1 Introduction
Worldwide prevalence rates of obesity are increasing, and this is not only restricted to adults. In the USA, 34.5% of adolescents (aged 12–19 years) were overweight (body mass index [BMI] for age ≥ 85th percentile) and 20.5% were obese (BMI for age ≥ 95th percentile) in 2011–2012 [
1]. In Western Europe, prevalence rates of obesity (based on International Obesity Task Force cut-offs) ranged from 3.8 to 13.5%, whereas percentages in North Africa and the Middle East ranged from 4.2 to 23.3% in boys and girls (aged < 20 years) in 2013 [
2]. Obesity is the most common cause of insulin resistance in children and adolescents [
3] and a known risk factor for type 2 diabetes mellitus (T2DM) [
4]. Metformin is a registered drug for the treatment of T2DM in pediatric patients aged > 10 years. Recently, a large increase in prescriptions of metformin or other oral antidiabetic drugs was observed in children [
5,
6], suggesting that metformin is prescribed not only in children with T2DM but also in obese children with type 1 diabetes mellitus in addition to insulin and for obesity with or without insulin resistance [
5]. To date, except for one abstract of a study in obese adolescents with T2DM [
8], the pharmacokinetics of metformin have only been studied in non-obese children who received metformin for early-normal onset of puberty [
7]. While tubular secretion is the primary route of elimination, conducted by organic cation transporters (OCT2) (basolateral membrane of the renal tubule cells; uptake of metformin in the renal tubule cells) and human multidrug and toxin extrusion transporters (i.e., MATE1 and MATE2K) (apical membrane of the renal tubules; transports metformin out of the tubule cells) [
9,
10], the influence of obesity on the expression or activity of these renal transporters is largely unknown. The OCT1 transporter is expressed on the apical membrane of the renal tubule and may play a role in the reabsorption of metformin [
9]. The aim of this study was to determine the pharmacokinetics of metformin in overweight and obese adolescents enrolled in a multicentre randomized double-blind controlled trial investigating the short- and long-term efficacy and safety outcomes of metformin [
11].
2 Methods
2.1 Patients
The clinical trial protocol of the multicentre randomized double-blind controlled study on short- and long-term efficacy and safety outcomes of metformin are described elsewhere [
11] and repeated briefly here as relevant to this pharmacokinetic sub-study. Obese adolescents were recruited at the pediatric outpatient clinic of the St. Antonius Hospital Nieuwegein and Jeroen Bosch Hospital in ‘s-Hertogenbosch, the Netherlands. Patients were eligible for inclusion if they were aged 10–16 years, obese (defined as BMI-standard deviation score [SDS] > 2.3 at pre-study screening), insulin resistant (defined as homeostatic model assessment—insulin resistance [HOMA-IR] ≥ 3.4 at pre-study screening), and of Caucasian descent. Patients were excluded if they had T2DM, polycystic ovary syndrome, or endocrine disorders for which treatment with corticosteroids was indicated. Also excluded were patients with a height ≤ 1.3 standard deviation (SD) of target height, (history of) alcohol abuse, impaired renal function (glomerular filtration rate [GFR] < 80 ml/min), or impaired hepatic function (alanine aminotransferase [ALT] > 150% of normal value for age) and patients who used anti-hyperglycemic drugs or who were pregnant. Before participation, parents and patients provided written informed consent and assent, respectively. The study was approved by the local human research and ethics committee of the St. Antonius Hospital (VCMO, NL34611.100.11) and was conducted in accordance with the principles of the Declaration of Helsinki and the Medical Research Involving Human Subjects Act (WMO) of the Netherlands.
2.2 Study Design
In this multicentre randomized double-blind trial (NCT01487993 and EudraCT 2010-023980-17), 23 overweight and obese adolescents of the metformin arm participated in the pharmacokinetic study, which was performed during an oral glucose tolerance test (OGTT). At the time of the pharmacokinetic sub-study, patients had received metformin 500 or 1000 mg (Centrapharm, Etten leur, the Netherlands) twice daily for 37 weeks [
11]. Patients were asked to fast overnight and postpone their morning metformin dose until arrival at the outpatient clinic. After insertion of a venous cannula, patients received a solution of glucose 1.75 g/kg body weight (maximum 75 g) dissolved in 200 or 300 ml of water. Directly after ingestion of the glucose solution, the metformin tablet was ingested (500 or 1000 mg). Blood samples were collected before the metformin dose (trough concentration) and at 60, 120, 240, 360, and 480 min, centrifuged at 4000 RPM (
g) for 5 min and stored at − 20 °C until analysis. In addition, blood samples for creatinine, ALT, and DNA to determine the genetic variation in the OCT1 and MATE1 transporters were collected.
