1 Introduction
Sodium-glucose co-transporter 2 (SGLT2) inhibitors were originally developed as oral anti-diabetic drugs that lower plasma glucose and glycated haemoglobin by promoting urinary glucose excretion [
1,
2]. In addition to improving glycaemic control, SGLT2 inhibitors exert beneficial effects on risk markers for kidney disease, such as body weight, systolic blood pressure and albuminuria [
3,
4]. Large outcome trials have shown that SGLT2 inhibitors improve cardiovascular outcomes and delay the progression of kidney function decline in patients with type 2 diabetes mellitus and chronic kidney disease [
5,
6]. These clinical benefits are unlikely explained by improvements in glycaemic control alone and are thought to be the result of direct effects on kidney and systemic vascular haemodynamics [
7]. This suggests that SGLT2 inhibitors may also have beneficial effects in patients without diabetes with cardiovascular or kidney disease.
As SGLT2 inhibitors were originally developed for the treatment of type 2 diabetes, earlier pharmacokinetic and dose-finding studies focused on characterising the pharmacokinetics and pharmacodynamics of SGLT2 inhibitors in patients with type 2 diabetes with or without kidney disease [
8]. These studies revealed that the plasma exposure of SGLT2 inhibitors increased in patients with impaired kidney function [
9‐
12]. In contrast, pharmacodynamic effects of SGLT2 inhibitors on glucose excretion were attenuated in patients with impaired kidney function due to less glucose filtration [
9‐
12]. The effects on body weight and blood pressure appear to be preserved [
13].
Emerging data suggest that the benefits of SGLT2 inhibitors on kidney outcomes likely extend to patients without diabetes as well [
14‐
17]. The DAPA-HF (Dapagliflozin And Prevention of Adverse Outcomes in Heart Failure) and DAPA-CKD (Dapagliflozin And Prevention of Adverse Outcomes in Chronic Kidney Disease) trials assessed the effects of the SGLT2 inhibitor dapagliflozin in broad cohorts of patients with heart failure or chronic kidney disease, respectively [
15,
17]. Patients with or without type 2 diabetes participated in both trials [
18]. Given the favourable results of the aforementioned outcome trials, it is likely that SGLT2 inhibitors will be prescribed in a large cohort of non-diabetic patients. In the design of these trials, it was assumed that the pharmacokinetic profile of SGLT2 inhibitors in patients without diabetes is similar to that in patients with diabetes. However, empirical data confirming this assumption are lacking.
We therefore aimed to characterise the pharmacokinetic profile of the SGLT2 inhibitor dapagliflozin in patients with non-diabetic kidney disease. Subsequently, we investigated the association between dapagliflozin plasma exposure and several risk markers for kidney disease.
2 Methods
2.1 Study Design and Patient Population
Data were used from the “Effects of the SGLT2 inhibitor dapagliflozin on proteinuria in non-diabetic patients with chronic kidney disease” (DIAMOND) trial (NCT03190694), a randomised, placebo-controlled, double-blind, cross-over trial that assessed the kidney protective effects of dapagliflozin in non-diabetic patients with albuminuria. The study design and primary results have been reported elsewhere [
19]. In short, the DIAMOND trial enrolled 53 participants with non-diabetic kidney disease, characterised by 24-h urinary protein excretion > 500 mg/day and ≤ 3500 mg/day, and an estimated glomerular filtration rate (eGFR) ≥ 25 mL/min/1.73 m
2. Participants had to be treated with a stable dose of an angiotensin-converting enzyme inhibitor or angiotensin II receptor blocker for at least 4 weeks prior to enrolment. Participants were randomly assigned, in a cross-over design, to placebo followed by dapagliflozin 10 mg once daily, or dapagliflozin 10 mg once daily followed by placebo. Each treatment period lasted 6 weeks, followed by a 6-week wash-out period to avoid carry-over effects. The primary endpoint of the trial was 24-h proteinuria and secondary endpoints included body weight, measured glomerular filtration rate (mGFR), systolic blood pressure and urinary albumin-to-creatinine ratio (UACR). The study was performed in accordance with the Declaration of Helsinki and good clinical practice guidelines and participants gave their written informed consent before any study-specific procedure commenced.
