Prediction of maternal pharmacokinetics using physiologically based pharmacokinetic models: assessing the impact of the longitudinal changes in the activity of CYP1A2, CYP2D6 and CYP3A4 enzymes during pregnancy.

Concerns over gestational effects on the disposition of drugs has highlighted the need for a better understanding of drug distribution and elimination during pregnancy. This study aimed at predicting maternal drug kinetics using a physiologically based pharmacokinetic (PBPK) modelling approach focusing on the observed gestational changes in three important Cytochrome P450 metabolizing enzymes, namely, CYP1A2, CYP2D6 and CYP3A4 at different gestational weeks (GWs). The Pregnancy PBPK model within the Simcyp Simulator V19 was used to predict the pharmacokinetics of sensitive probes to these enzymes; namely caffeine, theophylline, metoprolol, propranolol, paroxetine, midazolam, nifedipine and rilpivirine. PBPK model predictions were compared against clinical data collated from multiple studies for each compound to cover a wide spectrum of gestational ages. Pregnancy PBPK model predictions were within 2-fold error and indicated that CYP1A2 activity is approximately 0.70, 0.44 and 0.30 fold of the non-pregnant level at the end of the first, second and third trimesters, respectively. On the other hand, CYP2D6 activity increases by 1.36, 2.16 and 3.10 fold of the non-pregnant level at the end of the first, second and third trimesters, respectively. Likewise, CYP3A4 activity increases by 1.25, 1.75 and 2.32 fold of the non-pregnant level at the end of the first, second and third trimesters, respectively. The enzymes activity have been qualified throughout pregnancy. Quantified changes in drug dosing are most relevant during the third trimester, especially for drugs that are mainly eliminated by CYP1A2, CYP2D6 and CYP3A4 enzymes. The provided functions describing the continuous changes to the activity of these enzymes during pregnancy are important when modelling long term pharmacokinetic studies where longitudinal modelling or time-varying covariates are used.