Prediction of maternal and fetal pharmacokinetics of dolutegravir and raltegravir using physiologically-based pharmacokinetic modeling

Abstract

Background

Predicting drug pharmacokinetics (PK) in pregnant women including placental drug transfer remains challenging. This study aimed to develop and evaluate maternal-fetal physiologically-based PK (PBPK) models for two antiretroviral drugs (ARVs), dolutegravir (DTG) and raltegravir (RAL).

Methods

PBPK models were built with the Open Systems Pharmacology software suite (PK-Sim/MoBi). Different approaches to inform placental drug transfer were applied and compared. Model performance was evaluated using in vivo DTG and RAL maternal plasma concentrations during the 2^nd and 3^rd trimesters and umbilical vein concentrations at delivery. All clinical in vivo data were obtained from the International Maternal Pediatric and Adolescent AIDS Clinical Trials (IMPAACT) Network P1026s study.

Results

The PBPK models successfully predicted plasma concentration-time profiles of DTG and RAL in the 2^nd and 3^rd trimesters and most predicted PK parameters fell within a 1.33-fold error range. Predicted umbilical vein concentrations of DTG were in reasonable agreement with in vivo data but were sensitive to changes in the placental partition coefficient and transplacental clearance.

Conclusion

Maternal-fetal PBPK modeling reliably predicted maternal PK of DTG and RAL during pregnancy. For the fetal PK, data on the unbound fraction of highly protein-bound DTG has proven to be important to adequately capture changes in total clearance in silico. More research efforts, along with clinical data, are needed to verify the predictions of fetal PK of ARVs. Overall, the findings suggest that it may be possible to use PBPK models to assess the disposition of ARVs in pregnant women and their fetuses.

1. Introduction

The recommendation to provide all pregnant women living with HIV with antiretroviral therapy (ART) has had a major impact on the prevention of vertical transmission of HIV in the United States (U.S.) and globally. The incidence rate for perinatally acquired HIV infection in the U.S. has decreased from over 5% in 2002 to about 2% in 2013 and to 53 cases in 2015 [1, 2]. The goal of the U.S. Center for Disease Control and Prevention is the complete elimination of perinatal HIV transmission [3]. Perinatal HIV infection still persists globally due to challenges with timely identifying HIV infection and maintaining suppressive ART throughout pregnancy, postpartum and throughout breastfeeding. Providing efficacious ART in to pregnant women is also challenging as physiologic changes during pregnancy can considerably affect the pharmacokinetics (PK) of antiretroviral drugs (ARVs). Indeed, a number of ARVs require a dose adjustment during pregnancy or are not recommended for use in pregnant women because of decreased plasma concentrations or fetal risks [4].

The integrase strand transfer inhibitors (INSTIs) are recommended as first line ART agents in ARV-naïve pregnant women [5]. Their initiation in late pregnancy has also been reported to rapidly achieve viral suppression by the time of delivery [6]. A recent report of the association between dolutegravir (DTG) use at the time of conception with a higher risk of neural tube birth defects in newborns has led to the change in the global and U.S. guideline regarding DTG use in women of child-bearing potential, however still recommending the use of DTG during the 2^nd and 3^rd trimester of pregnancy [7, 8]. Physiological alterations in pregnancy may lead to reduced maternal plasma concentrations of DTG and RAL which increases the risks of viremia and vertical transmission of HIV.

Understanding the PKs of ARVs in pregnant women facilitates optimal dosing of ART and prevention of viremia and vertical transmission of HIV.

Physiologically-based pharmacokinetic (PBPK) modeling is a promising approach to investigate PK of xenobiotics in special populations where clinical trials are difficult to conduct, such as pregnant women and their fetuses. PBPK models are mechanistic models mapping the model structure to the circulatory system in a biologically plausible manner and integrating extensive information on the anatomy and physiology of the organism as well as physicochemical properties of the xenobiotic. While some pregnancy PBPK models have been previously reported [9], they are still not considered qualified for use in pregnancy because of limited experience. This study aimed to build a maternal-fetal PBPK model for DTG and RAL in order to predict their PK profiles in pregnant women during the 2^nd and 3^rd trimester, as well as in their fetuses at delivery. This study presents a new process to parameterize unknown parameters describing placental transfer (specifically, the transplacental clearance and drug partitioning between the fetus and the mother in the placenta) by combining recently proposed in silico techniques relying exclusively on in vitro information; additionally, drug partitioning between the fetus and the mother was estimated according to more conventional approaches and the results obtained from these methods were compared to clinical data. Predicted PK profiles were evaluated through comparison with in vivo data obtained from the International Maternal Pediatric and Adolescent AIDS Clinical Trials (IMPAACT) Network P1026s study (trial number: NCT00042289).

2. Materials and Methods

2.1. Software

PBPK models were developed using the open source software tool Open Systems Pharmacology (OSP) version 8.0 (http://www.open-systems-pharmacology.org/) which makes formerly commercial software PK-Sim® and MoBi® available as freeware under the GPLv2 License. All source code and the herein developed models will be made publicly available on GitHub (accessible via www.open-systems-pharmacology.org). WebPlotDigitizer (http://automeris.io/WebPlotDigiti-zer/) was used to extract data from published figures and convert them into digital format. The open source software PaDEL-Descriptor [10] was used to estimate molecular descriptors of DTG and RAL and the free software R (version 3.4.1, R Foundation for Statistical Computing, Vienna, Austria; http://www-.r-project.org) was used for non-compartmental analysis and graphics creation.

2.2. General Workflow

The workflow for the development of the pregnancy PBPK model has been previously described in detail [11] and is schematically shown in Fig S1. Briefly, a PBPK model was initially developed for a virtual non-pregnant population and evaluated by comparing simulation results with the observed in vivo PK data in non-pregnant subjects reported in the comparison studies. Thereafter, the non-pregnant PBPK model was translated to pregnancy by substituting the standard model structure with the pregnancy structure and parametrizing the model for the respective gestational age as described before. [11] PK predictions in pregnant women were evaluated by comparison with in vivo PK data obtained from clinical trials of IMPAACT P1026s.

