Novel natural and synthetic inhibitors of solute carriers SGLT1 and SGLT2

Abstract Selective analogs of the natural glycoside phloridzin are marketed drugs that reduce hyperglycemia in diabetes by inhibiting the active sodium glucose cotransporter SGLT2 in the kidneys. In addition, intestinal SGLT1 is now recognized as a target for glycemic control. To expand available type 2 diabetes remedies, we aimed to find novel SGLT1 inhibitors beyond the chemical space of glycosides. We screened a bioactive compound library for SGLT1 inhibitors and tested primary hits and additional structurally similar molecules on SGLT1 and SGLT2 (SGLT1/2). Novel SGLT1/2 inhibitors were discovered in separate chemical clusters of natural and synthetic compounds. These have IC50‐values in the 10‐100 μmol/L range. The most potent identified novel inhibitors from different chemical clusters are (SGLT1‐IC50 Mean ± SD, SGLT2‐IC50 Mean ± SD): (+)‐pteryxin (12 ± 2 μmol/L, 9 ± 4 μmol/L), (+)‐ε‐viniferin (58 ± 18 μmol/L, 110 μmol/L), quinidine (62 μmol/L, 56 μmol/L), cloperastine (9 ± 3 μmol/L, 9 ± 7 μmol/L), bepridil (10 ± 5 μmol/L, 14 ± 12 μmol/L), trihexyphenidyl (12 ± 1 μmol/L, 20 ± 13 μmol/L) and bupivacaine (23 ± 14 μmol/L, 43 ± 29 μmol/L). The discovered natural inhibitors may be further investigated as new potential (prophylactic) agents for controlling dietary glucose uptake. The new diverse structure activity data can provide a starting point for the optimization of novel SGLT1/2 inhibitors and support the development of virtual SGLT1/2 inhibitor screening models.


| INTRODUC TI ON
Type 2 Diabetes Mellitus (T2DM) is a metabolic disease characterized by prolonged hyperglycemia. In healthy individuals, insulin signaling controls blood glucose levels by inducing tissues to absorb glucose from the circulation. In T2DM, peripheral tissues and the liver gradually become insensitive to insulin. The resulting hyperglycemia can induce a range of pathologies, in particular cardiovascular disease. T2DM is a growing pandemic, 1 and there is a need for alternatives to insulin therapy. Powerful glucose-lowering drugs are available for T2DM. 2 However, these can cause hypoglycemia resulting in considerable morbidity and mortality. 3 Glycemic control therapies should therefore prevent both hyperglycemia and hypoglycemia. Accordingly, inhibitors of the Sodium Glucose Linked Transporters type 1 and 2 (SGLT1 and SGLT2) have emerged as insulin independent antihyperglycemic remedies. [4][5][6][7] SGLTs are members of the superfamily of Solute Carrier transporter proteins (SLC) and are cotransporters of monosaccharides that rely on the concentration gradient of sodium ions to transport molecules across cell membranes. SGLT1 is a high-affinity, low-capacity transporter that absorbs D-glucose in the small intestine.
SGLT2 has a lower affinity, but higher capacity for D-glucose and is expressed in the kidney proximal glomerular tubules where it reabsorbs ≥90% of glucose from the primary urine. 2,8 The natural glucosylated-dihydrochalcone phloridzin is a recognized SGLT inhibitor, [9][10][11] and its glucosuric effect has long been known. 12 Phloridzin was shown to correct hyperglycemia in rats. 13,14 Oral intake of phloridzin containing apple extracts caused blood glucose lowering and glucosuria in animal and human studies, but phloridzin was not detected in plasma or urine and the glucuronide metabolites of its aglycon phloretin positively correlated with glucosuria. 15,16 Thus, the hypoglycemic action of phloridzin involves metabolic and systemic effects that are not entirely understood. Furthermore, phloretin was shown to have undesirable bioactivities in vitro, including inhibition of various transporters, inhibition of oxidative phosphorylation and estrogenic activity. 17 SGLT2 selective, rapidly absorbed phloridzin analogs were synthesized to prevent discomfort from high colonic glucose due to intestinal SGLT1 inhibition. [18][19][20] These gliflozins efficaciously and safely reduced hyperglycemia in humans. 21 While the glucosuria resulting from taking these drugs prevented weight gain or even induced weight loss in T2DM-patients, 6 it can lead to urinary tract infections. 21 Moreover, SGLT2 inhibitors were shown to have a lower efficacy in individuals with impaired renal function. 22 As 40% of T2DM-patients have some degree of nephropathy, these drugs could have a reduced efficacy in individuals who would benefit the most. 7 Alternatively targeting intestinal SGLT1 to inhibit glucose absorption has therefore gained considerable interest. 4,23,24 The dual SGLT1/SGLT2 inhibitor sotagliflozin has shown efficacious and safe glycemic control in T2DM patients and healthy individuals. [25][26][27] Furthermore, sotagliflozin increased plasma levels of the incretin GLP-1 and the anorexigenic PYY. 28 Recently, a nonabsorbable sotagliflozin analog was developed, LX2761, that acts as a selective SGLT1 inhibitor 29 and improved glycemic control associated with increased circulating GLP-1 in rodents. 30 A natural, moderately active SGLT1 blocker causing no gastrointestinal discomfort would offer a good prophylactic for (pre-) diabetic individuals. 31 This study aimed to identify molecules outside the chemical space of phloridzin and structurally similar glycosides with moderate SGLT1 inhibitory activity. We screened a bioactive compound library, using an in vitro SGLT1 inhibition assay with a fluorescent glucose derivative as substrate. 32 Screening hits were explored by in vitro (re)testing of these and additional structurally similar compounds, on SGLT1 and SGLT2 (SGLT1/2). We identified various structurally diverse, novel natural and synthetic inhibitors of SGLT1/2 in distinct chemical clusters. The dataset of diverse SGLT1/2 inhibitors with varying activity was used to develop an in silico proteochemometric (PCM) SGLT1 screening model. 33 Finally, the activity of an identified novel natural inhibitor and a plant extract containing this compound were tested in a more physiologically relevant setting with SGLT1-expressing Caco-2 cells, 34 and the close glucose analog 14 C-α-methylglucose as substrate. This study provides starting points for the development and optimization of novel SGLT1/2 inhibitors.

