Identification of new GLUT2-selective inhibitors through in silico ligand screening and validation in eukaryotic expression systems

Glucose is an essential energy source for cells. In humans, its passive diffusion through the cell membrane is facilitated by members of the glucose transporter family (GLUT, SLC2 gene family). GLUT2 transports both glucose and fructose with low affinity and plays a critical role in glucose sensing mechanisms. Alterations in the function or expression of GLUT2 are involved in the Fanconi–Bickel syndrome, diabetes, and cancer. Distinguishing GLUT2 transport in tissues where other GLUTs coexist is challenging due to the low affinity of GLUT2 for glucose and fructose and the scarcity of GLUT-specific modulators. By combining in silico ligand screening of an inward-facing conformation model of GLUT2 and glucose uptake assays in a hexose transporter-deficient yeast strain, in which the GLUT1-5 can be expressed individually, we identified eleven new GLUT2 inhibitors (IC50 ranging from 0.61 to 19.3 µM). Among them, nine were GLUT2-selective, one inhibited GLUT1-4 (pan-Class I GLUT inhibitor), and another inhibited GLUT5 only. All these inhibitors dock to the substrate cavity periphery, close to the large cytosolic loop connecting the two transporter halves, outside the substrate-binding site. The GLUT2 inhibitors described here have various applications; GLUT2-specific inhibitors can serve as tools to examine the pathophysiological role of GLUT2 relative to other GLUTs, the pan-Class I GLUT inhibitor can block glucose entry in cancer cells, and the GLUT2/GLUT5 inhibitor can reduce the intestinal absorption of fructose to combat the harmful effects of a high-fructose diet.


Results
In silico ligand screening against GLUT2 inward-facing conformation structural model. Depending on which side of the cell membrane the substrate cavity opens to, GLUTs have two major conformations captured by the crystal structures of some isoforms and GLUT bacterial homologs 5,[12][13][14][15] . For Class I GLUTs, inward-facing conformations have been determined for GLUT1 12 , and outward-facing conformations for GLUT3 13 ; three-dimensional structures for GLUT2 and GLUT4 are not available. In silico ligand screening requires a structural model for the protein target. Even in the absence of crystal structures, the homology modeling of GLUTs based on the available crystal structures has been used successfully to identify new specific ligands. For instance, in silico ligand screening with a GLUT5 model in the inward-facing conformation, generated based on the bacterial GLUT homolog GlcP Se 5 , produced the first potent and specific GLUT5 inhibitor 17 . The structural model for the GLUT2 inward-facing conformation ( Fig. 1) was modeled based on the crystal structure of GLUT1 (PDB ID: 4PYP) with the Molecular Operating Environment (MOE) software (https:// www. chemc omp. com/). GLUT1 and GLUT2 share 52% and 68% protein sequence identity and similarity, respectively, as determined with Align function in MOE. The docking site, containing the substrate cavity without the substrate binding site, was prepared using OpenEye FRED software (https:// www. eyeso pen. com/). The Chem-Navigator library of over 6 million commercially available compounds was prepared for docking using Omega2 and FRED software (https:// www. eyeso pen. com/). The docking studies were conducted using OpenEye FRED software. Docked compounds were scored using Chemgauss4 scoring function. The compounds docked in sites distinct from that of glucose, closer to the substrate cavity entrance (Fig. 1B). Considering commercial availability and affordability, we purchased 163 out of the top 200 scored compounds for experimental validation.