2.3 Drug Assay
Serum metformin concentrations were quantified using high-performance liquid chromatography (HPLC). For sample preparation, 50 μl of internal standard (phenformin 500 mg/l in methanol) and 250 μl of acetonitrile were added to 100 μl of the sample. After vortex mixing, this mixed sample was centrifuged at 4000g for 10 min. Ten microliter of the supernatant was injected into the HPLC column (Zorbax SB-CN 5 μm 4.6 × 150 mm [Agilent]), protected by a precolumn (H3-10C5 [Hichrom]), which was kept at 30 °C and detected by a photodiode array detector (234 nm). Isocratic elution was performed with a mobile phase consisting of sodium sulphate (pH of 2.3) and acetonitrile in a 770:230 ratio at a flow rate of 1.2 ml/min. The elution times for metformin and the internal standard were 2.1 and 3.0 min, respectively. The assay was linear over 0.2–5.0, and 0.2 mg/l was the limit of quantification (LOQ). Intra- and inter-assay coefficients of variation were within 0.8–3.6 and 4–8.5%, respectively. The concentrations of the quality controls were 0.5, 2.5, and 4.5 mg/l (low, middle and high, concentration, respectively).
2.4 DNA Analysis
DNA was isolated from whole blood on the MagNA Pure LC 2.0 instrument (Roche®). Ready-made TaqMan assays were used for determination of OCT1 (rs72552763, rs12208357, rs34130495, rs34059508, and rs622342) and MATE1 (rs2289669, rs10735) polymorphisms on the ABI PRISM® 7500 Real-Time polymerase chain reaction system (Applied Biosystems®, Bleiswijk, the Netherlands). Violation of Hardy–Weinberg (HW) equilibrium was tested with the Chi squared test. In addition, observed minor allele frequencies were compared with the frequencies found in the single nucleotide polymorphism (SNP) database of the National Center for Biotechnology Information. The OCT1 haplotype, defined by the presence of one or more of the genetic variants rs72552763, rs12208357, rs34130495, and rs34059508 was estimated with the haplo.stats package (R, version 3.1.1), which uses the expectation–maximization logarithm and a posterior probability > 0.98.
2.5 Statistical Analysis
The area under the curve concentrations from 0 to 8 h after dose (AUC
0–8h) and AUC until infinity (AUC
inf) for metformin were calculated for each patient separately using the linear trapezoidal rule in R software (version 3.0.1) [
12]. Seven patients had metformin concentrations measured until 7 h post-dose and for these patients the AUC
0–8h was calculated based on extrapolation until 8 h. The calculated AUC
0–8h and AUC
inf were standardized for the administered dose (1000 mg), as patients received metformin 500 or 1000 mg. The Pearson correlation test was applied to test the correlation between total body weight (TBW) and dose-standardized AUC
0–8h and AUC
inf. The Mann–Whitney test or Kruskal–Wallis test was applied to test statistical differences in dose-standardized AUC
0–8h or dose-standardized AUC
inf and different genotype groups of the OCT1 and MATE1 transporter. Statistical analyses were performed using IBM SPSS software, version 22.