| Compounds
The Spectrum Collection bioactive compound library was obtained from Microsource (Gaylordsville, USA) and its compounds with simplified molecular-input line-entry specification (SMILES) are shown in Table S1. The additional compounds with their SMILES and suppliers are listed in Table S2. Stock solutions of 100 mmol/L were prepared in DMSO and stored at −20°C.

| Generation of stable CHO-hSGLT1 and CHO-hSGLT2 cell lines
For generation of CHO cell lines stably expressing the hSGLT1 (SLC5A1) gene, or hSGLT2 (SLC5A2) gene, CHO-wild type cells were seeded in a 12-well culture plate at 1 × 10 5 cells/well, grown overnight and transfected with 10 μg plasmid DNA and 4 μL Lipofectamine 2000 in 200 μL OptiMEM. The next day, the cells were collected by TrypLE Express treatment and transferred to 15 cm cell culture dishes in several dilutions in DMEM-F12 supplemented with 10% HI-FBS and 400 μg/mL geneticin to select for stable clones containing the hSGLT1 and hSGLT2 vectors that also carry a neomycin resistance gene. Cells were grown until small clones were visible and medium with geneticin was changed regularly. For each cell line, twenty clones were randomly selected, transferred to 48-well culture plates, propagated and tested for 1-NBD-glucose and 14 C-α-methylglucose uptake (data not shown). Clones with the highest substrate uptake levels were used for SGLT1 and SGLT2 inhibition assays.

| SGLT1 inhibition assay for screening of the Spectrum Collection compound library
Two days before the assay, CHO-wild type and CHO-hSGLT1 cells were seeded in maintenance medium at 25 000 cells/well in clear-bottom black 96-well cell culture plates. Before the assay, cells were washed 3 times with 150 μL/well D-glucose free DMEM with 0.3% (w/v) BSA. Library compounds at 50 μmol/L (singlicate) prepared in D-glucose free DMEM with 160 μmol/L 1-NBD-glucose and 0.3% (w/v) BSA and negative controls (triplicate) were added at 100 μL/well and placed in a humidified incubator at 37°C with 5% CO 2 for 90 minutes. Next, cells were washed 3 times with DMEM with 0.3% (w/v) BSA (150 μL/well). Care was taken to remove all medium after the last wash. Finally, 1-NBD-glucose was extracted from the cells by adding 10 μL/well isopropanol and orbital shaking for 5 minutes at 600 rpm, followed by incubation in the dark for 30 minutes. were screened for SGLT1 inhibiting activity. An activity threshold was arbitrarily set at ≤70% of 1-NBD-glucose uptake (or >30% inhibition) compared to negative control.

| SGLT1 and SGLT2 inhibition assays for validation and investigation of screening hits and structurally similar compounds
Two days before the assay, CHO-hSGLT1 or CHO-hSGLT2 cells were seeded in maintenance medium at 60 000 cells/well in clearbottom black 96-well plates, precoated with 100 μg/mL poly-L-lysine. Cells were washed with 240 μL/well D-glucose free DMEM.

