Comparison of Metformin PBPK Models Incorporating Placental Transfer to Predict Fetal and Maternal Exposure

ABSTRACT

Physiologically based pharmacokinetic (PBPK) modeling of placental drug transfer is an evolving tool for predicting fetal drug exposure. In this study, a pregnancy‐specific metformin PBPK model was developed, and the following four approaches were evaluated to predict metformin placental transfer: (1) perfusion‐limited model, and permeability‐limited models using (2) ex vivo cotyledon open system apparent clearance, (3) ex vivo cotyledon closed system data fit to a three‐compartment model to estimate clearance, and (4) active transport kinetics and passive clearance. Simulated metformin maternal plasma concentrations (MPCs) and umbilical cord venous plasma concentrations (UCCs) were compared to observed in vivo data from subjects with gestational diabetes mellitus taking metformin 500 mg twice daily. Model selection criteria were determined by the percentage of observed clinical data falling within the 5th to 95th percentiles of the simulated population. Among the approaches, the model that included passive permeability and in vitro intrinsic transporter clearances (Approach 4) best described placental metformin transfer, with 92% of UCCs falling within the 5th to 95th percentiles of the simulated population. Furthermore, maternal uptake transport had the largest influence on predicted UCCs. A two‐fold increase in maternal uptake transport increased the predicted population mean UCC C _max by 97%, whereas a 0.5‐fold decrease resulted in a 49% decrease in UCC C _max. This refined PBPK model offers a valuable framework for predicting placental transfer and fetal exposure of metformin when placental transporters are altered throughout pregnancy and/or with pathological conditions.

Study Highlights.

• What is the current knowledge on the topic?

  • ○
    Pregnant women are exposed to numerous drugs, and knowledge of proper dosing is required to optimize treatment of disease. Clinical studies are challenging and infrequently conducted for a variety of reasons, and methods to predict fetal exposure are limited.
• What question did this study address?

  • ○
    Which placental transfer approach incorporated in a pregnancy‐specific metformin PBPK model best predicts metformin maternal and umbilical cord concentrations (UCCs)?
• What does this study add to our knowledge?

  • ○
    A PBPK model incorporating placental transfer, including five transport proteins and passive diffusion, most accurately predicted observed metformin maternal and UCCs (92% within the 5th to 95th percentiles of predicted concentrations). Maternal uptake transport was a major determinant of fetal exposure.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    Refined metformin PBPK models incorporating placental transporter data may be used to predict placental transfer and fetal exposure at earlier gestational ages as knowledge of changes in placental transporters during pregnancy and/or pathologic conditions becomes available.

1. Introduction

In the United States, the incidence of pregnant patients diagnosed with gestational diabetes mellitus (GDM) was 7.8% in 2020, a 20% increase since 2016 [1]. The rising incidence of GDM parallels the ongoing obesity epidemic and increasing cases of type‐two diabetes mellitus (T2DM) [2]. GDM and T2DM both stem from an inability to compensate for the degree of insulin resistance, and metformin is the first‐line treatment for non‐pregnant patients with T2DM [3]. Metformin is often used as the first treatment option for GDM. However, caution is necessary when using metformin during pregnancy, as it can cross the placenta and reach the fetus [4].

Metformin undergoes extensive active transport and is a substrate for multiple transporters expressed in the intestine, kidney, liver, and/or placental tissue [5] including organic cation transporter (OCT)1 [6, 7], OCT2 [6, 8], OCT3 [6, 8, 9], multidrug and toxin extrusion protein (MATE)1 [8], MATE2 [6, 8], plasma membrane monoamine transporter (PMAT) [8], organic zwitterions/cation transporter (OCTN)1 [8, 10], OCTN2 [10], serotonin transporter (SERT) [8], and the human thiamine transporter (THTR)2 [8]. Metformin placental transfer has been reported in ex vivo cotyledon experiments [11, 12, 13, 14], and metformin concentrations in placental tissue were similar to those in the fetal compartment [15]. Assessing in vivo fetal drug exposure is challenging due to limited opportunities for sample collection during pregnancy without risking harm to the fetus. Fetal exposure is often assessed by opportunistic sampling of umbilical cord blood at the time of delivery. Clinical studies that collected metformin umbilical cord plasma concentrations (UCCs) at delivery [15, 16, 17, 18, 19, 20] have reported UCC to maternal plasma concentration (MPC) ratios in the range of 0.7–3. Liao et al. demonstrated a nonlinear relationship between the time of the last metformin dose and UCC to MPC ratios [15].

