The SGLT2 Inhibitor Dapagliflozin Increases the Oxidation of Ingested Fatty Acids to Ketones in Type 2 Diabetes

Sodium–glucose cotransporter 2 (SGLT2) inhibitors are a class of drug that promotes urinary glucose excretion to lower blood glucose levels. In type 2 diabetes inadequately controlled with metformin, dapagliflozin as monotherapy (1) and when added to metformin improves glycemic control (2,3), reduces weight, and results in less hypoglycemia than glipizide (4). In people whose type 2 diabetes was inadequately controlled with insulin therapy, adding dapagliflozin reduced HbA_1c levels and weight over 48 weeks without increasing overall insulin dose (5).

The observation that ketogenesis or episodes of euglycemic ketoacidosis are of higher frequency in people with type 1 and type 2 diabetes on SGLT2 inhibitors is of considerable worldwide interest (6,7). Ferrannini et al. (7) showed that after 4 weeks of treatment of type 2 diabetes with the SGLT2 inhibitor empagliflozin, lipid oxidation rate increased during a meal test. This finding suggests that the homeostatic response to losing glucose by urinary excretion is to switch metabolism in favor of nonesterified fatty acids (NEFAs). Animal models of type 2 diabetes and heart failure have shown that treatment with SGLT2 inhibitors increases exercise-induced ketogenesis and fatty acid oxidation (8,9). In a previous study, we found that treatment with dapagliflozin in type 1 diabetes after insulin withdrawal increased β-hydroxybutyrate (BOHB) concentration by 27%, despite no increase in NEFA concentrations and only a small increase in lipolysis (10). This finding suggests that mechanisms additional to increased delivery of fatty acids to the liver may drive ketogenesis with SGLT2 inhibitors.

In the current study, the contribution of ingested fatty acids to ketogenesis was investigated using tracer-labeled olive oil following 28 days administration of dapagliflozin in people with type 2 diabetes. In addition, stable isotopes of glucose and glycerol were infused to measure glucose flux and lipolysis, respectively.

The study was a randomized, double-blind, placebo-control, cross over trial of people with type 2 diabetes at the Cedar Centre, Royal Surrey NHS Foundation Trust. Ethics approval was granted from the Southern Central–Berkshire B Research Ethics Committee (Bristol, U.K.). The clinical trial was registered with the European Clinical Trials Database (identifier: 2016-004878-17) and clinicaltrials.gov (identifier: NCT04219124). The study was funded AstraZeneca Ltd. Inclusion criteria were diagnosis of type 2 diabetes >12 months; single-, dual-, or triple-therapy glucose-lowering agents comprising of sulfonylureas, biguanides, and dipeptidyl peptidase 4 but no previous exposure to SGLT2 inhibitors; age 18–75 years; BMI <40 kg/m²; and HbA_1c of ≥6.5% (48 mmol/mol) and <9% (75 mmol/mol) within 1 month of screening. Exclusion criteria included proliferative retinopathy requiring acute treatment within the previous 3 months, moderate to severe renal impairment (creatinine clearance <60 mL/min or estimated glomerular filtration rate <60 mL/min/1.73 m²), severe hepatic impairment, New York Heart Association class III–IV cardiac failure, uncontrolled cardiac arrhythmia, uncontrolled hypertension, mental incapacity, pregnancy or breastfeeding, women of child-bearing potential not taking adequate contraception precautions, and suspected allergy to trial products.

The primary outcome was the effect of dapagliflozin compared with placebo on plasma 3-hydroxybutyrate concentration. Secondary outcomes included the effect of dapagliflozin compared with placebo on hepatic ketone body metabolism, whole-body lipolysis, glucose metabolism, and the metabolome.

The primary end point was plasma BOHB concentration at the end of the meal study. Secondary end points were glucose production and disposal rates, glycerol production rate, palmitate enrichment in plasma NEFA and triacylglycerol (TAG), 3-hydroxybutyrate enrichment, glucagon and insulin concentrations, and plasma acylcarnitines measured by targeted metabolomics.

Participants gave written informed consent. Following screening and random assignment, participants received either 10 mg daily dapagliflozin or placebo for 28 days before the metabolic study visit. After a 28-day washout period, they switched to the other treatment (Supplementary Fig. 1). Participants were made aware of potential changes in glycemic control and were asked to record trial medication administration, any concomitant medication, hypoglycemia frequency (capillary glucose level <4 mmol/L), fasting ketone levels, and any adverse events.

