Informing the risk assessment related to lactation and drug exposure: A physiologically based pharmacokinetic lactation model for pregabalin

Abstract

Breastfeeding is important in childhood development, and medications are often necessary for lactating individuals, yet information on the potential risk of infant drug exposure through human milk is limited. Establishing a lactation modeling framework can advance our understanding of this topic and potentiate clinical decision making. We expanded the modeling framework previously developed for sotalol using pregabalin as a second prototypical probe compound with similar absorption, distribution, metabolism, and elimination (ADME) properties. Adult oral models were developed in PK‐Sim® and used to build a lactation model in MoBi® to simulate drug transfer into human milk. The adult model was applied to breastfeeding pediatrics (ages 1 to 23 months) and subsequently integrated with the lactation model to simulate infant drug exposure according to age, size, and breastfeeding frequency. Physiologically based pharmacokinetic (PBPK) model simulations captured the data used for verification both in adults and pediatrics. Lactation simulations captured observed milk and plasma data corresponding to doses of 150 mg administered twice daily to lactating individuals, and estimated a relative infant dose (RID) of approximately 7% of the maternal dose. The infant drug exposure simulations showed peak plasma concentrations of 0.44 μg/mL occurring within the first 2 weeks of life, followed by gradual decline with age after week four. The modeling framework performs well for this second prototypical drug and warrants expansion to other drugs for further validation. PBPK modeling and simulation approaches together with clinical lactation data could ultimately help inform infant drug exposure risk assessments to guide clinical decision making.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Adequate data to inform the “Risk Summary” of the Lactation subsection in the FDA drug labeling is lacking. PBPK modeling approaches integrating drug physicochemical properties and pediatric, lactation, and milk intake data is a viable method for predicting infant drug exposure.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Can an integrated, systematic, stepwise PBPK modeling platform previously introduced be subsequently employed to predict infant drug exposure for a second renally eliminated drug, pregabalin?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

Integrating known drug physicochemical properties with available lactation data into an overarching PBPK modeling framework to predict drug penetration into human milk and subsequent infant exposure was applied to assess the risk of drug exposure in breastfeeding infants for a second renally eliminated drug.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

This study highlights the importance of obtaining pediatric and lactation data and how its integration into an integrated PBPK modeling platform can foster a better understanding of infant drug exposure through lactation.

INTRODUCTION

The short‐ and long‐term benefits of breastfeeding to both mothers and infants are well documented. ¹ , ² In the United States, 83% of mothers are reported to have ever breastfed, with approximately 56% and 36% still breastfeeding at 6 and 12 months, respectively. ³ Several regions in Africa, Asia, and Eastern Europe have even greater breastfeeding rates, thus emphasizing the global importance of this topic. ⁴ Data from the 2015–2018 CDC Therapeutic Drug Use survey suggests that approximately 44% of women ages 18–44 years (i.e., the childbearing age range for adult females) took at least one prescription drug within 30 days prior to survey participation. ⁵ , ⁶ However, there is a considerable lack of data pertaining to medication use in lactating individuals and infant drug exposure via human milk.

Following the implementation of the Pregnancy and Lactation Labeling Rule (PLLR) by the United States Food and Drug Administration (U.S. FDA) in 2015, a review of 290 drug products approved between January 2010 and December 2019 found that only about half of the drug labels contained any lactation data. Most of these data were derived from animal studies as only 8 (5.4%) of those labels contained data from human studies. ⁷ , ⁸ In addition, to the paucity of data related to drug concentrations in human milk, pharmacokinetic (PK) and safety data are rarely available for infants of typical breastfeeding age (i.e., <2 years). Currently, the relative infant dose (RID)—the dose of drug an infant receives through breastfeeding as a percentage of the dose taken by the lactating individual—is a commonly adopted metric for estimating infant exposure from lactation data. ⁸ , ⁹ Thus, it is used to assess the risk of infant drug exposure via breastfeeding. The common RID risk categories reported in the literature are as follows: RID of <10% is considered safe, 10%–25% should be used with caution, and >25% is considered unsafe and/or should be avoided. ⁸ , ⁹ , ¹⁰ Translating these RID values to be clinically actionable becomes challenging when there is limited lactation data to do so, particularly due to the variability in ontogeny in absorption, distribution, metabolism, and elimination (ADME) pathways. ⁸

To overcome these challenges, the use of modeling and simulation, particularly physiologically based pharmacokinetic (PBPK) modeling and simulation, has been gaining popularity for this special patient population. This is due to the fact that PBPK modeling allows for the integration of drug‐specific and biological system‐specific information into an overarching model which, once developed and verified, can be used to simulate drug exposure in breastfeeding infants. Respective information can be integrated one drug at a time or in a more systematic fashion using drugs with similar ADME patterns to map out key processes relevant to maternal and infant drug exposure and, thus, for the establishment of a PBPK‐based lactation framework, which can be used for risk assessment. Similar model platform development efforts are underway from other groups, such as the ConcePTION Project. ¹¹ , ¹² To date, such a platform does not exist, as outlined by Dubbelboer et al. ¹³ in a systematic review of 52 published PBPK lactation models, which included a gap analysis discussing the dearth of generic models with translational applicability.

