Animal studies in Copenhagen were approved by the Danish Animal Experiments Inspectorate (2013-15-2934-00833) and the local ethics committee and were performed in accordance with the guidelines of Danish legislation governing animal experimentation (1987) and the National Institutes of Health (publication number 85-23). Animal studies in Cambridge, UK (isolation of mouse pancreatic tissue) were approved by the University of Cambridge Animal Welfare and Ethical Review Body, conformed to the Animals (Scientific Procedures) Act 1986 Amendment Regulations (SI 2012/3039) and were performed under the UK Home Office Project License 70/7824.
Human islets were isolated by the European Consortium for Islet Transplantation (ECIT) under local approval in Milan, Italy, and were received in fully anonymous form. Human kidney cDNA was derived from kidneys surgically removed due to renal cell carcinoma. Kidney tissue was obtained from Sahlgrenska Hospital (Gothenburg, Sweden) in accordance with relevant ethical guidelines and patient consent. Human kidney and pancreas tissue used for immunohistochemistry were anonymous archival tissue.
Male Wistar rats were obtained from Janvier (Saint Berthevin Cedex, France) and kept at animal facilities at the Faculty of Health and Medical Sciences, University of Copenhagen. Rats were housed two per cage, had ad libitum access to standard chow and water and were kept on a 12 h light–dark cycle. Animals were acclimatised for at least 1 week.
Chickens (Bowans Brown hens) were also kept at the animal facilities at the Faculty of Health and Medical Sciences, University of Copenhagen and housed with ad libitum access to standard chow and drinking water for at least 2 weeks before experiments.
Mice for expression analysis were bred in-house at the University of Cambridge, UK (Central Biomedical Services [CBS]) and had been bred for at least 8 generations on a C57BL/6 background. Kidney and whole islets were isolated from female Glu-Cre-ERT2 × GCaMP3 mice (29 weeks old). For RNA sequencing analysis, delta cells were isolated from SstCre × EYFP mice (4 females, 2 males; 18–39 weeks old). Alpha and beta cells were isolated from Glu-Venus mice (2 females, 2 males; 22–29 weeks old). Further information about the strains has been detailed previously .
Isolated perfused rat pancreas procedure
Non-fasted rats were anaesthetised with Hypnorm/midazolam (0.3 ml/100 g body weight, per ml: Hypnorm: 0.08 mg fentanyl, 2.5 mg fluanisone, 0.45 mg methyl parahydroxybenzoate, 0.05 mg propyl parahydroxybenzoate; midazolam: 1.25 mg; Matrix Pharmaceuticals, Hellerup, Denmark) and the pancreas was isolated by removing the stomach, spleen and small and large intestine (except for the part of duodenum connected to pancreas) and by tying off the renal stalks. The abdominal aorta was tied off just below the diaphragm and, immediately, a catheter was inserted retrogradely into the abdominal aorta just distally to the renal arteries so that the aorta now exclusively supplied the pancreas (through both the coeliac and the superior mesenteric arteries). The pancreas was perfused by use of a UP100 Universal Perfusion System from Hugo Sachs (Harvard Apparatus, March-Hugstetten, Germany) (5.0 ml/min, 37°C) with a Krebs–Ringer bicarbonate buffer gassed with 95% O2, 5% CO2 . A catheter was inserted into the portal vein for collection of venous effluent. After the start of the perfusion, the rat was killed by diaphragm perforation. The preparation was allowed to stabilise for 25 min and then 1 min samples were collected, transferred to ice and stored at −20°C. Glucose concentration was 3.5 and 10 mmol/l as indicated in Results. The following reagents were administered: the SGLT1/2-specific and metabolically inert substrate α-MGP (20 mmol/l); the SGLT1/2 inhibitor phloridzin dihydrate (100 μmol/l); the SGLT2 inhibitor dapagliflozin (0.5 μmol/l) and l-arginine (10 mmol/l, as positive control). Dapagliflozin was from Seleckchem.com (cat. no. S1548, Munich, Germany). The remaining reagents were from Sigma-Aldrich, Brøndby, Denmark (α-MGP; cat. no. M9376, phloridzin; 274313 and l-arginine; A92600). Pre-analysis inclusion criteria were quick insertion of catheter in aorta, stable perfusion output and perfusion pressure. Post-analysis inclusion criteria were stable hormone output before addition of test compounds as well as responsiveness to the positive control, l-arginine. Perfusion pressure and flow were closely monitored and remained stable during the perfusion.