Screening of the lead candidates in the GLUT2-expressing hxt 0 yeast system. To  www.nature.com/scientificreports/ system that expresses human GLUT2 38 . Similar GLUT-specific hxt 0 yeast systems are available for all Class I GLUTs and GLUT5 38,[40][41][42] , providing a convenient assay platform for these transporters' ligands 16 . For GLUT2, the applied yeast strain EBY.S7 is devoid of all its endogenous hexose transporters (hxt 0 ) and carries the fgy1 mutation 36 in the EFR3 gene, proven to be beneficial for the heterologous expression of human GLUTs 16 . The active expression of the transporter required a GLUT2 version with a truncated loop between transmembrane regions TM1 and TM2 and an additional point mutation (GLUT2 ∆loopS_Q455R ) 38 Fig. S2). Also, reported GLUT2 inhibitors, phloretin and quercetin 36 , inhibited similarly GLUT2 ∆loopS_Q455R 38 , confirming this system's applicability to screening GLUT2 inhibitors. GLUT2 transport activity was determined as previously described 38 . Pre-grown yeast cells were washed and resuspended in PBS buffer to an OD 600nm of ~ 10; 100 µl of this cell suspension constituted the assay mix. Uptake activity of GLUT2 was determined by adding C 14 -hexose (glucose or fructose), quenching after 10 min, filtering the cells, and measuring the radioactivity with a scintillation counter. Initial compound screening for GLUT2 inhibition was performed at 15 mM glucose concentration (i.e., ~ K M ) and 100 µM of each chemical. While none of the tested compounds mediated an increase in glucose uptake activity by GLUT2, several diminished it significantly ( Fig. 2A). Among these, 11 compounds decreased GLUT2 activity by at least 60% and were further examined to determine their respective IC 50 value (Fig. 2B). All compounds are effective inhibitors (IC 50 < 20 µM); for simplicity, we named them G2i (from GLUT2 inhibitor) A-K in the order of decreasing inhibition potency (Table 1, Fig. 2). G2iA showed the strongest GLUT2 inhibition with an IC 50 of 0.61 µM, almost twice as strong as phloretin and five times more potent than quercetin 38 . Effect of GLUT2 inhibitors on the other Class I GLUTs and GLUT5. Establishing the selectivity of GLUT2 inhibitors for other GLUT isoforms, particularly its closely related Class I GLUTs, is crucial for future application of these inhibitors. Often several GLUTs coexist in the same tissue, and being able to modulate selectively an individual GLUT provides a powerful tool in unraveling its pathophysiological role. Therefore, to determine the selectivity of the identified GLUT2 inhibitors, we tested them for their effect on the GLUT homologs GLUT1, 3, 4, and 5. For this, hxt 0 yeast cells actively expressing the respective transporter 38,[40][41][42] were incubated with 100 µM of the tested compound, and the transport activity was assayed in the same manner as for GLUT2 but at substrate concentrations close to the K M in the respective GLUT (i.e., 5 mM glucose for GLUT1 44 and GLUT4 44 , 1.5 mM glucose for GLUT3 43 , 10 mM fructose for GLUT5 41 ) (Fig. 3A). GLUT2 is more closely  Table S1) are identified by the ChemNavigator structure ID. Eleven compounds (designated as G2iA-K, in red, with the corresponding structure ID in bold, underlined font) inhibited GLUT2 relative activity by more than 60% (marked by the dotted line). (B) Dose-response curves for G2iA-G2iK (see also  www.nature.com/scientificreports/ related to the other Class I GLUTs (52-65% sequence identity) than GLUT5 (Class II GLUT, 40% sequence identity) 5 . Nevertheless, most GLUT2 inhibitors seem to have only negligible inhibitory effects on the other GLUTs ( Fig. 3A). Thus, only G2iF inhibits GLUT1, 3, and 4, whereas G2iI decreased just GLUT5 activity (Fig. 3A). However, G2iF IC 50 values were higher for other GLUTs (33 µM for GLUT1, 19 µM for GLUT3 and 14 µM for GLUT4) than for GLUT2 (7 µM); the same was found for the IC 50 of G2iI (23 µM for GLUT5 vs. 13 µM for GLUT2) (Fig. 3B-E). Importantly, all other tested compounds, including the most potent GLUT2 inhibitor G2iA appear not to significantly affect the other GLUTs tested, indicating that these are GLUT2-specific.