Therefore, our group previously set out to establish a PBPK lactation modeling framework by mapping out processes relevant to lactation. In a first publication, Pressly et al. conducted a systematic data review and identified sotalol as a first prototypical drug given that it: (1) has a milk‐to‐plasma ratio >1.9, (2) exerts linear PK, (3) is primarily renally cleared (>90%) with negligible hepatic metabolism, (4) shows high solubility and permeability (Biopharmaceutical Classification System [BCS] class 1), and (5) has negligible protein binding. ¹⁴ , ¹⁵ The goal of our current study is to expand this framework with a second prototypical drug with very similar ADME properties (i.e., pregabalin) in order to gain additional confidence in its general applicability. ¹⁴ , ¹⁵

METHODS

Pharmacokinetic and lactation data

PBPK modeling was performed following a stepwise workflow as outlined in Figure 1. This workflow is similar to the one employed by Pressly et al. ¹⁴ using PK‐Sim® and MoBi®, with the exception that pregabalin lacks intravenous (IV) administration data. All data acquired from literature sources were digitized using WebPlotDigitizer (version 4.6). Data management and calculations were performed in RStudio® (version 2022.12.0, build 353).

FIGURE 1.

Physiologically based pharmacokinetic (PBPK) modeling workflow (a) and stepwise process for randomly splitting internal adult data (b).

As a first step (Figure 1a), a preliminary adult oral (PO) model was developed using population‐level PK data from two literature sources. ¹⁶ , ¹⁷ This model was then further refined using individual‐level data covering a wide range of doses (Table 1) that were previously submitted to the U.S. FDA. Subjects with reduced renal function, defined as creatinine clearance (CL_Cr) < 60 mL/min, were first removed from the adult dataset, leaving 712 individuals in the database. Of these 712 adults, 284 females were of childbearing age and were used for the adult and childbearing‐age model development. One‐third of these females (n = 95) were randomly extracted and set aside for childbearing‐age female PBPK model verification, leaving 617 total adults for the adult dataset. These data were split randomly into a build (n = 205) and two equally‐sized refine (n = 206) and verify (n = 206) datasets, each containing approximately 95 childbearing‐age females (Figure 1b). Additionally, in vitro capsule dissolution profiles were digitized and used for initial estimates and optimization of the expected dissolution behavior. ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² All of the dosing regimens used for modeling were acquired from the corresponding literature and internal data, and are outlined in Table 1. ¹⁶ , ¹⁷ , ²³

TABLE 1.

Summary of dosing regimens used for adult (including childbearing‐age females), pediatric, and lactation PBPK models. ¹⁹ , ²⁰ , ²⁶ , ²⁷

PBPK model	Dosing regimens	Meal status	Frequency
All adult models a	1, 2, 5, 10, 25, 50, 75, 100, 125, 200, 300 mg	Fasted, fed	Single‐dose
	100 mg	Fed	Single‐dose
	25, 100, 200, 300 mg	Fasted	TID
	300 mg	Fasted, fed	BID
Lactation	150 mg	Fasted	BID
Pediatric	1.25 mg/kg/dose	Fasted, fed	Single‐dose, BID
	2.5 mg/kg/dose	Fasted, fed
	5 mg/kg/dose	Fasted, fed
	7.5 mg/kg/dose	Fasted, fed

Abbreviations: BID, twice daily (every 12 h); CR, controlled‐release; IR, immediate‐release; TID, three times daily (every 8 h).

^a Refers to the Adult and Childbearing‐age Female PBPK models.

To our knowledge, the study by Lockwood et al. ²³ is the only source of PK data for pregabalin in human milk. The clinical study report, available through the US FDA contained relevant demographic and PK data for plasma, human milk, and urine (n = 10), and was used in the next steps of the workflow, that is, for lactation PBPK model development. The study participants received a total of four 150‐mg PO doses given every 12 h to achieve steady‐state concentrations without any dietary restrictions.

Median concentration‐time profiles for 12 pediatric subjects ages 1–23 months (n = 3 for each dosing group) were extracted from the study by Mann et al. ²⁴ The model was further refined using individual‐level data from FDA, which were subsequently used for pediatric model refinement and were grouped by ages 3–6, >6–12, >12–18, and >18–23 months (a description of the model refinement process is provided in the Supplemental Information). The original dataset was also split by dosing regimen, regardless of age. All drug‐specific physicochemical properties and initial input parameters for PBPK modeling are outlined in Table 2. ²⁵ , ²⁶ , ²⁷ , ²⁸

TABLE 2.

Drug‐specific parameters used to establish PBPK models of pregabalin.