Physiologically based pharmacokinetic (PBPK) modeling provides a mechanistic framework capable of incorporating physiological changes that occur during pregnancy. PBPK models can incorporate different metformin placental transfer approaches to simulate fetal exposure. The simplest approach to modeling metformin placental transfer is to assume perfusion‐limited disposition where exposure in the feto‐placental compartment is determined by blood flow into and out of the compartment. More complex placental transfer models include the permeability‐limited approach, which incorporates a combination of passive permeability and active placental transport [21]. While numerous pharmacokinetic and PBPK models describing metformin disposition in non‐pregnant and pregnant subjects have been published [22, 23, 24], only a few include maternal‐fetal transfer [24, 25], and none have compared various approaches to modeling metformin placental transfer.

The aim of this study was to evaluate different approaches to predict metformin fetal exposure using PBPK models by comparing predicted MPCs and UCCs with observed clinical data. This study is important because it assessed the utility of incorporating ex vivo cotyledon experimental data and in vitro transporter kinetic data into metformin PBPK models to predict placental transfer using observed clinical data for comparison to predicted MPCs and UCCs. Our refined model provides a framework for assessing how alterations in placental transporters due to gestational age or disease state may impact metformin placental transfer and fetal exposure.

2. Methods

2.1. Software and Clinical Data

PBPK models were built using the Simcyp Simulator (version 23, Certara Predictive Technologies, Sheffield, UK). WebPlot Digitizer (Version: 4.4, WebPlotDigitizer, Pacifica, California, USA) was used to extract data from published figures for model development and verification. Observed data consisted of measured metformin concentrations from subjects with GDM taking 500 mg metformin immediate release tablets every 12 h until labor and delivery (n = 16) [15, 18]. Umbilical cord and maternal plasma samples were collected after delivery, 5.5–37.25 h after the last metformin administration. UCCs and MPCs below the limit of quantitation (LOQ) were excluded (n = 3 and n = 4, respectively). The UCC (fetal) to MPC (maternal) ratio (FM ratio) of paired samples was determined when both MPC and UCC were greater than LOQ (n = 12).

2.2. Development of Metformin PBPK Model

An overview of the modeling and simulation workflow is outlined in Figure 1. The default metformin compound file within the Simcyp Simulator was used for base model development and verified against data from non‐pregnant patients with T2DM [15] using the default Simcyp healthy volunteers population. After transitioning from the healthy volunteers population to the Simcyp virtual pregnancy population, the first‐order absorption rate (k_a) in the metformin model was adjusted to fit clinical data from the pregnant GDM population, consistent with literature reports of reduced gastric mucosal blood flow and delayed gastric emptying [26, 27]. The modified metformin PBPK model was verified against data from pregnant patients with GDM. A summary of input parameters in the metformin PBPK model and pregnancy population is presented in Table 1. Once the metformin pregnancy PBPK model was verified, four approaches to modeling placental transfer were evaluated. Steady‐state predictions for metformin MPC and UCC from each approach were compared to observed clinical data. For all simulations, the trial design and demographic characteristics were matched to published information from the corresponding clinical trial. Ten separate virtual trials were simulated to assess interstudy pharmacokinetic variability. A term pregnancy was assumed to be 38 weeks gestation. The performance of the metformin PBPK model was evaluated by visual inspection of the simulated plasma concentration versus time profiles compared to observed data from non‐pregnant and pregnant patients, and comparison of predicted versus observed pharmacokinetic (PK) parameters (maximal concentration (C _max) and area under the curve from 0 to 12 h after dosing at steady state (AUC_0‐12,ss)).

FIGURE 1.

Schematic workflow of metformin PBPK model development. CL_int,T, in vitro intrinsic transporter‐mediated clearance; CL_PDF, placental‐to‐fetal passive clearance; CL_PDM, maternal‐to‐placental passive clearance; GDM, gestational diabetes mellitus; MATE, multidrug and toxin extrusion protein; OCT, organic cation transporter; OCTN, organic zwitterions/cation transporter; PAMPA, parallel artificial permeability assay; PBPK, physiologically based pharmacokinetic; PMAT, plasma membrane monoamine transporter; SERT, serotonin transporter; THTR, the human thiamine transporter; T2DM, type 2 diabetes mellitus. Created with BioRender.com.

TABLE 1.

Select input parameters for the metformin PBPK model, virtual pregnancy population, and placental transfer modeling approaches.

