The peak incidence of many psychiatric illnesses such as schizophrenia in women occurs during their reproductive years (Kulkarni et al., 2015). The prescription of second-generation antipsychotics (SGAs) to pregnant women has been steadily increasing in the last 20 years. The latest statistics from ten countries show that up to 2% of pregnant women use SGAs (Reutfors et al., 2020). Though concerns about the safety of antipsychotics during pregnancy persist, some large-scale clinical studies in recent years suggested that exposure to SGAs does not confer an increased risk of congenital malformations (Huybrechts et al., 2016; Damkier and Videbech, 2018). Given the severe consequences without pharmacotherapy, off-label use of antipsychotics during pregnancy may be inevitable. Except for concerns about fetal safety, clinicians often face another major challenge, i.e., optimizing dosage regimens to obtain effective maternal treatment.

The potential benefits of therapeutic drug monitoring (TDM) to optimize pharmacotherapy are particularly obvious in psychiatry and neurology. The TDM task force of the German Society of Neuropsychopharmacology and Pharmacopsychiatry (Arbeitsgemeinschaft für Neuropsychopharmakologie und Pharmakopsychiatrie [AGNP]) had given many antipsychotics a high recommendation strength for conducting TDM (Hiemke et al., 2018). For olanzapine, a reference concentration range of 20–80 ng/ml was recommended. On the other hand, pregnancy introduces conspicuous changes in various anatomical, physiological, and biological properties, for instance, organ blood flow and hormone levels. Those alterations will influence drug disposition and further their system exposure (Kazma et al., 2020). According to a comprehensive review, gestation-associated changes in pharmacokinetics widely exist (Pariente et al., 2016). Blood concentrations of commonly prescribed antipsychotics perphenazine, quetiapine, and aripiprazole decrease sharply in late pregnancy, suggesting effective treatment may not be achieved in this period with the pregestational dosing regimen (Westin et al., 2018). However, a paucity in complete pharmacokinetic reports makes it challenging to implement dose adjustment for pregnant women.

Physiologically based pharmacokinetic (PBPK) modeling serves as a critical pharmacometrics tool to make reliable pharmacokinetic predictions in special populations. The number of new drug application submissions to the US Food and Drug Administration (FDA) that included PBPK modeling for pediatric drug development has continued to grow over the past decade (Corriol-Rohou and Cheung, 2019) and the role of PBPK modeling for pregnant women in a regulatory context has been discussed recently (Coppola et al., 2021; Green et al., 2021). Since PBPK is a mechanism-based modeling method, the combined effects of multiple gestation-related physiological changes on drug disposition can be incorporated. The confidence in current pregnant modeling tools is restricted by a lack of robust data around the understanding of some metabolic enzymes and transporters and how gestation and genotypes affect drug exposure jointly (Abduljalil and Badhan, 2020). Despite these shortages, PBPK modeling seems promising to address an imperative query: whether dose adjustment is required during pregnancy. In a previous study, we proposed optimized dosage regimens of quetiapine and aripiprazole for the pregnant population using PBPK modeling and simulation (Zheng et al., 2021).

Olanzapine undergoes extensive metabolism in the liver. Several enzymes, namely cytochrome P450 (CYP) 1A2, 2C8, 3A4, flavin monooxygenase 3 (FMO3), and glucuronyl-transferase (UGT) 1A4, are responsible for the metabolism. Six metabolic pathways of olanzapine have been identified, and some metabolic pathways such as N-demethylation are mediated by different enzymes (Supplementary Figure S1) (Kassahun et al., 1997; Korprasertthaworn et al., 2015). UGT1A4 catalyzes 10-N-glucuronidation and 4′-N-glucuronidation, whose metabolites account for 23% of an administered oral dose in non-pregnant adults (Kassahun et al., 1997). The precise proportions of other metabolites generated by oxidases have not been determined. The in vivo pharmacological effects are believed to be derived mainly from the parent drug (Hiemke et al., 2018). Though the principal metabolic enzyme CYP1A2 reveals a sharp reduction in its metabolic ability during pregnancy, plasma concentrations of olanzapine appear to be not markedly changed according to therapeutic drug monitoring data reported from Norwegian hospitals (Tracy et al., 2005; Westin et al., 2018). Thus, this study aims to develop a whole-body PBPK model for olanzapine to evaluate the change in systemic exposure of olanzapine throughout pregnancy. The results of this study will be beneficial for rational antipsychotic medication in this vulnerable population.

This study developed a PBPK model of olanzapine using a ‘middle-out’ strategy and gave pharmacokinetic predictions for the pregnant population. To our knowledge, this is the first pregnant PBPK modeling study for olanzapine.

Previous olanzapine PBPK modeling studies for non-pregnant adults described olanzapine clearance based on in vitro data (Polasek et al., 2018; Sun et al., 2020), while this study applied a different strategy. Because direct in vitro to in vivo extrapolation of the contribution of an enzyme to the total drug metabolism is often associated with great uncertainty (Zhang et al., 2020), we calculated this important parameter for major enzymes using in vivo data (mass balance and clinical DDI study). CYP1A2 has long been regarded as the primary enzyme for olanzapine metabolism, and its contribution was first determined to be 50%. Unlike typical highly polymorphic enzymes such as CYP2C9 and CYP2D6, genetic polymorphisms of CYP1A2 contribute little to the interindividual pharmacokinetic variability of olanzapine (Na Takuathung et al., 2019). Therefore, CYP1A2 polymorphisms were not considered in this study. The major circulating metabolite of olanzapine is the 10-N-glucuronide, whose formation can be attributed to UGT1A4 and UGT2B10 with the former having a much higher catalytic activity than UGT2B10 (Soderberg and Dahl, 2013). Besides, sensitivity analysis shows that fraction unbound, pKa, logP, and CYP1A2 and UGT1A4-mediated clearance are the most sensitive parameters, while logP is much more sensitive following multiple administrations. This impact leads to a visibly different time to reach the plateau under multiple doses, probably because of late back-distribution from compartments where olanzapine accumulates, that significantly affect AUC of the last dose (data not shown). As a result, a higher sensitivity to logP was observed after multiple administrations. Because the Charge-dependent Schmitt method for calculating cellular permeabilities considers the effect of electric charge, pKa becomes a relatively important parameter.