	Units	Value in PBPK model	Published value[ref]	Description
PBPK model parameter
MW	g/mol	159.23	159.23 28 , 29	Molecular Weight of Pregabalin
Solubility	mg/mL	32.1	32.1 31	Solubility at pH 7.4
pKa (acid)	–	4.2	4.2 28 , 29	−log10 of acid dissociation constant for weak acid
pKa (base)	–	10.6	10.6 28 , 29	−log10 of acid dissociation constant of weak base
Lipophilicity a	Log units	−1.35	−1.35 28 , 29	LogP
f u	–	1.00	1.00 28 , 29	Fraction not bound to plasma proteins
fGFR				Fraction of GFR describing renal clearance
Adult and lactation models b	–	0.77	1.00 19 , 30 , 32
Pediatric model	–	0.94	1.07 27 , 30
Tissue K ISF/plasma				Tissue K p between interstitial fluid and plasma
Adult and lactation models b	–	0.48	0.96 [calculated by PK‐Sim®]
Pediatric model	–	0.78	0.96 [calculated by PK‐Sim®]
P intestinal				Diffusion speeds for intestinal absorption
Adult and lactation models b	cm/min	1.59 × 10−5	8.79 × 10−8 [calculated by PK‐Sim®]
Pediatric		4.77 × 10−6	1.59 × 10−5 [Adult model parameters]
Dissolution (Weibull) parameters
α	min	1.11	1.67 21 , 22 , 23	Dissolution time (until 50% drug released)
β	–	0.55	0.70 21 , 22 , 23	Dissolution shape
T Lag				Lag time before dissolution begins
Adult models c (Fasted)	min	5.8	14.58 33 , 34
Adult models c (Fed)	min	23.2	48.72 33 , 34
Lactation model (Fasted) d	min	26.3	48.72 33 , 34
Lactation model (Fed) e	min	13.2	5.83 [Adult model parameters]
Parameter lactation model
f u,milk	–	1.00	1.00 28 , 29 , 35	Fraction unbound in human milk
K ISF/plasma				Tissue K p between:
Breast tissue	–	0.48	0.48 [Adult model parameters]	Breast tissue ISF and plasma
All other tissues	–	0.48	0.48 [Adult model parameters]	ISF and plasma of all other tissues
K milk/plasma	–	0.76	0.76 26 , 35	Milk‐to‐plasma partition coefficient
V milk	mL/24 hrs	238.9	238.9 (110–575) 26 , 35	Milk volume produced per 24‐h period
Permeabilities				Diffusion speeds for:
P plasma/milk	cm/min	108.8	100	Plasma to milk
P ISF/cell (breast tissue)	cm/min	1.71 × 10−3	7.88 × 10−5 [calculated by PK‐Sim®]	Interstitial‐to‐intracellular fluid of breast tissue
Q breast	mL/min/100 g tissue	1.93	1.79 10 , 16	Volume of blood supplied per minute per 100 g of breast tissue

Abbreviations: f _u, fraction unbound; fGFR, fractional glomerular filtration rate; GFR, glomerular filtration rate; K _p, tissue‐to‐plasma partition coefficient; K _ISF/plasma, interstitial fluid (ISF)‐to‐plasma K _p; P _intestinal, intestinal permeability; T _Lag, lag time; K _milk/plasma, milk‐to‐plasma K _p; V _milk, volume of milk produced per day; P _plasma/milk, plasma‐to‐milk permeability; Q _breast, blood flow rate through breast tissues; ref., reference.

^a Lipophilicity in PK‐Sim® is defined as the log₁₀ of membrane permeability (logMA).

^b Refers to the adult, childbearing‐age female, and lactation models.

^c Refers to the adult and childbearing‐age female models.

^d Lag time in the lactation model used to account for food effects (presumed fed state).

^e Lag time in the lactation model when meals are incorporated at each dosing time.

Adult and childbearing‐age female PBPK model development and evaluation

Given that pregabalin exists as a zwitterion at physiological pH, the Schmitt method, which considers electrostatic interactions between charged molecules at physiological pH and acidic phospholipids was used to estimate tissue partition coefficients (K _p) and cellular permeabilities. This choice was confirmed with the calculation methods variation option in PK‐Sim® (additional information can be found in the Supplementary Methods, the publication by Schmitt ²⁹ and at https://docs.open‐systems‐pharmacology.org/working‐with‐pk‐sim/pk‐sim‐documentation/pk‐sim‐compounds‐definition‐and‐work‐flow#adme‐properties).

Renal clearance (CL_renal) is the primary elimination pathway for pregabalin with a respective value of 67–80.9 mL/min. This suggests that there is transporter‐mediated tubular reabsorption since plasma protein binding is negligible (fraction unbound, f _u, ~1). ²⁵ , ²⁶ , ²⁷ Fractional glomerular filtration rate (fGFR) was used instead of GFR with drug‐specific clearance to parameterize renal clearance because specific reabsorption mechanisms are unknown. fGFR was set to a minimum of 50% (i.e., CL_renal equal to half of the GFR) based on available literature data and then further optimized. ¹⁶ , ²⁷ , ³⁰

Similarly, pregabalin permeability is expected to be low, as demonstrated in Caco‐2 cell experiments. However, oral absorption is thought to be facilitated by uptake transporters in the gut given pregabalin's high oral bioavailability (≥90%) and short time needed to reach peak plasma concentrations (T _max: 0.7–1.5 h). ¹⁶ , ²⁷ , ³¹ , ³² , ³³ L‐amino acid transporter 1 (LAT1) is highly expressed in the gastrointestinal (GI) tract, mammary epithelial cells, and other low‐permeability barriers, and has been identified as a likely uptake transporter for pregabalin owing to its similarity to zwitterionic amino acids, with respective pK_a values of 4.2 (carboxylic acid) and 10.6 (primary amine). ¹² , ²⁵ , ²⁶ , ³⁴ , ³⁵ , ³⁶ Intestinal permeability (P _intestinal) was selected as a surrogate to appropriately characterize pregabalin exposure data following oral absorption due to the unavailability of specific transporter data. A standard absorption model was used since PK‐Sim® does not utilize a mechanistic model like the advanced compartmental and transit (ACAT) or advanced dissolution, absorption, and metabolism (ADAM) models used in GastroPlus® and Simcyp™, respectively.