2.2 Measurements
Twenty-four-hour urine was collected to monitor proteinuria at the start and end of each treatment period. Body weight and systolic blood pressure were recorded at every visit to the clinic. Measured glomerular filtration rate was estimated by determining the plasma clearance of non-radioactive iohexol at the beginning and end of each treatment period. At the end of the treatment period, during GFR measurement, plasma samples of dapagliflozin were collected pre-dose, and every 30 min for 4 h after administration of dapagliflozin or placebo. Actual sampling and dosing times were recorded. The plasma concentration of dapagliflozin was measured using a liquid chromatography–tandem mass spectrometry method, which has been described elsewhere [
20]. This bioanalytical method was validated for selectivity, linearity, accuracy and precision, dilution integrity, stability and recovery. The accuracy was between 94.6 and 101.0% and precision (coefficient of variation) was between 0.0 and 13.7%.
2.3 Estimation of Individual Exposure to Dapagliflozin
A population pharmacokinetic model was used to estimate individual plasma exposure to dapagliflozin. Non-linear mixed-effects models were used to develop this population pharmacokinetic model. Model development was conducted using NONMEM version 7.3.0 (ICON Development Solutions, Ellicott City, MD, USA).
Different structural models with linear absorption and elimination processes were evaluated, including one- and two-compartment models with and without a lag time. Furthermore, the inclusion of transit compartments in the model to describe the absorption phase was also explored. A log-normal distribution was assumed for the inclusion of random effects in the stochastic model. Covariance between random effects was also evaluated. Additive, proportional and combined error models were explored to describe the residual variability. Covariate screening was performed for age, sex, race, ethnicity, eGFR, mGFR, body weight and region. We used correlation matrices of the empirical Bayes estimates of the parameters vs covariates to evaluate potential relationships. For discrete covariates, separate population parameters were estimated. For body weight, allometric scaling normalised by 70 kg was explored and, for other continuous covariates, the covariate was median normalised and a power coefficient was estimated.
First-order conditional estimation with interaction was used to obtain model parameters. Model selection and evaluation were based on the minimum objective function value (MOFV), standard goodness-of-fit plots, condition number, residual standard error of parameter estimates, and coefficient of variation of the random effects representing residual and random variability [
21]. The predictive performance of the model was evaluated using a visual predictive check.
2.4 Evaluation of the Association between Exposure and Kidney Response
Risk markers of interest were proteinuria, UACR, mGFR, systolic blood pressure and body weight, which are well-known risk markers for progression of kidney disease. The individual change from baseline was estimated for all risk markers in both the placebo as well as the active treatment period. For proteinuria and UACR, the change from baseline was log-transformed to approximate a normal distribution.
Using the population pharmacokinetic model, the plasma exposure, defined as the area under the plasma concentration–time curve (AUC0–inf), was estimated by dividing the 10-mg dose by the individual apparent clearance parameter. The association between exposure to dapagliflozin, in terms of AUC0–inf, and response was investigated using linear mixed-effects models. A random intercept model was fitted to the data to estimate the placebo response of each individual patient, which was compared to a random intercept model including AUC0–inf as fixed effect. A model comparison was performed using a likelihood ratio test, which assumes a chi-square distribution. A significant increase in the maximum likelihood indicates that the addition of AUC0–inf to the model explains residual and/or between-patient variability. Furthermore, a t test was performed to evaluate whether the fixed regression coefficient of AUC0–inf was significantly different from zero. All linear mixed-effects models were fitted using full maximum-likelihood estimation. The linear mixed-effects models were fitted using the lme function of the nlme package (version nlme_3.1-131) in R version 3.4.2 (R Foundation for Statistical Computing, Vienna, Austria).