Docking sites of GLUT2 inhibitors. The virtual ligand screening showed that all 11 GLUT2 inhibitors
docked to the inward-facing conformation of GLUT2 in sites distinct from that of glucose, closer to the substrate cavity entrance (Fig. 1B). The two most potent GLUT2 inhibitors, G2iA and G2iB, showed noncompetitive inhibition with glucose ( Supplementary Fig. S3), consistent with their binding site being distinct from that of the substrate. Protein-ligand interactions ( Fig. 4 and Supplementary Fig. S4) include hydrogen-bonds with charged residues from the cytosolic loops or transmembrane (TM) helix ends (D120, R124, E178, R181, R185, R244, K249, E279, R280, R432), backbone carbonyls (G177, P433) or polar residues from TM helices (S112, Q193); hydrophobic or Van der Waals interactions (M174, A283, L436); and cation-pi interactions with guanidinium groups (R244, R280). Among these, the residues that are not conserved in GLUT1-5 are D120, K249, R280, A283, W420, and L436 (Fig. 4L). G2iA is oriented in its pocket by hydrophobic interactions with M174 and L436, a hydrogen bond of its amino indole group with the sidechain of S112, a polar interaction of its phenyl fluorine with the R280 guanidinium group, as well as a cation-pi interaction of the fluorophenyl group with the R280 sidechain (Fig. 4A). Hydrophobic interactions with M174 and L436 also contribute to the pockets of G2iB (Fig. 4B), G2iE (Fig. 4E), G2iH (Fig. 4H), and G2iI (Fig. 4I). R280 sidechain makes hydrogen bond interactions with oxygens from the methoxyl group of G2iB (Fig. 4B) or the sulfamide group of G2iG (Fig. 4G), and the sulfurs of the G2iC thienyl group (Fig. 4C) or of the G2E thiazol group (Fig. 4E). It also has cation-pi interaction with the G2iC thienyl group and G2iF quinoline moiety. The α-carbon of A283 comes close (3 Å) to G2iD (Fig. 4D); this inhibitor has Van der Waals interactions with the large cytosolic loop. In G2iF, besides the cation-pi interaction with the quinoline, R244 also makes a hydrogen bond with the ligand's carbonyl, suggesting that positioning of R244 is essential for G2iF recognition.

Discussion
GLUT2 shares characteristic motifs and high similarity with the other Class I representatives GLUTs 1, 3, and 4 7 . In this group, it is the only transporter that also accepts fructose as a substrate 6 . This prompted us to test the identified GLUT2-inhibiting compounds against all Class I GLUTs and the Class II member GLUT5, a fructoseonly transporter, to elucidate their specificity. Except for G2iF, which also inhibits other Class I GLUTs (although to a lesser extent: IC 50 in GLUT2 = 7 µM, IC 50 in other Class I GLUTs ≥ 14 µM), and G2iI which inhibits GLUT5 (IC 50 = 23 µM) less potently than GLUT2 (IC 50 = 13 µM), the other nine GLUT2 ligands did not show a significant effect on the tested transporters (Fig. 3). Although so far unknown effects on other human transporters cannot be ruled out completely, these data indicate that, for the first time, very potent and specific GLUT2 inhibitors were identified. For example, similarity search (Tanimoto > = 0.9, across the entire database, all the other options default) in PubChem for G2iA and G2iB returned 60 compounds with 34 bioactivity records and 409 compounds  A possible explanation for the high prevalence of GLUT2-selective inhibitors among the identified GLUT2 inhibitors may be that the inhibitors target the upper portion of the substrate cavity. All 11 inhibitors dock at the substrate cavity entrance, separately from the glucose binding site (Figs. 1, 4). As noted above, G2iA and G2iB are noncompetitive with glucose, confirming that the inhibitors bind at a different site than the substrate (Supplementary Fig. S3). The substrate cavity base containing the glucose binding site is made up of residues mostly conserved in Class I GLUTs (Supplementary Fig. S1A). The cytosolic entrance of the substrate cavity surrounded www.nature.com/scientificreports/ by soluble loops, especially the large loop connecting the N-and C-domains of the transporter (GLUT2 amino acid residues 240 -298), has much more variability in the protein sequence ( Supplementary Fig. S1). Thus, the binding site location of the GLUT2 inhibitors is consistent with the GLUT2 selectivity exhibited by most of these www.nature.com/scientificreports/ ligands and suggests that targeting the entrance of the substrate cavity, whether in the inward-or outward-facing conformations, for ligand screening may increase the chances of producing GLUT-specific ligands. Also, given the separation between glucose and such inhibitor sites, bridging the two sites by attaching a glucosyl group to these types of inhibitors may substantially improve inhibitor potency while maintaining selectivity. Additionally, such compounds could help to crystallize GLUTs whose structures are yet unknown, including GLUT2, as the combination of substrate and inhibitor would greatly stabilize the transporter conformation. A significant difference in the substrate sites between Class I GLUTs and GLUT5 is W420 GLUT2 (W388 GLUT1 ), conserved in Class I GLUTs but replaced by a smaller residue in Class II GLUTs (e.g., A396 in GLUT5). This substitution creates more space in the substrate cavity, changing the binding mode of ligands and substrate specificity 31 . For instance, GLUT5 A396W mutant became a transporter of both glucose and fructose, while the wild-type can only transport fructose. Therefore, it is likely that GLUT2 inhibitors adopt different binding modes in GLUT5 than those described for GLUT2.