A preliminary adult PO PBPK model was built based on mean concentration‐time profiles from Bockbrader et al. that contained oral solution data ranging from 1 to 10 mg. ¹⁶ , ¹⁷ Based on a sensitivity analysis, P _intestinal, interstitial‐to‐plasma partition coefficients for all tissues (K _ISF/plasma), and fGFR were optimized to more accurately reflect the reported T _max, apparent volume of distribution (V _d/F) of 0.5 L/kg, and apparent oral clearance (CL/F) of 57.6–82 mL/min. ¹⁶ , ²⁵ , ²⁶ , ²⁷ The adult PO PBPK model was then further refined using the individual‐level in‐house FDA data that were previously split randomly into respective datasets. The parameter values from the “build” step were used as initial inputs for model refinement and then optimized using the “refine” datasets. Once developed and verified for oral solutions, the model was expanded to capsules using a Weibull function fit to in vitro dissolution data and PK data following the administration of single oral doses of 100‐mg capsules. Samples for the 100‐mg dose were collected more frequently compared to the other doses and were, therefore, used for the initial model fitting of the capsule data. ¹⁹ , ²⁰ The delay in oral absorption was characterized by a lag time (T _Lag), which was originally taken from the literature and then further optimized following single and multiple capsule administration, and potential confounding with P _intestinal, K _ISF/plasma, fGFR, and in vitro dissolution was evaluated. ³⁷ , ³⁸ No parameter differences were expected between the adult and childbearing‐age female models, so the same workflow was followed for the latter simply to verify this assumption. Once the fasted‐state model was established for single‐dose administration, data for 100‐mg single doses under fed conditions were also modeled. ¹⁶ , ¹⁷ , ³⁹

Lactation PBPK model development

Once developed and verified, the childbearing‐age female PO PBPK model was expanded to a lactation model by adding a breast compartment following the default setting provided in MoBi® based on a publication by Job et al. ⁴⁰ This compartment consists of blood cells, plasma, and interstitial as well as intracellular spaces similar to other tissues. In addition, there is a milk container, which is separated from the surrounding tissue by a barrier permeable to plasma (Figure 2). The milk container represents the alveoli clusters that are lined by myoepithelial cells which contain the milk‐producing lactocytes. Changes in the amount of pregabalin in milk over time (dA _milk/dt), which is converted to respective concentrations (dC _milk/dt) when divided by the milk volume, were characterized by passive diffusion between plasma and milk subcompartments using Equation 1 adapted from Job et al.,

$$\frac{{\mathrm{d}A}_{\text{milk}}}{\mathrm{d}t}=P_{\text{plasma}/\text{milk}}\times {\mathrm{SA}}_{\text{plasma}/\text{milk}}\times f_u\left(C_{\text{plasma}}-\frac{C_{\text{milk}}}{K_{\text{milk}/\text{plasma}}}\right)$$ (1)

where P _plasma/milk represents the plasma‐to‐milk permeability, SA_plasma/milk the surface area of the barrier separating milk from plasma, C _plasma the pregabalin concentration in plasma, C _milk the concentrations in human milk, K _milk/plasma the milk‐to‐plasma partition coefficient, and f _u the fraction unbound in plasma. ¹³ , ⁴⁰ Drugs may also transfer to milk from interstitial fluid (ISF); however, this pathway was assumed to be negligible since transport‐mediated processes are believed to be chiefly responsible. ³¹ , ³⁴ , ³⁶

FIGURE 2.

Lactation PBPK model schematic and basic structure of the breast tissue and milk compartment (The lactation PBPK model structure as well as the structure of the breast tissue and milk compartments were informed by Job et al. ⁴⁰ ). The basic PBPK model structure illustrates oral drug administration into the stomach with absorption occurring in the small intestine, followed by distribution into venous and arterial blood. Pregabalin then distributes into breast tissue and subsequently into human milk, in addition to other tissues including its primary site of pharmacological action in the brain. The net flux from plasma into human milk is illustrated by the larger half arrow (red) in the breast tissue compartment (center of figure). Transfer between interstitial fluid (ISF) and milk is also illustrated, but with thin black arrows as this pathway is thought to be of minor significance. On the far right is a simplified depiction of the actual anatomical structure of the barrier separating plasma from human milk that resides in the lumen of the alveoli. The lumen of the alveoli is lined with milk‐producing lactocytes on the apical side and is layered with myoepithelial cells along with a basement membrane through which the drug occurs. The equations listed are those incorporated into the MoBi® template by Job et al. ⁴⁰ to calculate the number of lobules per lobe in the breast, the volume of milk produced per day (V _milk) as a function of baby weight (W _baby), baby weight according to sex (W _baby,girl, W _baby,boy) and age (t), and the milk‐to‐plasma partition coefficient (K _milk/plasma) set equal to the milk‐to‐plasma AUC ratio. The equation for V _milk was not used in the lactation PBPK model discussed here because observed values were input into the model instead; therefore, W _baby also had no impact on the model. K _milk/plasma, milk‐to‐plasma partition coefficient; K _{milku/plasmau}, milk‐to‐plasma partition coefficient for unbound drug; f _u, fraction unbound in plasma; f _u,milk, fraction unbound in milk; AUC_τ,milk, area under the milk concentration‐time curve for dosing interval (τ); AUC_τ, area under the plasma concentration‐time curve for dosing interval (τ); θ ₁, θ ₂, θ ₃, fitted parameters (160.39, 0.232, and 0.00252, respectively) ⁴⁶ ; t, time after delivery (days); W _baby, calculated weight of baby (kg); V _milk, daily milk volume (mL/day); V _exp, fraction of V _milk expressed per feeding.