Participants were asked to refrain from drinking alcohol and strenuous exercise for 24 h before the metabolic assessment day. They were given a standardized 500-kcal meal the evening before the test and were asked to fast overnight. They were asked to return their unused investigational medicinal product (IMP) at this visit. Intravenous cannulae were inserted into the antecubital fossa of each arm: one for taking blood samples and the other for the infusion of the glucose and glycerol isotopes. At 0 min, the study medication was taken, and a stable isotope of [U-¹³C]palmitate (200 mg, 98% enriched; CK Isotopes Ltd., Leicestershire, U.K.) in 30 mL virgin olive oil and a few drops of lemon juice (to make it palatable) was ingested in <2 min. Dilution of [U-¹³C]palmitate with palmitate in olive oil resulted in a tracer:tracee ratio (TTR) of 0.0384. Fasting blood samples were taken at −10 and 0 min to measure baseline enrichments and concentrations, and then samples were taken at regular intervals until 480 min to measure the concentrations of NEFAs, BOHB, TAGs, LDL cholesterol, insulin, and glucagon, as well as stable isotope enrichment and concentration in plasma of TAGs, NEFAs, and BOHB. An additional blood sample was taken at 480 min for targeted metabolomics. A primed infusion of [6,6-²H₂]glucose and [1,1,2,3,3-²H₅]glycerol was administered from 360 to 480 min, and the concentration and enrichment of both glucose and glycerol measured from 450 to 480 min. Urine was collected to measure glucose excretion at 120, 360, 450, and 480 min. At the end of the study, participants were prescribed the IMP or placebo required for the second arm of the study, depending on the randomization code, to be taken after the washout period. The metabolic study was repeated after 28 days of treatment.

Plasma glucose concentrations were measured with a Roche cobas MIRA analyzer using the ABX Pentra Glucose Kit (Horiba ABX, Northampton, U.K.) and plasma glycerol and BOHB concentrations using Glycerol and Ranbut assays (Randox Laboratories, Co., Antrim, U.K.). Plasma NEFA concentrations were measured using an enzymatic kit (Wako Chemicals GmbH, Neuss, Germany). Plasma total cholesterol, LDL cholesterol, and TAG concentrations were measured using enzymatic calorimetric kits ABX Pentra Cholesterol CP, ABX Pentra LDL Direct CP, and ABX Pentra Triglyceride CP, respectively (Horiba ABX).

Glucagon concentrations were measured using radioimmunoassay (Merck Millipore, Merck Chemicals, Nottingham, U.K.). Insulin concentration was measured using an ELISA (DRG Instruments GmbH, supplied by Immunodiagnostic Systems, Tyne and Wear, U.K.).

The isotopic enrichment of plasma glucose was determined as the trimethylsilyl-O-methyloxime derivative (11) using gas chromatography-mass spectrometry (GC-MS) (5975C Series Inert XL EI/CI MSD; Agilent Technologies, Wokingham, U.K.). The isotopic enrichment of plasma glycerol was determined as the tert-butyltrimethylsilyl glycerol derivative (12) using the same GC-MS system.

The isotopic enrichment of [U-¹³C]palmitate was determined in the fatty acid methyl ester moieties in plasma TAGs and NEFAs. It was possible to determine the concentration of the palmitate in NEFAs and TAGs by addition of internal standards of heptadecanoic acid and triheptadecanoate to the samples before extraction. Extraction was by the Folch method, and TAGs and NEFAs were purified by thin-layer chromatography visualized by 8-anilino-1-napthalene-sulfonic acid in water. The scraped TAG and NEFA spots were collected into tubes and solubilized with toluene and esterified using sulphuric acid in methanol (6%) at 80°C for 2 h. The samples were then neutralized and extracted into hexane. Fatty acid methyl esters in hexane were analyzed by GC isotope ratio MS (GC-C-IRMS, Thermo Scientific).