With this study, we present a range of molecules that will serve as valuable tools to investigate the physiological role of GLUT2 in health and disease and may evolve to therapeutic drugs in GLUT2-related diseases. Given their selectivity against other hexose transporters, we believe these compounds could serve as chemical probes for the in-depth study of GLUT2. Indeed, GLUT2 may play a role in several important diseases 27,28,45,46 . It is upregulated in several cancer types like pancreatic, hepatic, micropapillary, or colon cancer 47 . Inhibition of GLUT2 via the non-specific inhibitor phloretin has been shown to diminish tumor growth in colon cancer 48 and hepatocellular carcinoma 49 . The Class I GLUTs 1 and 3 are also overexpressed in many cancer types and related to elevated tumor growth and poor survival 50 . For cancer treatment, the non-specific inhibitor G2iF that inhibits Class I GLUTs but not GLUT5 might join phloretin as a putative drug 49 . Furthermore, substantial overexpression of the fructose transporters GLUT2 and GLUT5 lead to the hypothesis that certain cancer cells use fructose as a preferential carbon source 47 . In these cases, the here presented GLUT2/GLUT5 inhibiting compound G2iI might be a promising candidate in the combat against cancer and other high-fructose diet-related diseases 51 . Importantly, a potent and GLUT2-specific effector (e.g., G2iA) might further elucidate the particular role of GLUT2 in tumor pathogenesis and facilitate studies targeting GLUT2, thereby contributing to unravel complex cancer behavior further.
In healthy individuals, GLUT2 traffics to the apical side of the brush border membrane only after a meal, when glucose concentrations in the lumen are high, to support SGLT1 and accelerate glucose uptake 28 . In morbidly obese humans, a consistent location of GLUT2 at the apical membrane, even in fasting states, was observed and related to insulin resistance 52 . This might result in higher glucose levels in the lumen in fasting states and an abnormal sugar supply could support bacterial growth which interferes with a healthy gut microbiome 52 . Specific inhibition of GLUT2 could mitigate such pathologies. An altered microbiome composition in mice with intestinal-specific GLUT2 deletion has been detected in previous studies 53 , supporting the gut microbiome as a possible field of application for GLUT2 inhibitors. Also, Schmitt et al. showed that GLUT2 deletion in the murine intestine causes favorable effects like improved glucose tolerance and diminished body weight gain 53 . This suggests that GLUT2 tailored inhibitors could lead to similar results and might be applied in morbidly obese patients or type 2 diabetic persons with beneficial health effects.
Interestingly, viral infections affect the expression of GLUT2. While the hepatitis C virus downregulates GLUT2 expression 54 , the transmissible gastroenteritis virus upregulates the transporter's expression, enhancing intestinal glucose absorption, which promotes viral replication 55 . Hence, GLUT2 inhibition could assist in the containment of certain viruses. Clearly, the role of GLUT2 in the metabolic processes is highly complex and not fully understood. Therefore, the application of GLUT2-specific inhibitors also bears high risks as it might have not only beneficial but also adverse effects, and more studies are necessary to increase our level of knowledge. However, accessibility of specific GLUT2 inhibitors represents a tremendous advantage over less-specific GLUT inhibitors in developing drugs with a defined effective spectrum and lower side effects.
