Gluconeogenesis, But Not Glycogenolysis, Contributes to the Increase in Endogenous Glucose Production by SGLT-2 Inhibition

Selective pharmacological inhibition of the sodium-glucose cotransporter 2 (SGLT-2) in the kidney by dapagliflozin results in improved glucose metabolism and was recently shown to be effective in the reduction of cardiovascular mortality in patients with and without type 2 diabetes mellitus (T2DM) (1,2). Moreover, SGLT-2 inhibitors are associated with slower progression of chronic diabetic kidney disease and markedly reduced renal end points, including worsening of nephropathy and incident albuminuria (3). In patients with T2DM and increased cardiovascular risk, the use of an SGLT-2 inhibitor is therefore recommended as the first-line oral treatment after metformin in recent clinical practice guidelines (4).

Dapagliflozin exerts its glucose-lowering effects by promoting glucosuria due to inhibition of renal glucose reabsorption by SGLT-2, resulting in a dose-dependent increase in urinary glucose excretion. These improvements in glycemic control ameliorate insulin sensitivity and lower insulin secretion in patients with T2DM, resulting in a reduction of HbA_1c of 0.7–1% (5). Surprisingly, these benefits in glucose metabolism are associated with a significant rise in endogenous glucose production (EGP), which might potentially counteract the glucose-lowering effects (6,7). However, the mechanisms via which glucosuria causes an increase in EGP are not clear. Higher levels of glucagon and, therefore, a drop in the hepatic insulin-to-glucagon ratio might offer a possible explanation and could be linked to the degree of glucose lowering (8). On the other hand, SGLT-2 is expressed in pancreatic α-cells, and its inhibition was shown to directly trigger glucagon secretion in vitro (9).

With regard to lipid metabolism, SGLT-2 inhibitor therapy results in a switch from glucose to lipid oxidation (10), which is associated with an increase in lipolysis and enhanced ketogenesis acutely after a single dose as well as after a treatment period of 28 days (11). In addition, longer-term administration of dapagliflozin reduces abdominal subcutaneous and visceral fat depots (12,13) as well as hepatocellular lipid (HCL) content (14,15). Of note, it cannot be discerned whether these changes in different fat depots are due to direct lipolytic effects of SGLT-2 inhibition or are more likely mediated by the modest weight reduction observed after some weeks of treatment. So far, however, acute direct effects of SGLT-2 inhibition on hepatic glucose, lipid, and energy metabolism have not been comprehensively analyzed. We therefore investigated the impact of a single dose of dapagliflozin 1) on the contribution of hepatic glycogenolysis to discern whether the observed increase in EGP is due to glycogenolysis or gluconeogenesis (GNG), and 2) on acute changes in HCL and mitochondrial energy turnover independently of changes in body weight.

This study was performed as a crossover, randomized-controlled, double-blind trial. All study-related experiments were performed at the Medical University of Vienna. The local ethical committee of the Medical University of Vienna approved this study, and written informed consent was obtained from all participating subjects. This study was registered on ClinicalTrials.gov (NCT02558270) and EudraCT (2014-002337-56).

The study included 10 healthy volunteers and 6 patients with well-controlled T2DM (control [CON] vs. T2DM). Exclusion criteria for healthy control subjects were BMI >26 kg/m², smoking, regular medication, history of renal or liver diseases, and general contraindications against performance of magnetic resonance (MR) measurements, including claustrophobia or metal devices. Inclusion criteria for patients with T2DM were BMI 23–35 kg/m², age between 18 and 75 years, and HbA_1c <7.5% during treatment with metformin monotherapy or a combination therapy of metformin and a dipeptidyl peptidase 4 (DPP-4) inhibitor. All included patients were on metformin therapy. Two patients were on additional treatment with a DPP-4 inhibitor. Exclusion criteria were impaired renal function (estimated glomerular filtration rate <60 mL/min), abnormal liver enzymes (AST or ALT more than three times the upper limit of normal), or history of diabetic ketoacidosis or cardiovascular events. These two groups were chosen to investigate acute effects after SGLT-2 inhibition in two distinctly different phenotypes: insulin-sensitive, lean, healthy volunteers and insulin-resistant, overweight, patients with diabetes.