The isotopic enrichment of BOHB was determined as tert-butyldimethylsilyl derivative of BOHB (13). Blood taken on the day of study into cooled fluoride/EDTA-containing tubes was centrifuged immediately at 1,600g for 10 min at 4°C. Plasma (2 mL) was then deproteinized with cooled 2 mL perchloric acid (10% w/v) mixed vigorously and stored at −80°C until analysis. On the day of analysis, the perchloric acid–containing samples were thawed on ice and centrifuged at 1,600g for 20 min at 4°C. The supernatant was transferred to a new tube and neutralized by dropwise addition of the neutralizing agent potassium bicarbonate/potassium carbonate (1.5 mol/L) and mixing after each drop to release trapped CO₂ in the solution during the neutralization step. The volumes of the sample were recorded before and after neutralization. The sample was then centrifuged to eliminate the salt formed, and the pH was checked to be between 7 and 8. Following the reacidification of the sample (0.5 mL) to pH 1 with HCl (1 mol/L), the metabolites were extracted with ice-cold mixed solution of ethylacetate:diethylether (1:1 v/v), vortexed for 15 min, and centrifuged at 710g for 20 min at 4°C. The solvent layer was transferred to another tube and derivatized by adding pyridine (20 μL) and N-methyl-N-tert-butyldimethylsilyltrifluoroacetamide containing 1% tert-butyldimethylsilyl (20 μL) overnight at room temperature in the dark. The solvent was evaporated and the sample reconstituted in decane for analysis by GC-MS.

Acylcarnitines were measured by liquid chromatography-MS (ACQUITY UPLC with Xevo TQ-S Mass Spectrometer; Waters Corporation, Milford, MA) using the AbsoluteIDQ p180 Kit (Biocrates Life Sciences AG, Innsbruck, Austria). Plasma samples were measured in duplicate. A total of 40 acylcarnitines were measured. Any that had >25% of concentrations below the limit of detection were excluded. Samples were run on 96-well plates in random order with three levels of quality control.

Glucose R_a, R_d, and glycerol production were calculated using standard isotope dilution equations (14) between 450 and 480 min and corrected for the baseline enrichment level. Glucose uptake was determined by subtracting the rate of glucose excretion from glucose R_d.

Areas under the curve (AUCs) were calculated between study points 0 and 480 min without and with correction for the baseline (incremental AUC). Mean concentrations and TTRs were calculated between study points 450 and 480 min. Tracer concentrations of BOHB and palmitate in NEFA and TAG pools were calculated as atom percent enrichment × measured plasma concentration.

A power calculation was based on the primary end point, which was the plasma BOHB concentration at the end of the meal study. It was calculated from a study that investigated the effect of 7 days of treatment with the SGLT2 inhibitor luseogliflusin on plasma BOHB, 11 h after the oral administration of the IMP, in participants with type 2 diabetes in a randomized, double-blind, placebo-controlled, crossover study (15). Based on the mean differences in this study, completing the current study in 12 participants should detect a minimum mean ± SD difference in plasma BOHB concentration of 0.34 ± 0.38 mmol/L (35.4 ± 39.8 ng/mL) between placebo and treatment in a crossover design with 80% power using a two-sided t test with 5% level of significance. The trial was terminated after nine patients because of the COVID-19 pandemic. A retrospective power calculation showed that with nine participants, there is 72% power to detect the observed mean difference of 0.255 in BOHB concentration (450–480 min) as statistically significant in a two-sided hypothesis test with 5% level of significance. The a priori power calculation assumed a large effect size of 0.90 (0.340/0.380) and based on a two-sided statistical test, required 12 subjects to demonstrate a significant difference with 80% power. Had a one-sided test been proposed, as might have been reasonable since it has been shown previously that ketogenesis is higher with SGLT2 inhibitors, 10 participants would have been required to demonstrate such an effect, and 9 participants would have been sufficient to demonstrate an effect size of ≥0.91 with the same power. The observed effect sizes were 0.99 and 0.97, respectively, so it is unlikely that these differences were due to chance. (The post hoc power calculations for the observed differences on the basis of a one-sided test are 85.7% and 84.0%, respectively.)

All results are mean ± SEM. The primary end point is the statistical evaluation of contrasts between placebo and treatment on plasma BOHB concentration, as measured at 480 min. Final plasma BOHB concentration was statistically analyzed as the response variable in a generalized linear mixed model (using the PROC Mixed procedure in SAS software), with treatment, period, and treatment-by-period interaction as fixed effects and the baseline glucose concentration as a covariate. The participant was the random effect in the model. The denominator df were adjusted using Kenward-Roger approximation.

All outcome variables measured at a single point were analyzed in the same way as the primary end point. All outcome variables measured on repeated time points within each period were statistically contrasted between the treatment and placebo using a generalized linear mixed model with treatment, period, time, and treatment-by-period and treatment-by-time interactions as fixed effects, baseline measurement as a covariate, participant as random effect, and time as a repeated measure with a spatial power variance-covariance structure. Denominator df were adjusted using Kenward-Roger approximation.