2.6 Population Pharmacokinetic Analysis and Internal Model Validation
Metformin data were analyzed using non-linear mixed effects modeling with NONMEM (version 7.2; ICON Development Solutions, Hanover, MD, USA) [
13]. Pirana (2.9.1) [
14], R (3.0.1) [
12], Xpose (4.5.0) [
14], and Psn (3.6.2) [
14] were used to evaluate and visualize the data. Of the 129 metformin samples, seven trough samples (5.4%) and one post-dose sample (at 8 h; 0.8%) were below the LOQ. One trough sample below the LOQ of 0.2 mg/l (i.e., 0.12 mg/l) was kept in the dataset, whereas no metformin could be detected for the other samples below LOQ, so these samples were subsequently removed from the analysis [
15,
16]. The first-order conditional estimation method with interaction was used for model development. Discrimination between different models was guided by the likelihood ratio test, by comparison of the objective function value (OFV, i.e., − 2 log likelihood [− 2LL]) between nested models. A
p value of < 0.05, representing a decrease of 3.84 in OFV for one degree of freedom, was considered statistically significant. In addition, goodness-of-fit plots for metformin (observed vs. individual predicted concentrations, observed vs. population predicted concentrations, conditional weighted residuals vs. time after dose, and conditional weighted residuals vs. population predicted concentrations plots) were used for diagnostic purposes. Furthermore, precision of parameter estimates, the correlation matrix, and visual improvement in the individual plots were used to evaluate the model. Pharmacokinetic models incorporating either one or two compartments with first-order, zero-order, or combined first- and zero order oral absorption were tested. Furthermore, the addition of one or more transit compartments [
17] or an oral absorption lag time was evaluated. Interindividual variability (IIV) was assumed to follow a log-normal distribution. Residual variability was tested using proportional, additive, or combined proportional and additive error models for metformin. For internal model evaluation, we used a bootstrap resampling method using 1000 replicates and prediction-corrected visual predictive check (pcVPC) [
18] using 1000 simulated datasets of individuals from the original dataset.
2.7 Covariate Model
Tested covariates were TBW, BMI, BMI-SDS, lean body weight (LBW) according to the equations of Janmahasatian et al. [
19], Foster et al. [
20] and Peters et al. [
21] and waist–hip ratio, age, sex, creatinine, and estimated GFR (eGFR) according to the bedside Schwartz formula [
22]. In addition, the influence of genetic polymorphisms of the OCT1 and MATE1 transporter were tested. Covariates were plotted independently against the eta (
η) estimates of the pharmacokinetic parameters to visualize potential relations. Continuous covariates were tested using linear and power equations (Eqs.
1,
2):
$$P_{i} = P_{p} \times (1 + V \times \;({\text{COV}} - {\text{COV}}_{\text{median}} ))$$
(1)
$$P_{i} = P_{p} \times ({\text{COV}}/{\text{COV}}_{\text{median}} )^{W} ,$$
(2)
where
Pi and
Pp represent individual and population parameter estimates, respectively, COV represents the covariate, COV
median represents the median value of the covariate for the population,
V represents a correlation factor between the population pharmacokinetic parameter and the change in covariate value for a linear function, and
W represents the exponent for a power function. The categorical covariates sex and genetic polymorphisms of the
OCT1 and
MATE1 transporter were examined by calculating a separate parameter for each category of the covariate. Based on
OCT1 (rs622342) and
MATE1 (rs2289669 and rs10735) genotype, subjects were categorized into two groups: variants (homozygous and heterozygous variants) or wild types. For
OCT1 haplotype, subjects were categorized as normal transporter activity or decreased and absent transporter activity. In addition, different combinations of the genotypes were tested, for example between the
OCT1 (rs622342) and
MATE1 (rs2289669) transporter [
23].
Potential covariates were entered into the model one at a time and statistically tested using the likelihood ratio test. In addition, if applicable, reduction in IIV (omega [
ω]) of the parameter was evaluated upon inclusion of the covariate on the parameter. Further, trends in the random effects of the parameter versus the covariate involved were observed. Finally, after forward inclusion (
p < 0.05), a backward exclusion procedure was applied to justify the inclusion of a covariate (
p < 0.01). The choice of the final covariate model was further evaluated as discussed in Sect.
2.6.
2.8 Excess Weight Covariate Model
To further analyze the influence of (over)weight on the pharmacokinetics of metformin, an excess weight model was tested for the parameters for which TBW proved a covariate given the criteria described in Sect.
2.6. Using this covariate model, the TBW of each individual patient was considered to be composed of two parts: developmental weight (WT
for age and length) and excess body weight (WT
excess) [
24], whereas separate functions for each of these weights were used.
For each individual patient in the study, WT
for age and length was derived from the Dutch TNO growth calculator [
25] on the basis of the length and age of the patient at a BMI-SDS score of 0 (no overweight) [
25]. WT
excess and relative WT
excess (%WT
excess) were calculated using Eqs. (
3) and (
4), respectively, for each individual patient:
$${\text{WT}}_{\text{excess}} \, = \,{\text{TBW}} - {\text{WT}}_{\text{for\,age\,and\,length}}$$
(3)
$$\% {\text{WT}}_{\text{excess}} \, = \,({\text{WT}}_{\text{excess}} /{\text{WT}}_{\text{for\, age\, and \,length}} )\, \times \, 100\% .$$
(4)
First, WTfor age and length, WTexcess, and %WTexcess were all plotted independently against the eta estimate of the pharmacokinetic parameter of interest to visualize the relation.