4 Discussion
We developed a population pharmacokinetic model that adequately described the individual plasma concentration–time profile of dapagliflozin in patients with non-diabetic kidney disease. We found that plasma exposure to dapagliflozin was higher in patients with impaired kidney function and in patients with relatively low body weight. Furthermore, we demonstrated that individual dapagliflozin plasma exposure was associated with changes in UACR, mGFR and systolic blood pressure.
As there is increasing interest in the use of dapagliflozin and other SGLT2 inhibitors in patients without diabetes with cardiovascular or kidney disease, it is important to characterise the plasma concentration–time profile of dapagliflozin in patients without diabetes and assess whether this profile is comparable to patients with diabetes. A population pharmacokinetic model for dapagliflozin in patients with type 2 diabetes and chronic kidney disease has been described by van der Walt et al., who reported that a two-compartment structural model, with first-order elimination and usage of multiple transit compartments for the absorption phase, best described the data [
22]. Similar to the model of van der Walt et al., the individual plasma concentration–time profiles in patients without diabetes were also best described using a two-compartment model with first-order elimination, and multiple transit absorption compartments. Additionally, the estimated model parameters of the structural model are in a similar range (apparent clearance 11.7 L/h vs approximately 10.5 L/h in the model of van der Walt et al.), indicating that both models are quite comparable. Indeed, the mean estimated exposure to dapagliflozin that we found with our model in patients without diabetes is similar to the estimated exposure reported by van der Walt et al. in diabetic patients [
22]. It can therefore be concluded that the plasma concentration–time profile of dapagliflozin in patients without diabetes is comparable to those with diabetes.
The variability between patients in the individual plasma concentration–time profiles after administration of 10 mg of dapagliflozin was high. The between-patient variability in the plasma concentration–time profile was in part explained by differences in kidney function and body weight between patients. The model predicted that dapagliflozin plasma exposure nearly doubled in patients with mGFR of 30 mL/min/1.73 m
2 and nearly tripled in patients with mGFR of 15 mL/min/1.73 m
2 compared with patients with normal kidney function. These estimates are consistent with pharmacokinetic studies in patients with type 2 diabetes and kidney impairment [
11]. Kidney function and body weight were also identified in the population pharmacokinetic model of van der Walt et al. as significant predictors for individual plasma exposure [
22]. An advantage of our study was that we measured GFR by iohexol plasma clearance whereas other studies have used less precise serum creatinine-based equations to estimate GFR. Measured glomerular filtration rate is considered to be a more accurate marker of the actual kidney function than eGFR and thus provides a better estimate of the influence of kidney function on plasma exposure [
23]. Indeed, in our model, dapagliflozin plasma concentrations were better described with mGFR compared with eGFR.
In the present study, we demonstrate that the plasma exposure to dapagliflozin is associated with beneficial changes in risk markers for kidney disease in patients with non-diabetic kidney disease. A higher plasma exposure to dapagliflozin is associated with a more pronounced decrease in UACR, mGFR and systolic blood pressure. Additionally, a trend for a decrease in body weight with increased exposure was observed. These findings are in keeping with those previously reported in patients with type 2 diabetes and suggests dapagliflozin has an effect on these risk markers [
24,
25]. Dapagliflozin did not decrease proteinuria in the DIAMOND trial [
19]. Between-patient variability in proteinuria change was also not associated with dapagliflozin plasma exposure. In contrast, the statistically significant association between exposure and UACR response suggests that the reduction in UACR in the DIAMOND trial is real. The lack of an association between plasma exposure and proteinuria response, in contrast with the association between plasma exposure and albuminuria response, suggests that albuminuria might be a more suitable risk marker for detecting drug effects in patients with kidney disease.
