Red wine and green tea flavonoids are cis-allosteric activators and competitive inhibitors of glucose transporter 1 (GLUT1)-mediated sugar uptake

The antioxidant- and flavonoid-rich contents of red wine and green tea are reported to offer protection against cancer, cardiovascular disease, and diabetes. Some studies, however, show that flavonoids inhibit GLUT1-mediated, facilitative glucose transport, raising the possibility that their interaction with GLUT1 and subsequent downstream effects on carbohydrate metabolism may also impact health. The present study explores the structure–function relationships of flavonoid–GLUT1 interactions. We find that low concentrations of flavonoids act as cis-allosteric activators of sugar uptake, whereas higher concentrations competitively inhibit sugar uptake and noncompetitively inhibit sugar exit. Studies with heterologously expressed human GLUT1, -3, or -4 reveal that quercetin–GLUT1 and –GLUT4 interactions are stronger than quercetin–GLUT3 interactions, that epicatechin gallate (ECG) is more selective for GLUT1, and that epigallocatechin gallate (EGCG) is less GLUT isoform–selective. Docking studies suggest that only one flavonoid can bind to GLUT1 at any instant, but sugar transport and ligand-binding studies indicate that human erythrocyte GLUT1 can bind at least two flavonoid molecules simultaneously. Quercetin and EGCG are each characterized by positive, cooperative binding, whereas ECG shows negative cooperative binding. These findings support recent studies suggesting that GLUT1 forms an oligomeric complex of interacting, allosteric, alternating access transporters. We discuss how modulation of facilitative glucose transporters could contribute to the protective actions of the flavonoids against diabetes and Alzheimer's disease.

efits of flavonoids may indeed derive from their antioxidant capacities, future studies should also consider how the dual actions of flavonoids on glucose transport may influence downstream glucose metabolism in tumors, in insulin-secreting and responsive tissues, and in the CNS.
We determined the sidedness of the action of flavonoids on GLUT1 by examining their effects on two modes of red cell sugar transport: 1) zero-trans 3MG uptake (influx into sugarfree cells) and 2) zero-trans 3MG exit (efflux from sugar-loaded cells into medium lacking sugar). Transport theory informs us (26,27) that a ligand competing with sugar for binding at the exofacial sugar binding site serves as a competitive inhibitor of sugar uptake and as a noncompetitive inhibitor of sugar exit. Conversely, a ligand competing with sugar for binding at the endofacial sugar binding site serves as a noncompetitive inhibitor of sugar uptake and as a competitive inhibitor of exit.
The effects of the flavonoids on the concentration dependence of initial rates of 3MG uptake are shown in Fig. 2A. Using   Table 1. Inset, Lineweaver-Burk transformation of the data. The lines were computed by linear regression, and the results are also summarized in Table 1

