Abstract Introduction Diabetes mellitus (DM) is a common endocrine disease resulting from interactions between genetic and environmental factors. Type II DM (T2DM) accounts for approximately 90% of all DM cases. Current medicines used in the treatment of DM have some adverse or undesirable effects on patients, necessitating the use of alternative medications. Methods To overcome the low bioavailability of plant metabolites, all entities were first screened through pharmacokinetic, network pharmacology, and molecular docking predictions. Experiments were further conducted on a combination of antidiabetic phytoactive molecules (rosmarinic acid, RA; luteolin, Lut; resveratrol, RS), along with in vitro evaluation (α-amylase inhibition assay) and diabetic mice tests (oral glucose tolerance test, OGTT; oral starch tolerance test, OSTT) for maximal responses to validate starch digestion and glucose absorption while facilitating insulin sensitivity. Results The results revealed that the combination of metabolites achieved all required criteria, including ADMET, drug likeness, and Lipinski rule. To determine the mechanisms underlying diabetic hyperglycemia and T2DM treatments, network pharmacology was used for regulatory network, PPI network, GO, and KEGG enrichment analyses. Furthermore, the combined metabolites showed adequate in silico predictions (α-amylase, α-glucosidase, and pancreatic lipase for improving starch digestion; SGLT-2, AMPK, glucokinase, aldose reductase, acetylcholinesterase, and acetylcholine M2 receptor for mediating glucose absorption; GLP-1R, DPP-IV, and PPAR-γ for regulating insulin sensitivity), in vitro α-amylase inhibition, and in vivo efficacy (OSTT versus acarbose; OGTT versus metformin and insulin) as nutraceuticals against T2DM. Discussion The results demonstrate that the combination of RA, Lut, and RS could be exploited for multitarget therapy as prospective antihyperglycemic phytopharmaceuticals that hinder starch digestion and glucose absorption while facilitating insulin sensitivity. Keywords: antihyperglycemia, combinatory chemistry, phytopharmaceutical, polyphenol, in silico, multitarget 1 Introduction Diabetes mellitus (DM) is a widespread metabolic disease with a rapidly growing global population; it is characterized by chronic hyperglycemia resulting from inadequate insulin secretion and/or insulin action ([45]Ozougwu et al., 2013; [46]Thompson and Kanamarlapudi, 2013) and is attributable to the interactions between genetic and environmental factors. Type II DM (T2DM) is also designated as non-insulin-dependent DM (NIDDM) and is often known to occur as a consequence of excess blood glucose (hyperglycemia) caused by dysfunctional β-cells and insulin resistance. T2DM is the major form of diabetes, as an estimated 90% of DM patients are diagnosed with this form. The global incidence of DM continues to increase, and it is expected that there will be more than 590 million patients with this disorder by 2035 ([47]Ozougwu et al., 2013; [48]Guariguata et al., 2014). It is known that DM can cause several other complications, such as cardiovascular diseases, ischemic heart disease, obesity, peripheral vascular disease, stroke, retinopathy, neuropathy, nephropathy, diabetic foot ulcers, and a variety of heterogeneous diseases, based on abnormalities in the relative metabolic pathways ([49]Boulton et al., 2005; [50]Punthakee et al., 2018; [51]Swilam et al., 2022), with T2DM causing over 95% of these comorbidities ([52]Punthakee et al., 2018). Although various types of oral hyperglycemic drugs (e.g., acarbose, miglitol, voglibose, metformin, and sulfonylureas) are available for the treatment of diabetes ([53]Spínola et al., 2019), these approved synthetic drugs cannot be applied for long-term glycemic control or reversal of comorbidity progression; their side effects or adverse reactions are also usually downplayed, coupled with the cost, which makes their access especially challenging for the low-income population in developing and underdeveloped countries. Therefore, there is an urgency for developing alternative therapeutics