Based on our models, patients with impaired kidney function have a higher plasma exposure to dapagliflozin and are therefore expected to have more pronounced effects on risk markers for kidney disease as compared with patients without impaired kidney function. Dapagliflozin is filtered by the kidney and binds to the SGLT2 transporter located in the apical membrane of the proximal tubule [
26]. The total amount of filtered dapagliflozin decreases in patients with impaired kidney function, which could result in a decrease in dapagliflozin concentration in the kidney. However, patients with impaired kidney function have fewer numbers of functioning nephrons, which balances the decrease in filtered dapagliflozin and may even lead to an increase in dapagliflozin in the proximal tubule. Furthermore, dapagliflozin is metabolised in both the liver and kidney to pharmacologically inactive dapagliflozin 3-O-glucuronide by UGT1A9. The ratio between dapagliflozin and its metabolite is reduced in patients with kidney impairment [
22,
27]. This suggests that dapagliflozin metabolism in the kidney is also decreased in patients with kidney disease, which seems plausible as UGT1A9 expression is eight-fold higher in the kidney as compared with the liver [
22]. The concentration of dapagliflozin could therefore locally be increased in patients with kidney impairment, which could explain the more pronounced effects of dapagliflozin on risk markers for kidney disease. Unfortunately, we were not able to measure dapagliflozin 3-O-glucuronide concentrations, thus this possibility needs to be further explored in future studies.
This study has a number of limitations. First, all participants were treated with dapagliflozin 10 mg. Therefore, our analysis is limited to explaining variability in kidney response between patients and cannot be used to make any dosing recommendations for patients with impaired kidney function. Second, model parameters were estimated with reasonable precision, but shrinkages in the parameters for the transit compartment were relatively high. This was also observed for patients with diabetic kidney disease [
22]. This is most likely caused by a relatively low amount of plasma samples that were collected in the absorption phase of patients that demonstrated a relatively fast absorption. Consequently, the model should not be used to predict the maximum concentration. Third, no data were available of patients with very low kidney function (mGFR ≤ 20 mL/min/1.73 m
2). The predicted effects in patients with severe kidney impairment might therefore be overestimated in our model. Additional data are needed to draw definitive conclusions for this subgroup of patients. Fourth, we did not measure plasma metabolites nor urine concentrations of both the parent and metabolite and were thus unable to estimate absolute bioavailability. Finally, it should be noted that our pharmacokinetic model has not been externally validated.
Declarations
Conflict of interest
Annemarie B. van der Aart-van der Beek, Abdul Halim Abdul Gafor, Claire C.J. Dekkers, Daniel C. Cattran, Jasper Stevens, Jeroen V. Koomen, Qiang Li, Sean J. Barbour and Sunita Singh have no conflicts of interest that are directly relevant to the content of this article. Hiddo J.L. Heerspink is a consultant to Abbvie, AstraZeneca, Boehringer Ingelheim, Bayer, Chinook, CSL Behring, Gilead, Janssen, Merck, Mundipharma, Mitsubishi Tanabe, Novo Nordisk and Retrophin. He received research support from AstraZeneca, Abbvie, Boehringer Ingelheim and Janssen. David W. Boulton is an employee and shareholder of AstraZeneca. David Z.I. Cherney has received honoraria from Boehringer Ingelheim-Lilly, Merck, AstraZeneca, Sanofi, Mitsubishi-Tanabe, Abbvie, Janssen, Bayer, Prometic, BMS and Novo-Nordisk and has received operational funding for clinical trials from Boehringer Ingelheim-Lilly, Merck, Janssen, Sanofi, AstraZeneca and Novo-Nordisk. Ron T. Gansevoort has consulting agreements with AstraZeneca, Bayer, Sanofi-Genzyme and Mundi Pharma; all fees are paid to his institution. Peter J. Greasley is employed by and owns shares in AstraZeneca. Gozewijn D. Laverman has received research grants and consulting fees from Sanofi and AstraZeneca, and research grants from Novo Nordisk. Marc G. Vervloet has received consulting fees from Amgen, Vifor, Fresenius Medical Care Renal Pharma, Medice, Kyowa Kirin and Astra Zeneca. Heather N. Reich has received consulting fees from Omeros and was involved in clinical trials supported by Omeros and Calliditas. Soo Kun Lim has received consulting fees or speaking honoraria from AstraZeneca, Boehringer Ingelheim, Novo Nordisk, Fresenius Kabi, Baxter and Sanofi.