Normality of the data was assessed using residual plots. If evidence of any nonnormality was found, then the data were log₁₀-transformed, and estimated treatment differences (and their CIs) were back-transformed into ratios before reporting.

Mixed-model analysis was performed using SAS version 9.4 software. Analysis of the acylcarnitines was by paired t test, and linear relationships were assessed by Pearson correlation using SPSS version 25.0 software.

Nine participants completed the study (age 61.7 ± 3.7 years, BMI 28.9 ± 1.0 kg/m², HbA_1c 7.7 ± 0.2% [60.2 ± 2.5 mmol/mol], three women, six men, eight Caucasian, one Asian). All participants were on metformin, and five were taking sulfonylureas of whom four were also taking dipeptidyl peptidase 4 inhibitors. Six participants were taking statins for hypercholesterolemia, and four were prescribed antihypertensives. One participant had documented ischemic heart disease, and one had known diabetic retinopathy. Duration of diabetes was 8.8 ± 0.9 years. Baseline demographics are shown in Supplementary Table 1, and baseline medications are shown in Supplementary Table 2.

At baseline, glucagon was higher after dapagliflozin (P = 0.02), but there was no difference in glucose and insulin concentrations (Table 1). Baseline BOHB, TAGs, and NEFAs were not different, but total and LDL cholesterol were significantly lower with dapagliflozin (P = 0.01 and 0.008, respectively).

With dapagliflozin, glucose AUC_0–480min was lower (P = 0.008). Insulin AUC_0–480min (P = 0.084) and glucagon AUC_0–480min (P = 0.089) tended to be different without reaching statistical significance. At 450–480 min, mean glucose concentration (P = 0.002) and peripheral glucose uptake (P = 0.002) were lower, glucose excretion was higher (P = 0.000), and glucagon concentration was higher (P = 0.01) with dapagliflozin than placebo (Table 2). The significantly higher glucagon concentration with dapagliflozin was lost when the data were corrected for the baseline concentration. Endogenous glucose production was not different. There was a trend for a higher mean glucagon:insulin ratio_450–480min (P = 0.055). Glycerol R_a, a measure of lipolysis at 450–480 min, was not different between treatments.

AUC_0–480min for TAG and NEFA concentration (Table 1) and mean concentration (450–480 min) for TAG palmitate TTR, NEFA palmitate concentration, and NEFA palmitate TTR were not different between treatments (Table 2). Plasma NEFA concentration was higher at 300 and 420 min (P = 0.04 and 0.034, respectively), but the mean NEFA concentration AUC_450–480min was not different (Fig. 1 and Table 2).

BOHB concentration did not change significantly in the placebo group after consuming olive oil with tracer. With dapagliflozin, there was a significant effect of time (P = 0.0002), and concentrations were significantly higher than placebo from 360 min onward (Fig. 1). By 450–480 min, the mean difference in BOHB concentration between treatments was 0.256 ± 0.087 mmol/L (P = 0.023). With both treatments, BOHB TTR increased more rapidly than NEFA palmitate TTR, peaking at 240 min with dapagliflozin treatment and then declining much more slowly than NEFA palmitate TTR (Fig. 2). Dapagliflozin versus placebo BOHB TTR AUC_0–480min (1.450 ± 0.129 vs. 1.207 ± 0.126 TTR ∗ min) was higher (P = 0.009), and mean BOHB TTR_450–480min was higher (P = 0.03) (Table 2). BOHB tracer concentration AUC_0–480min (0.399 ± 0.085 vs. 0.154 ± 0.030 mmol/L ∗ min) and mean BOHB tracer concentration_450–480min were higher with dapagliflozin than placebo (P = 0.019 and 0.017), respectively (Fig. 2 and Table 2). Acetylcarnitine, hydroxybutyryl carnitine, hydroxytetradecanoylcarnitine, octadecanoylcarnitine, and octadecadienylcarnitine were significantly higher with dapagliflozin (all P < 0.05) (Supplementary Table 3). Baseline BOHB concentration correlated with baseline NEFA concentration with both dapagliflozin (r = 0.68, P = 0.04) and placebo (r = 0.78, P = 0.015), but there was no correlation between mean plasma concentrations of BOHB_450–480min and NEFA_450–480min in either treatment group. BMI negatively correlated with both BOHB_450–480min concentration (r = −0.806, P = 0.009) and BOHB tracer concentration_450–480min (r = −0.810, P = 0.008) with dapagliflozin but not placebo.