with limited associated shortcomings ([54]Takayanagi et al., 2011). This is a major health issue in today’s society and is significantly linked with the socioeconomic difficulties experienced worldwide. The documentation of natural products that moderate blood glucose levels can possibly accelerate exploitation of mild interventions like folk or herbal medicines as well as functional foods in the treatment of chronic diseases, including diabetes. In comparison, active metabolites from synthetic sources, extracts, or natural products from natural sources such as botanical drugs and plant metabolites that are commonly safe, easy to locate, easily accessible, and reasonably inexpensive, with low incidence of adverse effects must be prioritized to facilitate the surging diversion into phytomedicine ([55]Wais et al., 2012; [56]Bizzarri et al., 2020). Polyphenols are a large class of metabolites deemed to have multiple biological properties, such as antioxidant, cytotoxic, anti-inflammatory, antihypertensive, and antidiabetic functions ([57]Rana et al., 2022). For instance, curcumin has been widely shown to be a popular antidiabetic, but its limitations like poor absorption, rapid metabolism and elimination, and low concentrations in the plasma and target tissue are considered to be obstacles in treatment ([58]Pivari et al., 2019). It has been noted that monotarget therapy with natural polyphenols failed to manage blood glucose levels and other comorbidities; therefore, the combined use (polytherapy) of polyphenols has become a common practice ([59]Boonrueng et al., 2022). The flavone luteolin (3′, 4′, 5, 7-tetrahydroxyflavone, Lut) is a naturally occurring secondary metabolite present in various plants, such as celery, chrysanthemum flowers, sweet bell peppers, carrots, onion leaves, broccoli, and parsley. It has anti-inflammatory and antioxidant activities, and it increases glucose metabolism by potentiating insulin sensitivity while enhancing β-cell function and mass during a hyperglycemic clamp ([60]Daily et al., 2021). However, one study revealed that Lut has limited bioavailability that consequently affects its biological properties and efficacy ([61]Taheri et al., 2021). Resveratrol (3,5,4′-trihydroxy-trans-stilbene, RS) is a naturally occurring polyphenolic stilbene compound found in more than 70 plant species and their products, such as grapes, peanuts, mulberries, bilberries, blueberries, cranberries, and spruce, as well as other plant roots, leaves, and fruits in response to biotic and abiotic factors. It has been shown to have numerous biological activities, such as antitumor, antioxidant, antiviral, and phytoestrogenic properties. The effective use of RS is restricted by its poor solubility, photosensitivity, and rapid metabolism, which strongly undermine its bioavailability and bioactivity ([62]Santos et al., 2019). Rosmarinic acid (ester of caffeic acid, RA) is a secondary metabolite and polyphenol present in many culinary plants, such as rosemary, mint, basil, and perilla, that presents various well-documented biological effects, such as anti-inflammatory, antioxidant, antidiabetic, and antitumor properties. Despite the high therapeutic potential of RA, its intrinsic properties of poor water solubility and low bioavailability have limited its translation to clinical settings ([63]Chung et al., 2020). Based on the low bioavailability and bioactivity of the abovementioned compounds, a putative reconstitution (combinatory chemistry) of RA, Lut, and RS could enhance the bioavailability and efficiency to achieve maximal response against the limitations. This combination is worth investigating as it can potentially help lower blood glucose levels in diabetes. In recent times, researchers have been keen on developing methods to induce antihyperglycemia by eliminating starch digestion, promoting metabolism (glucose uptake, absorption, and utilization), and triggering insulin sensitivity; approaches have been employed to explore good metabolite candidates and further determine the best combination ratio for phytotherapeutics. In this