The breast compartment is connected to the remaining PBPK model via venous and arterial blood flow, which can be divided by the blood‐to‐plasma ratio of 0.86 to determine plasma flow. The starting value for Q _breast as well as variability therein was derived from literature and the Annals of the International Commission on Radiology Protection (ICRP), Reference Values. ¹³ , ⁴¹ The f _u,milk was set to 1 due to negligible protein binding. Additionally, pregabalin remains a zwitterion at the pH of human milk (~6.5), which decreases its fraction ionized by only ~0.44%. It also has a relatively low inherent lipid permeability, thereby preventing partitioning into milk lipids, which comprise only ~4% of the total milk contents. ⁴¹ K _milk/plasma was derived from the milk‐to‐plasma area under the concentration‐time curve ratio (MPAUC). V _milk was set to 238.9 mL based on data by Lockwood et al., and the breast tissue volume was fixed at the observed mean value of 1.07 L per Job et al., as milk production is independent of breast tissue volume. ²³ , ⁴⁰ , ⁴² , ⁴³

A sensitivity analysis (Figure S3) was conducted to evaluate the robustness of the respective breast model parameters as outlined by Job et al. ⁴⁰ Based on the results of this sensitivity analysis, Q _breast, P _ISF/cell, and P _plasma/milk were optimized to better characterize available pregabalin human milk data. In addition, the effect of food on milk exposure was assessed in light of the absence of fed versus fasted data in breastfeeding individuals.

Pediatric PBPK model development

The adult PO PBPK model was expanded to pediatrics by accounting for the relatively higher (~40%) renal clearance in pediatrics weighing less than 30 kg compared to adults. ²⁴ , ²⁷ To this end, the estimated glomerular filtration rate (eGFR) was calculated based on Equation 2,

$$\text{eGFR}\left(\mathrm{mL}/\min \bullet 1.73{\mathrm{m}}^2\right)=\frac{k*\text{height}\left(\mathrm{cm}\right)}{\mathrm{SCr}\left(\mathrm{mg}/\mathrm{dL}\right)}$$ (2)

where a value of k = 0.413 was used for pediatrics ≥12 months or 0.45 if they were younger than 12 months, and SCr is the serum creatinine level, which was reported in the in‐house datasets. ⁴⁴ , ⁴⁵ All other parameters were calculated by PK‐Sim® for the respective pediatric population. P _intestinal, K _ISF/plasma, and fGFR were optimized from the adult model values due to age‐related physiological differences affecting drug absorption and distribution.

Breastfeeding inputs into infant PBPK drug exposure model

Pregabalin concentration‐time profiles for breastfeeding infants were simulated over the course of 2 weeks for infants at: birth, 2 weeks, 4.5, 8, and 16 months of age in order to capture the changes that occur early on during development by estimating the exposure to pregabalin expected at that time through breastfeeding. Breastfeeding inputs were informed from previous PBPK models and lactation data. Breastfeeding characteristics include the concentration of drugs in milk, the number of feeds per day, and the volume of feeds per day. The weight‐based milk volume intake per day was input from Yeung et al. ⁴⁶ The number of feeds per day was fixed based on the average reported by Yeung et al., and adjusted to account for more frequent feedings early in life. For milk concentrations, the concentration each day was sampled from a normal distribution with a mean of 2.1 μg/mL and a standard deviation of 0.54 μg/mL based on the lactation PBPK model for a dose of 150 mg BID.PBPK Model Evaluation.

PBPK model performance was evaluated through a visual predictive check of population simulations with 95% confidence intervals (CI) and respective verification datasets. PK parameter calculations were performed by PK‐Sim® and compared to literature values and FDA drug labeling. Goodness‐of‐fit (GoF) plots with a twofold range of acceptance were included as an additional evaluation metric. Additional verification of the lactation model was undertaken by comparing the simulated PK parameter values to those calculated by Lockwood et al., ²³ including peak concentrations (C _max) and area under the concentration‐time curve (AUC) in human milk and plasma, along with calculated bodyweight‐normalized maternal and infant doses. Pediatric simulations were evaluated via the same metrics both when grouped by age and dosing regimen as well as when grouped only by dosing regimen.

RESULTS

Adult and childbearing‐age female PBPK model development and evaluation

Simulated PK profiles for adult populations are shown with the verification dataset overlain in Figure S1. The simulated PK values for CL/F, elimination half‐life (t _½), V _d/F, and T _max calculated by PK‐Sim® are comparable to those reported in the literature and FDA labeling (Table S1) (details on PK parameter calculations in PK‐Sim® are available at https://docs.open‐systems‐pharmacology.org/working‐with‐pk‐sim/pk‐sim‐documentation/pk‐sim‐simulations). ¹⁶ , ²³ , ²⁷ The optimized physiological parameters for all final PBPK models are outlined in Table 2.

GoF plots show nearly all of the simulated plasma concentrations were contained within the twofold range of the observed data (Figure S2) with the exception of early time points following single‐dose administration in the fed state (i.e., the absorption phase). The parameter values obtained from the adult model were also used in the childbearing‐age female and lactation PBPK models as parameterization showed no significant improvement in model fits based on calculated sum‐of‐squares values.

