Reagent and chemicals

Canagliflozin and TA-1887 were synthesized by Mitsubishi Tanabe Pharma Corporation27,36. Dapagliflozin and sotagliflozin were purchased from Cayman Chemical Company. Phlorizin, phloretin and α-MG were purchased from Sigma-Aldrich.

cDNA constructs

hSGLT2 complementary DNA and human MAP17 cDNA were synthesized and codon-optimized for expression in human cell lines. Both cDNAs were cloned into the pcDNA3.4 vector. The hSGLT2 sequence was fused with an N-terminal signal sequence from human trypsinogen 1, a His10 tag, and sfGFP, followed by a human rhinovirus 3C protease (HRV3C protease) recognition site. Point mutations were introduced into this construct using site-directed mutagenesis. These plasmids were utilized for all experimental procedures conducted in this study.

Expression and purification of the hSGLT2–MAP17 heterodimer

Mammalian Expi293 cells (Thermo Fisher Scientific) were grown and maintained in Expi293 Expression Medium at 37 °C and 8% CO2 under humidified conditions. Cells were transiently transfected at a density of 2.0 × 106 cells ml−1 with the plasmids and FectoPRO (Polyplus). Approximately 320 μg of the hSGLT2 plasmid and 160 μg of the MAP17 plasmid were premixed with 720 μl of FectoPRO reagent in 60 ml of Opti-MEM (Gibco, Thermo Fisher Scientific) for 10–20 min before transfection. For transfection, 60 ml of the mixture was added to 0.6 liters of the cell culture and incubated at 37 °C in the presence of 8% CO2 for 72 h before collection. The cells were collected via centrifugation (800g, 10 min, 4 °C) and stored at −80 °C before use. The detergent-solubilized proteins were analyzed via FSEC using an ACQUITY UPLC BEH450 SEC 2.5 µm column (Waters).

To prepare the complex sample with phlorizin, the cells were solubilized for 1 h at 4 °C in buffer (50 mM HEPES–NaOH (pH 7.5), 300 mM NaCl, 2% (w/v) DDM (Calbiochem), protease inhibitor cocktail and 1 mM phlorizin). After ultracentrifugation (138,000g, 60 min, 4 °C), the supernatant was incubated with Affi-Gel 10 (Bio-Rad) coupled with a GFP-binding nanobody37, and incubated for 2 h at 4 °C. The resin was washed five times with three column volumes of wash buffer (50 mM HEPES–NaOH (pH 7.5), 300 mM NaCl, 0.05% DDM (GLYCON Biochemicals) and 1 mM phlorizin), and gently suspended overnight with HRV3C protease to cleave the His10–sfGFP tag. After HRV3C protease digestion, the flow-through was pooled, concentrated, purified via size-exclusion chromatography on a Superose 6 Increase 10/300 GL column (GE Healthcare) and equilibrated with SEC buffer (20 mM HEPES–NaOH (pH 7.5), 150 mM NaCl, 0.03% DDM (GLYCON Biochemicals) and 0.5 mM phlorizin). For the samples complexed with canagliflozin, TA-1887, dapagliflozin and sotagliflozin, the same procedure was performed, but at concentrations of 30 µM of each inhibitor. The peak fractions were pooled and concentrated to 6–10 mg ml−1.

α-MG uptake in hSGLT2-transfected HEK293 cells

HEK293 cells (ECACC 85120602) were maintained in Dulbecco’s modified Eagle medium (Gibco) supplemented with 10% fetal bovine serum (Thermo Scientific), 2 mM l-glutamine, 100 U ml−1 benzylpenicillin and 100 µg ml−1 streptomycin at 37 °C in a humidified atmosphere (5% CO2 in air). HEK293 cells were seeded at 1.0 × 105 cells per well in poly-l-lysine coated 24-well plates. The cells in each well were transiently transfected with 0.25 µg hMAP17 plasmid and 0.50 µg hSGLT2 plasmid using Lipofectamine 2000 (Life Technologies) and cultured for 48 h. The medium was removed, and the cells were washed twice then preincubated with extracellular fluid buffer without glucose (122 mM NaCl, 25 mM NaHCO3, 3 mM KCl, 1.4 mM CaCl2, 2 mM MgSO4, 0.4 mM K2HPO4 and 10 mM HEPES; pH 7.4) at 37 °C for 20 min. After preincubation, uptake was initiated by replacing the preincubation buffer with extracellular fluid buffer containing 500 μM α-MG in the absence or presence of inhibitors. Uptake was completed by removing the uptake buffer and washing with ice-cold buffer three times, followed by solubilization in 1 N NaOH at room temperature. The increase in α-MG uptake was observed over 60 min (Supplementary Fig. 10), and the incubation time of the inhibition assay was 30 min.