Parameter	Value	Source and assumption
Metformin PBPK model parameters
Absorption	First‐order absorption model	Metformin exhibits nonlinear absorption with declining bioavailability at higher doses [28]. However, for purposes of this study, a simple first‐order absorption model was used, and simulations were restricted to a single dosing regimen to prevent error introduced by lower absorption at higher doses
Dose	500 mg
Tau	12 h
Number of doses	8
Fraction absorbed, fa	0.5	fa = observed bioavailability for 500 mg metformin in pregnant population, 0.64 [15] × salt factor 0.78
First‐order absorption rate constant, ka (h−1)	0.15	Optimized to fit clinical data in pregnant population [15]
Absorption Lag Time (h)	0.29	Simcyp Default [29, 30]
Fraction unbound	1	With a measured plasma fu of 1 [29], it was assumed that binding to all blood‐based proteins and tissues was non‐significant
Distribution Model	Full PBPK Model (Prediction Method 3)	Metformin is a cation at physiologic pH. Therefore, the modified Rogers and Rowland Method [31] that includes ion permeability (Method 3) was used to predict Vss
V ss (L/kg)	0.68	Method 3 Predicted
K p Scalar	0.31	Optimized
CYP3A4 CLint (μL/min/pmol)	0.334	Simcyp Default [32]
fumic	1
Hepatic CLPD (μL/min/million cells)	5.88 × 10−5	Simcyp Default Cryopreserved hepatocytes [33]
Hepatic OCT1 CLint,T (μL/min/million cells)	0.316 (RAF/REF = 1.84)	Simcyp Default Cryopreserved hepatocytes [33]
Renal CLPD (μL/min/million cells)	4.26 × 10−7	Simcyp Default Scaled PAMPA permeability [34]
Renal OCT2 CLint,T (μL/min/million cells)	1.155 (RAF/REF = 25)	Simcyp Default; elimination for the metformin compound file was determined by validated EGD renal elimination [30, 35] with renal transporters OCT2 and MATE1
Renal MATE1 CLint,T (μL/min/million cells)	16.64 (RAF/REF = 0.128)	Simcyp Default Transfected HEK293 [36]
Virtual pregnancy population parameters
Renal OCT2 over the course of pregnancy	1 + 0.0406x − 0.0007x 2 where x = Gestational Week	Simcyp Default
Renal MATE1 over the course of pregnancy	1	Simcyp Default
Renal OCT2 and MATE1 CV%	50	Imputed variability consistent with measured CV% of renal transporters [37, 38, 39]
Hepatic CYP3A4 over the course of pregnancy	1 + 0.0129x + 0.0005x 2 where x = Gestational week	Simcyp Default
Placental SERT, MATE1, OCT3, PMAT, and THTR2 CV%	30	Imputed variability consistent with measured CV% of placental transporters for SERT and OCT3 [9]

Model	Approach 1	Approach 2	Approach 3	Approach 4
	Perfusion‐limited	Permeability‐limited	Permeability‐limited	Permeability‐limited
Placental transfer parameters
CLPDM (L/h/mL placental volume)	—	0.000692 (calculated from ex vivo apparent cotyledon clearance)	0.0023 (predicted from 3‐compartment model)	0.0003 (calculated from PAMPA, see Equation 6; File S1) [34, 40, 41]
CLPDF (L/h/mL placental volume)	—	0.000692 (calculated from ex vivo apparent cotyledon clearance)	0.039 (Predicted from 3‐compartment model)	0.00019 (calculated from PAMPA, see Equation 6; File S1) [34, 40, 41]
PMAT CLint,T (μL/min/million cells)	—	—	—	283.6 (RAF/REF = 1, see Equation 7; File S1) [40, 41, 42, 43]
SERT CLint,T (μL/min/million cells)	—	—	—	7.99 (RAF/REF = 1, see Equation 7; File S1) [40, 41, 44, 45]
THTR2 CLint,T (μL/min/million cells)	—	—	—	21.8 (RAF/REF = 1, see Equation 7; File S1) [8, 40, 41, 44]
MATE1 CLint,T (μL/min/million cells)	—	—	—	28.1 (RAF/REF = 1, see Equation 7; File S1) [40, 41, 44, 46]
OCT3 CLint,T (μL/min/million cells)	—	—	—	51.4 (RAF/REF = 1, see Equation 7; File S1) [40, 41, 44, 47]

