Phenolics from Chrozophora oblongifolia Aerial Parts as Inhibitors of α-Glucosidases and Advanced Glycation End Products: In-Vitro Assessment, Molecular Docking and Dynamics Studies

Simple Summary The chemical investigation of Chrozophora oblongifolia aerial parts resulted in the isolation of five phenolic compounds. The isolated metabolites were tested for their antioxidant and advanced glycation end-products (AGEs) formation, α-glucosidase, and lipase inhibitory activities. 1,3,6-Trigalloyl glucose exhibited the highest activity as an antioxidant and AGEs inhibitor as well as an α-glucosidase inhibitor. It showed promising binding affinity and stability towards the human intestinal maltase-glucoamylase α-glucosidases, as revealed through coupled molecular docking and dynamics studies that could encourage the utilization of this compound in the management of diabetes and its complications. Abstract Modern life is associated with low physical activity that leads to the accumulation of fats, gaining more weight, and obesity. Accumulation of fat in the abdomen region contributes to diabetes via insulin resistance and hyperglycemia. Polyphenols are major plant constituents that exert antidiabetic activity through different mechanisms, including radicle scavenging activity, regulation of glucose uptake, and inhibition of fat and polysaccharide hydrolysis in addition to their inhibitory role regarding the formation of advanced glycation end products (AGEs). Chemical investigation of C. oblongifolia aerial parts resulted in the isolation of five major compounds: apeginin-7-O-β-D-glucoside (1), quercetin-3-O-β-D-glucuronic acid (2), quercetin-3-O-β-D-galacturonic acid (3), rutin (4), and 1,3,6-trigalloyl glucose (5). The isolated compounds were tested for their antioxidant and AGEs formation, α-glucosidase, and lipase inhibitory activities. Compound 5 revealed the highest antioxidant and AGEs inhibitory activity in bovine serum albumin (BSA)-methylglyoxal, BSA-fructose, and arginine-methylglyoxal models. Moreover, it exhibited a potent inhibitory profile on Saccharomyces cerevisiae α-glucosidases compared to the positive control, acarbose. Compound (5) further depicted promising binding affinity and stability towards the human intestinal maltase-glucoamylase α-glucosidases, which is a diabetes-related therapeutic target, through coupled molecular docking and dynamics studies. The obtained results encourage the usage of 1,3,6-trigalloyl glucose in the management of diabetes and its complications. However, detailed in-vivo studies for this compound should be performed.


Introduction
Diabetes is a complex chronic metabolic disease resulting from insulin deficiency or resistance of receptors to released hormone [1]. The disturbance of the natural balance between glucose homeostasis, proteins, and fat metabolism is the main hallmark. Diabetes incidence increases dramatically by four-fold from 2014 compared to 1980 according to World Health Organization [2]. Diabetes is considered the seventh cause of mortality worldwide according to USA National Center for Health Statistics [3].
Protein glycation represents one of the critical pathways involved in the progress of diabetic complications. Chronic hyperglycemia triggers postsynaptic protein alteration through a series of reactions called Maillard reactions [4]. Non-enzymatic glycosylation of amino groups of essential proteins with the carbonyl group of sugars drives the formation of complex products called advanced glycation end products (AGEs) [5]. AGEs may exert a pivotal role in the formation of metabolic memory in diabetic complications with accumulating evidence to link glycation and insulin resistance [6,7]. The glycation process and AGEs formation are usually associated with increased production of reactive oxygen species (ROS) [8]. Unfortunately, synthetic antidiabetic drugs might exert many side effects [2,5]. In this regard, there is a scientific interest to identify natural AGEs inhibitors with a good antioxidant potential to serve as a safe alternative approach for treating diabetic complications.
Lifestyle and Medicinal plants could be helpful in the management of diabetic complications. The regular practice of physical activity results in greater energy expenditure and a lower risk of diabetic complications. Moreover, a plant-based diet that is based on using definite amounts of whole-grain cereals, legumes, nuts, and olive oil will be supportive. Legumes and cereals are able to reduce postprandial glucose levels by decreasing carbohydrate absorption due to their high complex carbohydrates and fiber contents [9]. Moreover, many medicinal plants are known for activity in diabetic complications, like green tea, Garcinia cambogia, Nigella sativa, Hibiscus sabdariffa, Caralluma fimbriata [10], C. tuberculata [11], and C. hexagona [12]. The spurge family, Euphorbiaceae, is a large flowering plant family with 300 genera and about 7500 species that has different traditional uses [13]. Chrozophora genus belongs to the Euphorbiaceae family, which contains seven to eight species, most of which are monoecious herbs. This species is found in the Mediterranean regions, as well as in Pakistan, India, and West Africa. The genus is characterized by the presence of phenolic compounds [14] that showed different biological activities including anti-microbial [15], α-glucosidase inhibitory [16], anti-inflammatory [17], and osteoprotective [18]. In continuation of our work on diabetic complications [4,[19][20][21] and based on the few data concerning active constituents of C. oblongifolia (L.) A. Juss; it was deemed important to carry out a biological investigation of its phenolic constituents and assess their effect on diabetic complications.

Extraction and Isolation
One Kg of air-dried powdered flowering aerial parts of C. oblongifolia was extracted with methanol (MeOH) (4 × 5 L) at room temperature using Ultraturrax, till exhaustion. The combined extracts were filtered and concentrated to give a dark brown residue of 100 g. The residue was suspended in 500 mL distilled water and extracted with chloroform (CHCl 3 ) (4 × 500 mL) to give 64 g of dry extract. The concentrated remaining mother liquor was fractionated on the Diaion HP-20 column, eluted with water (H 2 O) (1 L), and subsequently eluted with MeOH/H 2 O (50%, 2.4 L) and MeOH (100%, 2.7 L), respectively. The eluates were evaporated under vacuum to give 9.7 g (fraction A), 18 g (fraction B), and 7.8 g (fraction C). Fraction A contained free sugars and exhibited no polyphenol character, by tracing on TLC using aluminum chloride (AlCl 3 ) and ferric chloride (FeCl 3 ). Fraction C was fractionated over SiO 2 CC (50 × 5 cm, 180 g) using CHCl 3 :MeOH gradient elution to give three major fractions, I, II, and III. Fraction I (0.5 g) was fractionated on Sephadex LH-20 column chromatography (CC) with MeOH as an eluent to afford compound 1 (50 mg). Fraction II (1.5 g) was chromatographed on SiO 2 100 C 18 -reversed-phase CC using MeOH:H 2 O (3:7) as an eluent to afford compound 2 (40 mg). Fraction III (2 g) was subjected to repeated chromatographic fractionations on SiO 2 100 C 18 -reversed-phase CC using MeOH:H 2 O (3:7) as an eluent to yield compound 3 (20 mg). Fraction B was fractionated over SiO 2 CC chromatography (50 × 5 cm, 180 g) using CHCl 3 :MeOH gradient elution to give two major, I and II. Fraction I (0.5 g) was chromatographed on SiO 2 100 C 18 reversed-phase CC using MeOH:H 2 O, 4:6 as an eluent to afford compound 4 (50 mg). Fraction II (1.5 g) was submitted to SiO 2 100 C 18 -reversed-phase CC using MeOH:H 2 O (3: 7) as an eluent to give compound 5 (40 mg).