Participants were asked to refrain from intensive physical training, to stop regular moderate exercising, and to ingest an isocaloric diet (30 kcal ⋅ kg⁻¹ ⋅ day⁻¹; 55% carbohydrate, 15% protein, 30% fat) for 3 days before the study-related experiments. They were investigated on two different study days after the administration of 10 mg dapagliflozin (D) or placebo (P) at time point 0 min in random order with a minimum interval of 14 days, resulting in the following protocols 1) CON-D, 2) CON-P, 3) T2DM-D, and 4) T2DM-P. Each subject was studied after an overnight fast of at least 8 h. In T2DM, oral antidiabetic therapy was paused 48 h before each study day. After two venous catheters were placed into the left and right antecubital vein, initial blood samples were taken at baseline (−120 min). Thereafter, d-[6,6-²H]glucose (98% enrichment; Cambridge Isotope Laboratories, Andover, MA) was infused (bolus: 3.5 mg/kg lean body wt; continuous infusion: 0.035 mg ⋅ kg lean body wt⁻¹ ⋅ min⁻¹) for measuring EGP from −15 min to 390 min. MR examinations were performed at three time points: 1) MR-0 (baseline: −120 to 0 min), 2) MR-1 (90–210 min), and 3) MR-2 (270–390 min). In addition, liver volume was assessed by MRI of the abdomen at the start of the experimental protocol. Blood samples for determination of EGP and glucose levels were drawn at −15, 0, 60, 90, 180, 240, and 390 min. Concentrations of glucagon, insulin, C-peptide, epinephrine, norepinephrine, growth hormone, cortisol, ACTH, free fatty acids, and β-hydroxybutyrate were determined at −120, 60, 180, and 390 min. Urine volume and urinary glucose excretion was assessed at baseline and after 180 and 390 min.

During all protocols, concentrations of glucose, insulin, C-peptide, growth hormone, and urinary glucose were measured by routine laboratory methods at the Department of Laboratory Medicine of the Medical University of Vienna (https://www.kimcl.at). All other blood samples were immediately chilled and centrifuged, and the supernatants were stored at −80°C. Concentrations of glucagon, ACTH, lactate, adrenaline, and noradrenaline were analyzed after completion of all study-related experiments at the Department of Laboratory Medicine of the Medical University of Vienna by routine laboratory methods. Free fatty acids and ketone bodies were measured by ELISA (both Wako Chemicals, Neuss, Germany). Free glycerol was determined by using the free glycerol determination kit (Sigma-Aldrich, St. Louis, MO). Alanine was measured using a commercial kit from Chromsystems (Gräfeling, Germany), according to the manufacturer’s instructions, on a Shimadzu 8050 high-performance liquid chromatography-coupled tandem mass spectrometer.

Total amount of glucosuria during each study day (g/study day) was calculated by multiplying urine volume with urinary glucose concentrations during the study day.

MRS and MRI measurements were performed in a high-field 7-T whole-body MR system (Magnetom; Siemens Healthcare, Erlangen, Germany) equipped with appropriate surface coils (¹H/¹³C and ¹H/³¹P) allowing MRS measurements in the abdominal region. To assess individual liver volumes, conventional MRI of abdomen without any contrast agent was performed in a 3-T whole-body MR system (Tim Trio; Siemens Healthcare) at the start and at the end of every experimental protocol. Absolute concentrations of hepatic metabolites were assessed with ³¹P MRS using a phantom replacement method (16). Flux through ATP synthase in the liver tissue was measured by dynamic ³¹P saturation transfer MRS experiment, as previously described (17). Possible changes of HCL accumulation were assessed by the single voxel localized ¹H MRS method optimized earlier (18). HCL was calculated from ratio of the summed area of methylene and methyl resonance to that of water after the individual spin-spin relaxation correction as the percentage of total tissue MRS signal (water + methylene + methyl). ¹³C MRS for the assessment of tissue glycogen amount was localized on tissue of interest based on image-selected in vivo spectroscopy localization (19). Absolute glycogen concentrations were quantified by comparing the C1 glycogen peak (100.5 ppm) integral of tissue spectra with that of a glycogen standard taken under identical conditions. Corrections for loading of the coil and sensitive volume of the coil were performed.