2.3. Development of PBPK pregnancy models

2.3.1. Dolutegravir

DTG is dosed at 50 mg once daily as an orally administered tablet in both pregnant and non-pregnant treatment-naïve and treatment-experienced adult patients without INSTIs resistance. DTG is primarily eliminated by metabolism through various enzymes including UGT1A1, UGT1A3, UGT1A9 and CYP3A4 (~ 51%, ~ 2.8%, ~ 5.5% and ~ 21% of the dose, respectively[12]). In the developed model, the contribution of UGT1A3 and 1A9 to total glucuronidation was combined into the biotransformation pathway mediated by UGT1A1. Finally, to obtain dose fractions summing up to 1.0, the dose fraction metabolized via UGT1A1 was increased to 0.79 based on the assumption that the reported value (0.51) may be underestimated due to hydrolyzation and back conversion of the glucuronide to DTG in the feces, as discussed elsewhere[12]. PBPK model input parameters for DTG are listed in Tab. 1.

Table 1:

Summary of input data for PBPK models in non-pregnant subjects

	Dolutegravir		Raltegravir
Parameter [unit]	Value	Reference	Value	Reference
Molecular weight [g/mol]	419.38	Drugbank.ca	444.42	Drugbank.ca
Lipophilicity [log units]	0.98	fitted a	0.58	Moss et al. [71]
pKa (acid)	10.1	Drugbank.ca	6.67	Moss et al. [71]
Fraction unbound:
Non-pregnant	7.0E-3	Clinical Pharmacology and Biopharmaceutics Review[16]	0.170	Laufer et al. [28]
2nd trimester	8.4E-3	calculated	0.198	calculated
3rd trimester	8.8E-3	calculated	0.206	calculated
Major binding protein	Albumin	Drugbank.ca	Albumin	Laufer et al. [28]
Solubility (at pH 7) [mg/L]	0.172c; 1.98d; 0.0252e	fitted a	8900 f	Moss et al. [72]
Intestinal permeability (transcellular) [cm/min]	0.05	fitted a	1.71 · 10−5	fitted g
Model for estimating organ-to-plasma partition coefficients	Rogers & Rowland		Rogers & Rowland
GFR fraction			1.0	Kassahun et al. [27]
Km-UGT1A1 [uM]	149	Reese et al. [12]	99	Kassahun et al. [27]
Vmax-UGT1A1 [nmol/min/mg]	7.34	fitted a	2.74 d	fitted g
Km-UGT1A9 [uM]			296	Kassahun et al. [27]
Vmax-UGT1A9 [nmol/min/mg] CLspec/[Enzyme_CYP3A4] [1/umol/min]	0.05	fitted a	1.63 d	fitted g

^aValue simultaneously fitted to in vivo plasma concentration-time profiles of non-pregnant subjects and to the reported dose fractions metabolized [17-26].

^bSolubility for the suspension

^cSolubility for the tablet formulation administered in fasted state

^dSolubility for the tablet formulation administered in fed state

^eSolubility for the suspension

^fSolubility implemented as table: PH=1 to 4, solubility=40mg/L; PH=5, solubility=120mg/L; PH=6, solubility=980mg/L; PH=7, solubility=8900mg/L; PH=8, solubility=37300mg/L.

^gValue fitted to in vivo pharmacokinetic data of non-pregnant subjects following oral administration [27, 73-77].

Abbreviations: GFR: glomerular filtration rate; PBPK: physiologically based pharmacokinetic

In the pregnancy PBPK model, physiologic parameters were adjusted to the respective stage of pregnancy as described previously [13]. Additionally, the reference concentrations of UGT1A1 and CYP3A4 (quantifying the concentrations of these enzymes in the model) were increased to reflect induction of these enzymes. Specifically, CYP3A4 reference concentration was increased by a factor of 1.60 in the 2^nd and 3^rd trimesters and UGT1A1 reference concentration by a factor of 1.75 in the 2^nd trimester and 1.92 in the 3^rd trimester [14, 15]. The fraction unbound of DTG, averaging 0.0070 in non-pregnant subjects,[16] was also adjusted based on the albumin concentration measured in the herein investigated study subjects. Specifically, the mean albumin concentration measured in the 2^nd trimester, 3^rd trimester and 6 – 12 weeks postpartum was 34.4 g/L, 32.8 g/L and 41.4 g/L, respectively. Using a previously presented scaling approach [13], these measurements resulted in a fraction unbound of 0.0084 and 0.0088 in the 2^nd and 3^rd trimester, respectively. Additional information on model development and translation to pregnancy can be found in the Supplemental Material.

PK simulations in the non-pregnant population were evaluated by comparison with in vivo data obtained from eight clinical studies reported in the literature that investigated the PK of DTG in a total of 22 different groups of non-pregnant subjects after single and multiple oral administrations of 2 to 100 mg as granule suspension or 50 mg as tablet in fasted or fed state [17-26].

In pregnant women, the PK was predicted in 2 different gestational age groups of non-laboring pregnant women in the 2^nd trimester (median gestational age [range]: 23.5 [20 - 25] weeks) and 3^rd trimester (median gestational age [range]: 33 [30-37] weeks), and in women in labor (median gestational age [range]: 38 [35 - 42] weeks). Drug concentrations in the blood plasma of the umbilical vein were predicted in the laboring pregnant women group.

2.3.2. Raltegravir

RAL is dosed at 400 mg twice daily or 1200 mg once daily as an orally administered tablets in both pregnant and non-pregnant treatment-naive and treatment-experienced adult patients. RAL is primarily eliminated by metabolism catalyzed by UGT1A1 and UGT1A9 (~70% and 11% of the administered dose, respectively[27]). Additionally, approximately 9% is eliminated unchanged through the kidneys [27]. The input parameters and their values for the RAL PBPK model are listed in Tab. 1. Additional information can be found in the Supplemental Material. The non-pregnant PBPK model for RAL was obtained from the OSP GitHub repository (https://github.com/Open-Systems-Pharmacology/Raltegravir-Model/releases) where an extensive description and evaluation of the model can be found.