Abbreviations: CL_int, intrinsic clearance; CL_int,T, in vitro intrinsic transporter‐mediated clearance; CL_PD, passive diffusion clearance; CL_PDF, placental‐to‐fetal passive clearance; CL_PDM, maternal‐to‐placental passive clearance; CV, coefficient of variation; CYP, cytochrome P450; EGD, electrochemical gradient driven; fu_mic, fraction unbound in in vitro microsomal incubation; MATE, multidrug and toxin extrusion protein; OCT, organic cation transporter; PAMPA, parallel artificial permeability assay; PBPK, physiologically based pharmacokinetic; PMAT, plasma membrane monoamine transporter; RAF/REF, relative activity factor/relative expression factor; SERT, serotonin transporter; THTR, the human thiamine transporter; V_ss, volume of distribution at steady state.

2.3. PBPK Modeling in Pregnant Patients With GDM Incorporating Placental Transfer

Four different approaches to modeling placental transfer were evaluated to predict MPCs and UCCs of metformin at term. Approach 1 used the default perfusion‐limited model within the Simcyp Simulator, which assumes rapid equilibrium between the maternal circulation and placental tissue. In this approach, the entire placenta–fetal compartment was represented as a single, well‐stirred compartment. Therefore, the predicted placenta–fetal concentration was equal to the venous UCC.

Approaches 2 through 4 used a permeability‐limited approach where the placental transfer from the maternal plasma to the fetus was described by a permeability‐limited model where the placenta was described by three compartments (placental maternal blood, placental tissue, and placental fetal blood) as previously described [24]. Approach 2 used the apparent cotyledon clearance from an ex vivo cotyledon open system experiment [11]. The clearance (L/h) at the maternal‐placenta and fetal‐placenta barrier for the entire placenta was calculated and then converted into a clearance in L/h per mL placental volume based on the following equation:

$$\begin{array}{c}{\mathrm{CL}}_{\mathrm{PD},\text{placenta}} \mathrm{per} \mathrm{placental} \mathrm{volume} \left(\mathrm{L}/\mathrm{h}/\mathrm{mL} placental volume\right)\hfill \\ =\frac{{\mathrm{CL}}_{\mathrm{PD},\cot }\left(\mathrm{mL}/\min \right)\times \text{number cotyledons}\mathrm{per}\text{placenta}}{\text{placental volume}\left(\mathrm{mL}\right)}\hfill \end{array}$$ (1)

assuming 20 cotyledons per placenta and a placental volume of 659 mL [40]. CL_PD,cot represents the clearance (mL/min) at the maternal‐placenta and fetal‐placenta barrier for a single cotyledon A value of 0.000692 L/h/mL placental volume for both maternal‐to‐placental passive clearance (CL_PDM) and placental‐to‐fetal passive clearance (CL_PDF) were used for the simulation.

Approach 3 determined CL_PDM and CL_PDF by fitting a three‐compartment model using Phoenix 64 WinNonlin (version 8.3.4.295, Certara USA, Princeton, NJ) to metformin concentration data from an ex vivo cotyledon closed system experiment (Table S1) [11]. The three compartments represented the maternal circulation, syncytiotrophoblast, and fetal circulation as described by the following equations:

$$C_m=\frac{A_m}{V_m};C_s=\frac{A_s}{V_s};C_f=\frac{A_f}{V_f}$$ (2)

$$\frac{d{A}_m}{\mathit{dt}}={\mathrm{CL}}_{\mathrm{PDM},\cot }\times \left(C_s-C_m\right);\text{Initial condition}=0$$ (3)

$$\begin{array}{c}\frac{d{A}_s}{\mathit{dt}}={\mathrm{CL}}_{\mathrm{PDM},\cot }\times \left(C_m-C_s\right)-{\mathrm{CL}}_{\mathrm{PDF},\cot }\times \left(C_s-C_f\right);\hfill \\ \text{Initial condition}=0\hfill \end{array}$$ (4)

$$\frac{d{A}_f}{\mathit{dt}}=-{\mathrm{CL}}_{\mathrm{PDF},\cot }\times \left(C_f-C_s\right);\text{Initial condition}=0$$ (5)