We speculated that the reduction in CYP1A2 activity during pregnancy is counteracted by the induction of other enzymes, especially UGT1A4. To this date, different studies have reported controversial results on hepatic blood flow during pregnancy. Therefore, the gestation-related physiology engine for creating virtual pregnant populations assumed unchanged absolute liver blood flow (Dallmann et al., 2017a). Alterations in CL/F should be mainly attributed to changes in (unbound fraction) x (intrinsic clearance) especially when olanzapine has a relatively low extraction ratio (<0.3). Fraction unbound, CYP1A2 and UGT1A4-mediated clearances are the most significant ones among all tested parameters modified during pregnancy (Figure 4). CYP1A2 activity decreases by up to 60% to late pregnancy. Meanwhile, UGT1A4 activity increases by 85% on average, whereas there is a slight alteration in fraction unbound, as calculated according to Eq. 5. As a result, the mean plasma concentrations of olanzapine are generally stable. On the other hand, the reliability of model predictions is potentially affected by several factors that cannot be accurately clarified at the current stage. First, the sensitivity analysis indicated that attention should be paid to the calculated fraction unbound. Although the calculation method (Eq. 5) has been evaluated for other drugs, the results stress the importance of a correct value for fraction unbound during pregnancy. For an extensively metabolized drug like olanzapine, increase in fraction unbound contributes to a higher hepatic clearance with a great possibility. Therefore, it should be beneficial to measure a precise fraction unbound of olanzapine in future clinical studies to confirm or refine the value used in the PBPK model. Second, it is currently undetermined whether gestation changes drug absorption from the gastrointestinal tract. In this model, settings for drug absorption were not specifically modified compared to non-pregnant women. Drug absorption is indeed a challenge requiring further investigation. There might be several gestation-related factors affecting drug absorption, for instance, prolongation in gastrointestinal transit time, and enlargement in intestinal villi surface area (Dallmann et al., 2018b; Koren and Pariente, 2018). However, we haven’t developed mathematical explanations for these factors due to inadequate quantitative human data. Since clinical data for C_max/t_max are lacking, the simulated absorption cannot be evaluated. This stresses again the need for further clinical data during pregnancy, ideally full pharmacokinetic profiles instead of trough concentrations. Third, uncertainty in clearance contribution of some minor enzymes during model development, though we estimated it had a minimal impact. Besides, there are conflicting reports on the exact magnitude of CYP3A4 induction in pregnant women. Some studies suggest that a 2-fold induction in the third trimester is plausible, whereas others suggest a lower activity increase (such as 27%) (Nylen et al., 2011). A previous modeling study used a weighted mean of 60% induction (Dallmann et al., 2018a). Assuming a 60% activity increase in the third trimester, the model predicts a C_max of 26.3 ng/ml and AUC_τ-ss of 488.1 ngh/mL, which show negligible differences from the current data. Therefore, among all relevant enzymes, CYP1A2 and UGT1A4 activity during pregnancy are critical determinants of olanzapine clearance. More quantitative data reflecting activity changes of metabolic enzymes are needed to enhance the predictive performance of pregnant PBPK modeling.

According to the TDM data and PBPK predictions, dose adjustments appear to be not urgently needed for pregnant women. Neither has a report that pregnant women show a higher treatment failure rate under the same doses. But we should notice that TDM data have indicated a considerable interindividual variability in trough concentrations (Westin et al., 2018). Therefore, TDM has its unique strength in the individualized dosing that could not be replaced. Effective treatment before gestation is essential to olanzapine usage during pregnancy with an unchanged dosage regimen.

A limitation of this study is that fetal exposure to olanzapine has not been addressed. In order to estimate the placental transfer of drugs, data from in vitro cell models and ex vivo placental perfusion are preferred. Additionally, umbilical cord blood concentration data during delivery would be needed to validate the model predictions (Dallmann et al., 2019). Unfortunately, these pieces of information on olanzapine are lacking. Recent studies have made beneficial attempts to explore the maternal-fetal drug transfer and exposure ratio with the abovementioned approaches using acetaminophen as a model drug (Mian et al., 2020; Mian et al., 2021). These studies provide helpful references in the analysis of fetal pharmacokinetics of olanzapine in the future. We urgently need more long-term studies with large samples to clarify the efficacy and adverse impacts on fetuses and determine management strategies for antipsychotics.

In summary, this study developed a PBPK model of olanzapine to evaluate the maternal exposure of this commonly prescribed antipsychotic in the pregnant population. The predictive performance was validated with various clinical pharmacokinetic studies. According to the presented PBPK simulations, the steady-state pharmacokinetics of olanzapine is slightly, and probably not clinically significantly altered during pregnancy. Combined with the TDM data, the model suggests that dose adjustment cannot be formulated for pregnant women, at least at the tested stages of pregnancy, if effective treatment was achieved before the onset of pregnancy, while fetal safety certainly needs continuous surveillance.