Cell lysates were deproteinized by adding acetonitrile containing candesartan as the internal standard. The α-MG concentration was quantified via liquid chromatography–tandem mass spectrometry (LC–MS/MS) using the internal standard method.

Specific peaks of α-MG were observed in the lysates of mock and hSGLT2-expressing cells incubated with α-MG but not in those of mock cells in the absence of α-MG (Supplementary Fig. 10). Cellular protein content was determined using a bicinchoninic acid protein assay kit (Thermo Fisher Scientific). The uptake of α-MG was expressed as the ratio of concentration in the cells (in pmol per mg protein) to concentration in the medium (in pmol μl−1); this is known as the cell-to-medium ratio (in μl per mg protein).

In the inhibition study, the cell-to-medium ratio of cells transfected with the empty vector was used as the background. The specific α-MG uptake was calculated by subtracting this background from the total cell-to-medium ratio and normalized to the uptake achieved without the inhibitor. IC50 was calculated via nonlinear regression using GraphPad Prism 8.4.3.

SGLT2 inhibitor-binding assay via affinity selection–mass spectrometry

To examine the inhibition of binding to the crude membrane, mammalian Expi293 cells were co-transfected with the hMAP17 and hSGLT2 plasmids, as described. The cells were collected and disrupted by sonication in a hypotonic buffer (50 mM HEPES–NaOH (pH 7.5), 10 mM KCl and protease inhibitor cocktail) or Na+-free hypotonic buffer (50 mM Tris–HCl (pH 7.5), 10 mM KCl and protease inhibitor cocktail). The cell debris were removed by centrifugation (2,000g, 5 min, 4 °C). The membrane fraction was collected by ultracentrifugation (112,000g, 30 min, 4 °C) and stored at −80 °C before use. The crude membrane (250 μg per sample) was incubated with SGLT2 inhibitor in an assay buffer (100 mM NaCl and 10 mM HEPES/Tris, pH 7.4) or Na+-free assay buffer (100 mM choline chloride and 10 mM HEPES/Tris, pH 7.4) at room temperature for 2 h. Reactions were terminated by filtration through a GF/C filter plate (Corning) presoaked in assay buffer containing 0.1% bovine serum albumin. The sample in the filter plate was washed three times with the assay buffer and eluted with acetonitrile: water (80:20, v/v). The extract solution from the filter plate was diluted with water containing candesartan as an internal standard, and the SGLT2 inhibitor concentration was quantified via LC–MS/MS.

Nonspecific binding was measured using the crude membrane of nontransfected Expi293 cells. Specific binding was calculated by subtracting nonspecific binding from the binding of hSGLT2-expressing cells. Specific binding was normalized to hSGLT2 protein expression levels, measured via FSEC. The equilibrium dissociation constant (Kd) and maximum number of binding sites (Bmax) were calculated via nonlinear regression in GraphPad Prism 8.4.3. The specific binding of the hSGLT2 mutants was normalized to the Bmax of WT hSGLT2.