Flavonoids are exofacial GLUT1 ligands
the extra sum of squares F-test (28) to test the null hypothesis that 3MG uptake is described equally well by nonsaturable sugar uptake (uptake ϭ k ϫ [S]), by Michaelis-Menten uptake (Equation 2), and by Michaelis-Menten uptake plus nonsaturable uptake fails for all conditions. All uptakes are best described by simple Michaelis-Menten kinetics. Analysis of 3MG uptakes by nonlinear regression analysis using Equation 2 reveals that the flavonoids increase K m(app) for sugar uptake from 2.39 Ϯ 0.36 to 11.07 Ϯ 5.03 mM (quercetin), 10.64 Ϯ 2.63 mM (EGCG), and 7.14 Ϯ 2.63 mM (ECG), without significantly affecting V max for uptake (1.202 Ϯ 0.082 mmol/liter cell water/ min; Fig. 2A; Table 1). Lineweaver-Burk analysis results in lower estimates of V max and K m(app) ( Fig. 2A (inset) and Table  1), but the same conclusion (V max is unchanged but K m(app) is increased). This suggests that quercetin, EGCG, and ECG inhibit GLUT1-mediated sugar uptake by binding at the exofacial 3MG binding site or at a site whose occupancy is mutually exclusive with 3MG occupancy of the exofacial sugar binding site. Consistent with this idea, the flavonoids act as noncompetitive inhibitors of net 3MG exit; they are without effect on K m(app) (12.9 mM) for exit, but they decrease V max for exit by more than 2-fold from 2.03 mmol/liter cell water/min to 0.89 (quercetin), 0.72 (EGCG), and 0.60 mmol/liter cell water/min (ECG; Fig. 2B).
CB is a membrane-permeant GLUT1 inhibitor that binds at or close to the endofacial glucose-binding site (26,27). Extracellular maltose, but not glucose, inhibits equilibrium binding of the CB to the endofacial glucose-binding site of GLUT1 (29,31,32). We therefore asked whether the flavonoids, which appear to act as exofacial ligands, also interfere with GLUT1 equilibrium [ 4A). Whereas these curve fits produce good correlation coefficients (R 2 Ͼ 0.91 in all cases), the S.D. of the residuals of each fit (Ն0.09; excluding CB) is greater than 20% of the S.D. of the y values, suggesting that the fits are poor.
Closer examination reveals that inhibition by quercetin and EGCG increases more steeply, whereas inhibition produced by ECG increases less steeply than is expected for simple Michaelis-Menten inhibition. Inhibition of radiolabeled CB binding by unlabeled CB is well-described by simple Michaelis-Menten inhibition. We therefore asked if quercetin, EGCG, and ECG inhibitions of CB binding are better approximated by inhibition involving multiple cooperative ligand-binding sites and applied a simple Hill-type model (33) to analyze these results (Fig. 4B). This analysis produces fits with residuals that do not deviate significantly from zero, and the S.D. of the resid- Table 1 Analysis of flavonoid inhibition of 3MG uptake The data of Fig. 2A were analyzed by nonlinear regression analysis assuming Michaelis-Menten kinetics (Equation 2) or by Lineweaver-Burk analysis, which also assumes Michaelis-Menten kinetics ( Fig. 2A, inset). The resulting parameters (V m and K m(app) for 3MG uptake) are shown as mean Ϯ S.D. of the analysis. The fit statistics (Sy.x and R 2 ) are also shown.  To further test the effects of flavonoids on CB binding to erythrocyte GLUT1, we measured the concentration dependence of CB inhibition of 3MG (0.1 mM) uptake with and without flavonoids. The presence of quercetin (2 M), EGCG (20 M), or ECG (5 M) inhibits basal sugar uptake (uptake in the absence of CB) and increases K i(app) for CB inhibition of 3MG uptake by at least 2.5fold (Fig. 4C). Assuming simple competition between the flavonoids and CB for binding to GLUT1 (but see Fig. 4B), the computed These results indicate that exofacial inhibitors impair CB binding to the GLUT1 endofacial sugar-binding site and thereby reduce the potency of CB inhibition of sugar transport.
Incubation We investigated potential interactions of ␤-D-glucose, quercetin, EGCG, and ECG with the exofacial sugar-binding site by molecular docking using the homology-modeled GLUT1 outward-open structure (GLUT1-e2 ( (18)). The exofacial, interstitium-exposed cavity of GLUT1-e2 presents three potential ␤-D-glucose docking sites: peripheral, intermediate, and core (18,20). Benzene ring A ( The residues contributing to ␤-D-glucose docking at core, intermediate, and peripheral ␤-D-glucose sites have been described previously (20). Neuronal GLUT3 and insulin-sensitive GLUT4 share 93% and 85% sequence similarity with GLUT1, respectively (41). We Table 2 Inhibition of GLUT1-, GLUT3-, and GLUT4-mediated 2DG uptake in HEK293 cells by quercetin, EGCG, and ECG The data of Fig. 3 (3MG uptake at 4°C) and Fig. 7 (2DG uptake at 37°C) were analyzed by nonlinear regression, assuming that uptake inhibition is described by Equation 1 (to yield IC 50 ) or by Equation 3 (to yield Const [1][2][3][4]  Flavonoids are exofacial GLUT1 ligands therefore asked whether quercetin, EGCG, and ECG inhibitions of GLUT3 and GLUT4 resemble their inhibition of GLUT1 or if these inhibitors present an unanticipated selectivity toward these proteins. We transiently expressed hGLUT1, hGLUT3, or hGLUT4 into HEK293 cells and assayed for dosedependent inhibition of 100 M 2-deoxy-D-glucose (2DG) uptake at 37°C. Our previous studies have shown that heterologous expression of hGLUT1, hGLUT3, or hGLUT4 suppresses expression of endogenous GLUT3 message and protein in HEK293 cells (20). All experiments were paired, and relative 2DG uptake (v i /v c ) by HEK293 cells was expressed as a function of [flavonoid]. Over the range of concentrations used, quercetin and EGCG inhibit GLUT1-, GLUT3-, and GLUT4-mediated 2DG uptake in HEK293 cells (Fig. 7, A and B and Table 2). ECG has a bimodal effect on 2DG uptake, first stimulating and then inhibiting transport ( Fig. 7C; Table 2). The curves drawn through the hGLUT1 data (dashed lines) were computed using Equation 3 and the parameters computed for flavonoid inhibitions of RBC-mediated 3MG transport shown in Fig. 3. Notwithstanding the limited dose-response range of Fig. 7A, the use of a different substrate (2DG versus 3MG), and the elevated assay temperature (37 versus 4°C) where the flavonoids are susceptible to more rapid decomposition (42), the agreement between flavonoid inhibitions of heterologously expressed hGLUT1 and RBC-resident hGLUT1 is strong. The curves drawn through the hGLUT3 and hGLUT4 data (solid lines) were computed by nonlinear regression using Equation 3. Table  2 summarizes our findings. Quercetin shows higher affinity for GLUT1 and GLUT4 than it does for GLUT3. EGCG displays similar avidity for GLUT1, -3, and -4. Relative to quercetin and EGCG, ECG has lower affinity for GLUT1, -3, and -4. Docking of quercetin, EGCG, and ECG to exofacial GLUT3 and GLUT4 reveals that these ligands coordinate with equivalent residues when docked to the same GLUT isoform but differ significantly when their interactions are compared across GLUT1, GLUT3, and GLUT4 (Fig. 8).