Rates of EGP were calculated using the Steele equation. Data were first smoothed by spline fitting (Data Curve Fit Creator; SRS1 Software, Newton Centre, MA) to have a better estimation of dTTR(t)/dt at each time point (20). Total-body glucose disposal was calculated by correcting EGP for the changes in glucose concentrations. The rate of tissue glucose uptake was calculated by subtracting the rate of urinary glucose excretion from the total-body glucose disposal rate. Rates of net hepatic glycogenolysis (μmol kg⁻¹ min⁻¹) were calculated from the best linear fit of the liver glycogen (mmol/L liver) time curves by the method of least squares and multiplying the slope of that line by the liver volume (in liters) divided by the body weight (in kg). Rates of glycogenolysis were calculated for the complete time period (0–380 min). Rates of GNG are given as the difference between rates of EGP and glycogenolysis (21).

Statistical analysis was performed by using SPSS version 24 (IBM, Armonk, NY) and GraphPad Prism 8 for Mac. Data are given as means ± SEM. Differences between CON and T2DM during the study days at repeatedly measured time points were analyzed by two-way ANOVA with adjustment for multiple testing by using the Tukey correction, where applicable. Baseline comparison between CON and T2DM was performed by using unpaired Student t tests. Paired Student t tests were used for the comparison between T2DM-P versus T2DM-D and CON-P versus CON-D. Correlation analysis was performed with the Spearman correlation coefficient.

With regard to sample size calculations, an ∼30% increase in EGP after a single dose of 10 mg dapagliflozin was expected based on previous studies (7), wherefore a sample size of six subjects per group was necessary to show differences between matched pairs with α = 0.05 and a power of >80%.

Ten healthy control subjects and six patients with T2DM were investigated. Baseline data of all participants are given in Table 1. With regard to anthropometric characteristics, patients with T2DM were older and had a higher BMI. Glucose metabolism was well controlled by metformin monotherapy in four patients with T2DM and with a combination therapy of metformin + DPP-4 inhibitor in two patients.

Administration of dapagliflozin resulted in a significant rise of glucosuria in CON (CON-P: 0.01 ± 0.00 g/study day vs. CON-D: 12 ± 0.6 g/study day; P < 0.001) and in T2DM (T2DM-P: 1.1 ± 0.4 g/study day vs. T2DM-D: 6.2 ± 1.6 g/study day; P < 0.5). The rise in dapagliflozin-induced glucosuria was substantially higher in CON than in T2DM (P < 0.001). Plasma glucose was lower in CON compared with T2DM and decreased significantly during all study days in both groups (P < 0.01 compared with baseline) (see Fig. 1). No difference in plasma glucose between CON-P and CON-D or between T2DM-P and TDM-D was observed. Concentrations of insulin and C-peptide were significantly lower after 390 min compared with baseline (P < 0.05), without differences between CON-P and CON-D or between T2DM-P and TDM-D. Ketone bodies were significantly higher at the end of each study day compared with baseline after administration of dapagliflozin only (see Fig. 1).

Glucagon levels were ∼30% higher throughout the study day after the administration of dapagliflozin in T2DM and CON (see Fig. 1), whereas no differences were observed in cortisol, growth hormone, ACTH, epinephrine, and norepinephrine (see Fig. 2). The insulin-to-glucagon ratio at the end of each study day was about two times higher in CON-P compared with CON-D and in T2DM-P compared with T2DM-D.

Concentrations of alanine, lactate, and free glycerol, as substrates for GNG, did not differ during the study days. All data are shown in detail in Table 2.

EGP was significantly higher after the administration of dapagliflozin in CON-D compared with CON-P (12.4 ± 0.5 vs. 10.0 ± 0.3 μmol kg⁻¹ min⁻¹; P < 0.05), but not in T2DM-D compared with T2DM-P (10.4 ± 0.5 vs. 10.1 ± 0.7 μmol kg⁻¹ min⁻¹; P = not significant [n.s.]). Tissue glucose uptake was not affected by SGLT-2 inhibition in CON (CON-P: 9.61 ± 0.3 vs. CON-D: 9.63 ± 0.4 μmol kg⁻¹ min⁻¹; P = n.s.) or in T2DM (T2DM-P: 10.02 ± 0.7 vs. CON-D: 9.11 ± 0.7 μmol kg⁻¹ min⁻¹; P = n.s). The increase in EGP in CON-D was almost identical with the amount of glucose excreted in the urine (2.84 ± 0.15 μmol kg⁻¹ min⁻¹). In addition, a strong positive association between the area under the curve for EGP and glucosuria (r = 0.829; P < 0.001) was observed (see Fig. 2).