where concentrations for the maternal (C _m), fetal (C _f), and syncytiotrophoblast (C _s) compartments were defined as the amount of metformin in the respective compartment (A _m, A _f, and A _s) divided by the respective compartment volume (V _m, V _f, and V _s). The maternal and fetal compartment volumes were set as the reservoir volumes of 250 mL and 150 mL, respectively [48]. The syncytiotrophoblast volume was calculated as 3.395 mL (67.9 mL placental volume/20 cotyledon per placenta) [41]. A 1250 μg intravenous bolus dose was used for the simulation, based on the reported final concentration (5 μg/mL) added to the maternal reservoir (250 mL). CL_PDM and CL_PDF initial estimates were 0.37 and 0.47 mL/min, respectively, based on a previous metformin placental perfusion experiment [12]. The model assumed no loss of metformin in the system, and any metformin not accounted for in the maternal or fetal reservoirs at a given timepoint was assumed to be in the syncytiotrophoblast. The CL_PDM and CL_PDF final parameter estimates from the model were converted into a clearance in L/h/mL placental volume for use in the Simcyp model simulation as described for Approach 2, with a resulting clearance value of 0.0023 L/h/mL and 0.039 L/h/ml placental volume estimated for CL_PDM and CL_PDF, respectively.

Approach 4 used CL_PD values calculated using the parallel artificial permeability assay (PAMPA) for both CL_PDM and CL_PDF with the following equation:

$$\begin{array}{c}{\mathrm{CL}}_{\mathrm{PD}}\left(\mathrm{L}/\mathrm{h}/\mathrm{mL}\text{placental} volume\right)=\text{PAMPA permeability value}\left(\mathrm{cm}/\mathrm{s}\right)\hfill \\ \times \text{surface area}\left({\mathrm{cm}}^2\right)/\text{placental volume}\left(\mathrm{mL}\right)\hfill \end{array}$$ (6)

where the PAMPA permeability value was 0.5 × 10⁻⁶ cm/s [34], the surface area was 110,000 cm² for the placental maternal side surface (CL_PDM) and 69,400 cm² for the placental fetal side surface (CL_PDF) [41], and 659 mL for the placental volume [40]. In addition to the CL_PD, active uptake and efflux by placental transporters was incorporated in Approach 4. OCT3 was included as a placental uptake transporter from the fetal side to the placenta. SERT, PMAT, and THTR2 were included as placental uptake transporters and MATE1 as an efflux transporter on the maternal side as shown in Figure 1. In vitro intrinsic transporter clearance values (CL_int,T) were based on previously published data determined in Human Embryonic Kidney (HEK293) cells stably expressing hOCT3 [47], hMATE1 [46], hSERT [45], and hTHTR2 [8], and Madin‐Darby Canine Kidney (MDCKII) cells stably expressing hPMAT [42]. Values were scaled to units of μL/min/mL placental volume using the following equation:

$$\begin{array}{c}{\mathrm{CL}}_{\mathrm{int},\mathrm{T}}\left(\mathrm{\mu }\mathrm{L}/\min /\mathrm{mL}\text{placental volume}\right)\left({\mathrm{cm}}^2\right)\hfill \\ ={\mathrm{CL}}_{\mathrm{int},\mathrm{T}}\left(\mathrm{\mu }\mathrm{L}/\min /\mathrm{mg}\text{protein}\right)\times \text{in vitro}\frac{\mathrm{mg}\text{protein}}{\text{surface area}\left({\mathrm{cm}}^2\right)}\hfill \\ \times \text{placental}\frac{\text{surface area}\left({\mathrm{cm}}^2\right)}{\text{volume}\left(\mathrm{mL}\right)}\hfill \end{array}$$ (7)

assuming the following based on previously published data: 0.0684 mg protein/cm² for HEK293 cells [44], 0.133 mg protein/cm² for MDCKII cells [43], villous surface of 110,000 cm² placental maternal side surface (SERT, PMAT, THTR2, and MATE1), capillary surface of 69,400 cm² placental fetal side surface (OCT3) [41], and 659 mL placental volume [40]. Assuming linear kinetics, intrinsic transporter clearance values for SERT, PMAT, and THTR2 were summed and represented as a single maternal uptake transporter in the model (File S1). In addition, to reflect known population‐transporter level variability in the placenta tissue [9], a CV% of 30% (Table 1) was applied to the relative expression value of each placental transporter within the pregnancy population file.

2.4. Evaluation of PBPK Models

Metformin MPC and UCC versus time model predictions (mean and 5th to 95th percentiles) for each placental transfer approach in the virtual pregnancy population were compared to MPC and UCC versus time clinical data to determine the optimal model. Selection criteria for the model that best described the data were based on visual comparison of the predicted versus observed MPC‐ and UCC‐time profiles and evaluation of the number of observed data points within the simulated 5th to 95th percentiles. The ideal goodness of fit is 90% of individual data points within the 90th prediction interval (5% above and 5% below).