These compounds are valuable tools in the efforts of answering many open questions concerning GLUT2. For instance, it is still unclear how GLUT2 is mobilized in response to glucose in various cell types and different pathologies 56 . Possible players include the type of membrane lipids 57 , protein partners 56,58 , or glycosylation 3 . Distinct from other GLUTs, the extraordinary low affinity for glucose and fructose probably assigns special functions of glucose sensing 59 and signaling 21 to GLUT2, but the detailed molecular functions remain to be elucidated. Furthermore, the Fanconi-Bickel syndrome due to GLUT2 malfunction 27 has various symptoms that indicate yet undiscovered physiological roles for GLUT2, and the transporter's role in certain cancer types remains unclear 47 . Future studies will benefit from the existence of a range of easily accessible GLUT2-specific inhibitors with varying affinities. In silico ligand screening. GLUT2 homology models were build using Molecular Operating Environment (MOE) software (www. chemc omp. com). Based on sequence alignment between GLUT1 and GLUT2, with the crystal structure for GLUT1 inward-facing conformation (PDB ID: 4PYP) as a template, the initial model geometry was generated, followed by refinement of the sidechains and energy minimization with the MMFF94x force field. The model with the lowest interaction energy and RMSD was selected for docking studies. Molecular probing of inner cavities was done to identify potential binding sites. Two sites of interest were identified in the proximity of both ends of the transmembrane regions and used for receptor preparation with OpenEye FRED software 60 (https:// www. eyeso pen. com). ChemNavigator collection (MilliporeSigma, St. Louis, MO, USA) of commercially available compounds (~ 6 million) was processed for docking studies using the following protocol: (i) remove all compounds that are not small organic molecules, (ii) remove salts counterions, (iii) normalize charges and select the most likely tautomer at pH 7, (iv) generate an ensemble of up to 400 molecular conformers for each compound using Omega2 software (https:// www. eyeso pen. com).
After completing the preparation steps, the virtual docking screen was performed with OpenEye FRED software on a Linux cluster. All conformer ensembles were docked into the selected sites described above, retaining only the best scoring pose based on the Chemgauss4 score for each compound. The top 200 best scoring compounds were extracted and selected for purchase and experimental validation. Due to availability and affordability issues, only 163 compounds were sourced and submitted for experimental validation.
GLUT transport assay. Commercial providers for chemicals tested for GLUT2 inhibition are listed in Supplementary Table S1. C 14 -fructose and -glucose were from Moravek Inc (Brea, CA, USA). For transport activity assay, cells in the hexose media were centrifuged (1000×g, 5 min, room temperature), washed once with PBS buffer (10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , 2.7 mM KCl, 137 mM NaCl, pH 7.4), and resuspended in PBS buffer at an OD 600nm ~ 10; each assay contained 100 µl of this cell solution. The transport activity assay was started by adding C 14 -hexose (5 mM glucose for GLUT1 or GLUT4, 1.5 mM glucose for GLUT3, 10 mM glucose for GLUT2, and 10 mM fructose for GLUT5). When determining the K M for fructose and glucose in GLUT2, substrate concentrations were varied accordingly. Transport activity assay was stopped after 10 min by adding 3 ml ice-chilled Quench buffer (0.1 M KPi, 0.1 M LiCl, pH 5.5), followed by filtration through a glass fiber channel (GC50; Advantec, Tokyo, Japan) under vacuum and another wash with 3 ml Quench buffer and filtration. The filtration membranes were transferred into scintillation vials, combined with 10 ml of Scintillation Solution (BioSafeII; Research Products International, Mount Prospect, IL, USA), and vortexed briefly. The radioactivity was determined with a scintillation counter (Tri-carb 2900TR, Perkin Elmer, USA). The compounds were dissolved in dimethyl sulfoxide (DMSO) at 100× (i.e., ~ 10 mM) the final assay concentration. Controls for determining the relative transport activity included 1% (v/v) DMSO, representing the normal GLUT2 activity (100%), and known inhibitors 200 µM phloretin for GLUT1-4 38,44 , and 100 µM N-[4-(methylsulfonyl)-2-nitrophenyl]-1,3-benzodioxol-5-amine (MSNBA) for GLUT5 17 , representing fully inhibited activity. Primary screening was done at 100 µM compound concentration (see Supplementary Table S1 for a list of all tested compounds). The IC 50 values were further determined for the compounds that diminished the relative transport activity by at least 60%. When determining the inhibition mode for the most potent GLUT2 inhibitors (IC 50 < 2 µM), G2iA and G2iB, transport activity at 7, 15, and 30 mM glucose concentrations were determined in the absence or presence of different inhibitor concentrations (0, 0.66 and 2 µM for G2iA or 0, 1.66, 5 µM for G2iB; Supplementary Fig.  S3). Data were analyzed with GraphPad Prism (San Diego, CA, USA).