The separate impact of WT
for age and length and WT
excess on clearance in adolescents was evaluated using Eqs. (
5) and (
6):
$${\text{CL}}_{{{\text{non-obese}}\,{\text{adolescent}}}} \, = \,{\text{CL}}_{{ 70\,{\text{kg}} \,{\text{adult}}}} \, \times \,({\text{WT}}_{\text{for\, age\,and\, length}} / 70)^{0. 7 5} $$
(5)
$${\text{CL}}_{{\left( {\text{obese}} \right)\,{\text{ adolescent}}}} \, = \,{\text{CL}}_{{{\text{non-obese}} \,{\text{adolescent}}}} \, + \,(Z\, \times \,{\text{WT}}_{\text{excess}} )$$
(6)
in which CL
non-obese adolescent represents the clearance estimate of adolescents without overweight, CL
70kg adult represents the population clearance of an adult weighing 70 kg, WT
for age and length represents developmental weight, 0.75 is the scaling factor previously proposed by the US FDA for scaling clearance from adults to adolescents [
26], CL
(obese) adolescent is the individual clearance estimates of (obese) adolescents, and
Z is the linear influence of WT
excess in this function.
4 Discussion
This study aimed to determine the pharmacokinetics of metformin in overweight and obese adolescents in view of the increasing use of metformin for obesity in children, in whom adult dosages are often used. This study shows that the AUC decreases and the CL/
F of metformin increases significantly with TBW in obese adolescents (Fig.
1). The increase in CL/
F can be explained by both WT
for age and length and WT
excess (Fig.
3a, b), for which an excess weight covariate model was proposed in which WT
for age and length scaled allometrically to the power of 0.75 and a separate function for WT
excess was estimated. This excess covariate weight model was applied because the interrelation between growth, age, and obesity in obese children and adolescents complicates a systematic covariate analysis [
24,
27].
When the current pharmacokinetic parameters of metformin in overweight and obese adolescents are compared with literature values of non-obese children and (obese) adults (Table
3), the dose-standardized AUC value we report here in overweight and obese adolescents proved about two times lower than in non-obese children [
7]. However, this AUC is around the same as in obese T2DM adolescents [
8] and healthy adult subjects [
8], and similar to or higher than in obese adults [
28] (Table
3). These results indicate that obese adolescents have a higher clearance than their non-obese counterparts [
7] (Table
3), whereas, within the obese adolescent population, CL/
F increased even further in more severely obese individuals (Fig.
1). The CL/
F value in obese adolescents seems comparable to the value in non-obese adults [
29] and similar to or lower than that in obese adults [
28] (Table
3).
Table 3
Pharmacokinetic parameters of oral metformin in overweight and obese adolescents from this study compared with literature values from different patient populations
Obese adolescents from this study | Fasted | 79.3 ± 13.9 | 14.5 ± 1.8 | AUC0–inf 460.5 ± 223 AUC0–8 316.5 ± 101.6 | 1.17b | 485b | 0.0248b |
| Dinner | 33.3 ± 2.7 | 9.5 ± 0.1 | AUC0–12 816.2 ± 57.8 | 0.55 ± 0.55c | 185 ± 11.7d | 0.0175 ± 0.0045 |
Obese adolescents with T2DM [ 8] | Breakfast | 98 ± 25 | (12–16) | AUC0–inf 378.7 ± 1.6 | 1.32e | – | – |
| Breakfast | 91 ± 21 | (20–45) | AUC0–inf 398 ± 1.7 | 1.26e | – | – |
Healthy adults and adults with T2DM [ 29] | – | 71 ± 14 89 ± 16 | 27 ± 6 56 ± 5 | – | 1.27 ± 0.27 1.32 ± 0.11 | 559 ± 163 648 ± 13.8 | – |
| Fasted | 114.6 ± 26.1 | 43.5 ± 11.7 | AUC0–inf 342 ± 108 | 1.46e | 114.6 ± 45.8f | – |
Explanations for the higher CL/
F in overweight and obese adolescents versus non-obese children may differ. First, a higher tubular secretion of metformin by induced OCT2 and MATE1/MATE2K transporters in the kidney [
9,
10] can be postulated in obese patients. This hypothesis is supported by our excess weight model, which shows an increase in CL/
F with WT
excess (Fig.