Hepatic glycogen concentrations were comparable between all groups at baseline in CON and T2DM (see Table 2). Glycogenolysis throughout the whole study day was lower in T2DM compared with CON but did not differ between CON-P and CON-D (CON-P: −3.28 ± 0.49 vs. CON-D: −2.53 ± 0.56 μmol kg⁻¹ min⁻¹; P = n.s.) or between T2DM-P and T2DM-D (T2DM-P: −0.74 ± 0.23 vs. T2DM-D: −1.21 ± 0.33 μmol kg⁻¹ min⁻¹; P = n.s.). GNG was significantly higher after the administration of dapagliflozin in CON (CON-P: 9.9 ± 0.6 vs. 6.7 ± 0.6 μmol kg⁻¹ min⁻¹; P < 0.01), but not in T2DM (9.2 ± 0.32 vs. 9.36 ± 0.57 μmol kg⁻¹ min⁻¹; P < 0.01). Thus, the observed rise in EGP in CON after SGLT-2 inhibition by dapagliflozin is entirely mediated by an acute increase in GNG, whereas hepatic glycogen breakdown remains unchanged (see Fig. 2).

HCL was higher in T2DM compared with CON (7.5 ± 1.7 vs. 2.8 ± 0.8% of water signal; P < 0.01). HCL trended to increase compared with baseline after SGLT-2 inhibition in CON-D compared with CON-P (P = 0.080) and in T2DM-D compared with T2DM-P (P = 0.061) but did not reach statistical significance. kATP was higher in T2DM compared with CON at baseline (0.37 ± 0.03 vs. 0.23 ± 0.02 s⁻¹; P < 0.01), but no significantly different changes between MR-0 and MR-1 and between MR-0 and MR-2 could be found between the groups. With regard to concentrations of hepatic phosphorus-containing metabolites, glycerophosphoethanolamine, glycerophosphocholine, and phosphoethanolamine were higher in T2DM compared with CON but also did not show any significant dynamics throughout the individual study days. Data of HCL, kATP, and all phosphometabolites are given in detail in Table 2.

Consistent with previous studies, increased glucosuria induced by SGLT-2 inhibition is paralleled by a significant rise in EGP in healthy subjects, which instantly counteracts the glucose-lowering effects of dapagliflozin. Of note, we do not observe changes in glycogenolysis after the administration of dapagliflozin, wherefore we conclude that the observed 25% increase in EGP is driven by acute stimulation of hepatic and/or renal GNG.

These findings are of special interest, because glycogenolysis was previously shown to be rapidly stimulated by a standardized increase in circulating glucagon concentrations (21). In line with previous reports, glucagon concentrations during the study days and the insulin-to-glucagon ratio at the end of each study day are higher after dapagliflozin administration compared with placebo in CON and T2DM. However, in contrast to our initial expectations, the acute rise in glucagon levels after SGLT-2 inhibition paradoxically does not promote glycogenolysis.

The relative contribution of GNG to total EGP was substantially higher in T2DM compared with CON but was not different between T2DM-P and T2DM-D. The rates of glycogenolysis and GNG in healthy control subjects and patients with diabetes measured in the current study are in the expected range. We and others have previously shown ∼50% lower rates of glycogenolysis in patients with diabetes (19,22). In combination with lower amounts of glucosuria in T2DM compared with CON after a single dose of dapagliflozin, this might not have been sufficient to further stimulate the already excessive GNG in T2DM (contribution of GNG to EGP: 94% vs. 68% in T2DM-P vs. CON-P) (Fig. 2).