Quantification of SGLT2 substrate and inhibitors via LC–MS/MS

The concentrations of the extract solution from the filter plate and of the cell lysate were quantified using a tandem mass spectrometry QTRAP6500 System (SCIEX) coupled with an ACQUITY UPLC system (Waters) using the internal standard method. Mobile phases A and B used 10 mM of ammonium bicarbonate and acetonitrile, respectively. Chromatographic separation was performed on an ACQUITY UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 μm; Waters) at 50 °C, with the following gradient of mobile phase B: 1% (at 0.00 to 0.50 min), 1% to 95% (0.50 to 2.00 min), 95% (2.00 to 2.50 min) and 1% (2.51 to 3.00 min); the flow rate was 0.4 ml min−1. Mass spectrometric detection was performed by multiple reaction monitoring in the electrospray-ionization negative-ion mode controlled by Analyst 1.6.2, using m/z 443.1/364.9 for canagliflozin; 425.9/264.1 for TA-1887; 407.0/328.8 for dapagliflozin; 423.0/387.0 for sotagliflozin; 435.0/273.0 for phlorizin; 273.0/148.9 for phloretin; 192.9/100.9 for α-MG; and 439.0/309.1 for candesartan.

Electron microscopy sample preparation

The purified protein solution of hSGLT2–MAP17 was mixed with the inhibitor solutions (except for phlorizin), at final concentrations of 0.5 mM dapagliflozin, TA-1887, sotagliflozin or canagliflozin. After incubation for 1 h on ice, the grids were glow-discharged in low-pressure air at a 10 mA current in a PIB-10 (Vacuum Device). The protein solutions containing 0.5 mM of the inhibitors were applied to a freshly glow-discharged Quantifoil Holey Carbon Grid (R1.2/1.3, Cu/Rh, 300 mesh) (SPT Labtech) using a Vitrobot Mark IV system (Thermo Fisher Scientific) at 4 °C, with a blotting time of 4–6 s under 99% humidity; the grids were then plunge-frozen in liquid ethane.

Electron microscopy data collection and processing

The grids containing phlorizin, TA-1887, dapagliflozin and sotagliflozin were transferred to a Titan Krios G3i system (Thermo Fisher Scientific) running at 300 kV and equipped with a Gatan Quantum-LS Energy Filter (GIF) and a Gatan K3 Summit direct electron-detector in correlated double-sampling mode. Imaging was performed at a nominal magnification of 105,000×, corresponding to a calibrated pixel size of 0.83 Å per pixel, at the University of Tokyo, Japan. Each movie was dose-fractionated to 64 frames at a dose rate of 6.2–9.0 e− per pixel per second at the detector, resulting in a total accumulated exposure of 64 e− Å−2 of the specimen. The data were automatically acquired using the image-shift method in SerialEM software38, with a defocus range of −0.8 to −1.6 μm.

The grid with canagliflozin was transferred to a Titan Krios G4 device (Thermo Fisher Scientific) running at 300 kV and equipped with a Gatan Quantum-LS Energy Filter (GIF) and a Gatan K3 Summit direct electron-detector in correlated double-sampling mode. Imaging was performed at a nominal magnification of 215,000×, corresponding to a calibrated pixel size of 0.4 Å per pixel, at the University of Tokyo, Japan. Each movie was recorded for 1.4 s and subdivided into 64 frames. Electron flux was set to 7.5 e− per pixel per second at the detector, resulting in an accumulated exposure of 64 e− Å−2 of the specimen. The data were automatically acquired via the image-shift method using EPU software (Thermo Fisher Scientific), with a defocus range of −0.6 to −1.6 μm. The total number of images is described in Table 1.

For all datasets, the dose-fractionated movies were subjected to beam-induced motion correction using RELION24, and the contrast transfer function (CTF) parameters were estimated using CTFFIND4 (ref. 39).

For the canagliflozin-bound state dataset, 2,364,108 particles were initially selected from 19,943 micrographs using the topaz-picking function in RELION-4.0 (ref. 25). Particles were extracted by downsampling to a pixel size of 3.2 Å per pixel. These particles were subjected to several rounds of 2D and 3D classification. The best class contained 221,701 particles, which were then re-extracted with a pixel size of 1.60 Å per pixel and subjected to 3D refinement. The second 3D classification resulted in three map classes. The best class from the 3D classification contained 179,761 particles, which were subjected to 3D refinement. The particles were subsequently subjected to micelle subtraction and non-aligned 3D classification using a mask (without micelles), resulting in three map classes. The best class, containing 65,919 particles, was subjected to 3D refinement, reversion to the original particles, and 3D refinement. The particle set was resized to 1.00 Å per pixel and subjected to Bayesian polishing, 3D refinement and per-particle CTF refinement before the final 3D refinement and post-processing, yielding a map with a global resolution of 3.1 Å, according to the FSC 0.143 criterion. Finally, local resolution was estimated using RELION-4. The processing strategy is illustrated in Supplementary Fig. 2.