In the pregnancy PBPK model, physiologic parameters and the reference concentrations of UGT1A1 were adjusted to the respective stage of pregnancy as described above. Since no information on the effect of pregnancy on UGT1A9 could be found, this enzyme was not induced in the presented model. Similar to DTG, the fraction unbound of RAL, averaging 0.17 in non-pregnant adults,[28] was adjusted based on the mean albumin concentration measured in the 2^nd trimester, 3^rd trimester and 6 – 12 weeks postpartum (34.1 g/L, 32.4 g/L and 41.4 g/L, respectively) resulting in a fraction unbound of 0.198 and 0.206 in the 2^nd and 3^rd trimester, respectively. Additional information on model development and translation to pregnancy can be found in the Supplemental Material.

In pregnant women, the PK were predicted in 2 different gestational age groups of non-laboring pregnant women in the 2^nd trimester (median gestational age [range]: 23.5 [21 - 26] weeks) and the 3^rd trimester (median gestational age [range]: 34 [30 - 38] weeks), and in women in labor (median gestational age [range]: 38 [36 - 40] weeks). Drug concentrations in the blood plasma of the umbilical vein were predicted in the laboring pregnant women group.

2.3.3. Clinical in vivo data in pregnant women living with HIV

The clinical in vivo data were from the IMPAACT network P1026s study. Intensive steady state blood samples for PK assessment were collected in the 2^nd trimester (optional), 3^rd trimester, and postpartum, and single maternal and cord blood samples were collected at delivery. Further information can be found in Tab. S1 in the supplement. The performance of the PBPK model to predict drug concentrations in the umbilical cord was evaluated via comparison with clinical in vivo data collected at delivery. The protocol for this study was approved by the responsible Institutional Review Boards. The IMPAACT P1026s study is an ongoing (trial number: NCT00042289), multicenter, phase IV prospective study and a part of the herein reported data has been previously published [29, 30]. In addition, clinical PK data in pregnant women reported by Waitt et al [31] and Blonk et al.[32] were used to evaluate the steady-state PK predictions of DTG around gestational week 31 and of RAL around gestational week 33, respectively.

2.4. Parameterization of placental transfer

Placental transfer kinetics of DTG and RAL was mathematically described as reported previously [33] (also described in detail in the supplement) and informed by in silico methods. Specifically, the transplacental clearance (D_pl) was estimated from the approach suggested by Zhang et al. [34] which estimates the transplacental clearance from the permeability measured in Caco2 cell lines. For DTG, Caco2 cell permeability (2.5 E-6 cm/s) was obtained from Grießinger et al. [35] resulting in a transplacental clearance of 0.43 L/min. For RAL, a Caco2 cell permeability of 7.3E-6 cm/s was reported [36] resulting in a transplacental clearance of 1.24 L/min.

Four different methods were used to estimate the partition coefficient between the fetal intracellular space and the maternal blood plasma of the placenta in the model (K_fc:mp), in particular: i) the “PK-Sim Standard” calculation method;[37] ii) the method proposed by Poulin & Theil;[38, 39] iii) the method proposed by Rodgers & Rowland;[40, 41] and vi) a quantitative structure-activity relationship (QSAR) model for the fetal-maternal blood concentration ratio which was used as surrogate for K_fc:mp. The fetal-maternal blood concentration ratio was calculated according to the QSAR model suggested by Takaku et al. [42] which uses molecular weight, polar surface area, and maximum E-state of hydrogen atom in the compound (H_max) to estimate the fetal: maternal blood concentration ratio. Using the open source software PaDEL-Descriptor [10], the polar surface area was estimated to be 95.9 Ų and 147 Ų for DTG and RAL, respectively; and H_max 0.93 and 0.82 for DTG and RAL, respectively. Physiological values required for calculating the partition coefficients, e.g. tissue composition, are published elsewhere[13]. The values for K_fc:mp calculated according to each of the four methods are listed in Tab. 2.

Table 2:

Values for the partition coefficient between the fetal intracellular space and the maternal blood plasma (K_fc:mp) calculated according to different method

Drug	Method PK-Sim Standard	Poulin & Theil	Rodgers & Rowland	QSAR
Dolutegravir	0.0085	0.43	0.04	0.40
Raltegravir	0.19	0.52	0.26	0.42

2.5. Parameterization of placental metabolism

In the model, the amount of CYP3A4, UGT1A1 and UGT1A9 in the fetal part of the placenta and the fetal body was informed based on reported data from previous studies that quantified the mRNA level of CYP3A4 [43] as well as protein levels and the activity of UGT1A1 [44, 45] in various human tissues, including the adult liver, placenta and fetal liver. Further information can be found in the supplement. The protein amounts listed in Tab. 3 were then incorporated in the pregnancy PBPK model. As discussed further below, the difference of fraction unbound between fetus and mother was not considered because of the high uncertainty in fetal protein binding.

Table 3:

Enzymes amount included in fetal compartments of the PBPK model

Tissue	Enzyme amount [μmol]
	CYP3A4	UGT1A1	UGT1A9
Fetal part of the placenta	3.98E-4	0.06	NA
Fetus	0.12	0.03	0.024

2.6. Evaluation of PBPK models

The PBPK models were evaluated through visual comparison of observed in vivo plasma concentration-time profiles with the concentrations simulated in non-pregnant subjects or in pregnant women. Additional visual assessments included goodness-of-fit (GOF) and residuals vs time plots. Ratios of simulated to observed PK parameters were estimated and the number of ratios falling within a 1.33-fold error range (i.e. 0.75 ≤ ratio ≤ 1.33) was given. An extensive evaluation of the non-pregnant PBPK model for RAL has been previously published on GitHub (https://github.com/Open-Systems-Pharmacology/Raltegravir-Model).

2.7. Sensitivity analysis

Local sensitivity analyses were conducted to assess how the uncertainty in specific parameters might propagate to the final model output (plasma concentration-time profiles in the 2^nd and 3^rd trimester of pregnancy or at delivery). The following parameters were included in local (univariate) sensitivity analyses: UGT1A1 induction, the transplacental clearance, and the gastric emptying time of the mother at delivery.

3. Results

3.1. Non-pregnant PBPK models

3.1.1. Dolutegravir:

The simulated plasma concentration-time profiles of DTG in non-pregnant populations following administration of 50 mg tablet QD in fed state (i.e. the same dosing regimen than in pregnant women) are shown in Fig. 1, while Fig. S2 shows simulated plasma concentration-time profiles following other dosing regimens. The ratios of simulated to observed PK parameters in non-pregnant subjects are listed in Tab. 4 together with the absolute simulated and observed values. All simulated AUC values fell within a 1.33-fold error range. Eleven out of 13 (85%) C_max values fell within a 1.33-fold error range. Time to C_max (t_max) was somewhat less accurately simulated, but still 10 out of 13 (77%) simulated values fell within a 1.33-fold error range. This model was subsequently translated to pregnant women.