Of note, the mechanisms underlying the observed rise in EGP are up to now only poorly understood. Previously published studies using a single dose of empagliflozin, as well as dapagliflozin, yielded similar results compared with our data (6,7). Therefore, we assume that the observed effects of SGLT-2 inhibition are due to a class effect. The SGLT-2 receptor is expressed on pancreatic α-cells, and dapagliflozin treatment triggers glucagon release through direct activation in vitro, which initially emphasized the importance of glucagon secretion in the stimulation of EGP (9). In addition, the observed rise in glucagon concentrations due to SGLT-2 inhibition is paralleled by a decrease in circulating insulin levels, resulting in an altered hepatic insulin-to-glucagon ratio, which is in line with data from our experiments and was also initially suggested to be an important driver for EGP upregulation (6,7). On the other hand, this hypothesis is contrasted by in vivo studies in humans, showing that the rise in glucagon is mitigated when the plasma glucose concentration is clamped at a stable level by continuous glucose infusion (8).Interestingly, the EGP stimulation after dapagliflozin persists not only during conditions of stable plasma glucose levels but also when endogenous glucagon and insulin secretion is blunted by somatostatin infusion, which was recently demonstrated in elegant pancreatic clamp studies by Alatrach et al. (23). In line with these reports, coadministration of the glucagon-like peptide 1 analog liraglutide, which is a well-known insulin secretagogue and a potent inhibitor of glucagon secretion, prevents the SGLT-2 inhibitor-induced insulin decline and glucagon rise, whereas EGP stimulation is unaltered (24). Alterations in the hepatic insulin-to-glucagon ratio might therefore rather reflect a physiological counterregulatory response defending the established glucose levels, although the observed decline in plasma glucose is far away from a hypoglycemic range.

Because of the rapidity of changes in EGP within the 1st h after dapagliflozin administration, an involvement of the sympathic-adrenergic system was hypothesized to play a role, although the classical clinical signs, such as increased blood pressure or higher pulse rate, are missing (24). However, we could not find differences in the concentrations of epinephrine and norepinephrine after SGLT-2 inhibition. Also, no alterations in the secretion of other glucoregulatory hormones involved in hypoglycemia counterregulation were observed during our study experiments.

Another, but speculative, explanation for the rise in EGP might offer a yet unknown renal mediator triggered directly by SGLT-2 inhibition or glucosuria. This putative signal might act locally in the kidneys as a direct stimulator of renal GNG, which might be supposed to strongly contribute to the stimulation of EGP.

Our data clearly demonstrate that the rise in EGP is mainly driven by GNG, whereas hepatic glycogenolysis is unaltered after administration of an SGLT-2 inhibitor. It is known that hepatic glucose production accounts for ∼80% of EGP, of which GNG plays the major role after a prolonged fasting period (25). Of the 80% of oxygen consumption coupled to ATP synthesis in the liver, 7–10% is used by hepatic GNG (26). Therefore, increased hepatic GNG should be paralleled by a rise in ATP turnover. However, we did not observe a significant rise in kATP due to dapagliflozin in our experiments, suggesting that renal GNG might significantly contribute to enhanced EGP. In line with this hypothesis, studies in animal models report an activation of GNG gene expression in the kidney, but not in the liver, after 8 weeks of SGLT-2 inhibitor therapy (27). In addition, dapagliflozin increases the expression of gluconeogenic enzymes and GNG in human renal proximal tubular cells (28). With regard to the liver, mice models show that the expression of genes involved in hepatic GNG is upregulated after short-term treatment with an SGLT-2 inhibitor (9). On the other hand, a single dose of SGLT-2 inhibitor promotes hepatic GNG only in lean but not in obese mice with hepatic insulin resistance (29). Of course, our study design does not allow discerning the individual contribution of hepatic or renal GNG separately to the increase in total GNG; therefore, it can only be speculated on the actual source of increased GNG based on our data.

Another explanation for unaltered hepatic kATP, despite increased GNG, might be that it offers changes in substrate utilization after administration of dapagliflozin. It was previously reported that SGLT-2 inhibitor therapy promotes a shift from glucose to fatty substrate utilization and increases ketogenesis (11). Also in our study, ketone bodies were significantly higher after the administration of dapagliflozin in CON and T2DM. Higher levels of β-hydroxybutyrate, which is taken up in the liver in proportion to its circulating plasma concentration, might economize energy metabolism (5,30) and therefore probably counterbalance higher demands of increased hepatic GNG. However, because we are not able to trace individual substrate utilization by the applied MRS methods, this assumption cannot be confirmed by our study experiments.