2.5. Model Application

Exploratory simulations were performed to investigate how potential alterations in placental transporters due to gestational age and/or disease may affect metformin fetal exposure. Virtual trial simulations were performed (13 subjects × 10 trials) where the intrinsic clearance values for placental uptake transporters (SERT, PMAT, and THTR2) from the maternal side, placental efflux transporters (MATE1) into the maternal compartment, and the placental uptake transporter (OCT3) from the fetal side were individually adjusted by two‐fold (File S1). Subsequent changes in population mean UCC‐ and FM ratio versus time profiles from the Approach 4 baseline model were evaluated.

3. Results

3.1. PBPK Model Development and Verification

Model development was performed using the metformin plasma concentration data from individual T2DM non‐pregnant and GDM pregnant patients [15]. The modified metformin PBPK model adequately described observed steady‐state plasma concentration‐time data from non‐pregnant and pregnant patients based on visual inspection and comparison of predicted versus observed PK parameters (Figure 2, Table S2).

FIGURE 2.

PBPK model predicted steady‐state concentration versus time profiles for (a) non‐pregnant patients with type 2 diabetes mellitus (T2DM) and (b) pregnant patients with gestational diabetes mellitus (GDM) taking metformin 500 mg every 12 h. Simulated mean (line) and 5th to 95th percentiles (shaded area) are shown. Observed data (open circles [15], open squares [18]) are plotted as individual data points. Simulation results are from (a) 7 subjects × 10 trials and (b) 38 subjects × 10 trials.

3.2. Incorporation of Placental Transfer Approaches in PBPK Model

All four placental transfer modeling approaches accurately predicted MPCs, with 92%–100% of MPCs falling within the 5th to 95th percentiles of the model simulation (Figure 3, Table 2). Notably, MPC predictions were similar for all four placental transfer modeling approaches, indicating that the choice of placental transfer model did not have a major influence on the volume of distribution and subsequent MPC predictions. Using the perfusion‐limited placental transfer model (Approach 1), only 31% of UCCs fell within the 5th to 95th percentiles of the model simulation (Figure 3, Table 2). Additionally, the FM ratio versus time profile was not adequately described. Placental transfer modeling Approaches 2 and 3 both used a permeability‐limited approach and CL_PDM and CL_PDF values derived from ex vivo cotyledon data. These approaches performed better than Approach 1, with 54% of UCCs falling within the 5th to 95th percentiles of the model simulation (Table 2). Approach 3, which incorporated the higher passive diffusion clearance of the two approaches, provided a slightly better prediction of the UCC profile shape. However, the FM ratio versus time profile was not well described. Placental transfer modeling Approach 4, the permeability‐limited model incorporating placental transporters and PAMPA passive clearance, performed best with 92% of UCC observations falling within the 5th to 95th percentiles of the model simulation. Additionally, the predicted FM ratio versus time profile using Approach 4 best described the observed clinical data (Figure 3). Among all of the modeling approaches, Approach 3 predicted the highest UCC C _max and AUC_0‐12,ss of 0.64 mg/L and 7.20 mg h/L, respectively (Table 2).

FIGURE 3.

Simulated steady‐state maternal plasma concentration versus time, umbilical cord concentration versus time, and fetal to maternal (FM) ratio versus time profiles using a PBPK model incorporating placental transfer based on (a) Approach 1 (perfusion‐limited model), (b) Approach 2 (permeability‐limited model using ex vivo open system apparent cotyledon clearance), (c) Approach 3 (permeability‐limited model using clearance estimated from a three‐compartment model of ex vivo closed system cotyledon data), and (d) Approach 4 (permeability‐limited model using in vitro placental transporter data and PAMPA passive clearance). Simulated mean (line), 5th to 95th percentiles (shaded area), and individual observed data points (closed circles [15]) are shown.

TABLE 2.

Comparison between predicted population mean pharmacokinetic parameters at steady state using a PBPK model incorporating four different approaches for metformin placental transfer.