Discussion
Red wine and green tea flavonoids inhibit the facilitative glucose transporter, GLUT1, by interacting at its exofacial sugarbinding site. Quercetin, EGCG, and ECG competitively inhibit

Flavonoids are exofacial GLUT1 ligands
net sugar uptake by human erythrocytes but are noncompetitive inhibitors of red cell sugar exit. Molecular docking studies using homology-modeled GLUT1 reveal that quercetin, EGCG, ECG, and D-glucose share overlapping interaction sites in the exofacial ligand-binding cavity of GLUT1. Whereas docking studies suggest that only one flavonoid can bind to GLUT1 at any instant, sugar transport and ligand binding studies indicate that the erythrocyte sugar transporter can bind at least two flavonoid molecules simultaneously.
The sidedness of flavonoid action (exofacial) is compatible with previous reports suggesting that quercetin (21, 24) and EGCG (16) act as competitive inhibitors of GLUT1-mediated sugar uptake. Consistent with this conclusion, molecular docking analysis suggests that quercetin, EGCG, and ECG share overlapping interaction envelopes in the exofacial ligand binding cavity of GLUT1-e2, including the previously defined core ␤-D-glucose interaction site (20) (Fig. 6). Docking analysis also suggests that flavonoid binding is coordinated by the side chains of amino acids that form the previously defined exofacial intermediate and peripheral ␤-D-glucose interaction sites. The validity of our docking analysis is bolstered by the finding that the coordination of ␤-D-glucose at the core site in homologymodeled GLUT1-e2 involves the same amino acid residues coordinating ␤-D-glucose binding in the ␤-D-glucosehuman