With regard to hepatic lipid metabolism, HCL is higher in T2DM compared with CON, as expected. The higher kATP in T2DM is best explained by the increased mitochondrial lipid oxidation capacity observed in early stages of nonalcoholic fatty liver disease, which is consecutively lost with liver disease progression (17,31). Interestingly, HCL trended to increase after a single dose of dapagliflozin in CON and T2DM in our experiments. This might be due to the previously reported increase in lipolysis after SGLT-2 inhibition (11), probably increasing substrate availability and thereby promoting de novo lipogenesis in the liver. Thus, it might be assumed that the observed reduction of HCL after long-term administration of an SGLT-2 inhibitor (14,15) is rather secondary to weight reduction than mediated by direct antisteatotic effects.

The small sample size of patients with T2DM might be considered as the major limitation of our study. However, in sample size calculations, the number of included patients yielded a sufficient power to detect the estimated differences in EGP, according to previous protocols from our study group investigating short-term changes in hepatic glycogenolysis and GNG after alterations in plasma glucagon levels (21). In contrast to previous reports, we were not able to observe the expected increase in EGP in T2DM, which might be linked to a lower urinary glucose excretion. Because EGP strongly correlates with glucosuria, this most likely explains the missing increase after SGLT-2 inhibition in patients with diabetes in our study. Urinary glucose excretion was higher in CON compared with T2DM, probably due to a better kidney function. This is in line with observations from routine clinical care that the glucose-lowering effects of SGLT-2 inhibitors are enhanced in pronounced hyperglycemia and attenuate with a reduction in the glomerular filtration rate (32). In addition, a recently published study investigating the acute effects of empagliflozin on glucagon secretion was also not able to replicate the EGP increase in patients with diabetes (33). Therefore, a single-dose treatment with dapagliflozin might not have been adequate to achieve the full effects in the T2DM group. However, we do not assume that there might be a fundamental difference on the contribution of GNG to EGP in patients with diabetes compared with the control subjects, when the SGLT-2 receptor is inhibited sufficiently, because of comparable underlying pathophysiological mechanism.

It is important to mention that our study design does not allow us to draw conclusions regarding long-term treatment with an SGLT-2 inhibitor in routine clinical care in combination with other glucose-lowering drugs, including metformin, because only the acute effects after a single dose of dapagliflozin were investigated. However, this setting was chosen to investigate the direct impact of SGLT-2 inhibition, which might not be possible after long-term therapy, because optimization of glycemic control itself was previously shown to distinctly affect hepatic glycogen breakdown and synthesis (34).

Taken together, the results of our study clearly demonstrate that an acute rise in GNG, but not glycogenolysis, causes the increase in EGP after a single dose of dapagliflozin in healthy control subjects. Higher glucagon concentrations and a lower insulin-to-glucagon ratio after SGLT-2 inhibition are not associated with an increased hepatic glycogen breakdown. HCL insignificantly trends to increase in the short-term, and no changes in hepatic ATP turnover are observed, speaking against direct antisteatotic effects in the liver.

Acknowledgments. The authors want to thank Marie-Bernadette Aretin and Zivadinka Nikolic from the Hospital Pharmacy, Medical University of Vienna, for performing the blinding of the study medication, as well as all subjects for participating in study-related activities.

Funding and Duality of Interest. This study was partly funded by AstraZeneca through an investigator-initiated research grant and was partly funded by an unrestricted research grant from AstraZeneca to M.Kre. M.Kre. has received research support from Sanofi and AstraZeneca as well as speaker and consulting fees from AstraZeneca, Novartis, Novo Nordisk, Lilly, Merck, Boehringer, and Sanofi. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. P.W., P.F., L.P., H.B., P.K., C.B., A.G., J.H., M.M., T.S., M.Z., S.B.-P., R.M., M.Krš., and M.Kre. performed experiments. P.W., P.F., L.P., H.B., P.K., C.B., A.G., J.H., S.B.-P., R.M., S.T., A.K.-W., M.Krš., and M.Kre. wrote the manuscript. P.W., P.F., H.B., J.H., and M.Kre. contributed to subjects’ recruitment. P.W., C.B., A.G., M.M., T.S., M.Z., S.B.-P., R.M., M.Krš., and M.Kre. contributed to data analysis and interpretation. P.W., M.Krš., and M.Kre. designed the study. P.W. and M.Kre. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.