Characterization of Isolated Compounds
Structure elucidation of isolated compounds was performed by 1D and 2D NMR on Bruker Avance DRX 850 MHz spectrometers (Bruker BioSpin/Billerica/MA/USA).

Biological Evaluation
2.4.1. Inhibition of α-Glucosidase α-Glucosidase from Saccharomyces cerevisiae (Sigma-Aldrich, St. Louis, MO, USA cat#G003) was used in this assay. Solutions of the different compounds were prepared in 1000 µM final concentrations in dimethyl sulfoxide (DMSO). Compounds that exceeded 50% inhibition at this concentration were serially diluted to determine their half-maximal inhibitory concentration (IC 50 ). As a positive control, acarbose was prepared in MeOH at a concentration of 1 mM. α-Glucosidase inhibition was assessed colorimetrically using p-nitrophenyl-α-D-glucopyranoside as the substrate (pNPG, Sigma-Aldrich, St. Louis, MO, USA cat#N1377) as previously reported with slight modifications [22].
Briefly, 25 µL of samples/blank were incubated for 10 min at 37 • C with 50 µL of α-glucosidase (0.6 U/mL) in phosphate buffer (0.1 M, pH 7) in 96-microwell plates. The mixture was then incubated at 37 • C for 5 min with 25 µL of 3 mM pNPG in phosphate buffer (pH 7). A microplate reader (TECAN, Männedorf, Switzerland) was used to measure p-nitrophenol release from pNPG substrate at 405 nm to determine enzyme activity. Then, % α-glucosidase inhibition was estimated.

Inhibition of Pancreatic Lipase
The inhibition of pancreatic lipase activity was assessed using p-nitrophenyl dodecanoate (p-NPD, Sigma-Aldrich, St. Louis, MO, USA cat#N0252) as substrate and porcine pancreatic lipase type II (PPL, sigma Aldrich, cat#L3126) which was described previously [23] with slight modification. Solutions of the provided compounds were prepared in final concentrations of 50 µM in 5% DMSO. Orlistat was used as positive control and was Biology 2022, 11, 762 4 of 24 prepared in assay buffer. In a Tris-HCl buffer (100 mM, pH 8); 50 µL of PPL (1 mg/mL) was incubated with 25 µL of samples/blank for 10 min at 37 • C. Then, 10 µL of solution p-NPD (10 mM, in isopropanol) was added followed by dilution of the final volume to 200 µL with Tris-HCl buffer. The mixture was incubated for 20 min at 37 • C and absorbance was then measured at 405 nm on a TECAN microplate reader (Männedorf, Switzerland). After that, the percentage of lipase inhibition was assessed (Supplementary Table S1 and Figure S1).

Inhibition of the Formation of Advanced Glycation End Products (AGEs)
The isolated compounds were prepared at final concentrations of 500 µM in DMSO in the BSA-fructose assay, and 1000 µM in all other AGEs assays. Compounds that exceeded 50% inhibition at this concentration were serially diluted to determine their IC 50 . All tests were conducted using quercetin (positive control).

BSA-Fructose Assay
According to the previous report [8], the inhibition test for AGE formation was assessed using the BSA-fructose model. Briefly, BSA (50 mg/mL) and fructose (1.25 M) (in 200 mM phosphate buffer, pH 7.4, with 0.02% sodium azide) were incubated with Sample/blank in the dark for 72 h at 37 • C, (1:1:1 by volume). On a microplate reader (Fluostar Omega BMG Labtech, Ortenberg, Germany), fluorescent AGEs were detected using excitation and emission wavelengths of 355 nm and 460 nm, respectively. Triplicates of all experiments were performed, and the percentage of AGEs inhibition was calculated.

Arginine-Methylglyoxal Assay
The determination of the inhibitory effect of the isolated compounds on the formation of AGEs in the arginine methylglyoxal model was used as described previously with slight modification [8]. Briefly, in black 96-microwell plates, a sample was incubated for 72 h at 37 • C in the dark with methylglyoxal (53.3 mM, in distilled water) and 106.6 mM arginine (in 200 mM phosphate buffer, pH 7.4, with 0.02% sodium azide), (1:1:1 v/v/v). On a microplate reader (Fluostar Omega/BMG Labtech/Germany), fluorescent AGEs were detected using excitation and emission wavelengths of 355 nm and 460 nm, respectively. Triplicates of all experiments were performed.

BSA-Methylglyoxal Assay
AGEs inhibitory activity of the isolated compounds was determined according to the reported procedures [8] with slight modification. Briefly, in black 96-microwell plates, samples were incubated for 72 h at 37 • C in the dark with bovine serum albumin (BSA, 50 mg/mL) (in 200 mM phosphate buffer, pH 7.4, with 0.02% sodium azide) and methylglyoxal (53.3 mM, in distilled water), (1:1:1 v/v/v). On a microplate reader (Fluostar Omega/BMG Labtech/Germany), fluorescent AGEs were detected using excitation and emission wavelengths of 355 nm and 460 nm, respectively. Triplicates of all experiments were performed.

Antioxidant Activity 2,2-diphenyl-1-picrylhydrazyl (DPPH) Assay
Isolated compounds were prepared in final concentrations of 1000 µM in DMSO. Compounds that exceeded 50% inhibition at this concentration were serially diluted to determine their IC 50 . Ascorbic acid was used as a positive control, prepared in a concentration of 100 µM in water from which six concentrations were prepared to calculate its IC 50 . The method is based on the reduction of free radical of 2,2-diphenyl-1-picrylhydrazyl (DPPH) [22]. Briefly, in a 96-well plate, 20 µL of the isolated compound was incubated for 20 min in the absence of light with 180 µL of DPPH reagent (100 µM). The reduction of intensity in the color of DPPH was assessed on a microplate reader (TECAN, Männedorf, Switzerland) at 540 nm (Supplementary Table S2). 2,2 -azino-bis (3-ethylbenzothiazoline-6-sulfonic acid) (ABTS) Assay DMSO was used for the preparation of 0.5 mg/mL concentrations of isolated compounds, meanwhile, the positive control (ascorbic acid) was used to prepare a stock solution of 1000 µM in water; from which seven concentrations were prepared including 7.812, 15.625, 31.25, 62.5, 125, 250 and 500 µM to construct the concentration-response curve (Supplementary Data Figure S2). The activity of the tested compounds to scavenge the free radicals of ABTS (2,2 -azino-bis (3-ethylbenzothiazoline-6-sulfonic acid)) was assessed according to the methodology reported previously [20] with slight modifications.
Five milliliters of aqueous solution (7 mM) of ABTS was mixed with 88 µL potassium persulfate (140 mM) and the mixture was kept in the dark for 16 h to produce ABTS radical cation (ABTS •+ ). Then methanol was added to the reaction mixture to dilute the formed ABTS •+ solution to an absorbance of 0.700 (1:50) at 690 nm.
Ten microliters of the tested compound were incubated with 190 µL of ABTS •+ solution in a 96-well plate for thirty minutes in the dark and the absorbance was measured later using a microplate reader (TECAN, Männedorf, Switzerland) at 690 nm. Data are represented as means (n = 3) ± SD and the antioxidant effect of the tested compounds was calculated as µM Ascorbic acid equivalents by substitution in the linear regression equation (Supplementary Figure S2).