Dilutions of test compounds and controls prepared in D-glucose
free DMEM with 350 μmol/L 1-NBD-glucose, 0.3% (w/v) BSA and 2 mmol/L probenecid (to inhibit efflux of 1-NBD-glucose) were added at 90 μL/well in duplicate and placed in a humidified incubator at 37°C with 5% CO 2 for 30 minutes. The assay incubation media were removed and stored at −80°C for subsequent cytotoxicity analysis. Then, the cells were washed once with ice-cold DMEM-F12 and once with ice-cold HBSS, both at 240 μL/well.
Care was taken to remove all medium after each wash. Finally, 1-NBD-glucose was extracted from the cells with 100 μL/well isopropanol for 10 minutes at 600 rpm using an orbital shaker.  Table 1.
If applicable, statistical outliers were determined using Grubbs' test (alpha = 0.05) and excluded for calculation of the mean ± SD normalized SGLT1/2 activity and mean ± SD IC 50 -values.

| Cytotoxicity assay
The cytotoxicity of representative SGLT1/2 inhibitors and compounds from the different chemical clusters was tested.
Cytotoxicity was determined using the ToxiLight bioassay kit according to the supplier's instructions. This nondestructive assay measures leakage of the enzyme adenylate kinase (AK) from damaged cells into the CHO-SGLT1/2 inhibition assay media, ie the degree of cytolysis. Briefly, 20 μL of CHO-SGLT1/2 inhibition assay medium was added to 100 μL reconstituted AK detection reagent in white 96-well Cellstar plates and incubated for 5 minutes at room temperature. Next, bioluminescence was measured using a  Table S2. If applicable, statistical outliers were determined using Grubbs' test (alpha = 0.05) and excluded for calculation of the mean ± SD. CHO-SGLT1/2 inhibition assay medium was not stored for all assays and cytotoxicity of some compounds was not analyzed or analyzed for two biological replicates only. Mean values ≥20% were considered toxic (arbitrary threshold). Most compounds showed cytotoxicity values between −20% and 20% of vehicle control. Only quinine showed a mean cytotoxicity value above the 20% threshold, albeit with a high standard deviation (SD), 21 ± 38% (Table S2). Non SGLT1 inhibiting stilbenoids from the viniferin-like cluster, but not (+)-ε-viniferin or (−)-ε-viniferin, produced low negative cytotoxicity values ranging from −42% to −89%, as well as the flavonoid glycoside wistin with −51%.

| NMR sample preparation and data acquisition
Dried P p.-extract and (+)-pteryxin were dissolved in deuterated chloroform with 0.03% (v/v) TMS using an Eppendorf Thermomixer C at room temperature. Subsequently, the samples were centrifuged for 5 minutes at 17 000 g at room temperature and 650 µL of supernatant was transferred to a 5-mm NMR tube for analysis. 1D 1 H-NMR spectra were recorded with a ZG pulse sequence using a Bruker Avance III HD 700 spectrometer, equipped with a 5-mm BBI probe.
The probe was tuned to detect 1 H resonances at 700.13 MHz. The internal probe temperature was set to 298 K and 64 scans were collected in 64k data points with a relaxation delay of 1 second and a IC 50 -values of the most active novel inhibitors from chemical clusters presented in Figure 2. Results are means with SD from biological replicates, or individual measurements. n.a. = not applicable. + = original dataset contained a statistical outlier that was excluded from calculation of the mean and SD.
spectral width of 20 ppm. The data were processed in Topspin v3.5 pl 1 (Bruker BioSpin GmbH, Rheinstetten, Germany). An exponential window function was applied to the free induction decay with a line-broadening factor of 0.15 Hz prior to the Fourier transformation.
Manual phase and baseline correction was applied to all spectra. The spectra were referenced against the methyl signal of TMS (δ 0.0 ppm).