Lactation PBPK model development and evaluation

Simulated pregabalin concentration‐time profiles fit observed concentrations in plasma very well but significantly overpredicted concentrations in human milk when using the lactation template and respective default parameters provided by Job et al. ²³ , ⁴⁰ However, the overall curve shape was similar suggesting that drug input was too high. The results of our sensitivity analysis (Figure S3) showed that SA _plasma/milk did not significantly impact pregabalin exposure in human milk, leaving P _plasma/milk and Q _breast as sensitive parameters given that f _u = 1. Decreasing Q _breast from the default value of 62 mL/min/100 g to 1.79 mL/min/100 g of breast tissue, which is in line with the ICRP average values and reflects perfusion‐limited distribution, largely resolved this model mismatch. ¹³ , ⁴¹ In addition, P _plasma/milk and P _ISF/cell were optimized to further improve the model fit for milk as shown in Figure 3, where an optimized value of T _Lag accounted for the difference in fasted (Figure 3a,c) versus fed (Figure 3b,d) state. Lastly, the permeability from plasma to milk (P _plasma/milk) was optimized to 108.8 cm/min (vs. the default 100 cm/min in the opposite direction), suggesting net flux into human milk, which may be largely transporter dependent, whether from plasma or ISF. ¹² , ³⁵

FIGURE 3.

Lactation PO PBPK model performance with visual predictive check and simulated populations. Evaluation of lactation PBPK model predictive performance in plasma (red and orange simulation curves and corresponding shaded regions) and human milk (blue and cyan simulation curves and corresponding shaded regions) for fasted (a) and fed (b) state simulations, and for virtual populations in the fasted (c) and fed (d) states. Solid lines represent the simulated mean concentration and the respective shaded regions represent the 95% confidence interval (CI, 2.5%–97.5% range) of the simulated value. For the fasted‐state simulations (a) and (c), mean plasma concentration is shown in red with corresponding observed data overlain as red squares (mean ± standard deviation [SD]), and mean milk concentration is shown in blue with corresponding data as blue circle (mean ± SD). For the fed‐state simulations (b) and (d), mean plasma and milk concentrations are shown in orange and cyan, respectively, with corresponding data overlain as red squares and blue circles as with the fasted‐state simulations. Semi‐log plots of simulation means are shown in the inset plots in the upper right corners. The observed data correspond to all individuals from the Lockwood et al. ²³ study.

Performance of the lactation PBPK model was further assessed via GoF plots (Figure S4) and by comparing the simulated PK parameter values for plasma and human milk to those calculated from the study data (Table 3). ²³

TABLE 3.

Comparison of simulated vs. observed PK and infant dose parameters for lactation PBPK model.

PK parameter	Units	Plasma		Human milk
		Simulated a	Observed b	Simulated a	Observed b
AUCτ	μg h/mL	32.1 (23)	32.5 (24)	25.9 (21)	24.6 (27)
C max,ss	μg/mL	5.17 (11)	4.67 (18)	2.57 (1)	2.47 (30)
C av,ss	μg/mL	3.12 (34)	2.71 (24)	2.11 (21)	2.05 (27)
T max,ss	h	1.5 (1.2–1.75)	2.01 (1.00–3.00)	4.2 (3.7–5.9)	4.63 (3.08–6.16)
Half‐life (t ½)	h	6.2 ± 1.06	5.62 ± 0.66	7.43 ± 1.3	8.12 ± 3.09 b
CLss/F	mL/min	73.3 (23)	76.90 (24)	–	–
CLbm	mL/min	–	–	0.14 (7)	0.15 (60)

Milk‐specific parameter	Units	Geometric mean (% CV)		Range
		Simulated a	Observed b	Simulated a	Observed b
Aeτ,bm	μg	255.9 (7)	286.9 (60)	–	–
Aeτ,bm	%	0.1701 (6)	0.1913 (60)	–	–
Ae24,bm	μg/day	511.6 (7)	574.0 (60)	414.7–814.9	270–1720
MPAUCτ	–	0.83 (34)	0.76 (18)	0.49–1.07	0.49–0.96
MPC max	–	0.56 (15)	0.53 (22)	0.43–0.72	0.34–0.76
BWNID	μg/kg/day	314.6 (32)	307.9 (27)	215–514	202–458
BWNMD	μg/kg/day	4800.1 (10)	4288.8 (17)	3989–5589	3444–5676
BWNIDPCM	%	6.55 (39)	7.183 (23)	3.9–10.9	4.4–9.8

Abbreviations: AUC_τ, area under the concentration‐time curve for dosing interval τ (12 h); C _max,ss, maximal drug concentration at steady‐state; C _av,ss, average steady‐state concentration; T _max, time to achieve maximal concentrations at steady‐state; CL_ss/F, apparent plasma clearance at steady‐state; CL_bm, clearance from milk; CL_r, renal clearance; Ae_τ,bm, amount (or %) excreted in milk over dosing interval τ; Ae_24,bm, amount excreted in milk per day; MPAUC_τ, milk‐to‐plasma ratio for AUC_τ; MPC _max, milk‐to‐plasma ratio for C _max; BWNID, body weight normalized infant dose; BWNMD, body weight normalized maternal dose; BWNIDPCM, body weight normalized infant dose as a percentage of maternal dose.

^a Simulated values are expressed as geometric mean (% CV), except for T _max,ss, which is expressed as median (range).