Ferric Reducing Antioxidant Power (FRAP) Assay
Isolated compounds were dissolved in DMSO to prepare a 0.5 mg/mL concentration, meanwhile, the positive control (ascorbic acid) was used to prepare a stock solution of 1000 µM in water; from which seven concentrations were prepared including 15.625, 31.25, 62.5, 125, 250 and 500 µM to construct the dose-response curve. The method of the assay depends on the reduction of Fe 3+ to Fe 2+ , which is chelated by 2,4,6-tris(2-pyridyl)-s-triazine (TPTZ) to form Fe 2+ -TPTZ complex which is known as FRAP (ferric reducing antioxidant power) [1]. Briefly, 190 µL of FRAP reagent (300 mM sodium acetate buffer pH 3.6, 10 mM TPTZ, and 20 mM ferric chloride, (10:1:1)) was incubated in the dark with 10 µL of the isolated compound in a 96-well plate for 30 min at 37 • C and absorbance was determined at 620 nm using microplate reader (Tecan, San Jose, CA, USA). Data are represented as means (n = 3) ± SD and the antioxidant effect of the extract was calculated as µM Ascorbic acid equivalents by substitution in the linear regression equation (Supplementary Figure S3).

Metal Chelation Assay
Isolated compounds were dissolved in DMSO to prepare a concentration of 0.5 mg/mL, meanwhile, the positive control (EDTA) was used to prepare a stock solution of 100 µM in water; from which six concentrations were prepared including 5, 10, 20, 40, 50 and 80 µM to construct the dose-response curve. The assay was carried out as previously reported [24], with minor modifications. Briefly, in 96 wells plate (n = 6) 20 µL of the freshly prepared ferrous sulfate (0.3 mM) was mixed with 50 µL of the tested compound and incubated with 30 µL of ferrozine (0.8 mM) for 10 min at room temperature. Finally, reduction in color intensity was assessed at 540 nm using a microplate reader (Tecan, San Jose, CA, USA). Data are represented as means (n = 3) ± SD and the antioxidant effect of the extracts was calculated as µM EDTA equivalents by substitution in the linear regression equation (Supplementary Figure S4).

Data Analysis
All IC 50 data were depicted as means ± SE and analyzed utilizing Microsoft_Excel ® and IC 50 s were estimated by Graph-pad-Prism 8 ® . One-way ANOVA followed by Tukey's test was utilized for significant differences between means.
Guided by a triangular matcher, several ligand conformations were generated and initially ranked via London_dG scoring. Subsequently, a refinement step proceeded where retained top-scored poses were rescored through a Generalized Born-solvation-VI/Weighed Surface-Area_dG (GBVI/WSA_dG)-directed protein residue-tethered minimization step for providing the final ligand/target binding modes [30]. This final score was based on electrostatics of explicit solvation, currently assigned partial charge, the surface area of exposure, and Coulomb's electrostatics by protein-ligand van der Waals scoring [31,32]. Significant docking energy scores (kCal/mol), 2.0 Å root-mean-square deviation (RMSD) cut-off relative to co-crystalline ligand superimposition, as well as significant contacts with reported important pocket-binding residues were all considered during the selection of the best docking pose. Visualization and ligand-target binding analysis were done using PyMol_V2.0.6 (Schrödinger TM , New York, NY, USA) [33].

Molecular Dynamics (MD) Simulations
Conformational stability, protein contacts, and predicted binding scores of ligand-hMGAM complexes, of both investigated gallotannin derivative as well as co-crystalline ligand antagonist, were explored through 200 ns explicit MD simulations using GROMACS-2019 under CHARMM36m.forcefield for protein [34] and CHARMM-General_forcefield program for ligand modeling [35]. A periodic box of TIP3P solvent was added around the protein ensuring at least a 10 Å distance between the protein and the box [36]. Standard ionization states of hMGAM amino acids were assigned at pH 7.4 keeping the whole system neutralized through added potassium and chloride ions [37]. Each system was individually minimized (0.005 ns; steepest-descent method), double equilibrated (0.1 ns, Biology 2022, 11, 762 7 of 24 303.15 K at NVT ensemble; then 0.1 ns, 303.15 K, 1 atm. P at NPT ensemble) before the MD simulation production was conducted for subsequent 200 ns runs under NPT ensemble and Particle-Mesh Ewald protocol for modeling long-ranged electrostatic contacts [38]. linear constraint LINCS was applied to maintain covalent bond distances at 2 fs integration timestepped sizes [39], while non-bonded interactions were truncated via the Verlet scheme at 10 Å cut-offs [40]. Stability analysis of the simulated models was performed through rootmean-square deviation (RMSD) and difference RMS fluctuation (∆RMSF) analysis while the latter was estimated for each ligand-bound protein (holo) in relation to its unliganded (apo) state (PDB entry: 2QLY; 2.00 Å atomic resolution), where ∆RMSF = RMSF (apo−holo) . Finally, Molecular Mechanics/Poisson-Boltzmann Surface Area (MM/PBSA) was applied for binding-free energy calculations as well as for estimating the residue-wise energy contributions [41]. Ligand-protein conformational analysis was represented and analyzed using the PyMol_V2.0.6 software (Schrödinger TM , New York, NY, USA).

Results and Discussion
Chemical investigation of C. oblongifolia resulted in the isolation of five major compounds ( Figure 1) that were identified as: (3), rutin (4), and 1,3,6-trigalloyl glucose (5). The identity of isolated compounds was assessed by different spectroscopic methods including one-and two-dimensional NMR (Supplementary Figures S6-S23).

Inhibition of α-Glucosidase
α-Glucosidase is a membrane-bound enzyme of the small intestine that represents the key enzyme in carbohydrates metabolism. Enzyme inhibition is the main target for the treatment of type II diabetes and can significantly attenuate the elevated glucose level especially postprandial blood glucose [22]. In the present study, acarbose inhibited αglucosidase at IC 50 of 0.161 mM. Compound 5 showed the most potent inhibition of the enzyme with IC 50 of 0.0045 mM with a more potent inhibitory effect than acarbose. Other compounds revealed inhibitory effect on α-glucosidase in the following order 3 > 1 > 2 > 4 with IC 50 of 0.385, 0.408, 0.545, and 0.754 mM, respectively ( Figure 2).

Inhibition of the Formation of Advanced Glycation End Products (AGEs)
Evaluation of cell-free in-vitro anti-glycation activity was established using three models viz: BSA-fructose, BSA-methylglyoxal, and arginine-methylglyoxal assays. Tight control of intracellular glucose concentration plays an important role in the pathogenesis of microvascular diabetes complications. The overproduction of ROS by mitochondrial stimulates the formation of AGEs as a piece of the puzzle [42].
Indeed, the generation of AGEs was established in three stages; in an early stage, a series of cascade reactions between the glucose carbonyl group and the terminal α-amino-

Inhibition of Pancreatic Lipase
Orlistat inhibited pancreatic lipase with an IC 50 of 0.70 µM, while all tested compounds did not show significant inhibition till 50 µM concentration (Supplementary Figure S1 and Table S1).