For the dapagliflozin-bound-state dataset, 3,692,950 particles were initially selected from 4,841 micrographs using the topaz-picking function in RELION-4.0. Particles were extracted by downsampling to a pixel size of 3.32 Å per pixel. These particles were subjected to several rounds of 2D and 3D classification. The best class contained 569,516 particles, which were then re-extracted at a pixel size of 1.30 Å per pixel and subjected to 3D refinement. Non-aligned 3D classification using a soft mask covering the proteins and micelles resulted in four map classes. The best class from the 3D classification contained 197,695 particles, which were subjected to 3D refinement, per-particle CTF refinement, and 3D refinement. The resulting 3D model and particle set were resized to 1.11 Å per pixel and subjected to Bayesian polishing, 3D refinement and per-particle CTF refinement. Final 3D refinement and post-processing yielded maps with global resolutions of 2.8 Å, according to the FSC 0.143 criterion. Finally, the local resolution was estimated using RELION-3. The processing strategy is illustrated in Supplementary Fig. 3.

For the TA-1887-bound state dataset, 3,395,470 particles were initially selected from 4,383 micrographs using the topaz-picking function in RELION-4. Particles were extracted by downsampling to a pixel size of 3.32 Å per pixel. These particles were subjected to several rounds of 2D and 3D classification. The best class contained 274,477 particles, which were then re-extracted with a pixel size of 1.30 Å per pixel and subjected to 3D refinement. Non-aligned 3D classification using a soft mask covering the proteins and micelles resulted in three map classes. The best class from the 3D classification contained 103,853 particles, which were subjected to 3D refinement, per-particle CTF refinement, and 3D refinement. The resulting 3D model and particle set were resized to 1.11 Å per pixel and subjected to Bayesian polishing, 3D refinement and per-particle CTF refinement. Final 3D refinement and post-processing yielded maps with global resolutions of 2.9 Å, according to the FSC 0.143 criterion. Local resolution was estimated using RELION-4. The processing strategy is illustrated in Supplementary Fig. 4.

For the sotagliflozin-bound state dataset, 5,242,427 particles were initially selected from 5,499 micrographs using the topaz-picking function in RELION-4. Particles were extracted by downsampling to a pixel size of 3.32 Å per pixel. These particles were subjected to several rounds of 2D and 3D classifications. The best class contained 823,369 particles, which were then re-extracted with a pixel size of 1.30 Å per pixel and subjected to 3D refinement. Non-aligned 3D classification using a soft mask covering the proteins and micelles resulted in four classes of maps. The two good classes from the 3D classification contained 227,811 particles, which were subjected to 3D refinement. The resulting 3D model and particle set were resized to 1.11 Å/px and subjected to Bayesian polishing, 3D refinement and further non-aligned 3D classification using a soft mask covering the proteins and micelles. The best class from the 3D classification contained 72,773 particles, which were subjected to 3D refinement, per-particle CTF refinement, 3D refinement, Bayesian polishing, 3D refinement and per-particle CTF refinement. The final 3D refinement and post-processing yielded maps with global resolutions of 3.1 Å, according to the FSC 0.143 criterion. Finally, the local resolution was estimated using RELION. The processing strategy is illustrated in Supplementary Fig. 5.

For the phlorizin-bound state dataset, 3,013,029 particles were initially selected from 3,159 micrographs using the Laplacian-of-Gaussian picking function in RELION-3.1 (ref. 24) and were used to generate 2D models for reference-based particle picking. Particles were extracted by downsampling to a pixel size of 3.32 Å per pixel. These particles were subjected to several rounds of 2D and 3D classification. The best class contained 324,355 particles, which were then re-extracted with a pixel size of 1.66 Å per pixel and subjected to 3D refinement. The particles were subsequently subjected to micelle subtraction and non-aligned 3D classification using a mask (excluding the micelles), resulting in three map classes. The best class contained 76,485 particles, which were then subjected to 3D refinement and reversion to the original particles. The particle set was resized to 1.30 Å per pixel, and subjected to Bayesian polishing, 3D refinement and per-particle CTF refinement before the final 3D refinement and post-processing, yielding a map with a global resolution of 3.3 Å, according to the FSC 0.143 criterion. Finally, the local resolution was estimated using RELION-3. The processing strategy is illustrated in Supplementary Fig. 6.