Figure 1:

Plasma concentration-time profiles of dolutegravir following oral administration once a day of 50 mg in fed state in non-pregnant subjects. Circles represent observed in vivo data from following studies: A: Ford et al [23]; B: group 1 of the study from Johnson et al [24]; C: group 2 of the study from Johnson et al [24]. D: Wang et al [25]; E: Song et al [20]; The solid line represents the simulated plasma concentration in different population and the shaded area is the predicted 5^th – 95^th percentile range. Semi-log scale figures are given as inset figure in the top right corners.

Table 4:

Comparison of simulated or predicted and observed PK Parameters

	AUC0-t/∞ [mg ·h/L]a Simulated/ observed (ratio)	Cmax [mg/L] Simulated/ observed (ratio)	tmax [h] Simulated/ observed (ratio)
Dolutegravir
Non-pregnant women
Castellino study [17]	34.0/35.9 (0.95)	2.29/2.53 (0.86)	0.95/0.50 (1.90)
Dooley study cohort 1 [18] (steady state)	39.0/36.1 (1.08)	2.91/2.65 (1.10)	2.25/1.5 (1.50)
Dooley study cohort 2 [18] (steady state)	41.1/42.1 (0.98)	2.96/2.91 (1.02)	2.20/2.00 (1.10)
Ford study [23] (steady state)	63.1/52.5(1. 20)	4.00/3.43 (1.17)	4.25/4.00 (1.06)
Johnson2014, Cohort 1 (steady state) [24]	65.2/71.9 (0.91)	4.14/4.35 (0.95)	4.30/3.00 (1.43)
Johnson2014, Cohort2 (steady state) [24]	67.4/71.9 (0.94)	4.30/4.78 (0.90)	4.30/3.50 (1.23)
Song2012_high_fat [21]	66.1/83.6 (0.79)	2.97/4.19 (0.71)	4.90/5.00 (0.98)
Song2012_low_fat [21]	59.8/66.7 (0.90)	2.83/3.81 (0.74)	4.00/3.00 (1.33)
Song2012_moderate_fat [21]	64.7/71.0 (0.91)	2.94/3.86 (0.76)	4.75/4.00 (1.19)
Song2016_moderate_fat_meal [20] (steady state)	62.2/55.4 (1.12)	3.99/3.83 (1.04)	4.00/3.00 (1.33)
Song2013 study [19] b	47.2/40.3 (1.17)	1.82/1.90 (0.96)	2.50/3.00 (0.83)
Weller study [22]	44.1/37.1 (1.19)	1.89/1.84 (1.03)	2.40/2.50 (0.96)
Wang2019 [25]	63.8/51.62 (1.23)	4.30/3.81(1.13)	4.10/4.00 (1.03)
Pregnant women
2nd trimester (steady state)	34.70/42.38 (0.82)	2.77/3.00 (0.92)	4.20/2.00 (2.10)
3rd trimester (steady state)	31.91/47.59 (0.67)	2.57/3.00 (0.86)	4.20/4.00 (1.05)
Raltegravir
Non-pregnant women	See OSP GitHub	See OSP GitHub	See OSP GitHub
Markowitz2006[75]	8.86/7.96 (1.11)	3.04/2.24 (1.36)	0.80/1.00 (0.80)
Iwamoto2009[78]	8.66/4.90 (1.77)	3.10/1.28 (2.42)	0.75/1.50 (0.50)
Rhee2014 [76]	9.13/8.53 (1.07)	3.11/2.22 (1.40)	0.75/2.00 (0.38)
Wenning2009 [77]	8.99/12.25 (0.73)	3.12/3.82 (0.81)	0.75/1.50 (0.50)
Brainard2011_fasted b [79]	9.66/6.47 (1.49)	3.42/1.59 (2.15)	0.75/2.00 (0.38)
Brainard2011_high fat b [79]	8.83/ 11.37 (0.78)	1.48/1.59 (0.93)	2.45/2.00 (1.23)
Brainard2011_moderate_fat b [79]	8.86/6.44 (1.38)	1.54/0.74 (2.08)	2.20/4.00 (0.55)
Brainard2011_low_fat b [79]	9.01/3.39 (2.66)	1.76/0.59 (2.98)	1.95/3.50 (0.56)
Taburet2015_moderate_fat_1 c[80]	8.46/8.24 (1.03)	1.65/2.03 (0.82)	2.20/1.00 (2.20)
Taburet2015_moderate_fat_2 c [80]	8.95/ 11.00 (0.81)	1.54/2.77 (0.56)	2.20/2.00 (1.10)
Pregnant women
2nd trimester (steady state)	4.10/3.90 (1.05)	0.834/0.67 (1.22)	2.55/2.00 (1.28)
3rd trimester (steady state)	3.71/4.44 (0.84)	0.763/0.85 (0.89)	2.50/2.00 (1.25)

Data expressed as geometric mean values unless indicated otherwise.

^aAUC_0-t was used in the studies by Dooley et al. [18], Ford et al. [23], Johnson et al. [24], Song et al. [20], Markowitz et al.,[75] Brainard et al.[79] and Taburet et al.[80]; whereas AUC_0-∞ was used in the studies by Castellino et al. [17], Song et al. [21], Song et al. [19], Weller et al. [22], Iwamoto et al.[78], Rhee et al.[76] and Wenning et al.[77].

^bData expressed as arithmetic mean values.

^cData expressed as median values.

Abbreviations: AUC_0-t: area under the concentration-time curve from zero to the time point of the last observed plasma concentration (in case of multiple dose studies, time refers to the time after last dose); AUC_0-∞: area under the concentration-time curve from zero to infinity; C_max: peak plasma concentration; t_max: time at which peak plasma concentration is reached.