We used a novel application of a stable isotope test to demonstrate that in type 2 diabetes, dapagliflozin compared with placebo results in a greater increase in ketone concentration after the ingestion of olive oil. Olive oil was ingested in the absence of carbohydrate to minimize an increase in insulin secretion and preserve a steady state at the end of the time course. The tracer in the oil was designed to label the plasma NEFA pool, with [U-¹³C]palmitic acid as a precursor for hepatic BOHB production to measure ketogenesis. The test was designed to initially label the plasma TAG pool, and with subsequent hydrolysis by the enzyme lipoprotein lipase located on capillary endothelium, the systemic NEFA pool would become labeled (16). When part of a mixed meal, fatty acids have previously been shown to be precursors for BOHB synthesis in healthy people (17). The approach here was successful, as indicated by the respective enrichment profiles. The rise in [¹³C₂]BOHB tracer concentration with both treatments demonstrates the oxidation of the ingested [U-¹³C]palmitate into ketones. This finding was similar in both groups at the early time points, but at the end of the study, the higher BOHB tracer concentration with dapagliflozin suggests a higher oxidation rate to ketones. The higher concentration of both short-chain and long-chain acylcarnitines with dapagliflozin is consistent with an increase in fatty acid oxidation. The accumulation of acylcarnitines has previously been documented in fasting (18) and following a liquid low-calorie diet (19) as a result of inefficient β-oxidation. The urinary excretion of acetylcarnitine has been shown to be correlated with blood BOHB concentration in normal-weight patients during fasting and in patients with diabetic ketoacidosis (20).

Glycerol R_a, a measure of lipolysis, was not different between treatments. Further evidence for a lack of effect of dapagliflozin on lipolysis was the similar plasma palmitic acid TTR profile with the two treatments. An increase in lipolysis will dilute the tracer, and TTR would be expected to be lower with dapagliflozin. Interestingly, the rise in palmitic acid TTR was much slower than the BOHB TTR, which suggests that this is not the direct source of ketones at the early time points. High-fat diets stimulate ketogenesis in the intestine, and ketones have been measured in the portal vein. Intralipid gavage in mice fed a high-fat diet has been shown to result in elevated ketone levels in portal blood within 30 min and in venous blood after 120 min (21), demonstrating that ketogenesis in the intestine can increase circulating ketone levels. Thus, it is possible that some of the ingested [U-¹³C]palmitic acid was directly oxidized to ketones in the enterocyte. BOHB TTR was similar at the early time points in both treatments, which suggests that dapagliflozin did not increase enterocyte ketogenesis. However, although enterocyte ketogenesis may initially be the primary source, it seems unlikely that this would still be the case by 480 min. Hepatic ketogenesis from plasma palmitate is the most likely source for the rise in BOHB TTR from 120 min. The fall in plasma NEFA palmitate TTR below BOHB TTR from 450 min suggests that storage of the labeled palmitate within the liver, presumably in TAGs, provides an additional source for ketogenesis.

The robust increase in the isotopic enrichment of the plasma TAG pool, which would be expected to recycle back to the liver as chylomicron remnants, could also provide additional potential precursor fatty acyl moieties for BOHB synthesis. The TTR profile was similar to NEFA palmitate TTR, peaking at 180 min, albeit at a lower enrichment. This fraction is a mixture of chylomicrons and VLDL, and the chylomicron TAG TTR would be diluted by VLDL TAG.

The correlations are interesting. Although as expected baseline BOHB concentration correlated with baseline NEFA concentration with both placebo and dapagliflozin treatment, there was no correlation between plasma BOHB and NEFA concentrations at the end of the study, further suggesting that plasma NEFA concentrations are not the major driver of ketogenesis. The negative relationship between BMI and BOHB concentration in the current study was also shown in our previous study of dapagliflozin in type 1 diabetes (10). This was a small group but may suggest a greater risk of BOHB and, therefore, ketosis in people with type 2 diabetes and a lower BMI and may be of clinical significance.