Inhibition of the Formation of Advanced Glycation End Products (AGEs)
Evaluation of cell-free in-vitro anti-glycation activity was established using three models viz: BSA-fructose, BSA-methylglyoxal, and arginine-methylglyoxal assays. Tight control of intracellular glucose concentration plays an important role in the pathogenesis of microvascular diabetes complications. The overproduction of ROS by mitochondrial stimulates the formation of AGEs as a piece of the puzzle [42].
Indeed, the generation of AGEs was established in three stages; in an early stage, a series of cascade reactions between the glucose carbonyl group and the terminal αamino-groups of proteins produce an unstable Schiff's base. In an intermediate stage, intermolecular rearrangements of Schiff's base through the Maillard reaction form more stable Amadori products. In the late stage, Amadori products undergo oxidative modifications of reactive dicarbonyl compounds as glyoxal or methylglyoxal to yield irreversible AGEs compounds [43,44].
The reactive metabolite (methylglyoxal, MGO) is the permeant precursor of AGEs that are directly associated with neurodegenerative diseases and diabetic complications [45]. Regarding the glycation process, MGO reacts with different protein residues such as arginine, lysine, and cysteine to generate irreversible AGEs [42].
BSA was selected as the model protein to mimic human serum albumin. The fluorescence intensity indicated the formation of AGEs in all BSA models. The inhibition of BSA glycation prevents the cascade reactions that yield AGEs [8].
In the BSA-fructose model compound, 5 showed potent inhibition at IC 50 of 0.035 mM, followed by compound 4 with an IC 50 of 0.037 mM. Whilst compounds 1 and 2 significantly cence intensity indicated the formation of AGEs in all BSA models. The inhibition of BS glycation prevents the cascade reactions that yield AGEs [8].
Finally, in the BSA-methoxyglyoxal assay, quercetin revealed inhibitory IC 50 of 0.941 mM, while compounds 5, 4, and 1 had IC 50 at 0.657, 0.493, and 0.668 mM, respectively. Compound 2 showed the lowest inhibition at 1.877 mM and like the BSA fructose assay, compound 3 did not reveal any inhibitory effect with IC 50 > 1.200 mM (Figure 3).

Antioxidant Activity
For assessment of free radical scavenging activity of isolated compounds, at least more than two different methods are highly recommended [46]. In the current study, five different assays (DPPH, ABTS, FRAP, metal chelation, and ORAC) have been established utilizing different mechanisms to assess more realistic measurements [47]. They were established for measuring the total antioxidant capacity (TAC) through a significantly different and specific chemical reaction. The results of the antioxidant activity of different compounds are shown in Figures 4 and 5. highest activity with an IC50 of 0.226 mM, followed by compounds 4, 1, and 2 (IC50 of 0.474, and 1.143 mM, respectively) compared to quercetin (IC50 0.536 mM). Compo did not reveal any inhibitory effect (IC50 > 1.200 mM) ( Figure 3).
Finally, in the BSA-methoxyglyoxal assay, quercetin revealed inhibitory IC50 of mM, while compounds 5, 4, and 1 had IC50 at 0.657, 0.493, and 0.668 mM, respec Compound 2 showed the lowest inhibition at 1.877 mM and like the BSA fructose compound 3 did not reveal any inhibitory effect with IC50 > 1.200 mM (Figure 3).

Antioxidant Activity
For assessment of free radical scavenging activity of isolated compounds, a more than two different methods are highly recommended [46]. In the current stud different assays (DPPH, ABTS, FRAP, metal chelation, and ORAC) have been estab utilizing different mechanisms to assess more realistic measurements [47]. They we tablished for measuring the total antioxidant capacity (TAC) through a significant ferent and specific chemical reaction. The results of the antioxidant activity of dif compounds are shown in Figures 4 and 5.
In the DPPH assay, ascorbic acid showed IC50 of 0.028 mM. Compound 5 poss the most potent inhibition of DPPH radical with an IC50 of 0.004 mM, followed by pounds 4 and 2 with IC50 of 0.019, and 0.029 mM, respectively. While compounds 3 revealed much lower inhibition at 0.123 and 0.366 mM, respectively ( Figure 4). Generally, preventive antioxidants are enzyme-based reactions, such as catala roxidase, and superoxide dismutase, which suppress ROS. On the other side, chaining antioxidants are compounds that are able to scavenge tissue ROS and break the tion chain like vitamin E, vitamin C, and phenolic compounds [25,47]. The mech underlying the function of chain-breaking antioxidants can be explained based on h gen atom transfer (HAT), single electron transfer (SET), or the combination of both and SET mechanisms.
DPPH assay depends on both SET or HAT while ABTS depends only on SET Consequently, the ABTS assay may be considered a confirmatory assay to the DPP say. However, ABTS has an advantage over DPPH since it is pH-independent and plicable for antioxidants of lipophilic or hydrophilic nature [25]. In the DPPH assay, ascorbic acid showed IC 50 of 0.028 mM. Compound 5 possessed the most potent inhibition of DPPH radical with an IC 50 of 0.004 mM, followed by compounds 4 and 2 with IC 50 of 0.019, and 0.029 mM, respectively. While compounds 3 and 1 revealed much lower inhibition at 0.123 and 0.366 mM, respectively ( Figure 4).
Generally, preventive antioxidants are enzyme-based reactions, such as catalase, peroxidase, and superoxide dismutase, which suppress ROS. On the other side, chain-breaking antioxidants are compounds that are able to scavenge tissue ROS and break the reaction chain like vitamin E, vitamin C, and phenolic compounds [25,47]. The mechanism underlying the function of chain-breaking antioxidants can be explained based on hydrogen atom transfer (HAT), single electron transfer (SET), or the combination of both HAT and SET mechanisms.
Conversely, the ORAC assay measures the capacity of hydrophilic antioxidants to capture peroxyl radicals depending on the hydrogen atom transfer (HAT) mechanism [46]. Compound 4 showed the highest activity in the ORAC assay revealing a 2.16 mM TE/mg compound. Compounds 5 and 1 revealed significant activity with 1.11-and 1.09-mM TE. Finally compounds 2 and 3 showed the lowest activity at 0.74-and 0.54-mM TE, respectively. Finally, metal chelation is based on the capacity of antioxidant compounds to compete with ferrozine for Fe 2+ ion in a process called the Fenton reaction. Interestingly, as illustrated in Figure 5; compounds 5, 4, 1, and 3 showed high activity with 17.52, 28.96, 19.26, and 28.26 µM EDTA eq./mg compound, respectively. While compound 2 showed the lowest activity at 8.29 µM EDTA eq./mg. Antioxidant activity plays a pivotal role to maintain cellular physiology by scavenging ROS and preventing its damaging capacity [50]. In terms of biological activity, the generation of free radicals triggers the spontaneous production of non-enzymatic AGEs due to rearrangements of Amadori products.
The aforementioned data revealed that compound 5 is the most potent with regard to all observed activities. Compound 5 revealed more potent inhibition on α-glucosidases than acarbose. Nevertheless, acarbose was renowned to be more efficient on human αglucosidase than other sources of α-glucosidases such as Saccharomyces, utilized in this assay, which necessitates comparing compound 5 and acarbose on human glucosidase through docking and molecular dynamic studies. Furthermore, compound 5 did not show DPPH assay depends on both SET or HAT while ABTS depends only on SET [25]. Consequently, the ABTS assay may be considered a confirmatory assay to the DPPH assay. However, ABTS has an advantage over DPPH since it is pH-independent and is applicable for antioxidants of lipophilic or hydrophilic nature [25].
Unlike DPPH, the FRAP assay is a non-specific redox colorimetric assay based on the ability of the antioxidant sample to donate electrons to reduce a colorless Fe 3+ -TPTZ complex to an intense blue Fe 2+ -TPTZ complex under acidic conditions [48]. In the current study, unlike the ABTS inhibition assay, compounds 5 and 4 revealed the highest activity with 7.23 and 2.66 mM AAE/mg of the compounds. On the other hand, compounds 2, 3, and 1 showed lower activity at 0.73, 0.49, and 0.23 mM AAE/mg compound. Although both ABTS and FRAP are based on the SET mechanism, FRAP can measure only the reducing power of samples [48,49].
Conversely, the ORAC assay measures the capacity of hydrophilic antioxidants to capture peroxyl radicals depending on the hydrogen atom transfer (HAT) mechanism [46]. Compound 4 showed the highest activity in the ORAC assay revealing a 2.16 mM TE/mg compound. Compounds 5 and 1 revealed significant activity with 1.11-and 1.09-mM TE. Finally compounds 2 and 3 showed the lowest activity at 0.74-and 0.54-mM TE, respectively.
Finally, metal chelation is based on the capacity of antioxidant compounds to compete with ferrozine for Fe 2+ ion in a process called the Fenton reaction. Interestingly, as illustrated in Figure 5; compounds 5, 4, 1, and 3 showed high activity with 17.52, 28.96, 19.26, and 28.26 µM EDTA eq./mg compound, respectively. While compound 2 showed the lowest activity at 8.29 µM EDTA eq./mg. Antioxidant activity plays a pivotal role to maintain cellular physiology by scavenging ROS and preventing its damaging capacity [50]. In terms of biological activity, the generation of free radicals triggers the spontaneous production of non-enzymatic AGEs due to rearrangements of Amadori products.
The aforementioned data revealed that compound 5 is the most potent with regard to all observed activities. Compound 5 revealed more potent inhibition on α-glucosidases than acarbose. Nevertheless, acarbose was renowned to be more efficient on human αglucosidase than other sources of α-glucosidases such as Saccharomyces, utilized in this assay, which necessitates comparing compound 5 and acarbose on human glucosidase through docking and molecular dynamic studies. Furthermore, compound 5 did not show any inhibitory effect on pancreatic lipase which could eliminate any possibility of non-specific action of this compound on enzymes such as precipitation or denaturation.