| Principal component analysis of compound clusters and Spectrum Collection compound library
Principal

| Screening of the Spectrum Collection bioactive compound library for SGLT1 inhibitors
A screening of 1956 compounds from the Spectrum Collection library for SGLT1 inhibitors resulted in identification of 108 primary hits and 1848 noninhibitors at an activity threshold of ≤70% of negative control, corresponding to a hit rate of 5.5%. A complete overview of screened compounds and their SGLT1 inhibitory activity can be found in Table S1. triadimefon-like (2 compounds). The intra-cluster compound similarities, SGLT1 and SGLT2 inhibitory activities, and SGLT2/SGLT1 activity ratios to indicate selectivity are presented in Table S2.

| Clusters of structurally similar molecules with SGLT1 and SGLT2 inhibitory activity
To  Table S2. For means of n = 2 biological replicates SDs are used indicatively. + = original dataset for this compound contained a statistical outlier which was excluded from calculation of the mean and SD described in the next paragraphs and shown in Figure 2. Unless stated otherwise, all results described are mean inhibition percentages obtained with 50 μmol/L inhibitor. Compounds with ≥30% inhibition were considered active inhibitors and clusters are described in order of increasing SGLT1 inhibition.

| Angular pyranocoumarins
The cluster of the natural APCs contained compounds that were second to phloridzin regarding SGLT1 inhibitory activity and showed SGLT1 selectivity. Substitution at the 9' and 10' C-atom of the APC backbone with an alkyl or alkene ester of 2 to 5 C-atoms was essential for SGLT1 inhibition, ranging from 89% for (+)-pteryxin to 41% for one praeruptorin B ( Figure 2B, Table S2). For one APC, the ester side chain at C-atom 10' contained a 2,3-dimethyl-2-oxiranecarboxylate group which reduced SGLT1 inhibition to 27%. Selinidin, lacking the alkene substitution at the C-10', showed minimal inhibition of SGLT1 and SGLT2 (12% and 25% respectively). Lomatin and cis-khellactone, which both lack alkene substitutions but have either a hydroxy-group at C-9', or at C-9' and C-10' respectively, displayed no SGLT inhibition. Notably, for several of the SGLT1 inhibiting APCs substituted at C-9' and C-10' with alkyl or alkene ester side chains that inhibited SGLT1 with 50% to 72%, no to minimal inhibition of SGLT2 was observed indicating SGLT1 selectivity

| Trimipramine-like compounds
In the trimipramine-like cluster, bepridil caused the strongest SGLT1 inhibition of 74% and SGLT2 inhibition of 70%, followed by verapamil (SGLT1 61%, SGLT2 59%) ( Figure 2C, Table S2). These two compounds were the least similar to the cluster center, with S xc 47% and 49% respectively. Results from a subcluster of most similar compounds suggested that SGLT1 inhibition is strongest with the N,N,2-

| Diphenhydramine-like compounds
Most (active) diphenhydramine-like compounds showed similar inhibition of SGLT1 (31%-72%) or SGLT2 (34%-79%). The inhibitory activity is related to two phenyl-ring structures connected via a single C-atom and an extending N-containing alkyl group attached via an O-atom ( Figure 2D, Table S2). Accordingly, cloperastine displayed the strongest SGLT1 inhibition of 72%. Caramiphen deviates structurally (S xc of 57%) from the stronger SGLT1 inhibitors in this cluster by a substitution of one of the phenyl groups for a cyclopentyl group and attachment of the N-containing group via an ester bond instead of an O-atom and, notably, showed stronger inhibition of SGLT2 (58%) than of SGLT1 (33%) (SGLT2/SGLT1-50μmol/L of 0.6).

| Trihexyphenidyl-like compounds
Most active trihexyphenidyl-like compounds were good inhibitors of both SGLT1 (43%-69%) and SGLT2 (40%-61%) ( Figure 2E, Table   S2). Trihexyphenidyl, the central compound of this cluster, caused a considerable reduction in SGLT1 substrate uptake of 69%. For this cluster, a central 1-phenyl-propan-1-ol structure is important for SGLT inhibition, with the C-1' atom connected to an additional ring structure like cyclohexyl, or bicycloheptenyl, and the C-3' atom connected to an extending N-containing alkyl group, like pyrrolidinyl, piperidine, or diethylamine, as seen for the active diphenhydraminelike compounds. Compounds in this cluster with a lower S xc showed lower SGLT1 inhibition. Minimal differences were observed between SGLT1 or SGLT2 inhibition.