Ussing chamber experiments
Experiments were carried out using fresh colonic tissue collected from Bowans Brown hens, which had been used for other purposes. Hens (n = 6) were killed by cervical dislocation and immediately after the colon was isolated, opened along the mesenteric border and cleaned by rinsing in cold Krebs–Ringer buffer. Tissue was placed in Krebs Ringer buffer (room temperature) and the mucosa was isolated by a careful scrape within 10 min after collection and placed in Ussing chambers (exposed mucosal area ~0 .6cm2) with oxygenated (95% O2 and 5% CO2) Krebs–Ringer buffer (37°C). The preparation was left for ~35 min to stabilise and then clamped at a potential difference of 0 mV from 10 min before the start until the end of the experiment. Short circuit currents (IScc), potential difference (mV) and resistance (Ω) were recorded every 2 s and shown in real-time by use of Clampdata software (Clamp version 2.14, Mussler Scientific Instruments, Aachen, Germany). Glucose, dapagliflozin or phloridzin was added to the luminal side of the mucosal layer (final concentration 100 mmol/l, 55 μmol/l and 100 μmol/l, respectively, corresponding to ~50 × IC50 on human SGLT1  in case of dapagliflozin). l-Lysine (10 mmol/l) was included as a positive control. Inclusion criteria were stable resistance throughout the experiment, stable IScc before addition of test compounds, and responsiveness to both d-glucose and l-arginine.
Formalin-fixed tissue from rat, mouse and human kidney and pancreas (n = 3 for all tissue types) was embedded in paraffin. Five micrometre dewaxed tissue sections were pre-treated by boiling for 15 min in a microwave oven in EGTA buffer at pH 9 (antigen retrieval). Sections were then pre-incubated in PBS with 10% (wt/vol.) BSA for 10 min and incubated with the primary antibody (rabbit anti-human, -rat and -horse SGLT2 antibody; cat. no. LS-A2810; LSBio Seattle, Seattle, WA, USA) diluted 1:1000 in PBS with 10% (wt/vol.) BSA overnight at 4°C. Next day, the tissues were incubated with biotin goat anti-rabbit IgG 1:200 (Vector Labs, Burlingame, CA, USA), ABC (PK4000; Vector labs) and DAB, lightly counterstained with haematoxylin and coverslipped with Pertex mounting medium. In initial validation and optimisation experiments, the primary SGLT2 antibody was tested diluted 1:100, 1.1000 and 1:10000 times on human kidney tissue and with and without microwave antigen retrieval treatment both at pH 6 and 9. The optimal dilution was found to be 1:1000. The antibody yielded positive immunostaining in kidney only after microwave pretreatment at pH 9. The antibody was tested in different tissues, with human and rat kidney as the positive controls (known to be localised on apical membrane of proximal tubules). No SGLT2 immunoreactive stain was found in other parts of the kidney or in the exocrine pancreas (negative control).
The SGLT2 antibody used is a commercially available antibody and the immunisation fragment was not available for pre-absorption experiments. The sections were viewed under a Leitz orthoplan microscope (Wetzlar, Germany) and images were taken with an AxioCam IcC5 camera (Zeiss, Jena, Germany) mounted onto the microscope.
Rat islets, human donor islets and rat kidney tissue were lysed on ice with NP-40 lysis buffer containing protease inhibitor cocktail (Life Technologies, Nærum, Denmark, cat. no. FNN0021) and stored at −20°C. Total protein concentration was measured by Bradford protein assay kit (cat. no. 5000001, Bio Rad Laboratories, Copenhagen, Denmark). Ten micrograms of protein lysate/sample was loaded onto gels, separated by 4–20% SDS-PAGE and blotted on a PDVF membrane. To avoid stripping membranes when staining for multiple proteins, membranes were cut according to the desired molecular mass range. Membranes were then stained separately with rabbit anti-human, -rat and -horse SGLT2 antibody (cat. no. LS-A2810-50, LSBio) or anti-α-tubulin antibody (loading control; cat. no. T6074; Sigma-Aldrich, Denmark), both diluted to a concentration of 0.5 μg/ml. Blots were developed with the chemiluminescence detection system using Super Signal (Life Technologies, cat no. 34075). Light emission was captured using an Alpha imager system (Alpha-Innotech, Broager, Denmark).
Gene expression procedures
Pancreatic islets from non-diabetic 1-week-old rats (n = 5) were isolated as described previously . Rat and donated human islets (anthropometric data supplied in electronic supplementary material [ESM] Table 1) were cultured for 72 h at 37°C (95% O2, 5% CO2) in RPMI-glutamax medium supplemented with 10% (vol./vol.) FBS, 1% (vol./vol.) P/S (cat. no. P4333, Sigma-Aldrich, Denmark) and 5.6 mmol/l glucose. Total RNA was extracted by use of Nucleo-Spin kit (Macheray-Nagel, Bethlehem, PA, USA). Quality and quantity of the extracted RNA were assessed using a NanoDrop-1000 (Thermo Scientific). Five-hundred nanograms of total RNA were used for cDNA synthesis with the iScript-cDNA Kit (BioRad). Real-time reverse-transcriptase quantitative PCR (RT-qPCR) was performed on 12 ng cDNA with SYBR Green PCR mastermix (Life Technologies, cat. no. 4309155) using specific primers (ESM Table 2) and run in a real-time PCR machine (Applied Biosystems, Naerum, Denmark). Gene expression levels were normalised to the housekeeping mRNA Hprt1 (encoding hypoxanthine-guanine phosphoribosyltransferase-1) through the −∆Ct method. Methods are described in detail elsewhere . As positive control, kidney tissue from rats and human donors were subjected to the same procedure.