Molecular Docking Analysis
From a therapeutic standpoint, the above profound activity of the C. oblongifoliaisolated 1,3,6-trigalloylglucose (5) on Saccharomyces α-glucosidase prompted the further evaluation of the compound's affinity towards human α-glucosidase through comprehensive in-silico study. Note that acarbose has been reported to be more effective on human α-glucosidase than α-glucosidases from other sources, such as Saccharomyces [51]. In this regard, the molecular docking affinity of this top-active compound was investigated against the crystalline human intestinal maltase-glucoamylase α-glucosidase enzyme (hMGAM; PDB entry: 2QMJ), in comparison with the canonical small molecule inhibitor, acarbose [52]. The adopted 101.02 kDa biological target is a monomer-A1 hydrolase glycoprotein solved at 1.90 Å atomic resolution in a complex with different glycans in addition to acarbose. The overall architecture of this α-glucosidase enzyme ( Figure 6A) is of a trefoil P-type domain at the N-terminus which is then followed by a β-sandwich sheet domain and [α/β] 8barrel catalytic domain bearing two inserts (insert-I and -II) protruding after the β3 and β4-strands, respectively. At the hMGAM carboxy terminus, the structure ends with subsequent proximal and distal domains having distinct β-sandwich topologies comparable to the other homologous glycoside hydrolase (GH)-31 family [51,[53][54][55]. The substratebinding pocket is formed mainly by the [α/β] 8 -barrel catalytic residues, in addition to the N-terminus loop (residue range 200-217) as well as portions of catalytic inserts-I and -II near the opening of [α/β] 8 -barrel catalytic domain.
The non-cleavable crystalline ligand, acarbose, shows favored anchoring at the hMGAM active-site pocket via its two initial rings, the acarvosine unit, being positioned at the non-reducing terminal of this pseudo-α-1,4-tetrasaccharide ( Figure 6B). In the X-ray structure, this acarvosine scaffold exhibited extended polar interactions with the target pocket residues at the −1 and +1 carbohydrate subsites including Asp203, Asp327, Arg526, Asp542, and His600 sidechains. Water-bridge hydrogen bonding was also depicted for the acarvosine moiety towards Asp443 and Asp571. Based on mutagenesis, substrate trapping studies, and sequence alignment with GH-31 protein members, the Asp443 and Asp542 were identified as the respective catalytic nucleophile and acid/base catalysis residue, making them the key ligand-binding amino acids for halting the hydrolase catalytic machinery [54]. The aglycone valienamine, the first ring of acarbose, adopted a 2 H 3 -half-chair conformational structure as being lodged within the −1 subsite. This depicted orientation has directed the non-hydrolyzable inter-glycosidic nitrogen atom towards the catalytic center for hydrogen bond pairing with the Asp542 sidechain. The stability of acarbose at +1 subsite was mediated via the ligand's C2 and C3 hydroxyl groups through polar interactions with Arg526, Asp542, and Asp203, where the last residue originated from the N-terminal loop at the β-sheet domain. Notably, loop residues at such β-sheet domain exhibited invariable interactions with several carbohydrate ligands within all available GH-31 crystalline structures [51,[53][54][55][56]. Unlike the acarvosine scaffold, the acarbose's two maltose moieties at +2 and +3 carbohydrate subunits showed limited polar interactions with the lining residues. The stability of these glycon scaffolds was rather mediated through the crystal lattice packing, hydrogen-bond pairing between the terminal maltose ring and Thr205/Asn207, as well as water-mediated binding with penultimate maltose and Tyr605 at the pocket's rim. Further stabilizing interactions are provided via the hydrophobic contacts with Tyr299, Ile328, Ile364, Trp406, Trp441, Phe450, Trp539, Phe575, Ala576, Leu577, and Tyr605. Redocking acarbose into the hMGAM substrate-binding site was performed through the rigid docking protocol since the superposition correlation analysis between the protein's apo-and holo-states illustrated insignificant protein conformational changes (RMSD = 0.1 Å), either locally or globally. The latter suggested an irrelevant impact of local ligand inducedfitting on hMGAM's holo-structure, at least within the macromolecular crystalline states [57]. This could be also the reason for the surface located shallow nature of the hMGAM substratebinding pocket with possible accommodation of two carbohydrate subunits [51][52][53]. Findings from the acarbose redocking analysis revealed good docking binding energy (−6.8 kCal/mol) and low RMSD (1.5 Å) relative to the co-crystalline ligand. Depicting well-matching superimposition ( Figure 6B), the above-cited ligand-residue hydrophobic and polar binding interactions were conserved for the redocked acarbose molecule following the molecular docking protocol (Table 1). Additionally, the polar functionalities of the redocked acarvosine and +2 maltose molecule illustrated close proximity (~4.7 Å) towards Asp203, Asp443, Asp571, and Tyr605 allowing a good chance of forming water-mediated hydrogen-bonding. Based on these redocking findings, the adopted docking protocol was confirmed appropriate since depicting RMSD values below 2.0 Å indicated that both the adopted docking parameters and algorithms were sufficient for determining the best docking pose [58]. Thus, the adopted directed docking protocol can ensure the biological relevance of the obtained docking binding modes and in turn their respective docking energies. Moving towards the molecular docking analysis of 1,3,6-trigalloylglucose (5), a more favorable docking score (−7.3 kCal/mol) was assigned for this gallotannin derivative as compared to the co-crystalline hMGAM inhibitor, acarbose. This compound predicted significant anchoring at the −1 and +1 subsites of the target protein ( Figure 6C). Having its 1-O-galloyl ring directed into the −1 subsite, the investigated gallotannin derivative exhibited an extended hydrogen bond network with several key pocket residues including the sidechains of Asp327, His600, and Tyr299, as well as the catalytic Asp542 being crucial for the enzyme's acid/base catalysis (Table 1). On the other hand, the 3-O-galloyl ring predicted polar contact with the N-terminal β-sheet loop residue, Asp203, near the rim of the binding site. Whereas, the 6-O-galloyl moiety exhibited a single hydrogen bond pairing with the mainchain of Leu577 lining the +1 subsite. Both the catalytic residue, Asp542, and N-terminal β-sheet loop residue, Thr205, further stabilized the ligand's central D -glucopyranose scaffold within the canonical pocket through strong hydrogen bond pairing with its C2 hydroxy group. Further stabilizing interactions were provided via comparable acarbose-based hydrophobic contact profiles with Tyr299, Ile328, Ile364, Trp406, Trp441, Phe450, Trp539, Phe575, Ala576, Leu577, and Tyr605. Nevertheless, only the docked gallotannin derivative predicted significant π-mediate hydrophobic interactions with Tyr299 and Phe575 owing to its aromatic nature.
Overlayed conformational analysis of the docked gallotannin derivative and cocrystalline acarbose showed a relevant orientation of the ligand's 1-O-galloyl ring at the acarbose aglycone moiety relevant to the −1 subsite. Regarding the central D -glucopyranose scaffold of the docked gallotannin derivative, relevant orientation was depicted for this flexible core structure being at the intermediate distance of both the 6-deoxyglucosyl and first maltose moieties of the acarbose corresponding to the +1 and +2 carbohydrate subsites, respectively. Interestingly, both C3 and C6 galloyl rings showed no typical orientation at the +3 subsite, this was due to the extended and branched topology of the docked gallotannin compound allowing minimal steric clashes with the lining residues of the substrate-binding pocket. It is worth mentioning, that only a 3-O-galloyl ring was in close proximity to the latter subsite residues. Depicting such a favored docking pose, the docked gallotannin compound depicted a V-shaped/arrow-head conformation within the hMGAM binding site having its 1-O-galloyl ring as its anchor while both other galloyl rings were flanked at both sides of the compound. Accumulated evidence from the crystalline structures of GH-31 family homologs and lysosomal α-glucosidase enzymes revealed no direct interactions between co-crystallized inhibitors and +3 subsite residues while suggesting the paucity of only −1 and +1 subsites as the productive substrate-binding sites for these biological targets [51,[53][54][55][56]. Thus, the lack of predicted direct polar contacts for the isolated gallotannin derivative with +3 carbohydrate subsite residues was suggested to not greatly influence the ligand-target affinity and neither its enzyme inhibition potentiality.