3b) within the population of obese adolescents. Moreover, studies with intravenous procainamide, ciprofloxacin, and cisplatin, which are all primarily eliminated via tubular secretion, also show a higher clearance in obese adults than in non-obese adults [
30]. These drugs are also renally transported by the OCT system: procainamide by OCT1–3 [
31], ciprofloxacin by OCT and OAT (organic anion transporter) [
31], and cisplatin by OCT2 [
32]. In contrast, obese rodent models provide no basis for this hypothesis. High-fat-diet-induced mice do show a trend in increased OCT2 renal messenger RNA (mRNA) expression but no difference in OCT1 expression [
33]. Ob/Ob and db/db mice also show no difference in renal mRNA expression of the OCT1 transporter and even a decrease of the renal OCT2 and MATE1 transporter [
34‐
36]. However, these ob/ob and db/db mice models may not be representative for obese patients, since these mice do not produce leptin or have no functional leptin receptor, and leptin may act on renal cells to regulate gene expression [
33]. It is difficult to determine the net result of the various effects of obesity on the regulation of transporter expression in these rodent models [
37], as pro-inflammatory cytokines, insulin, leptin, and cholesterol all have opposing effects on mRNA expression of the transporters [
37]. In addition, renal transporters (OCT1-2 and MATE1/MATE2K) should be considered as a whole because they work together in the excretion of metformin, and one transporter can compensate the other [
34]. Hence, our suggested increase in active tubular secretion by increased activity of the renal transporters in obese adolescents warrants further study. Another explanation for the higher CL/
F in obese adolescents could be a decreased bioavailability (
F) in obese adolescents as a result of a decreased expression or activity of the intestinal transporters of metformin (i.e., OCT1, PMAT, OCT3). However, no literature is available about changes in intestinal transporter expression or activity with obesity, so this hypothesis remains rather speculative. Lastly, the higher CL/
F in obese adolescents versus non-obese children may also to some extent be explained by difference in age (11–17.5 vs. 9 years, respectively). However, as this difference in age is small, this study suggests, in our opinion, that the increase in CL/
F of metformin in obese adolescents may be explained by an increase in clearance or a decrease in oral bioavailability resulting from an increase or decrease in the activity in renal or intestinal transporters, respectively. Since the value for CL/
F we report in obese adolescents is higher than in non-obese children [
7] and comparable to the value in healthy adults and adults with T2DM [
29], the maximum recommended dose of metformin 2 g for children may be considered to be increased to the maximum dose in adults (i.e., 3 g per day) where clinical response is lacking.
In this study, we could not identify an influence of genetic polymorphisms in the
OCT1 and
MATE1 transporter on the pharmacokinetics of metformin. It is likely that a larger number of patients is needed to accurately test this influence. Stocker et al. [
38] was unable to identify an influence of the
MATE1 promoter variant on the pharmacokinetics of metformin, whereas two other studies did find an influence of genetic variation in the
OCT1 transporter on the pharmacokinetic parameters of metformin [
39,
40]. Stocker et al. [
38] concluded that MATE1 and MATE2 transporters work together in the renal elimination of metformin and that genetic variants in
MATE and
OCT transporters should be considered together when determining the genetic determinants of renal elimination of metformin. For future studies, it would be interesting to evaluate the genetic variants of
OCT1-3 and
MATE1-2 together. In addition, it is of special interest to evaluate the genetic variants of the intestinal transporters of metformin (i.e.,
OCT1, OCT3 and
PMAT) on the bioavailability [
39] and oral absorption of metformin, since the oral absorption of metformin showed large interindividual variability in our data.
There are some limitations to this study. First, no non-obese adolescents were included because this pharmacokinetic study was part of a randomized double-blind controlled trial on metformin in obese adolescents with insulin resistance [
11], precluding a head-to-head comparison with the obese adolescents. Second, our patients received metformin in a fasted state because, for ethical reasons, the pharmacokinetics study was performed during an OGTT for which a venous line was inserted for clinical purposes. In clinical practice, patients take metformin with food and as such our Ka may differ from that in other studies. Oral absorption varied widely between the obese adolescents. Nine adolescents had a very fast absorption, with a
Tmax of 60 min, whereas the mean
Tmax in non-obese children was reported to be 2.4 h [
7]. This difference may be explained by the fasted state of the obese adolescents or by the state of obesity.