Å) (Figure 7B). Furthermore, the Akt protein docked region with stevioside demonstrated the Akt–stevioside-binding location (Figure 7A). Stevioside with Akt using PyRx software revealed higher binding energy at −6.7 kcal/mol, with one salt bridge, and formed five hydrogen bond interactions with amino acids (GLY-175, ARG-176, TYR-177, GLU-230, and TRP-479) with the hydrogen bond lengths of 3.09 Å, 2.98 Å, 1.77 Å, 2.22 Å, and 2.03 Å, respectively (Table 6).

The docking of the IR protein with stevioside by the HDOCK server with the docking score was recorded as −244.64, along with a confidence score of 0.8691 and ligand RMSD of 74.83(Å) (Figure 8B). The IR protein docked site with stevioside accurately indicated the stevioside binding region to the IR protein (Figure 8A). Utilizing IR protein analysis using PyRx software, stevioside revealed a greater binding energy of −8 kcal/mol, with three salt bridges, and established 10 hydrogen bond connections with amino acids, namely, CYS-1056, HIS-1057, GLU-1077, GLN-1111, GLU-1115, HIS-1142, ASP-1143, LYS-1147, HIS-1268, and SER-1270 with hydrogen bond distances of 2.49 Å, 2.33 Å, 2.02 Å, 3.18 Å, 3.30 Å, 2.77 Å, 1.98 Å, 3.20 Å, 2.77 Å, and 2.50 Å correspondingly (Table 6).