Molecular Dynamics Simulation Analysis
Both the docked gallotannin derivative and co-crystallized acarbose in complex with hMGAM were subjected to 200 ns molecular dynamics (MD) simulations. The estimated RMSD deviations for the target hMGAM proteins, in reference to the Cα backbone (Cα-RMSD), illustrated an overall typical behavior for MD simulations ( Figure 7A). Following the release of constraining at the start of the MD production, the proteins' Cα-RMSDs were elevated over initial frames and for the first 30 ns of the MD runs. Beyond that timeframe, steady protein Cα-RMSD tones were obtained for more than half the simulation runs (i.e., >140 ns) with no significant fluctuations along with the entire MD timeframes. Notably, the hMGAM proteins within both simulated models were leveled-off at comparable Cα-RMSDs throughout the trajectory plateau and till the MD simulation end (2.8 ± 0.1 Å, and 2.7 ± 0.2 Å for acarbose and gallotannin derivative models, respectively. These illustrated dynamic behaviors confer successful convergence of the hMGAM proteins throughout the designated MD simulation timeframes as well as successful system minimization, relaxation, and thermal equilibration prior to the MD production step. In these regards, no further extension of the MD simulation beyond the 200 ns-period was needed. For investigating the stability of the docked gallotannin derivative and co-crystalline acarbose within the hMGAM substrate-binding site, the Cα-RMSDs of both the sole ligand as well as the combined ligand-hMGAM complex were monitored along with the whole MD timeframes in relation to the reference protein backbone frame ( Figure 7B,C). Backbone Cα-RMSD plateaus were illustrated for both models, despite limited fluctuations, conferring significant confinement of the simulated ligands within the target's pocket. Steady complex Cα-RMSDs tones were depicted for both ligands around averages of 2.7 ± 0.2 Å and 2.9 ± 0.2 Å for the investigated gallotannin derivative and acarbose, respectively. Comparable acarbose-based dynamic findings were also reported by Zhang et al. for unraveling possible inhibitors of the hMGAM biological target [59]. Concerning the sole ligand's Cα-RMSDs, higher average values were assigned for the docked gallotannin compound in relation to the co-crystalline acarbose (5.2 ± 0.5 Å versus 2.9 ± 0.6 Å) till the half of the MD simulation run. However, the simulated gallotannin derivative depicted more steady and equilibrated trajectories beyond the 100 ns timeframe and till the end of the simulation as it managed to converge at lower Cα-RMSD around the final frames (3.8 ± 0.3 Å). On the other hand, ligand's Cα-RMSDs fluctuations with higher standard deviation values were shown for the acarbose across the whole MD simulation run, highlighting possible conformational changes for the pseudosugar within the hMGAM substrate-binding pocket. The presented Cα-RMSD trajectories ensured successful convergence, as well as ligandpocket confinement and the suitability of 200 ns MD simulation, runs without the need for further extensions.  Examining the models at trajectories of the start, midway, and final timeframes (0 ns 100 ns, and 200 ns), was beneficial for monitoring the possible conformational changes of the simulated ligand-protein models across the MD simulation run. Designated frames for each ligand-protein model were extracted and minimized to a 0.001 kcal/mol.A 2 gradient using the MOE system preparation package. Notably, stable ligand-hMGAM binding profiles were assigned for both the simulated gallotannin derivative and co-crystalline ligand (Figure 8). Ligands showed favored orientation/conformation at the hMGAM substrate-binding pocket, while limited ligand orientation alterations were depicted at the end of the MD simulation runs. Throughout the entire MD simulation run, the co-crystalline ligand acarbose showed significant conformational alterations for its terminal +3 maltose sugar as compared to the rest of the ligand's scaffold rings ( Figure 8A). This dynamic behavior could be reasoned for the reported limited binding interaction at this terminal subsite which matches the above molecular docking investigation as well as the reported crystalline structures [51,[53][54][55][56]. The reasonably flexible +3 maltose ring could Examining the models at trajectories of the start, midway, and final timeframes (0 ns, 100 ns, and 200 ns), was beneficial for monitoring the possible conformational changes of the simulated ligand-protein models across the MD simulation run. Designated frames for each ligand-protein model were extracted and minimized to a 0.001 kcal/mol.A 2 gradient using the MOE system preparation package. Notably, stable ligand-hMGAM binding profiles were assigned for both the simulated gallotannin derivative and co-crystalline ligand (Figure 8). Ligands showed favored orientation/conformation at the hMGAM substrate-binding pocket, while limited ligand orientation alterations were depicted at the end of the MD simulation runs. Throughout the entire MD simulation run, the co-crystalline ligand acarbose showed significant conformational alterations for its terminal +3 maltose sugar as compared to the rest of the ligand's scaffold rings ( Figure 8A). This dynamic behavior could be reasoned for the reported limited binding interaction at this terminal subsite which matches the above molecular docking investigation as well as the reported crystalline structures [51,[53][54][55][56]. The reasonably flexible +3 maltose ring could account for the significant acarbose's Cα-RMSDs fluctuations having higher standard deviation values across the MD simulation run. On the one hand, the ligand's acarvosine scaffold (aglycone valienamine and 6-deoxyglucosyl moieties) was highly confined at the −1/+1 sugar subsites showing limited conformational/orientation shifts. Moving towards the simulated gallotannin derivative, the ligand's V-shaped/arrow-head conformation was conserved showing the 1-O-galloyl moiety stably anchored at the −1 sugar subsite. On the other hand, both the central glucosyl scaffold and the other two flanking galloyl moieties showed reasonable conformational alterations yet maintained conserved orientations and confinements at the +1 and +2 sugar subsites. For further investigation of the hMGAM's local flexibility and residue-wise contribution within the ligand-protein binding stability, the difference root-mean-square fluctuations (ΔRMSF) of the apo hMGAM relative to its holo state (RMSFApo−Holo) were monitored along the whole MD runs. Applying the cut-off mobility threshold at ΔRMSF of 0.3 Å, higher flexibility/mobility patterns were assigned for the N-terminal residues as compared to those of the carboxy end ( Figure 9). This was consistent with the MD dynamic behavior of the hMGAM proteins reported by Zhang and his research group [59] as well as the reported conformational stability and B-factor analyses regarding the hMGAM crystalline structures [51,[53][54][55][56]. The latter would highlight the validity of the presented MD simulation study and adopted protocol. Regarding the core protein residues, comparable flexibility/immobility profiles were depicted for both gallotannin derivative and acarbose models across several residue ranges. Notably, the residue ranges along 200-210 and 830-850 showed significantly high mobility profiles for both ligands with ΔRMSF down to −2.6 Å. On the contrary, residues around 650-665 and 800-820 ranges were fewer mobiles (ΔRMSF up to 1.4 Å). On the other hand, trends of less negative and/or more positive ΔRMSF values were assigned for the amino acids in complex with gallotannin derivative relative to those of the simulated acarbose reference ligand. This was most recognized across 55-60 and 105-117 residue ranges where the gallotannin-bound protein residues exhibited highly positive (ΔRMSF ~2.7 Å versus 0.3 Å) or either much less negative ΔRMSF (~−0.4 Å versus −2.6 Å) across the 118-124, and 330-350 residue ranges. The latter findings further highlight the impact of significant acarbose fluctuations on the protein tertiary structure stability.
Notably, the 330-350 residue range originated from the ligand-binding domain (catalytic GH-31) was almost three-fold higher immobility profile assigned for the gallotannin-protein model in relation to that of acarbose. This differential fluctuation/immobility pattern infers the significant role of these residues for gallotannin binding at the hMGAM substrate-binding site. Focusing on the specific flexibility of the pocket's key lining residues in relation to bounded ligand, interesting findings were also illustrated. The simu- For further investigation of the hMGAM's local flexibility and residue-wise contribution within the ligand-protein binding stability, the difference root-mean-square fluctuations (∆RMSF) of the apo hMGAM relative to its holo state (RMSF Apo − Holo ) were monitored along the whole MD runs. Applying the cut-off mobility threshold at ∆RMSF of 0.3 Å, higher flexibility/mobility patterns were assigned for the N-terminal residues as compared to those of the carboxy end ( Figure 9). This was consistent with the MD dynamic behavior of the hMGAM proteins reported by Zhang and his research group [59] as well as the reported conformational stability and B-factor analyses regarding the hMGAM crystalline structures [51,[53][54][55][56]. The latter would highlight the validity of the presented MD simulation study and adopted protocol. Regarding the core protein residues, comparable flexibility/immobility profiles were depicted for both gallotannin derivative and acarbose models across several residue ranges. Notably, the residue ranges along 200-210 and 830-850 showed significantly high mobility profiles for both ligands with ∆RMSF down to −2.6 Å. On the contrary, residues around 650-665 and 800-820 ranges were fewer mobiles (∆RMSF up to 1.4 Å). On the other hand, trends of less negative and/or more positive ∆RMSF values were assigned for the amino acids in complex with gallotannin derivative relative to those of the simulated acarbose reference ligand. This was most recognized across 55-60 and 105-117 residue ranges where the gallotannin-bound protein residues exhibited highly positive (∆RMSF~2.7 Å versus 0.3 Å) or either much less negative ∆RMSF (~−0.4 Å versus −2.6 Å) across the 118-124, and 330-350 residue ranges. The latter findings further highlight the impact of significant acarbose fluctuations on the protein tertiary structure stability.
showed immobility values beyond the cut-off ΔRMSF values only for the gallotannin de-rivative model. Based on the hydrophobic/hydrophilic nature of the depicted immobilized residues, trends of polar residues were dominant for stabilizing the acarbose-protein model, while both polar and hydrophobic amino acids were shown important for gallotannin stability. Finally, the above residue-wise immobility profiles inferred the important role of catalytic GH-31 domain residues, with the assistance of particular residues of the N-terminal β-domain and both inserted loops, for stabilizing the simulated ligands at the hMGAM substrate-binding site.