Model building and validation

The models of the phlorizin-bound inward state of hSGLT2–MAP17 were manually built, de novo, using the cryo-EM density map tool in COOT40, facilitated by an hSGLT2-homology model generated using Alphafold2 (ref. 41). After manual adjustment, the models were subjected to structural refinement via the Servalcat pipeline in REFMAC5 (ref. 42) and manual real-space refinement in COOT. The models of the dapagliflozin-, TA-1887-, sotagliflozin- and canagliflozin-bound outward states were built using the Alphafold2-derived hSGLT2-homology model as the starting model. The 3D reconstruction and model refinement statistics are summarized in Table 1. All molecular graphics figures were prepared using CueMol (http://www.cuemol.org) and UCSF Chimera43.

Thermostability measurement

The thermostability of the detergent-solubilized proteins was analyzed using an FSEC-based thermostability assay44. Mammalian Expi293 cells (Thermo Fisher Scientific) were transiently transfected with the plasmids and with ExpiFectamine (Thermo Fisher Scientific). For each 1 ml transfection, 1 ml of cells (2.4 × 106 cells) was transferred to each well in a 96-well MASTERBLOCK (Greiner Bio-One). For the hSGLT2–MAP17 heterodimer, 0.8 μg hSGLT2 plasmid and 0.4 μg MAP17 plasmid were mixed in Opti-MEM (total volume 60 μl) and 3.2 μl ExpiFectamine 293 Reagent was added to 56.8 μl Opti-MEM. For hSGLT2 alone, 1.2 μg hSGLT2 plasmid was added to Opti-MEM (total volume 60 μl). After incubating for 5 min at room temperature, the diluted plasmid was added to the diluted ExpiFectamine 293 Reagent, gently mixed, and incubated for 20 min at room temperature. The reagent–plasmid mixture was added to each well and incubated at 37 °C in the presence of 8% CO2 on a Maximizer MBR-022UP bioshaker (TAITEC) at 1,200 rpm. After 48 h incubation at 37 °C, the cells were collected from 6 ml of the culture via centrifugation. The cell pellet was resuspended in 600 µl of buffer (50 mM HEPES–NaOH (pH 7.5), 300 mM NaCl, 1% (w/v) DDM and protease inhibitor cocktail) and shaken for 60 min at 4 °C. After clearing the cell lysate via centrifugation (20,000g, 30 min, 4 °C), 45 μl portions of the cell lysate were incubated at 4, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55 or 60 °C for 10 min in a PCR Thermal Cycler SP (Takara Bio). The sample was again centrifuged (20,000g, 60 min, 4 °C) to clear the lysate, and a 1 μl portion of the supernatant was placed in an ACQUITY UPLC BEH450 SEC 2.5 µm column (Waters), pre-equilibrated with buffer containing 50 mM Tris, pH 7.6, 150 mM NaCl and 0.05% DDM (GLYCON Biochemicals). The GFP fluorescence of the eluent was monitored, and the peak heights of heat-treated samples were normalized to that of the untreated sample. For each mutant, the measurement was performed at least three times, and the melting temperatures were determined by fitting the curves to a sigmoidal dose–response equation, using GraphPad Prism 7.

Water analysis using 3D-RISM

The structure was preprocessed using the Protein Preparation Wizard45 in Schrödinger Suite v2021-4 (Schrödinger). The default operation flipped the carbamoyl group of Q457 in the dapagliflozin complex; this was therefore corrected manually. All water in the pocket was retained. Finally, restrained minimization was performed using the OPLS4 force field46. Water analysis was performed on the prepared structures using 3D-RISM (as implemented in MOE (Molecular Operating Environment)). The supplementary program required for water site analysis is available directly from Chemical Computing Group ULC.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.