study, the best or gold combination of known antidiabetic metabolites, specially RA, Lut, and RS, was tailored on the basis of IC[50], solubility, and network pharmacology studies. Then, molecular docking of these selected metabolites was examined toward diabetic multitarget proteins, such as α-amylase, α-glucosidase, pancreatic lipase, SGLT-2, AMPK, glucokinase, aldose reductase, acetylcholinesterase, acetylcholine M2 receptor, GLP-1R, DPP-IV, and PPAR-γ, which are extensively considered therapeutic targets in clinical treatments for maximal response to reducing blood glucose levels. Furthermore, in vitro α-amylase inhibition assays with the lone metabolites and combination as well as in vivo tests (oral starch tolerance test, OSTT, and oral glucose tolerance test, OGTT) in diet-induced obese diabetic mice were conducted to verify the combination for putative antidiabetic effects. Finally, the molecular mechanism of the combination was predicted through a network pharmacology study. 2 Materials and methods 2.1 Materials Purified Lut (98% purity), RS (98%), and RA (98%) were obtained from Shaanxi Dongshuo Biotechnology Co., Ltd. (Shaanxi, China). 2.2 Pharmacokinetics and ADME/toxicity profiling The pharmacokinetic properties of the materials, such as ADMET behaviors of the ligands in the human body, were screened using the SwissADME ([64]http://www.swissadme.ch/index.php) and admetSAR prediction tool webserver ([65]http://lmmd.ecust.edu.cn/admetsar2). This step is significant for identifying the drug likeness, medicinal chemistry, lead likeness, and toxicity potential of new candidate drugs, phytochemicals, food additives, and industrial chemicals; it is also a prerequisite for establishing valid complementary methods before in vivo/in vitro analyses ([66]Cheng et al., 2012; [67]Sharma et al., 2020; [68]Suchitra et al., 2020). 2.3 Exploring the potential targets of RA, Lut, and RS The Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) ([69]https://tcmsp-e.com/) ([70]Liu et al., 2013) were used to obtain the potential targets of RA, Lut, and RS (RLR). The target names were standardized using the UniProt database ([71]https://www.uniprot.org/) for the status criterion of “Reviewed” and organism category of “Human”. 2.4 Acquisition of diabetic hyperglycemia and T2DM related targets as well as construction of Venn diagrams Diabetic hyperglycemia targets were screened using the DrugBank database; further, the RCSB database was searched for the target protein database (PDB) using qualifiers such as “Homo sapiens,” “X-ray,” and “no mutation.” These targets were compared with the RLR targets, and Venn diagrams were constructed to identify the targets related to both diabetic hyperglycemia and RLR. T2DM-related targets were obtained from four databases, namely, the Comparative Toxicogenomics Database (CTD) ([72]http://ctdbase.org/), GeneCards ([73]https://www.genecards.org/), OMIM ([74]https://www.omim.org), and DisGeNET ([75]https://www.disgenet.org/). Threshold scores were simultaneously set in the CTD, GeneCards, and DisGeNET to filter the targets; all targets from the four databases were merged, and duplicate values were deleted to obtain the corresponding T2DM targets. Venn diagrams were also constructed for the T2DM targets using the tools in Hiplot Pro ([76]https://hiplot.com.cn/), a comprehensive web service for biomedical data analysis and visualization. The common targets between RLR and T2DM (i.e., T2DM-related targets treated by RLR) were obtained from the Venn diagrams. 2.5 Construction of regulatory networks between RLR and the intersecting targets Cytoscape (version 3.9.1; [77]https://cytoscape.org/) ([78]Shannon et al., 2003) was used to visualize the interactions of RLR with diabetic hyperglycemia and T2DM as a regulatory network. The intersecting targets and different types of metabolites were displayed using various shapes. The degree value represents the number of interactions generated by a node, and a higher degree value denotes a more significant status in the network. These higher degree values were represented using deeper colors. 