Ligand-induced loss of mobility
Ligand-induced gain of mobility Figure 9. Protein ΔRMSF analysis, in complex with investigated gallotannin derivative and acarbose, across 200 ns MD simulation run. The ΔRMSF values, in reference to protein backbone Cαatoms, were estimated and represented in terms of residue sequence numbers.  Notably, the 330-350 residue range originated from the ligand-binding domain (catalytic GH-31) was almost three-fold higher immobility profile assigned for the gallotanninprotein model in relation to that of acarbose. This differential fluctuation/immobility pattern infers the significant role of these residues for gallotannin binding at the hMGAM substrate-binding site. Focusing on the specific flexibility of the pocket's key lining residues in relation to bounded ligand, interesting findings were also illustrated. The simulated gallotannin compound depicted a wider residue range of immobility/inflexibility as compared to acarbose ( Table 2). The Pocket's residues; Arg298, Asp327, Asp542, and His600 showed recognized inflexibility profiles (∆RMSF ≥ 0.3 Å) for both ligands which highlight their crucial role in ligand anchoring at −1 sugar subsite. Only Arg5276 showed a significant immobility value at acarbose being higher than at the galloyannin model. On the other hand, residues like Thr211, Phe575, Ala576, Leu577, Gln603, Trp406, and Phe450 showed immobility values beyond the cut-off ∆RMSF values only for the gallotannin derivative model. Based on the hydrophobic/hydrophilic nature of the depicted immobilized residues, trends of polar residues were dominant for stabilizing the acarbose-protein model, while both polar and hydrophobic amino acids were shown important for gallotannin stability. Finally, the above residue-wise immobility profiles inferred the important role of catalytic GH-31 domain residues, with the assistance of particular residues of the N-terminal βdomain and both inserted loops, for stabilizing the simulated ligands at the hMGAM substrate-binding site.