3.1.2. Raltegravir:

An extensive evaluation of the non-pregnant PBPK model for RAL can be found in the model repository on GitHub (https://github.com/Open-Systems-Pharmacology/Raltegravir-Model); here, only a limited number of results is shown. The simulated plasma concentration-time profiles of RAL in non-pregnant populations following administration of 400 mg tablet BID in fed state (i.e. the same dosing regimen as in pregnant women) are shown in Fig. 2. In addition, Fig. S3 shows simulated plasma concentration-time profiles following other dosing regimens. The ratios of simulated to observed PK parameters in non-pregnant subjects are listed in Tab. 4 together with the absolute simulated and observed values. Five out of ten (50%) simulated AUC value fell within a 1.33-fold error range (i.e. 0.75 ≤ ratio ≤ 1.33). For C_max and t_max, 3 out of 10 (30%) simulated values were within this range. While there were some difficulties to describe the data of few clinical studies, the model was overall deemed adequately to describe RAL PK when also considering additional studies with different posology (see Fig. S3 and additional information available on https://github.com/Open-Systems-Pharmacology/Raltegravir-Model). Hence, this model was subsequently translated to pregnant women.

Figure 2:

Plasma concentration-time profiles of raltegravir following oral administration twice a day of 400mg with moderate fat meal in non-pregnant subjects. Circles represent observed in vivo data from following studies: A: Brainard et al [79]; B: group 1 of the study from Taburet et al [80]; C: group 2 of the study from Taburet et al[80]. The solid line represents the simulated plasma concentration in different population and the shaded area is the predicted 5^th – 95^th percentile range. Semi-log scale figures are given as inset figure in the top right corners.

3.2. Pregnancy PBPK models in the 2^nd and 3^rd trimester

3.2.1. Dolutegravir:

The predicted DTG plasma concentration-time profiles in the 2^nd and 3^rd trimesters of pregnancy are shown in Fig. 3 and in Fig. S5 with clinical data reported by other research groups. Fig. 4 shows the GOF plot for the model-predicted DTG plasma concentrations in non-pregnant and pregnant women. All but 2 out of 16 (87.5%) geometric mean DTG concentrations were predicted within a two-fold error range. The absolute values as well as ratios of predicted to observed AUC_0-24, C_max and t_max in the pregnant populations are listed in Tab. 4. The AUC_0-24 of 2^nd trimester and C_max of 2^nd trimester and 3^rd trimester were predicted within a 1.33-fold error range. The AUC_0-24 in 3^rd trimester was predicted within a 1.5-fold error range; similar to simulations in non-pregnant subjects, t_max was less accurately predicted in the 2^nd trimester, but again in the 1.33-fold error range in the 3^rd trimester. Tab. S2 provides an overview of additional PK parameters simulated in non-pregnant and pregnant populations. Variability was reasonably described by the model; specifically, the predicted 5^th – 95^th percentile range contained 76% of all observed concentration values in the 2^nd trimester and 69% of all observed concentration values in the 3^rd trimester.

Figure 3:

Plasma concentration-time profiles of dolutegravir following oral administration of 50 mg once a day in pregnant women. Semi-log scale figures are given as inset figure in the top right corners. Observed steady-state in vivo data were taken from in vivo study of IMPAACT P1026. [29] A: dolutegravir 50 mg once a day in pregnant women in 2^nd trimester. Empty circles represent individual concentrations taken from in vivo study of IMPAACT P1026. [29] Black circles represent geometric mean concentrations taken from in vivo study of IMPAACT P1026. [29] The solid line represents the predicted mean concentration and the shaded area the predicted 5^th – 95^th percentile range; The dotted line represents the predicted mean concentration of non-pregnant population; B: dolutegravir 50 mg once a day in pregnant women in 3^rd trimester. Empty circles represent individual concentrations taken from in vivo study of IMPAACT P1026. [29] The solid line represents the predicted mean concentration and the shaded area the predicted 5^th – 95^th percentile range; The dotted line represents the predicted mean concentration of non-pregnant population; C: dolutegravir 50 mg once a day in pregnant women with an average gestational age of 38 weeks at delivery. Empty circles represent individual concentration data in the maternal plasma taken from in vivo study of IMPAACT P1026. [29]; the lines represent the predicted mean concentration in the maternal plasma using different partition coefficients calculated by different methods.; The solid line represents the QSAR method; The dotted line represents Poulin and Theil method; The dash-dot line represents the Rodgers and Rowland method; The dash line represents the PK-sim standard method. The shaded area represents the predicted 5^th – 95^th percentile range of the prediction by using QSAR method. D: dolutegravir 50 mg once a day in pregnant women with an average gestational age of 38 weeks at delivery. Empty circles represent individual concentration data in the umbilical vein taken from in vivo study of IMPAACT P1026; [29] the lines represent the predicted mean concentration in the umbilical vein using different partition coefficients calculated by different methods: the solid line represents the QSR method; the dotted line represents Poulin and Theil method; the dash-dot line represents the Rodgers and Rowland method; and the dash line represents the PK-sim standard method. The shaded area presents the predicted 5^th – 95^th percentile range of the prediction by using QSAR method.

Figure 4:

Goodness-of-fit and residuals vs time plots of dolutegravir concentrations in non-pregnant subjects (A) and pregnant women (B). The solid line represents the line of identity and the dotted lines the 2-fold error range. A panel: Upper plot: GOF plot of geometric mean dolutegravir concentrations in non-pregnant populations. single dose study: blue circles indicate the study by Castellino et al. (20mg single dose) [17]; green circles indicate the study by Song et al. (50mg single dose) [19]; light blue circles indicate the high fat meal group in the study by Song et al. (50mg single dose) [21]; black circles indicate the study by Weller et al. (50mg single dose) [22]; bright green circles indicate the moderate fat meal group in the study by Song et al. (50mg single dose) [21]; grey circles indicate the low fat meal group in the study by Song et al. (50mg single dose) [21]. multiple dose study: dark blue circles indicate the study by Ford et al. (50mg once daily) [23]; orange circles indicate the group 1 in the study by Dooley et al. (50mg once daily)[18]; red circles indicate the group 2 in the study by Dooley et al. (50mg once daily)[18]; dark green circles indicate the group 1 in the study by Johnson et al. (50mg once daily) [24]; sky blue circles indicate the group 2 in the study by Johnson et al. (50mg once daily) [24]; light grey circles indicate the study by Song et al. (50mg once daily) [20]; yellow circles indicate the study by Wang et al. (50mg once daily)[25]. Lower plot: residuals vs time plot of dolutegravir in non-pregnant population. empty circles represent the geometric mean concentrations of non-pregnancy reference studies. B panel: Upper plot: GOF plot of dolutegravir in pregnant population. Blue circles indicate geometric mean concentrations in 2^nd trimester and orange circles indicate geometric mean concentrations in 3^rd trimester; Lower plot: residuals vs time plot of dolutegravir in pregnant population. Empty circles represent the individual concentrations in 2^nd and 3^rd trimesters.