Flavonoids are exofacial GLUT1 ligands
GLUT3-e2 crystal complex (42). Core ␤-D-glucose, quercetin, EGCG, and ECG all form hydrophobic interactions with GLUT1 Ile-164, Val-165, Ile-168, Phe-291, and Phe-379 and form a hydrogen bond with Glu-380, suggesting that these residues are integral to ligand binding in the exofacial cavity. The chemical structures of EGCG and ECG are almost identical (the one difference being the addition of a hydroxyl group at C3Ј in EGCG), yet our docking analysis reveals no striking differences in EGCG and ECG coordination to GLUT1-e2. Notwithstanding, EGCG inhibits GLUT1 with 5-fold lower affinity than ECG, illustrating the limitations of this docking analysis.
Prior equilibrium ligand-binding studies demonstrate that extracellular maltose inhibits CB binding at the GLUT1 endofacial sugar binding site (27,29). Here, we show that exofacial quercetin, EGCG, and ECG have the same effect. K i(app) for inhibition of CB binding is 1.6 -3-fold lower than K i(app) for transport inhibition. This is not explained by competition between inhibitor and transported sugar for binding at the exofacial site because GLUT1 saturation by 0.1 mM 3MG is less than 5% (K m(app) for 3MG ϭ 2.4 mM; see Fig. 2A). Closer review of the curve fit statistics for flavonoid inhibition of CB binding indicates that the fits to the simple inhibition model are poor. Fitting the data to a Hill model comprising multiple cooperative binding sites (33) (Equation 5) produces significantly better fits for quercetin, EGCG, or ECG but not for CB. Moreover, using the extra sum of squares F-test (28) to test the null hypothesis that the simple inhibition and Hill models provide equally good fits of the binding data fails for quercetin, EGCG, or ECG. These results suggest that GLUT1 CB binding is affected by at least two positively cooperative binding sites each for quercetin and EGCG and two or more (possibly negatively cooperative) binding sites for ECG (32,33). If multiple exofacial flavonoid-binding sites exist, docking analysis suggests that they are unlikely to co-exist within the same GLUT1 molecule. Rather, they must be present in adjacent molecules that interact with each cytochalasin B-binding GLUT1 protein.
We also considered that flavonoids compete directly with CB for binding at the endofacial sugar-binding site. This would explain their displacement of CB from GLUT1 and would also be consistent with demonstrations of cooperativity between endofacial ligand-binding sites present in adjacent GLUT1 subunits of the GLUT1 tetramer (29,30,44,45). Molecular docking of CB and quercetin to the endofacial orientation of GLUT1 (GLUT1-e1) (46) supports the hypothesis that CB and quercetin could compete for high-affinity binding to GLUT1 (not shown). However, inhibition of transport by a molecule that can bind with equal avidity to both exo-and endofacial sugar binding sites should noncompetitively inhibit both net uptake and net exit (26,27). Because flavonoid inhibition of net uptake is competitive, this possibility is refuted. If an inhibitor were to bind with significantly lower affinity to the endofacial binding site versus the exofacial site, it would not only increase K m(app) for net sugar uptake but also reduce V max for uptake by Ͼ65% (26,27), and this is not observed.

Flavonoids are exofacial GLUT1 ligands
GLUT1-mediated sugar uptake at low inhibitor concentrations and then inhibit transport as their concentration is raised. This reinforces the idea that at least two flavonoid-binding sites modulate GLUT1 function. Two possibilities exist: 1) one of these sites is presented by GLUT1 and the second by a non-GLUT1 but nevertheless GLUT1-interacting protein, or 2) the Residues within a 4-Å distance to docking ligand are shown for quercetin (A), EGCG (B), and ECG (C). Residues and contacts are shown with the following color code: blue, polar; green, hydrophobic; orange, negatively charged; violet, positively charged; purple arrow, hydrogen bond and its directionality; green arrow, -stacking; gray-shaded circles, solvent-exposed regions of ligands.