^b Observed values are expressed as geometric mean (% CV), except for T _max,ss and t _½ values, which were reported as median (range) and arithmetic mean (± SD), respectively. ²⁶

Pediatric PBPK model development and evaluation

Pediatric PO PBPK model performance is displayed by visual predictive check (with 95% CI ranges) of virtual populations, ages 3 to 23 months, grouped by dosing regimen, indicating that the model fits the data reasonably well since simulated concentrations (Figure S5) are within a twofold range of the observed values (Figure S6). The eGFR values calculated by inputting the reported SCr values into the equations by Schwartz et al. and Brion et al., with k = 0.45 for individuals <12 months and k = 0.413 for those >12 months of age, were consistent with measured CL_Cr values and similar to most of the GFR values generated by PK‐Sim®. ⁴⁴ , ⁴⁵ After accounting for renal function, the pediatric model was similar to adults in that P _intestinal and K _ISF/plasma had the most significant impact on PK parameters relevant to absorption and distribution (i.e., C _max, T _max, and V _d,ss/F). A difference from the adult model was expected due to the tendency for absorption—and possibly transporter expression—to be lower in pediatrics, while V _d is generally higher due to greater body fat and water compositions. PK‐Sim® calculations for T _max, t _½, V _d/F and CL/F were in line with published values (Table S1). ²⁴ , ²⁷

Breastfeeding infant PBPK drug exposure model

For the simulations of the infants across the first 2 years of life, the optimized pediatric PBPK model and the outputs of the milk concentration of the childbearing‐age female PBPK were combined. In Figure 4, the concentrations are shown over a 2‐week simulation period for an infant at 0, 2 weeks, 4.5, 8, and 16 months of age. From the simulations, the first 2 weeks of life show an increase to a steady state that is maintained for the 2‐week‐old infant at the highest level over the timespans tested. The trend seen in the steady‐state concentrations decreases by 4.5 months and continues to decrease for 8 and 16 months.

FIGURE 4.

Infant drug exposure simulations. Simulation showing the infant drug exposure expected from milk consumption over 2 weeks for an infant 0 weeks old, 2 weeks old, 4.5 months old, 8 months old, and 16 months old, with a maternal dose of 150 mg administered twice daily. “Toxicity Threshold” (dark red horizontal line) corresponds to the observed C _max for the highest dose tested in pediatric clinical trials (i.e., 15 mg/kg/day in two divided doses, or 7.5 mg/kg twice daily), which was ~16 μg/mL. “Therapeutic Threshold” (blue horizontal line) corresponds to the observed C _max for the lowest dose tested in pediatric clinical trials (i.e., 2.5 mg/kg/day in two divided doses, or 1.25 mg/kg twice daily), which was ~1.5 μg/mL.

DISCUSSION

In this study, we were able to further verify the lactation modeling workflow that was proposed for sotalol by Pressly et al. ¹⁴ through the inclusion of pregabalin, a second prototypical drug with similar ADME properties. Furthermore, we were able to expand the lactation modeling framework by Pressley et al. ¹⁴ by adding an actual lactation compartment rather than relying on the availability of human milk data. This expansion provides the scientific basis for assessing lactational exposure of compounds, where actual clinical data was previously missing.

The model for pregabalin was developed and verified in a stepwise fashion starting with a PO model for adults due to the absence of IV data before expanding it to females of childbearing age, pediatrics, and ultimately lactation. It became evident during model development and verification that P _intestinal, tissue K _ISF/plasma, and fGFR were the primary drivers for pregabalin exposure in plasma, with P _intestinal being the most influential one.

The parameterization of fGFR was informed by three aspects: (1) renal clearance of pregabalin is lower than GFR despite the absence of plasma protein binding, which suggests the involvement of tubular reabsorption, (2) clearance is proportional to CL_Cr, thereby allowing fGFR to serve as a correction factor, ¹⁶ , ²⁷ , ³⁰ , ⁴⁷ and (3) and the value of 77% for fGFR that we used in our PBPK model is consistent with values calculated from the literature, which range from ~60% to 90%. Even with P _intestinal and fGFR being optimized, the model still underpredicted pregabalin exposure in plasma. Therefore, tissue K _ISF/plasma was optimized to more accurately reflect tissue distribution and to correct the initial V _d/F values calculated by PK‐Sim®, which were approximately 40% higher than the established literature value of 0.5 L/kg. ¹⁶ , ²⁷ Since there were no IV data available, the optimization was performed using oral solution data to avoid further confounding with formulation‐related factors, such as oral capsule dissolution. Once these optimization steps were performed, the PO PBPK model was able to capture corresponding observations well (Figure S1a) as observations fell within a twofold range of the predictions (Figure S2).

The inclusion of a short T _Lag of less than 6 min, which was informed by population PK studies, further improved the model's performance. ³⁷ , ³⁸ Further refinement of P _intestinal in conjunction with the Weibull parameters resulted in additional improvement of model performance following single‐ and multiple‐dose administration (Figure S1b). An increase in T _Lag to 23 min allowed us to capture the 100‐mg hard gelatin capsule data also in the fed state. ¹⁹ , ²⁰ , ²¹ , ²² Our fed‐state simulations verified that food has a negligible impact on the absorption and PK of IR pregabalin formulations as is to be expected for a BCS class I drug. While overall exposure is not impacted by food, it can shift the pregabalin plasma concentration–time course, resulting in a 25%–30% decrease in C _max and an up to 3‐h increase in T _max, which was considered for the pediatric and lactation models (Figure 3). ¹⁶ , ¹⁷ , ²⁷

The primary drivers for pregabalin exposure in human milk were Q _breast and P _ISF/cell. The default Q _breast value of 62 mL/min/100 g of tissue, which was also the default value for cardiac output in the MoBi® template is at least 10 times higher than the respective physiological value. As a result, initial simulated milk concentrations were nearly equal to concentrations in plasma. Using a more physiological value of 1.93 mL/min/100 g of tissue, which is within the reported range of 1.2–4.1 mL/min/100 g of tissue in the literature and ICPR, significantly improved the simulation outcomes. ¹³ , ⁴¹ This was expected because tissue blood flow rate helps determine how much drug is delivered to, and potentially distributed into, tissues over time. Similar to P _intestinal, P _ISF/cell was optimized to account for transporter‐mediated uptake of pregabalin into human milk. ¹¹ , ¹² , ³⁵ None of the other adult model parameters were changed in the lactation model. The lactation model was verified with data for 10 lactating women (Figure 4a,b) from the study by Lockwood et al. ²³ A PK analysis of the simulated populations was performed and the results were similar to the calculated PK parameters reported in the study (Table 3). ²³ The simulated milk concentrations were then used as PK inputs into the pediatric PO PBPK model to develop the infant exposure PBPK model.