2.6 Network analysis of protein–protein interactions (PPIs) of the intersecting targets The intersecting targets were imported into the STRING 11.0 platform and visualized using Cytoscape 3.9.1 software to construct the PPI network. The CytoHubba plugin was applied to analyze and obtain the top-10 hub genes ranked by degree value ([79]Chin et al., 2014). The MCODE plugin was then used to analyze the most significant module and determine the top-10 hub targets ranked on the basis of the MCODE score in the module ([80]Saito et al., 2012). The K-means clustering algorithm was used with the MCODE plugin, and the module with the highest score was considered to be the most significant module. Nodes with higher MCODE scores were assigned more significant statuses in the general PPI network. 2.7 GO and KEGG pathway enrichment analyses The R packages “clusterProfiler,” “org.Hs.eg.db,” “ggplot2,” and “DOSE” were used to perform gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analyses, and the results were visually displayed using R 4.2.3 software. The statistical significance for the enrichment analysis was an adjusted p value ≤0.05. Three aspects of the GO analysis, namely, molecular function (MF), biological process (BP), and cellular component (CC), which were most significantly associated with the top-10 GO functional terms were selected in each field. Correspondingly, the KEGG pathway enrichment analysis was conducted to investigate the intersecting genes, and results were obtained for the top-20 pathways. 2.8 Molecular docking All structures of the tested target proteins were first downloaded from the RCSB repository with PDB numbers ([81]https://www.rcsb.org/), namely, α-amylase (5U3A; 4GQR), α-glucosidase (3TOP; 3L4Y), pancreatic lipase (1LPA), SGLT-2 (7VSI), AMPK (6C9F), glucokinase (3A0I), aldose reductase (1IEI), acetylcholinesterase (4BDT), acetylcholine M2 receptor (4MQT), GLP-1R (7C2E), DPP-IV (4N8D), and PPAR-γ (1WM0; 4CI5). The criteria for target selection were as follows: “H. sapiens” and “no mutation.” The key residues used as constituents of the putative binding pockets were found from the corresponding literature in the RCSB repository based on PDB numbers. For each protein structure, the selected key residues were applied to constitute the putative binding pocket ([82]Supplementary Table S1 footnotes). Originally, the ligands and water molecules were removed from the protein crystal, to which hydrogen and the desired electric charge were added using Discovery Studio 2019. Next, the protein structures were subjected to energy minimization before docking. In the Simulation | Change Forcefield tools, CHARMM36 was one of the versions applied to minimize the energy. Second, the test chemicals (ligands) were sourced from PubChem ([83]https://pubchem.ncbi.nlm.nih.gov/; RA (5281792), Lut (5280445), and RS (445154)) and were also charged with hydrogen atoms. Finally, molecular simulations were performed, whose results showed that the ligand was located inside the grid box of the receptor. The lowest binding energy (kcal/mol) was calculated using AutoDock Vina 1.2.0, and the 2D and 3D images were visualized and analyzed using Discovery Studio 2019. 2.9 α-Amylase activity assay The α-amylase inhibitory evaluations of the selected metabolites (RA, Lut, and RS) were performed using the 3,5 dinitrosalicylic acid (DNSA) colorimetric assay ([84]Chen T.-H. et al., 2021; [85]Chen S.-P. et al., 2021). The α-amylase activity was quantified based on the reduction of sugars from the breakdown of starch, using DNSA dissolved in 2 M NaOH/5.3 M Na^+-K^+-tartaric acid, as previously described ([86]Luyen et al., 2018). In brief, the substrate solution was prepared by dissolving starch (4 mg/mL) in 20 mM of phosphate-buffered saline (PBS 1×, pH 6.8). Approximately 50 μL of each sample solution (various concentrations in each of the test chemicals) and 100 μL of 16 unit/mL α-amylase were added to 1.5-mL Eppendorf tubes and incubated for 10 min. Next, 100 μL of the substrate solution was added to each mixture and incubated at 37°C for an additional 30 min. Finally, 50 μL of the DNS reagent was added to each mixture and boiled for 10 min. The optical densities of the samples were detected at 540 nm (OD[540]) using the Infinite^® M200 PRO multimode microplate reader (Tecan, Switzerland) with acarbose as the positive control. 