Flavonoids are exofacial GLUT1 ligands
flavonoid-binding sites exist on individual GLUT1 molecules whose adjacency in the GLUT1 homotetramer (20,29,30) results in cooperative interactions. In both instances, flavonoid binding at the first, high-affinity site stimulates sugar uptake by flavonoid-free GLUT1 molecules. As flavonoid concentration is raised, ligand and sugar now compete for binding at the GLUT1 exofacial sugar-binding site, and transport is inhibited. Whereas the former hypothesis cannot be eliminated by the current study, our previous ligand-binding (18, 20, 27, 29 -32, 43), hydrodynamic analysis (44,47), biochemical cross-linking (48), freeze-fracture EM (47), and co-immunoprecipitation (49) studies of membrane-resident and purified GLUT1 support the latter hypothesis.
Studies with heterologously expressed GLUT1, GLUT3, and GLUT4 indicate that quercetin inhibits GLUT1 and GLUT4 with comparable avidity but is up to 9-fold less potent against GLUT3. EGCG shows similar inhibitory potency toward GLUT1, GLUT3, and GLUT4. ECG appears to be only a very poor inhibitor of sugar uptake in GLUT1-, GLUT3-, or GLUT4-expressing HEK293 cells, but this conclusion should be tempered by the ligand's reported instability at 37°C (42).
GLUT1 inhibition at high flavonoid concentrations could explain the anti-cancer action of flavonoids. Small molecule inhibition of cellular sugar transport results in a cascade of downstream events, including down-regulation of glycolytic enzymes, cell cycle arrest, and, ultimately, cell death (49). GLUT1 stimulation by low concentrations of flavonoids could explain their protective actions against diseases such as diabetes (50 -53) and neurodegenerative diseases (54 -57), where enhanced cellular glucose uptake could be ameliorative. However, flavonoids also directly or indirectly modulate other cellular targets, including mitogen-activated protein kinase, cyclin-dependent kinases, HIF-1␣, and vimentin (14). Whereas some studies suggest that flavonoids may not readily cross the plasma membrane (58), others suggest that flavonoids enter cells via carrier proteins, including SGLT1 (37,38,59,60), MCTs (39), GLUT1 (15), and GLUT4 (40), or via transbilayer diffusion (35,36). The current study indicates that, whereas quercetin may cross the plasma membrane, it does so via a CB-insensitive, GLUT1-independent pathway.
Whereas some health benefits of the flavonoids may derive from their antioxidant capacities, the dual actions of flavonoids on glucose transport (interaction with the exofacial sugar-binding site and at regulatory sites that modulate sugar transport) present new tools to explore glucose transporter function and downstream glucose metabolism in tumors, in insulin-secreting and -responsive tissues, and in the central nervous system.

Cells
De-identified whole human blood was purchased from Biological Specialty Corp. (Colmar, PA). HEK293 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 100 units/ml penicillin, and 100 g/ml streptomycin in a 37°C humidified 5% CO 2 incubator.

Heterologous expression of GLUTs
Heterologous expression of hGLUT1, hGLUT3, and hGLUT4 in HEK293 cells was as described previously (49,61). Both hGLUT1 and hGLUT4 contain a Myc epitope in exofacial loop 1 (61), whereas the 13 C-terminal amino acids in hGLUT3 are replaced by the corresponding residues of hGLUT4 (49) to facilitate detection of heterologously expressed transporter.

Red blood cell sugar transport measurements
All human erythrocyte sugar transport experiments were performed at 4°C as described previously (20,49). Red blood cells were isolated from whole blood and glucose-depleted as described previously (62). Sugar transport and ligand-binding experiments reported below typically involve the addition of 2-10 volumes of uptake or ligand-binding medium with or without inhibitor to 1 volume of a 50% suspension of red cells. Our measurements show that preincubating RBCs with increasing volumes of medium containing the test inhibitor progressively reduces K i(app) for inhibition of sugar transport or cytochalasin B binding. The explanation (44) is that the very high GLUT1 content of RBCs depletes [inhibitor] free as it interacts with GLUT1, resulting in an [inhibitor] free Յ [inhibitor] total at the time of transport or ligand-binding assay. The most practical solution to this problem is to preincubate RBCs with an excess volume of uptake or ligand binding medium lacking sugar or CB but containing the inhibitor at the requisite concentration. We observe that K i(app) for test compound inhibition of sugar transport or CB binding approaches its minimum asymptote when preincubation conditions are 1 volume of RBCs to 50 -400 volumes of preincubation assay medium (the necessary dilution falls with increasing [inhibitor] total ). We therefore preincubated RBCs with 50 -400 volumes of assay medium prior to centrifugation and resuspension in uptake or ligand binding medium to ensure optimal equilibration of inhibitor with GLUT1 before performing transport or ligandbinding measurements. Flavonoids are exofacial GLUT1 ligands adding 10 volumes (100 l) of uptake medium with or without inhibitor to 1 volume (10 l) of sugar-depleted, 50% hematocrit red cells, and sugar uptake was allowed to proceed for 30 -60 s at 4°C. Uptake was stopped by adding 50 volumes (1 ml) of ice-cold stop solution containing 50 M WZB117 and 100 M phloretin. Cells were washed one more time in stop solution and lysed in 3% perchloric acid, and radioactivity was assayed in clarified lysates using liquid scintillation counting. Radioactivity measurements were done in duplicates.