Approach to model metformin placental transfer	Approach 1	Approach 2	Approach 3	Approach 4
	Perfusion‐limited model	Permeability‐limited model with clearance estimated from
		Ex vivo open system apparent cotyledon clearance	3‐compartment model of ex vivo closed system cotyledon data	In vitro transporter data + PAMPA passive clearance
Maternal parameters
Observations within the 5th to 95th percentile (%)	92 (11/12)	100 (12/12)	100 (12/12)	100 (12/12)
C max (mg/L)	0.84	0.88	0.87	0.87
T max (h)	2.36	2.28	2.25	2.28
AUC0‐12,ss (mg h/L)	6.92	7.28	7.29	7.30
Fetal parameters
Observations within the 5th to 95th percentile (%)	31 (4/13)	54 (7/13)	54 (7/13)	92 (12/13)
C max (mg/L)	0.64	0.37	0.64	0.42
T max (h)	2.49	7.29	4.17	4.19
AUC0‐12,ss (mg h/L)	5.28	4.41	7.20	4.63

Abbreviations: AUC_0‐12,ss, area under the curve from 0 to 12 h after dosing at steady state; C _max, maximum concentration; T _max, time of maximum concentration.

3.3. Model Application

A sensitivity analysis was performed with the PBPK model incorporating placental transfer modeling Approach 4 to evaluate the impact of placental transporter alterations on UCC and FM ratio versus time profile predictions. Among the transporters analyzed, SERT, PMAT, and THTR2 (maternal uptake) exhibited the most significant influence on UCC and FM ratio versus time profiles. A two‐fold increase in SERT, PMAT, and THTR2 intrinsic transporter clearance increased the predicted population mean UCC C _max by 97%, whereas a 0.5‐fold decrease resulted in a 49% decrease in UCC C _max (Figure 4). Increasing OCT3 intrinsic transporter clearance by two‐fold led to a 45% reduction in predicted UCC C _max, while a 0.5‐fold decrease caused a 74% increase in UCC C _max. Changes in MATE1 intrinsic transporter clearance also significantly affected UCC predictions, with a two‐fold increase reducing UCC C _max by 45% and a 0.5‐fold decrease increasing it by 70%. As expected, MPC predictions remained relatively unchanged across varying intrinsic transporter clearances (data not shown); therefore, the observed changes in FM ratio versus time profiles were primarily driven by changes in the UCC versus time profiles.

FIGURE 4.

Sensitivity analysis using placental transfer modeling Approach 4 to investigate how potential alterations in placental transporters due to gestational age and/or disease may affect metformin fetal exposure. Population mean steady‐state (a) umbilical cord concentration versus time and (b) fetal to maternal (FM) ratio versus time profiles were simulated in virtual pregnant patients taking metformin 500 mg every 12 h (13 subjects × 10 trials). The black line represents population mean model predictions from the Approach 4 baseline model, and colored/dashed lines represent population mean model predictions when intrinsic clearance values for placental transporters were individually adjusted by twofold (2× increase or decrease). MATE; multidrug and toxin extrusion protein; OCT, organic cation transporter; PMAT, plasma membrane monoamine transporter; SERT, serotonin transporter; THTR, the human thiamine transporter.

4. Discussion

The use of safe and effective anti‐diabetic medications during pregnancy is essential and has become a critical issue due to the growing prevalence of T2DM in patients of child‐bearing age and the increasing incidence of GDM. Understanding fetal exposure to anti‐diabetic medications such as metformin is vital to making appropriate treatment decisions for pregnant patients. In this study, a PBPK model for metformin was refined and verified using maternal data from T2DM non‐pregnant patients and GDM pregnant patients to predict fetal exposure in pregnant patients with GDM. Four different approaches to model placental transfer of metformin were evaluated: (1) a perfusion‐limited model, and permeability‐limited models using (2) ex vivo cotyledon open system apparent clearance, (3) ex vivo cotyledon closed system data fit to a three‐compartment model to estimate clearance, and (4) active transport kinetics and PAMPA passive clearance.

The perfusion‐limited model used in Approach 1 resulted in a static FM ratio that did not describe the observed data, which had a clear time‐dependent change in FM ratio. This was expected considering metformin has a low permeability and is most likely permeability‐limited at the maternal‐placental barrier. Furthermore, the ex vivo cotyledon open system experimental data used in Approach 2 was designed with conditions that mimic continuous removal following placental transfer, which may have skewed the estimation of clearance when compared to in vivo data and would not reflect bidirectional concentration‐gradient driven transport. The ex vivo cotyledon closed system data used in Approach 3 better described the transfer of metformin, most likely due to the recirculation of concentrations from both the maternal and fetal compartments. However, this approach underestimated the variability with only 54% of the observed UCCs falling within the 5th to 95th percentiles. This could be due to limitations of the ex vivo placental perfusion system, which does not account for maternal and fetal drug clearance or the distribution of compounds within the maternal and fetal compartments [49]. Due to lack of information on transport protein abundance in the tissue used for placental perfusions, individual uptake and efflux clearance values on both membranes were not included in this model.