3.2.2. Raltegravir:

The predicted RAL plasma concentration-time profiles in the 2^nd and 3^rd trimesters of pregnancy are shown in Fig. 5 and in Fig. S5 with clinical data reported by other research groups. Fig. 6 shows the GOF plot for the model-predicted RAL plasma concentrations in non-pregnant and pregnant women with the residuals versus time. All but 1 of out 14 (92.9%) geometric mean RAL concentrations in pregnant populations were predicted within a 2-fold error range. The observed and predicted values for AUC_0-24, C_max and t_max as well as their ratios are listed in Tab. 4. These geometric mean PK parameters were all predicted within a 1.33-fold error range in the 2^nd and 3^rd trimester. Tab. S2 provides an overview of additional PK parameters simulated in non-pregnant and pregnant populations. Variability was underestimated by the model; specifically, the predicted 5^th – 95^th percentile range contained only 64% of all observed concentration values in the 2^nd trimester and 55% of all observed concentration values in the 3^rd trimester.

Figure 5:

Plasma concentration-time profiles of raltegravir following oral administration of 400 mg twice a day in pregnant women in steady state. Semi-log scale figures are given as inset figure in the top right corners. Observed steady-state in vivo data were taken from in vivo study of IMPAACT P1026.[30] A: raltegravir 400 mg twice a day in pregnant women in 2^nd trimester. Empty circles represent individual concentrations taken from in vivo study of IMPAACT P1026. [30] Black circles represent geometric mean concentrations taken from in vivo study of IMPAACT P1026. [30] The solid line represents the predicted mean concentration and the shaded area the predicted 5^th – 95^th percentile range; The dotted line represents the predicted mean concentration of non-pregnant population; B: raltegravir 400 mg twice a day in pregnant women in 3^rd trimester. Empty circles represent individual concentrations taken from in vivo study of IMPAACT P1026. [30] The solid line represents the predicted mean concentration and the shaded area the predicted 5^th – 95^th percentile range; The dotted line represents the predicted mean concentration of non-pregnant population; C: raltegravir 400 mg twice a day in pregnant women with an average gestational age of 38 weeks at delivery. Empty circles represent individual concentration data in the maternal plasma taken from in vivo study of IMPAACT P1026; [30] the lines represent the predicted mean concentrations in the maternal plasma using different partition coefficients calculated by different methods.; The solid line represents the QSAR method; The dotted line represents Poulin and Theil method; The dash-dot line represents the Rodgers and Rowland method; The dash line represents the PK-sim standard method. The shaded area represents the predicted 5^th – 95^th percentile range of the prediction by using QSAR method. D: raltegravir 400 mg twice a day in pregnant women with an average gestational age of 38 weeks at delivery. Empty circles represent individual concentration data in the umbilical vein taken from in vivo study of IMPAACT P1026; [30] the lines represent the predicted mean concentrations in the umbilical vein using different partition coefficients calculated by different methods: the solid line represents the QSAR method; the dotted line represents Poulin and Theil method; the dash-dot line represents the Rodgers and Rowland method; and the dash line represents the PK-sim standard method. The shaded area represents the predicted 5^th – 95^th percentile range of the prediction by using QSAR method.

Figure 6:

Goodness-of-fit and residuals vs time plots of raltegravir concentrations in non-pregnant subjects (A) and pregnant women (B). The solid line represents the line of identity and the dotted lines the 2-fold error range. A panel: Upper plot: GOF plot of geometric mean dolutegravir concentrations in non-pregnant populations. single dose study: bright green circles indicate the study by Iwamoto et al. (400mg single dose) [78]; dark green circles indicate the study by Rhee et al. (400mg single dose) [76]; grey circles indicate the study by Wenning et al. (400mg single dose) [77]. multiple dose study: blue circles indicate the fasted group in the study by Brainard et al. (400mg multiple dose) [79]; orange circles indicate the high fat group in the study by Brainard et al. (400mg multiple dose)[79]; dark blue circles indicate high moderate fat group in the study by Brainard et al. (400mg multiple dose) [79]; red circles indicate the low fat group in the study by Brainard et al. (400mg multiple dose) [79]; black circles indicate the study by Markowitz et al. (400mg multiple dose) [75]. sky blue circles indicate the group 1 in the study by Taburet et al. (400mg multiple dose) [80]; light blue circles indicate the group 2 in the study by Taburet et al. (400mg multiple dose) [80]; B panel: Upper plot: GOF plot of raltegravir in pregnant population. blue circles indicate geometric mean concentrations in women in the 2^nd trimester; orange circles indicate geometric mean concentrations in women in the 3^rd trimester. Lower plot: residuals vs time plot of raltegravir in pregnant population. Empty circles represent the individual concentrations in 2^nd and 3^rd trimesters.

3.3. Pregnancy PBPK models for delivery

3.3.1. Dolutegravir

Maternal and umbilical cord plasma concentrations of DTG predicted at delivery are shown in Fig 3C & 3D, respectively. Fourteen out of 20 maternal samples at delivery fell within the 2-fold error range of the predicted mean concentration, and 8/20 maternal samples fell within the 1.5-fold error range. Seventeen out of 20 cord samples fell within the 2-fold error range of the predicted mean concentration and 11/20 cord samples fell within the 1.5-fold error range.

The predicted steady-state exposure was higher in the fetus compared to the mother; specifically, the predicted geometric AUC_tau was 40.99 mg h/L in the umbilical vein and 26.84 mg h/L in the maternal plasma.