Catalytic GH-31 domain
To investigate the binding affinity of the simulated ligands towards hMGAM as well as grasp the nature of ligand interactions, the binding-free energy was estimated by MDdirected MM/PBSA calculations [60]. This approach can account for more accurate ligandprotein affinity as compared to static or highly sophisticated flexible docking techniques but is computationally less expensive than free energy perturbation [41]. The SASA-only model of the free-binding energy calculation (∆G Total = ∆G Molecular Mechanics + ∆G Polar + ∆G Apolar ), and single trajectory approach, were adopted. Interestingly, a more favorable total binding affinity was estimated for the simulated gallotannin derivative as compared to the reference inhibitor (−69.6 ± 24.9 versus −40.7 ± 22.8 kJ/mol). The latter binding free energy pattern matches the trend from the preliminary docking investigation showing preferential higher docking scores for 1,3,6-trigalloylglucose in relation to acarbose.  Dissecting the obtained ligand-hMGAM binding-free energy into its contributing energy terms (Table 3) has shown a dominant energy contribution of the electrostatic interactions (∆G Electrostatic ) within the free-binding energy calculation for acarbose. This electrostatic energy term was more than five folds higher than that of the van der Waals (∆G Van der Waal ) contributions for the acarbose-protein model. On the other hand, the simulated gallotannin derivative showed balanced electrostatic/van der Waal energy term contributions. It is worth mentioning that these energy findings were consistent with the above described ∆RMSF hydrophobic/polar contact preferentiality. Higher polar solvation energy (∆G Solvation; Polar ) was assigned for acarbose, whereas the non-polar solvation energy (∆G Solvation; SASA ) was higher for the simulated gallotannin derivative. The high ∆G Solvation; Polar depicted with acarbose can be reasoned since the ligand possesses several polar oxygen-related functionalities. While these groups can serve in furnishing polar contact with −1 and 1 sugar subsites, they can act as double blades increasing solvation entropy particularly since the hMGAM pocket is shallow and solventexposed. The latter could compromise the acarbose binding at the target site as binding is considered a solvent-displacement process. On the contrary, the simulated gallotannin derivative was assigned polar and hydrophobic functionalities of comparable magnitude (∆G Electrostatic ≈ ∆G Van der Waal ). The ligand's aromatic functionalities would provide reasonable hydrophobic compensation against the potential solvation entropy. For gaining more insights regarding ligand-residues interactions, the binding-free energy was further decomposed for identifying the key residue-wise binding energy contributions ( Figure 10) [41]. Depicting high binding energy contribution (≥−3.0 kJ/mol), pocket's residues including Tyr299, Asp327, Asp443, Asp542, Asp571, and His600, were considered significant for both ligand-protein complex stability. However, the energy contribution of several hydrophobic pocket residues; Thr211, Tyr299, Trp406, Met444, Phe450, Leu473, Phe575, and Leu577, were only significant for the simulated gallotannin derivative (up to −8.3 kJ/mol). The highest residue-wise energy contribution was assigned for the catalytic Asp542 pocket residue (up to −15.0 kJ/mol and −10.6 kJ/mol for acarbose and gallotannin compounds, respectively). Nevertheless, a few residues like; Arg334, Lys480, and Arg562 in the gallotannin model and Asp334 and Gly604 within the acarbose model, showed sizeable positive energy contribution. The latter infers a repulsion effect and an unfavored impact on the stability of respective ligand-target complexes. mise the acarbose binding at the target site as binding is considered a solvent-displacement process. On the contrary, the simulated gallotannin derivative was assigned polar and hydrophobic functionalities of comparable magnitude (ΔGElectrostatic ≈ ΔGVan der Waal). The ligand's aromatic functionalities would provide reasonable hydrophobic compensation against the potential solvation entropy. For gaining more insights regarding ligand-residues interactions, the binding-free energy was further decomposed for identifying the key residue-wise binding energy contributions ( Figure 10) [41]. Depicting high binding energy contribution (≥−3.0 kJ/mol), pocket's residues including Tyr299, Asp327, Asp443, Asp542, Asp571, and His600, were considered significant for both ligand-protein complex stability. However, the energy contribution of several hydrophobic pocket residues; Thr211, Tyr299, Trp406, Met444, Phe450, Leu473, Phe575, and Leu577, were only significant for the simulated gallotannin derivative (up to −8.3 kJ/mol). The highest residue-wise energy contribution was assigned for the catalytic Asp542 pocket residue (up to −15.0 kJ/mol and −10.6 kJ/mol for acarbose and gallotannin compounds, respectively). Nevertheless, a few residues like; Arg334, Lys480, and Arg562 in the gallotannin model and Asp334 and Gly604 within the acarbose model, showed sizeable positive energy contribution. The latter infers a repulsion effect and an unfavored impact on the stability of respective ligand-target complexes.

Conclusions
Five major phenolic constituents: apeginin-7-O-β-D-glucoside (1), quercetin-3-O-β-Dglucuronic acid (2), quercetin-3-O-β-D-galacturonic acid (3), rutin (4), and 1,3,6-trigalloyl glucose (5) were purified and characterized from C. oblongifolia. Their antioxidant and AGEs formation, α-glucosidase, and lipase inhibitory capacities were estimated. Compound 5 possessed the highest antioxidant and AGEs inhibitory potential. The furnished molecular docking and dynamics simulations illustrated the preferential stability and binding affinity of compound 5 towards human α-glucosidase in relation to its reported potent inhibitor, acarbose. The latter findings suggest the potential therapeutic utility of the C. oblongifolia-isolated compound for managing diabetes mellitus as well as other obesity-related conditions. Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/biology11050762/s1, Figure S1. Pancreatic lipase inhibitory effect of Orlistat; Figure S2. Concentration-response linear response of Ascorbic acid in ABTS assay; Figure S3. Concentration-response linear response of Ascorbic acid in FRAP assay; Figure S4. Concentrationresponse linear response of EDTA in metal chelation assay; Figure S5. Concentration-response linear response of Trolox in ORAC Assay; Figure S6 Table S1: Pancreatic Lipase inhibition assay by isolated compounds from C. oblongifolia and Orlistat; Table S2: Antioxidant activity of isolated compounds.