Approach 4 best described the data with 100% of the observed MPCs and 92% of the observed UCCs within the simulated 5th to 95th percentiles. This approach incorporated passive clearance derived from PAMPA and five placental transporters using the relevant influx or efflux clearance values [OCT3, PMAT, THTR, SERT, MATE1] (see Table 1, File S1). Even though PMAT and MATE1 gene and protein expression in placental tissue have been reported to be low [6], we included these transporters in the model alongside the others given the availability of metformin kinetic data for these transporters. Additionally, while localization of PMAT and THTR2 in the placenta is unknown, they are apical uptake transporters in the intestine [50, 51], and thus, were incorporated as apical (maternal) uptake transporters in the placental transfer model. UCCs were accurately predicted when the following were included: PMAT, THTR2, and SERT as placental uptake transporters on the apical (maternal) side combined with MATE1 efflux, OCT3 as a placental uptake transporter on the fetal side, and passive clearance calculated using PAMPA. Although metformin is reported to be a substrate for other placental transporters (e.g., OCTN1, OCTN2), these specific transporters were not incorporated into the model due to a lack of available metformin kinetic data. Overall, predictions of metformin fetal exposure based on all four placental transfer modeling approaches indicated that fetal exposure remained well below metformin concentrations reported to be toxic (> 100 mg/kg in pediatric patients) [52].

A limitation of this model is that there is no validation for the model's ability to predict fetal exposure prior to term because there are currently no available data on metformin UCCs throughout pregnancy. Furthermore, the model assumes first‐order absorption, and since metformin displays nonlinear pharmacokinetics (e.g., saturation of intestinal transporters), model application is limited to a 500 mg dose and cannot be extrapolated to higher doses. In addition, the current framework does not include physiological changes associated with T2DM and/or GDM that may impact drug disposition such as reduced gastric mucosal blood flow and delayed gastric emptying [26, 27]. While a decreased absorption rate was included in the metformin PBPK model to account for these changes, future work should focus on including these physiological changes in the virtual pregnancy population.

In Approach 4, transporter clearance values were estimated using overexpressed cell systems, which may not be representative of in vivo clearance. Ideally, knowing protein abundance in the overexpressed cell system relative to placental tissue could inform a relative expression factor (REF) and enhance the accuracy of in vitro‐to‐in vivo extrapolation. However, these data are not currently available. As more information emerges on placental transport protein abundance [9] and localization, the PBPK model incorporating Approach 4 can be refined to include scaled clearance values and/or additional transporters. Furthermore, as more data become available on how placental transporter activity is influenced by gestational age and disease, this framework may be applied to predict metformin fetal exposure and potential toxicity throughout pregnancy and in patients with pathologic conditions.

In conclusion, modeling metformin placental transfer for this permeability‐limited drug was best described by Approach 4 where five placental transporters and PAMPA passive clearance were incorporated into the model. This method holds promise for extrapolating to earlier stages of pregnancy to predict fetal exposure as knowledge of transporter expression throughout pregnancy becomes available.

Author Contributions

T.L.W., J.B.T., K.L.R.B., and M.F.H. designed the research. T.L.W. and M.E.H. performed the research. M.E.H., J.B.T., T.L.W., M.D.S.M., K.L.R.B., and M.F.H. analyzed the data. M.E.H., J.B.T., and T.L.W. wrote the manuscript. K.L.R.B., M.F.H., M.E.H., J.B.T. revised the manuscript.

Conflicts of Interest

M.D.S.M. is a full‐time employee of Certara Predictive Technologies, UK (Simcyp Division). All other authors declared no competing interests for this work.

Supporting information

Table S1: psp470136‐sup‐0001‐TableS1‐S2.docx.

Table S2: psp470136‐sup‐0001‐TableS1‐S2.docx.

Data S1: psp470136‐sup‐0002‐DataS1.xlsx.

Tiley J. B., Hartauer M. E., Whigham T. L., Mendes M. D. S., Brouwer K. L. R., and Hebert M. F., “Comparison of Metformin PBPK Models Incorporating Placental Transfer to Predict Fetal and Maternal Exposure,” CPT: Pharmacometrics & Systems Pharmacology 15, no. 1 (2026): e70136, 10.1002/psp4.70136.