The lower glucose concentration with dapagliflozin caused by a rise in glucose excretion is well documented (1). Glucose uptake into peripheral tissues was reduced with dapagliflozin, and this is the likely mechanism for the drive in ketogenesis. The brain is a major user of glucose. In the fasting state, the brain accounts for at least 50% of whole-body glucose uptake (22). The participants were fasted overnight, and only olive oil was ingested, so by the end of the study, no carbohydrates had been ingested for 20 h. The brain plays a central role in metabolic homeostasis, sensing glucose levels and regulating endogenous glucose production and lipolysis partly through efferent pathways to the periphery (23). BOHB and acetoacetate are the brain’s main physiological alternative fuels to glucose, and brain ischemia in mice induces ketogenesis in liver, mediated by β-adrenergic receptors (24). That there is higher ketogenesis with dapagliflozin compared with placebo in the absence of any difference in lipolysis and only minor changes in NEFA concentrations suggests that there is a metabolic switch in the liver to increase ketogenesis rather than this being driven by increased NEFAs and may possibly be mediated by the brain. However, participants in the current study did not develop ketoacidosis. It is likely that while there may be a metabolic switch in the liver to increase ketogenesis in response to dapagliflozin, increased rates of white adipocyte lipolysis are still necessary to develop ketoacidosis.

It is possible that the study was underpowered for NEFA concentrations. Ferrannini et al. (25) studied the effect of 28 days of empagliflozin treatment in 66 patients with type 2 diabetes in an open-label study. Fasting mean BOHB increased from 0.3 mmol/L at baseline to 0.6 mmol/L after treatment, an increase comparable to the current study. Although fasting NEFAs significantly increased (P < 0.05), the increase was small at only 7.8%. However lipolysis measured by glycerol R_a was not different as in the current study.

We cannot exclude an effect of glucagon to increase ketosis, since this was higher following dapagliflozin treatment. However, the role of glucagon in promoting ketogenesis has recently been challenged. In mice, neither fasting nor SGLT2 inhibitor–induced ketosis was altered by interruption of glucagon signaling (26), and in a rat model of type 2 diabetes, the effect of dapagliflozin to promote ketosis was independent of hyperglucagonemia (27).

Previous studies have suggested that the rise in ketones with SGLT2 inhibitors may be due to reduced renal excretion or reduced uptake of BOHB by peripheral tissues. SGLT2 inhibition can reduce glomerular filtration rate in type 1 and 2 diabetes and reduce renal ketone excretion (28). While we cannot rule out a contribution to increased BOHB through these mechanisms, our study clearly shows that dapagliflozin increases ketone synthesis.

The finding of lower total cholesterol and LDL cholesterol with dapagliflozin is interesting. Most previous studies have shown dapagliflozin to either increase LDL cholesterol (29) or have no effect in patients treated with statins (30). One study found a reduction in small dense LDL (31), while real-life data reported a decrease in LDL cholesterol after 6 months of treatment (32).

In summary, this study shows that dapagliflozin increases the oxidation of ingested fatty acids to ketones. There is no evidence that this is driven by increased plasma NEFA concentrations as in diabetic ketoacidosis, and we suggest that a metabolic switch within the liver increases ketogenesis. We also demonstrate that the initial rise in BOHB may be due to ketogenesis in the intestine.

Acknowledgments. The authors thank the Metabolomics Core Facility (University of Surrey) and Dr. Dovile Lingaityte for the targeted liquid chromatography-MS analysis, and Rachael Gribbin (University of Surrey) for laboratory assistance.

Funding and Duality of Interest. AstraZeneca Ltd. funded the study (ESR-15-11442) and provided the IMP. M.D. has acted as consultant, advisory board member, and speaker for Novo Nordisk, Sanofi, Eli Lilly, Merck Sharp & Dohme, Boehringer Ingelheim, AstraZeneca, and Janssen; as an advisory board member for Servier and Gilead Sciences; and as a speaker for NAPP, Mitsubishi Tanabe Pharma Corporation, and Takeda Pharmaceuticals International. M.D. also has received grants in support of research trials from Novo Nordisk, Sanofi, Eli Lilly, Boehringer Ingelheim, AstraZeneca, and Janssen. R.A.H. and D.L.R.-J have received research funding or advisory board or lecture fee honoraria from AstraZeneca. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. R.A.H., A.M.U., B.A.F., M.D., and D.L.R-J. designed the study. F.S.-M. and M.S. carried out the metabolic studies. I.P. carried out participant-related study procedures as aligned to study protocol. F.S.-M. and N.J. carried out sample analyses. F.S.-M., A.M.U., and B.A.F. interpreted the data and drafted the manuscript. J.M. completed the statistical analysis. B.M. conducted the metabolomic analysis. All authors reviewed and edited the manuscript. R.A.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.