| Bupivacaine-like compounds
The SGLT1 inhibitory activity of the bupivacaine-like compounds is related to the presence of a piperidine group and the length of the attached N-alkyl chain, with inhibitory activity of N-butyl (bupivacaine, 64% inhibition)> N-propyl (ropivacaine, 30% inhibition)> N-methyl (mepivacaine, 11% inhibition) ( Figure 2F, Table S2). No difference in inhibition was observed between a 2,6-dimethylphenyl, or a 2,4,6trimethylphenyl group. Notably, unlike the other compounds in this cluster the structurally least similar compound ritanserin showed selectivity towards SGLT2 with an SGLT2/SGLT1-50μmol/L of 0.3.
Dihydroquinine showed somewhat stronger inhibition than quinine.

| SGLT1/2-IC 50 -values of identified novel inhibitors
The inhibitory activity of the most potent compound from relevant clusters was analyzed more accurately by determining IC 50 values for 1-NBD-glucose uptake by SGLT1 and SGLT2 (Table 1). SGLT1/2-IC 50 values of found novel inhibitors were in the 10-100 μmol/L range.

| Caco-2 SGLT1 inhibition by (+)-pteryxin and Peucedanum praeruptorum extract
The most potent identified novel natural inhibitor (+)-pteryxin and an extract of the root of the plant Peucedanum praeruptorum (P p.-extract), known to contain the APCs (+)-pteryxin, peucedanocoumarin I and praeruptorins A to E, 36 were tested for inhibition of uptake and transport of 14 C-α-methylglucose by Caco-2 cells expressing endogenous SGLT1. Qualitative NMR analyses confirmed the presence of (+)-pteryxin in the P p.-extract ( Figure 3B). At 50 μmol/L and 500 μmol/L (+)-pteryxin, 14 C-α-methylglucose uptake was inhibited by 18% and 91% respectively, while the transcellular transport of 14 C-α-methylglucose was inhibited by 37% and 89%. Dilutions of 1/2000 and 1/200 of the P p.-extract inhibited cellular uptake of 14 C-α-methylglucose by 20% and 90% respectively, while the transcellular transport was inhibited by 43% and 88% ( Figure 3A).    (Table S4). 58 We applied a cellular screening assay to analyze the SGLT1/2 inhibitory activity of structurally diverse compounds using the fluo-  (Table S5). However, these drugs are designed to be rapidly absorbed and further study is required to determine whether effective concentrations are obtained and maintained at the intestinal SGLT1 target site. 58 Possibly, applying a reversed pharmacokinetics optimization approach, these compounds provide a scaffold to develop nonabsorbable SGLT1 inhibitors analogous to the phloridzin-like inhibitor LX2761. 29 Moreover, to enhance the potency and selectivity of the identified leads we suggest an active learning strategy based on PCM modeling. 33,67 This type of modeling incorporates data of both ligand and known protein targets without requiring 3D structural information making it particularly useful as a virtual screening model for transmembrane transporters like SGLT1/2. The PCM model is trained on existing data and used to virtually screen a library of novel compounds (or modified previously identified hits). Subsequently, a selection is made for experimental validation based on a high predicted activity but with a relatively modest probability. 68 This is contrary to ordinary virtual screening wherein data points to be selected have a high predicted activity and probability. Hence, the model is used to identify data points that lead to the highest information gain (exploration) as opposed to identify newly active data points (exploitation). Previously this approach was shown to lead to a quick improvement in biological activity. 69 Accordingly, using the current and a public dataset, a first PCM SGLT1 screening model was developed that effectively predicted moderately active SGLT1 inhibitors outside the chemical space of the training set. 33 We expect further iterations of in silico and in vitro testing to improve this PCM SGLT1 model and the activity of its predicted hits.

| D ISCUSS I ON
In conclusion, we discovered novel natural and synthetic SGLT1/2 inhibitors beyond the chemical space of phloridzin and its analogs. The natural inhibitors are promising leads that may be further investigated as (prophylactic) agents to control dietary glucose uptake, for example as functional food ingredients or supplements. The synthetic inhibitors are mainly registered drugs indicated for conditions other than hyperglycemia. A blood glucose lowering (side) effect may be further investigated for these drugs. The new structure activity data from this study expands the existing public dataset to support further development of virtual SGLT1 inhibition screening models. Additional in vitro mechanistic studies are required to elucidate molecular interactions between the detected inhibitors and SGLT1/2. Finally, the new diverse structure activity data in this study provides starting points for development and optimization of novel, potent and selective