The pediatric model was able to capture all data within twofold of the observed concentration values and did so when grouped into a wide age range from 3 to 23 months. Simulated C _max and AUC values were in line with those published by Mann et al., ²⁴ which provided insight regarding toxicity thresholds. The optimized pediatric model parameters and simulated milk concentrations were integrated with weight‐based milk volume intake per day from Yeung et al. ⁴⁶ to perform drug exposure simulations (Figure 4), which predicts the greatest exposure occurring within the first 2 weeks of life with a plasma C _max of approximately 0.44 μg/mL. This was likewise observed in the simulations from the first prototypical drug, sotalol, that applied this simulation approach.

One limitation of the pediatric model is that the single‐dose fasted‐state data were limited to only one individual in each dosing group (Figure S5a)—all of whom were over 12 months of age. Even so, the models for both sotalol and pregabalin perform well and led to findings that agree with one another, including nonlinearity in human milk concentrations that may be related to transport mechanisms, and variability related to ontogenetic/developmental differences among pediatrics (e.g., renal function and gastric emptying time). In this study, drug transfer into human milk was heavily perfusion‐limited and primarily from plasma (as opposed to ISF); however, physicochemical properties remain important considering that pregabalin enters milk at concentrations comparable to sotalol despite existing primarily as a zwitterion with inherently low permeability. This is likely due to transporter involvement, which highlights that there are considerable gaps between compounds that require further exploration. For ease of comparison in Figure 4, horizontal lines show concentration thresholds according to dose, including: the minimum labeled dose (3.5 mg/kg/day, blue line), the maximum labeled dose (14 mg/kg/day, red line), and the highest dose tested in pediatric clinical trials (15 mg/kg/day, dark red line). ²⁴ , ²⁷ The majority of discontinuations due to adverse effects in the clinical trial were precipitated by the 15 mg/kg/day dose with a corresponding C _max of ~13 μg/mL, considered the toxicity threshold for the purposes of this study. ²⁴ Simulations show that peak concentrations lie below the lowest approved treatment dose of 3.5 mg/kg/day, which has an estimated average steady‐state concentration (C _av,ss) of 1.2 μg/mL, and that the estimated RID is ~7% for maternal doses of 300 mg/day, which is consistent with observed data.

In summary, our work further supports the value and applicability of integrating modeling and simulation techniques with pediatric and lactation data to bridge the knowledge gaps in infant drug exposure through breastfeeding. This, in turn, provides a scientific basis for knowledge and evaluation that others may build upon using a range of additional compounds to map out physiological processes involved in lactational drug exposure. Once sufficient confidence in the PBPK modeling platform is achieved with simpler compounds (i.e., small molecule drugs that are primarily renally eliminated, have limited plasma protein binding and metabolism, linear PK, etc.), it should be expanded to compounds with greater complexity and more significant PK considerations (i.e., non‐BCS Class 1, alternative routes of metabolism and elimination, significant protein binding, etc.). Applying this framework to more complex drugs will require additional considerations in terms of model parameterization and the dynamic processes necessary to describe corresponding ADME characteristics. For example, changes in human milk composition over time, for example, from protein‐rich colostrum to mature milk within a few days postpartum or from aqueous foremilk to lipid‐rich hindmilk over the course of feeding, are also worth considering because they can potentially impact homogeneous drug distribution throughout the milk, which might affect drug quantification in addition to infant drug exposure during the course of breastfeeding. ⁴¹ , ⁴⁸ , ⁴⁹ , ⁵⁰ Nevertheless, the type of lactation modeling results yielded by this workflow demonstrates the potential of PBPK modeling for risk assessments in special populations and could support pediatric lactation risk assessment when combined with toxicity thresholds identified in pediatrics to better inform drug labels and breastfeeding recommendations. In fact, these efforts could also have valuable utility in drug discovery and development by facilitating the exploration of drug characteristics that provide adequate therapeutic exposure in lactating individuals with minimal transfer to breastfeeding infants.

AUTHOR CONTRIBUTIONS

C.H., M.P., Z.L., Sh.S., and St.S. wrote the manuscript. E.P.F., M.P., D.G., Z.L., and St.S. designed the research. C.H. performed the research. C.H. and M.P. analyzed the data.

FUNDING INFORMATION

This work was supported by the FDA Perinatal Health Center of Excellence grant and was conducted under an Academic Memorandum of Understanding between FDA and University of Florida.

CONFLICT OF INTEREST STATEMENT

The authors declared no competing interests for this work.

DISCLAIMER

The opinions expressed in this article are those of the authors and should not be interpreted as the position of the US Food and Drug Administration.

Supporting information

Data S1.

Humerickhouse C, Pressly M, Lin Z, et al. Informing the risk assessment related to lactation and drug exposure: A physiologically based pharmacokinetic lactation model for pregabalin. CPT Pharmacometrics Syst Pharmacol. 2024;13:1953‐1966. doi: 10.1002/psp4.13266