2.10 Diet-induced diabetic model preparation 2.10.1 Animal care and diet-induced obesity induction Sixty male ICR mice (6 weeks old) were obtained from Wu’s Laboratory Animals (Fujian, China) and housed in controlled environmental conditions at room temperature (22°C ± 2°C) and humidity (50% ± 10%). A 12/12 h light/dark (6 a.m. to 6 p.m.) cycle was maintained throughout the study period. The mice had free access to food as well as tap water and were maintained on a standard laboratory diet (Rodent feed 1022, BEIJING HFK BIOSCIENCE Co., Ltd., Beijing, China). The animal experiments were approved by the Xiamen Medical College Animal Ethics Committee (SYXK, 2018-0010) and were conducted in accordance with the “Guide for the Care and Use of Laboratory Animals” of Xiamen Medical College. 2.10.2 Insulin intolerance test (IGT) and determination of diabetic mice The protocols for the diet-induced obese (DIO) mice and glucose tolerance (GT) were followed as per our previous study ([87]Huang et al., 2019a). First, DIO mice were fasted for 10 h prior to oral gavage of 4 g/kg bodyweight (Bwt) glucose solution. At the beginning of the test, the fasting blood glucose levels of the mice were measured from tail-vein samples using an Accu-Chek blood glucose analyzer (Hoffmann-La Roche AG, Basel, Switzerland). 2.10.3 OGTT The OGTT of the diabetic mice were performed as per the procedures described in a previous report ([88]Huang et al., 2019a), with slight modifications. For this test, a total of 48 DIO diabetic mice were divided into eight groups (n = 6) as control (water), positive control (125 mg/kg Bwt of metformin dissolved in water), Lut (12.5 mg/kg Bwt dissolved in edible oil), RA (66.0 mg/kg Bwt dissolved in edible oil), RS (91.0 mg/kg Bwt dissolved in edible oil), RLR mixture (RA: Lut: RS = 5:1:4; fasting blood glucose >7 mmol/L; 100 mg/kg Bwt), RLR-H (fasting blood glucose >10 mmol/L; 100 mg/kg Bwt), and insulin (0.2 U/kg intraperitoneally). Then, 4 g/kg Bwt of glucose solution (fresh preparation, dissolved in water) was provided during the test at various times (30, 60, 90, and 120 min). The blood glucose levels of the mice were lastly measured from tail-vein samples using a blood analyzer. 2.10.4 OSTT For this test, a total of 24 DIO diabetic mice were divided into four groups (n = 6) as control (water only), positive control (10 mg/kg acarbose dissolved in water), RLR mixture (fasting blood glucose >7 mmol/L; 100 mg/kg Bwt), and RLR-H (fasting blood glucose >10 mmol/L; 100 mg/kg Bwt). The protocols for the OSTT were the same as those for OGTT but with three modifications: a) 3 g/kg Bwt of corn starch solution (fresh preparation, dissolved in water) was administered; b) the positive control used was acarbose (10 mg/kg); c) the test time was extended to 180 min. The blood glucose levels of the mice were again measured from tail-vein samples using a blood analyzer. 2.11 Statistical analysis The data were expressed in terms of means ± SEM for the in vivo results and means ± SD for all other cases. Statistical comparisons of the results were conducted using one-way analysis of variance (ANOVA). The means in each column followed by different letters indicate the significant differences at p < 0.05 based on the post hoc Tukey’s test. 3 Results 3.1 ADMET The in silico ADMET profiling of metabolites characteristically illustrates the latent to be profitable interactions between the potential drug candidates with specific protein targets for successful drug discovery and development. In this study, the deviations of RA, Lut, and RS are all within acceptable ranges. It was found that two of the chemicals, Lut and RA, behaved as HOB+, while RS behaved as HOB-. Therefore, the drug candidates reflect this blood–brain barrier (BBB) crossing with a topological polar surface area (TPSA) values as follows: RS <60.7 Å[2]; RA, Lut >90 Å[2]. All of the WLogP values were also found to be less than 6 ([89]Ishola et al., 2021). Thus, the three phytochemical candidates examined in this study (RA, Lut, and RS) can cause transversion of the BBB; however, two of the candidates exhibit carcinogenic properties. The BBB is essential for restricting in/outflux to the CNS microenvironment to ensure adequate neuronal function ([90]Ishola et al., 2020). Although RA, Lut, and RS possess considerable binding affinities to the three antidiabetic protein targets, their significant BBB transversion could be employed to develop drugs against neurodegenerative diseases. Caco2 cells are widely used as a model of the intestinal epithelial cells exposed to the intestinal lumen, which is the location of action for pancreatic α-amylase and α-glucosidase ([91]Lea, 2015). Furthermore, all three candidates are readily absorbed at the intestine and are negative. The accessibility of a drug candidate through a membrane is determined by its Caco2 permeation, and this attribute is especially notable in the case of RA, Lut, and RS. These three phytochemicals as potential drug candidates therefore substantially pass the profiling tests ([92]Table 1). We assessed the bioavailability and toxicity of the metabolites using Lipinski’s rule-of-five and ADMET analysis, and the metabolites fulfil all the listed criteria, similar to the findings of a previous study ([93]Sharma et al., 2020), suggesting their suitability for the development of potent antidiabetic drugs. TABLE 1. (A) Physicochemical properties and (B) in silico ADME/toxicity profiles of rosmarinic acid, luteolin, and resveratrol. Property name Luteolin Resveratrol Rosmarinic acid (A) Physicochemical properties Molecular weight 286.24 228.24 360.3 XLogP3 1.4 3.1 2.4 Hydrogen bond donor count 4 3 5 Hydrogen bond acceptor count 6 3 8 Rotatable bond count 1 2 7 Topological polar surface area (TPSA) 107 Å ^2 60.7 Å ^2 145 Å ^2 Heavy atom count 21 17 26 Formal charge 0 0 0 Complexity 447 246 519 Defined atom stereocenter count 0 0 1 Defined bond stereocenter count 0 1 1 Covalently bonded unit count 1 1 1 Compound is canonicalized Yes Yes Yes (B) ADMET Human intestinal absorption HIA- HIA+ HIA- Human oral bioavailability HOB+ HOB- HOB+ Blood–brain barrier BBB+ BBB+ BBB+ Caco2 permeability Caco2+ Caco2+ Caco2+ Acute oral tox log (1 mol/kg) Nil Nil Nil Carcinogenic - - - CYP2C9 - + - CYP2D6 + - - CYP1A2 + + - CYP2C19 - - - CYP3A4 + + - Hepatotoxicity - - - Lipinski rule violation Nil Nil Nil Lead likeness violation Nil 1 1 Solubility LogS −2.588 −2.439 −3.154 [94]Open in a new tab Footnote: ADMET, Absorption, Distribution, Metabolism, Excretion, Toxicity; +, positive; -, negative; solubility normal range: −6.5 to 0.5. HIA% < 30% = HIA-; HIA% > 30% = HIA+; CYP2C9 inhibitor, CYP2D6 inhibitor, CYP1A2 inhibitor, CYP2C19 inhibitor. 3.2 IC[50] and solubility Based on a review of the half-maximal inhibitory concentrations (IC[50]) from literature, the solubility and IC[50] (in vitro and in vivo) are determined for the various proteins of RA, Lut, and RS ([95]Table 2). Based on the retrieved IC[50] and solubility values, the ratio of RA: Lut: RS is determined to be 5:1:4 for the subsequent experiments. TABLE 2. Solubility and IC[50] (in vitro and in vivo) values of rosmarinic acid, luteolin, and resveratrol. DMSO: dimethyl sulfoxide, DMF: dimethyl formamide. Ligands Solubility IC[50] (μM) EC[50] (μM) α-Amylase^1–4 α-Glucosidase^2,5–9 DPP-IV^2,10 PTP-1B^11,12 Aldose reductase^13,14 Pancreatic lipase^15,16 SIRT1^17 PPAR-γ^18 Luteolin In methanol and alkaline solutions; slightly in water, DMSO (57 mg/mL), and ethanol (6 mg/mL) 147–360 26.41–172 0.12 136.3 0.6 63 - 2.3 C[15]H[10]O[6] CID: 5280445 Resveratrol In water (3 mg/100 mL); in ethanol, DMSO, and DMF (65 mg/mL) 32.23 47.93–123 5.638 - 117.6 - 7 - C[14]H[12]O[3] CID: 445154 Rosmarinic acid In ethanol, DMSO, and DMF (25 mg/mL) 103 33 - 137 11.2 51.28 - - C[18]H[16]O[8] CID: 5281792 [96]Open in a new tab References: 1. Kusano G, Takahira M, Shibano M, et al. Studies on