For whole islet studies, mouse islets were isolated as described previously  and RNA from homogenised tissues was extracted using TRI reagent (Sigma-Aldrich, Gillingham, UK) or RNeasy Plus Micro Kit (Qiagen, Manchester, UK). Genomic DNA was removed via treatment with DNAse1 (Thermo Fisher Scientific, Winsford, UK) or through the RNeasy Plus gDNA eliminator spin columns. Purified RNA was reverse transcribed using Superscript II kit (Thermo Fisher Scientific). RT-qPCR was performed with a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). The PCR reaction mix consisted of first-strand cDNA template, 6-carboxyfluorescein/quencher probe/primer mixes (Thermo Fisher Scientific) and PCR Master Mix (Thermo Fisher Scientific) and was amplified for 40 cycles. Expression of selected targets was normalised to that of the housekeeping mRNA Actb, and expression levels calculated as above. Primer information is provided in ESM Table 2. RNA sequencing was performed as described previously  on FACS separated transgenic mice expressing fluorescent marker proteins in either the delta cells (Sst-EYFP strain) or the alpha cells (Glu-Venus strain). Beta cells were isolated based on their size, using forward and side scatter characteristics distinguishing them from other nonfluorescent islet cells in Glu-Venus mice. Methods are described in detail elsewhere .
Single-cell RNA sequencing of human pancreatic islet cells
Single-cell expression levels of SGLT1 (also known as SLC5A1), SGLT2 (SLC5A2) and GLUT1–5 (SLC2A1–5) in human alpha, beta and delta cells were determined by RNA-sequencing-based analysis by two different independent groups of investigators, using their own methodology and populations.
In population 1, expression was determined by reanalysing published human islet single-cell sequencing data, from donors with type 2 diabetes and non-diabetic donors (n = 4 with diabetes, n = 6 without diabetes) . FastQ files were downloaded from ArrayExpress (accession E-MTAB-5061). Data were analysed with bcbio-nextgen (https://github.com/chapmanb/bcbio-nextgen), using the hisat2 algorithm  to align reads to human genome version hg38 and the Salmon algorithm  for quantification of gene counts. Only cells that passed the quality control in the original study  were maintained for further analysis: non-diabetic donors/diabetic donors: alpha cells 443/443; beta cells 171/99 and delta cells 59/55. Cell type classification from the original study was maintained. Log2 transformed counts per million (CPM) were plotted for individual cells.
Publicly available RNA sequencing data from two different population sets (GSE85241, GSE81608) [16, 17] were used for this analysis. Only non-diabetic individuals were included in the analyses, with a combined sample size of 26. Reads per kilobase million (RPKM) values for SGLT1, SGLT2 and GLUT1–5 from the respective data bases were uploaded to the Jupyter Notebook (http://jupyter.org/, accessed 1207–2018). Data were then log2 transformed, and means were calculated for SGLT1, SGLT2 and GLUT1–5 in alpha, beta and delta cells, respectively. Further details about the donors, isolation of cells and RNA sequencing methods can be found in the original studies [16, 17].
The in-house glucagon RIA employed a C-terminal directed antibody (code no. 4305 ). Insulin and somatostatin (total: SST-14 and SST-28) concentrations were determined by in-house RIA (code no. 2006-3 and 1758-5 [19, 20]). Sensitivities of the assays were around 1 pmol/l, allowing detection of secretory outputs as low as 5 fmol/min.
Data presentation and statistical analysis
Data are presented as means ± SEM. For perfusion data, averaged basal and response outputs were calculated by taking an average of output values over the entire stimulus administration (response, 15 min) and the period leading immediately up to stimulus administration (15 min, baseline). Output (fmol/min) was the product of peptide concentration (pmol/l) and flow rate (ml/min). Averaged IScc at baseline, and after glucose and dapagliflozin, was calculated over at least 50 consecutive recordings within the terminal part of the respective periods. Statistical significance was assessed by one-way ANOVA for repeated measurements followed by Bonferroni multiple comparison test using GraphPad Prism software (version 6, La Jolla, CA, USA). Graphs were constructed in GraphPad and edited in Adobe Illustrator (San Jose, CA). p < 0.05 was considered significant.