Significant impact of time-of-day variation on metformin pharmacokinetics

Aims/hypothesis The objective was to investigate if metformin pharmacokinetics is modulated by time-of-day in humans using empirical and mechanistic pharmacokinetic modelling techniques on a large clinical dataset. This study also aimed to generate and test hypotheses on the underlying mechanisms, including evidence for chronotype-dependent interindividual differences in metformin plasma and efficacy-related tissue concentrations. Methods A large clinical dataset consisting of individual metformin plasma and urine measurements was analysed using a newly developed empirical pharmacokinetic model. Causes of daily variation of metformin pharmacokinetics and interindividual variability were further investigated by a literature-informed mechanistic modelling analysis. Results A significant effect of time-of-day on metformin pharmacokinetics was found. Daily rhythms of gastrointestinal, hepatic and renal processes are described in the literature, possibly affecting drug pharmacokinetics. Observed metformin plasma levels were best described by a combination of a rhythm in GFR, renal plasma flow (RPF) and organic cation transporter (OCT) 2 activity. Furthermore, the large interindividual differences in measured metformin concentrations were best explained by individual chronotypes affecting metformin clearance, with impact on plasma and tissue concentrations that may have implications for metformin efficacy. Conclusions/interpretation Metformin’s pharmacology significantly depends on time-of-day in humans, determined with the help of empirical and mechanistic pharmacokinetic modelling, and rhythmic GFR, RPF and OCT2 were found to govern intraday variation. Interindividual variation was found to be partly dependent on individual chronotype, suggesting diurnal preference as an interesting, but so-far underappreciated, topic with regard to future personalised chronomodulated therapy in people with type 2 diabetes. Graphical abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-05898-4.


Statistical analysis
Individual plasma measurements were analysed separately for differences between trough plasma concentration (Ctrough) values measured immediately before the next dose in the morning ('Ctrough,morning') and the evening ('Ctrough,evening'). Ctrough,morning and Ctrough,evening were calculated average for each individual, and differences were compared using a paired t test. Additionally, a linear mixed model with random effect on study participant was applied to account for intra-and interindividual variability, including unequal numbers of measurements in the morning and the evening for each individual. The same procedure was applied to compare maximum plasma concentration (Cmax) values measured after the morning dose ('Cmax,morning') and the evening dose ('Cmax,evening'). For all statistical analyses, the significance level α was set to 0.05 (5%). where predicted PK parameteri = predicted Ctrough or Cmax ratio, observed PK parameteri = corresponding observed Ctrough or Cmax ratio and m = number of studies. Overall GMFEs of ≤ 2 were considered reasonable predictions.

1-16
Virtual twins of study individuals were generated according to the demographic information, with corresponding ethnicity, sex, age, body weight, height and GFR, if reported. Metformin transporters were implemented in agreement with current literature, utilising the PK-Sim® expression database [9] to define their relative expression in the different organs of the body. Details on the expression of drug transporters implemented to model the pharmacokinetics of metformin are summarised in ESM Table  6. In all virtual individuals, enterohepatic circulation (EHC) was enabled (EHC continuous fraction set to 1) by assuming a continuous flow of the bile to the duodenum.
Model performance was evaluated by (1)   RT-PCR, reverse transcription-polymerase chain reaction measured expression profile a mean reference concentration µmol/l in the tissue of highest expression b relative expression in the different organs (PK-Sim expression database profile) c calculated from transporter per mg membrane protein x 26.2 mg human kidney microsomal protein per g kidney [10] d calculated from transporter per mg membrane protein x 37.0 mg membrane protein per g liver [14] e large intestinal mucosa → 0 f apical in enterocytes g transport rate constant (k cat ) was optimised according to [18] 2-17

Statistical analysis Fig 6.
Intra-and interindividual variability of trough plasma concentration (Ctrough) measurements from studies I and II [1,2], including paired t test results. Boxes represent the distance between first and third quartiles (IQR). Whiskers range from smallest to highest value (< 1.5 × IQR).

Fig 7.
Intra-and interindividual variability of maximum plasma concentration (Cmax) measurements from study I [1], including paired t test results. Boxes represent the distance between first and third quartiles (IQR). Whiskers range from smallest to highest value (< 1.5 × IQR).

NLME pharmacokinetic modelling
The volumes of distribution were 41.1 l for central (V2) and 129 l for peripheral (V3) compartments, respectively. The physiologically described rhythm of the GFR was implemented by multiplication of a sine function with clearance. The amplitude and a time shift of the sine function were estimated as 21% and 12.3 h (acrophase at 17:43 hours), respectively. Interindividual variability was found on the clearance, the volume of distribution and the bioavailability. Implementation of food, formulation and dose as significant covariates partly explained the interindividual variability and the related parameters were reduced about 14.2%, 75.0% and 51.5% for clearance, volume of distribution and bioavailability, respectively. Administration after food intake leads to a 1.91-fold higher bioavailability and a 0.63-fold lower absorption rate constant but 5.10-fold increased release duration for the extended-release formulation. The bioavailability of the extended-release formulation was 1.09-fold higher compared to the immediate-release formulation. The dose was implemented as a covariate using an exponential function according to ESM Equation 3, leading to a decreased bioavailability by administration of higher doses metformin. All estimated parameter values are summarised in ESM Table 7. All model parameters were precisely estimated with residual standard errors < 25%. Observed versus model predicted metformin concentrations are randomly distributed around the line of identity, indicating good descriptive properties of the final pharmacokinetic model.

Final PBPK model parameters (with daily oscillation)
Individual plasma concentration-time profiles show high interindividual variability. Therefore, organic cation transporter (OCT) 2 transport rate constant (kcat) values were optimised for every individual. Additionally, OCT2 amplitude and time shift were optimised (after inclusion of rhythmic GFR and RPF) with the help of individual profiles (n=26) (ESM Table 10). A correlation plot of OCT2 kcat values, amplitude and time shift (sine function describing daily oscillation) is shown in ESM Fig 20, where no correlation has been detected.
ESM Table 9. Drug-dependent parameters of the metformin PBPK model adopted from Hanke et al. [8] Parameter