3.3.2. Raltegravir

Maternal and umbilical cord plasma concentrations of DTG predicted at delivery are shown in Fig 5C & 5D, respectively. Five out of 24 maternal samples fell within the 2-fold error range of the predicted mean and 7/23 fetal samples fell within the 2-fold error (one fetal sample was missing). The predicted steady-state exposure was higher in the fetus compared to the mother; specifically, the predicted geometric AUC_tau was 4.73 mg h/L in the fetus and 3.16 mg h/L in the maternal plasma.

3.4. Sensitivity analyses

Local sensitivity analyses are shown in Figs S4, S6 and S7. The sensitivity analysis of UGT1A1 induction of DTG and RAL shown in Fig. S4 indicated that, within the tested range, UGT1A1 induction had a moderate influence on the predicted PK of DTG and RAL during pregnancy. The sensitivity analysis for the placenta diffusion clearance of DTG shown in Fig. S6 suggests that it was a sensitive model parameter for DTG but not for RAL. Although no sensitivity analysis was conducted for K_fc:mp, Fig. 3D and 5D show the predicted PK in the umbilical vein when different values for K_fc:mp are incorporated in the model. Further sensitivity analysis on the gastric emptying time of the mother are shown in Fig S7 indicating that a prolonged gastric emptying time increases both maternal and fetal exposure.

4. Discussion

In this study, maternal-fetal PBPK models were developed for DTG and RAL and evaluated by comparing predicted concentrations to those observed in the maternal plasma during the 2^nd and 3^rd trimester and in the maternal and umbilical cord plasma at delivery. Development of these models followed a standard workflow comprising the initial establishment of a non-pregnant PBPK model and the subsequent translation to pregnancy.

The PK of DTG predicted in pregnant, non-laboring women was in good agreement with clinical data obtained at different stages of pregnancy (Fig. 3 and Tab. 4). While most PK parameters were adequately predicted, t_max in the 2^nd trimester was somewhat overestimated and AUC_tau underestimated in the 3^rd trimester (Tab. 4, Fig. 3). The reason for the decrease in t_max in the 2^nd (but not the 3^rd) trimester in vivo is yet unknown and further clinical studies strictly controlling for food intake could help to elucidate that point. The underestimation of AUC_tau in the 3^rd trimester could mainly be ascribed to an overestimation of total body clearance. Combined with the results from the sensitivity analysis on UGT1A1 induction (Fig. S4), this finding emphasizes that UGT1A1 induction may be lower than expected here and that changes in the fraction unbound are, at least for DTG, the main driver of increased total clearance which is consistent with recent findings in the literature [46]. The PK of RAL was generally well predicted in pregnant, non-laboring women (Fig. 5) and all mean PK parameters were predicted within a 1.33-fold error range (Tab. 4). However, variability was underestimated, especially in the first hours after drug administration indicating that the model did not fully capture variability related to drug absorption.

UGT1A1 plays an important role in the metabolism of both DTG and RAL. While in vitro experiments support an increase in UGT1A1 expression mediated by rising progesterone levels during pregnancy [47], there is scarce information on quantitative changes in UGT1A1 expression during pregnancy in vivo. In a previous pregnancy PBPK model for acetaminophen (paracetamol), UGT1A1 was assumed to be induced by a factor of 1.75 in the 2^nd trimester and 1.92 in the 3^rd trimester [15] and these induction factors were incorporated here. Due to the relatively high uncertainty in these factors, a local sensitivity analysis was conducted (Fig. S4). While maternal PK of DTG was moderately sensitive to alterations in UGT1A1 expression, maternal PK of RAL was rather weakly affected by these alterations. Additional studies dedicatedly investigating UGT1A1 activity changes in pregnant women are needed to better define whether and to which extent UGT1A1 is induced during pregnancy. Clinical studies on different UGT1A1-metabolized drugs, e.g. carvedilol and irinotecan, could be helpful to answer this question.

A key objective of this study was to predict fetal PK of DTG and RAL in the maternal plasma and venous blood plasma of the umbilical cord at delivery. To this end, different approaches to inform placental transfer kinetics were tested. Specifically, D_pl (the placental diffusion clearance) was estimated from Caco-2 permeability [34] and K_fc:mp (the fetal intracellular-to-maternal plasma partition coefficient in the placenta) was calculated either by previously reported methods for estimation of organ-to-plasma partition coefficients [37]’[38, 39]’[40, 41] or by a previously reported QSAR approach [42] assuming that the fetal-maternal blood concentration ratio can be used as surrogate for K_fc:mp. Hence, placental transfer was exclusively informed by in vitro and in silico methods. This has the strength to bypass the need for additional ex vivo experiments, such as the placental cotyledon perfusion experiment, but it also displays several shortcomings. For example, DTG is a substrate of P-glycoprotein (P-gp; multidrug resistance protein 1 [MDR1]), one of the major efflux transporters expressed in the placenta[48], and the effect of this transporter may not be consistently accounted for by the proposed approach for estimation of D_pl. Although drug transporters are present in Caco-2 cell lines, their expression may differ from that in placental cells and may not correlate with the placental protein amount in vivo [48], thereby distorting the estimated value for D_pl. Indeed, the limited data available seem to indicate that the expression of P-gp normalized to that of the housekeeping gene GAPDH (glyceraldehyde-3-phosphate dehydrogenase) is slightly higher in Caco-2 cells [49-52] than in the human term placenta [53, 54]. The higher P-gp expression in Caco-2 cells may consequently have contributed to an underestimation of D_pl. As shown in Fig. S6, D_pl was a sensitive model parameter for DTG but not for RAL. Unfortunately, the low number of observed data and the high variability therein, especially in the terminal phase, preclude any sound conclusion on whether the estimated D_pl for DTG should be higher. This illustrates the importance of developing new modeling approaches that delineate passive diffusion and active transport across the placenta.

The presented findings in Fig. 3D and 5D show that the various methods to estimate K_fc:mp yield largely different predictions of umbilical vein concentrations. Given the relatively small size of the fetus, maternal plasma concentrations were barely affected by different K_fc:mp values (Fig. 3C and 5C). In general, the small amount of clinical data hindered a thorough evaluation of these predictions and more data are clearly needed, ideally from different tissues (e.g. the maternal plasma, placenta and umbilical vein) to better asses the predictive performance of these models. Nonetheless, keeping this limitation in mind, those models that informed K_fc:mp via the Poulin & Theil method or the QSAR approach appeared to predict umbilical vein concentration better than the other models. Umbilical concentrations of DTG appeared to be adequately predicted by these models (Fig 3D), but RAL concentrations were generally underestimated, predominantly because maternal concentrations were also underestimated (Fig 5C and D). This finding stresses the importance of the maternal PK on fetal drug exposure indicating that the main elimination pathway of the fetus is transfer over the placenta back in the mother.