Zero-trans exit
Glucose-depleted, packed RBCs were loaded with 10 mM 3MG by incubating 1 volume of cells with 20 volumes of 20 mM 3MG (containing 1 Ci of [ 3 H]3MG/ml of cold 3MG) for 1 h at 37°C. Immediately following 3MG loading, cells were transferred to 4°C and preincubated with or without inhibitors for 10 -15 min. Cell suspension were spun at 10,000 ϫ g for 1 min, and supernatant was discarded. One volume (0.5 ml) of sugarloaded RBCs were added to 50 volumes of KCl medium with or without inhibitor on a shaker with magnetic stirrer. Aliquots (0.5 ml) of the suspension were withdrawn at the indicated time intervals and immediately added to 1 ml of ice-cold stop solution. Cells were washed again in stop solution, lysed in 3% perchloric acid, and assayed in duplicate for radioactivity.

Equilibrium CB binding
CB binding to human red cells was performed as described previously (20,31). Briefly, 50 l of sugar-depleted RBC (50% hematocrit) with or without inhibitors were mixed with 50 l of ice-cold KCl medium containing 40 nM [ 3 H]CB and 10 M cytochalasin D for 15 min at 4°C, with constant end-over-end rotation. Total [CB] was obtained from 2 ϫ 10 l of the cell suspension lysed in 100 l of 3% perchloric acid, and radioactivity was assayed by liquid scintillation counting. To obtain free [CB], cell suspension was centrifuged at 10,000 ϫ g for 30 s, and 2 ϫ 10 l of clarified supernatant were assayed for radioactivity. Bound [CB] was calculated as total [CB] Ϫ free [CB].

Homology modeling
The homology models of the outward-open (e2) conformations of GLUT1 and GLUT4 were generated using the maltosebound human GLUT3 structure (Protein Data Bank code 4ZWC) (42). Maltose was removed from the GLUT3 structure, and chain A was used as the template for modeled structures. Sequence alignments were generated using ClustalX (64). Homology models were built using Modeler version 9.9 (65) and analyzed using PROCHECK (66).

Stochastic docking
The crystal structure of outward-open hGLUT3-e2 (4ZWC) (42) was obtained from the Protein Data Bank. The structures for ␤-D-glucose, quercetin, EGCG, and ECG were obtained from Pubchem (https://pubchem.ncbi.nlm.nih.gov). Docking was performed using the Schrödinger software suite. The protein structure was preprocessed with the Protein Preparation Wizard, bond orders were assigned, hydrogens were added, and the hydrogen bond network was optimized. The system was energy-minimized using the OPLS 2005 force field. Ligand structures were prepared with the LigPrep module, and the pK a of the ligands was calculated using the Epik module. Molecular docking was performed by the GLIDE module in standard precision (SP) mode using default values for grid generation. Cavities for docking were calculated using the CastP server (http:// sts.bioe.uic.edu/castp/) 3 (67), and the grid was centered on the residues forming the cavity. No restraints were used during the docking.

Data analysis
Linear and nonlinear regression analysis of data sets and statistical tests were performed using GraphPad Prism version 7.0a (GraphPad Software, Inc., La Jolla, CA).
Michaelis-Menten inhibition of sugar transport is assumed to be described by Equation 1, where v c is v measured in the absence of inhibitor I, [I] is the concentration of inhibitor, and K i(app) is that [I] producing 50% inhibition of uptake.
Michaelis-Menten transport is assumed to be described by Equation 2, where V max is the maximum rate of 3MG transport, [3MG] is the concentration of 3MG, and K m(app) is the [3MG] where the rate of uptake is V max /2. Sugar exit was analyzed by nonlinear regression analysis using Mathematica version 10.4.1.0 (Wolfram Research), assuming that exit follows Michaelis-Menten kinetics and that the first derivative of the exit progress curve represents d[S]/dt at any given [3MG] (20).
Transport stimulation followed by inhibition by inhibitors was approximated first by normalizing all uptake to v c and then using the following model, where v c is uptake measured in the absence of inhibitor I, v i is uptake measured in the presence of inhibitor, [I] is the concentration of inhibitor, and Const 1 through Const 4 are model-dependent (19 CB binding was also analyzed using the Hill equation for inhibition of equilibrium binding in which the transporter is allowed to bind more than one molecule of competing ligand, I,