RAL concentrations in the maternal plasma and umbilical vein were especially underestimated after 12 h (i.e. the dosing interval). One reason for the underestimation of maternal RAL plasma concentrations at delivery appeared to be the relatively fast absorption in the model. While the PBPK models for non-laboring women incorporated a fed state to reflect the fact that no restrictions on food intake were imposed in the clinical study, the PBPK model for laboring women incorporated the fasted state (assuming that food intake prior to the onset of labor and delivery is implausible). However, gastric passage of RAL (as well as DTG) was nonetheless delayed in the PBPK model at delivery because there is some evidence that gastric emptying and drug absorption from the gastrointestinal tract are slowed during labor [55]’[56, 57]. Specifically, Whitehead et al. reported a 3-fold delay in acetaminophen t_max in pregnant women (n=36) during labor when compared with 2 h post-delivery women (n=17) [57]. Based on this observation, a 3-fold delay in gastric emptying time was incorporated in the presented PBPK models for RAL and DTG which may not have been enough to reflect the delay in drug absorption. Unfortunately, little clinical data is available for the first few hours after drug administration which complicates a proper assessment of alterations in drug absorption during delivery. As discussed elsewhere in greater detail [9, 58], the effect of pregnancy on drug absorption is poorly understood. The potential effect of delivery – on top of that during pregnancy – further complicates the situation. Once informative clinical data are available, PBPK models applied to the peripartum period could investigate such effects and refine the understanding of physiological changes affecting drug absorption as well as other PK processes.

Ex vivo cotyledon perfusion experiments constitute another important source of information for parameterization of placental transfer in PBPK models [59-61]. One of the advantages is possible to estimate the fraction unbound of the compound in placenta. Yet, data from these experiments are not always available for the studied drug and in other cases they may not translate into meaningful PBPK predictions. For example, although DTG was used in a previous ex vivo cotyledon perfusion experiment, the transfer was observed to be very slow and apparent equilibrium concentrations were not reached after 3 hours [62]. These authors discussed that one reason for the slow transfer could have been equal albumin concentrations in the maternal and fetal compartments. Indeed, differences in maternal and fetal protein binding can be critical determinants for placental transfer and equilibrium concentrations reached in steady state [63].

Since only the free drug fraction crosses the placenta, it can be expected that, for highly protein-bound drugs, an increase in the fetal fraction unbound is associated with a decrease in total concentrations at steady-state because the bound drug concentration diminishes. Therefore, maternal-fetal PBPK should ideally consider such differences between the drug’s fraction unbound in the maternal and fetal plasma. However, although technically possible in the herein used model, little is known about changes in the fetal fraction unbound. While umbilical cord concentrations of fetal albumin and α-fetoprotein (AFP) can in principle be easily measured from plasma samples obtained at delivery, it is unclear whether fetal albumin and α-fetoprotein display the same affinity to drugs than adult albumin. There is evidence that, compared to adult albumin, fetal albumin has a different binding affinity to several drugs [64, 65] and that AFP lacks specific drug-binding sites [66, 67]. This complicates an estimation of a drug’s unbound fraction in the fetal plasma. To correctly parameterize placental transfer of highly protein-bound drugs in PBPK models, clinical data of the fetal fraction unbound in vivo are clearly needed. These data could then either be directly applied to inform specific PBPK models or to develop and train novel in silico approaches for prediction of the fetal fraction unbound.

On May 18^th, 2018, the US FDA released a warning letter [68] that DTG may cause serious neural tube birth defects involving the brain, spine, and spinal cord. The preliminary observations were found in a study in Botswana in women who received DTG at the time of conception [68, 69]. Although the mechanism leading to teratogenicity of DTG is not yet understood, one hypothesis is that DTG affects folic acid binding to the folate receptor α, thereby reducing folic acid concentrations in the fetus [70]. Since folic acid is essential for neural tube development, a reduction in folic acid can potentially cause neural tube defects in the fetus. According to in vitro results presented by Zamek et al. [70], free DTG concentrations of approximately 37 μM are associated with a 36% inhibition of folate receptor α. To set these figures in the in vivo context, the presented PBPK model was extrapolated to the 6^th gestational week while assuming an induction of UGT1A1 by 33% in the first trimester, as suggested previously [15]. Unbound DTG concentrations were then predicted in the maternal blood of the placenta. The maximum unbound DTG concentration predicted in steady state was 0.06 μM at the 6^th gestational week. Using a simple Emax-model fitted to the data reported by Zamek et al. [70] (fitted values are Emax: 1.0; EC50: 1276 μM) , this DTG concentration translates into an inhibition of the folate receptor α by approximately 7 %. It should be noted, though, that this value is based on predicted concentrations that cannot be evaluated because of lacking clinical data and it should therefore not be used to guide dosing decisions. Yet, this example illustrates how PBPK modeling can theoretically contribute to support decision making for the use of DTG during pregnancy. However, this example also emphasizes the need for clinical data to support the confidence of model-based predictions.

5. Conclusion

The developed PBPK models successfully predicted the mean PK profile at different stages of pregnancy by integrating prior knowledge of the pregnancy-related effects on relevant physiological parameters and apparent enzyme activity. Importantly, umbilical vein concentrations were predicted by integrating information solely from in vitro or in silico techniques. This is the first study evaluating the applicability of standard equations for predicting the fetal-maternal partition coefficient in the placenta. The presented models provide new mechanistic insights in the PK of RAL and DTG during pregnancy which can be conceptually generalized and applied to other drugs. The findings also stress the importance of measuring the unbound fraction for highly protein-bound drugs in both the maternal and fetal plasma when clinical trials are conducted in pregnant populations to facilitate the proper parameterization of PBPK models. Ultimately, verified PBPK models may be used to support informed decision making when clinical trials are designed in this frequently ignored population or when sound and consistent information from trials is lacking.