A tutorial on physiologically based pharmacokinetic approaches in lactation research

Abstract

In breastfeeding mothers, managing medical conditions presents unique challenges, particularly concerning medication use and breastfeeding practices. The transfer of drugs into breast milk and subsequent exposure to nursing infants raises important considerations for drug safety and efficacy. Modeling approaches are increasingly employed to predict infant exposure levels, crucial for assessing drug safety during breastfeeding. Physiologically‐based pharmacokinetic (PBPK) modeling provides a valuable tool for predicting drug exposure in lactating individuals and their infants. This tutorial offers an overview of PBPK modeling in lactation research, covering key concepts, prediction approaches, and best practices for model development and application. We delve into milk composition dynamics and its influence on drug transfer into breast milk, addressing modeling considerations, knowledge gaps, and future research directions. Practical examples and case studies illustrate PBPK modeling application in lactation studies. We demonstrate how prediction algorithms for Milk‐to‐Plasma (M/P) ratios within a PBPK framework can support scenarios lacking clinical lactation data or extend the utility of available lactation clinical data to support further untested clinical scenarios. This tutorial aims to assist researchers and clinicians in understanding and applying PBPK modeling to understand and support clinical scenarios in breastfeeding mothers. Advances in PBPK modeling techniques, along with ongoing research on lactation physiology and drug disposition, promise further insights into drug transfer during lactation.

INTRODUCTION

Breastfeeding plays a crucial role in promoting the health and well‐being of both infants and mothers. The benefits of breastfeeding for the infant are widely recognized. Thus, the World Health Organization (WHO) recommends exclusive breastfeeding for the first 6 months of life with partial breastfeeding continued to up to 2 years of age and beyond. ¹ This recommendation aims to optimize infant nutrition, support healthy growth and development, strengthen the bond between mother and child, and contribute to the reduction of infant morbidity and mortality. Malaria, tuberculosis, and neglected tropical diseases (NTDs) are among the leading causes of morbidity and mortality globally. These diseases are prevalent (more than 300 million cases in 2021 alone) in low‐ and middle‐income countries (LMICs), where exclusive breastfeeding is the norm. ² In regions with high disease burden, breastfeeding serves as a critical public health strategy to enhance infant and maternal health outcomes. However, for breastfeeding mothers, the management of medical conditions during this period poses unique challenges, particularly concerning the use of medications and breastfeeding practices. The transfer of drugs into breast milk and subsequent exposure to nursing infants raises important considerations regarding drug safety and efficacy. The prevalence of these chronic conditions requiring pharmacotherapy among lactating mothers, coupled with the growing awareness of the benefits of breastfeeding, emphasizes the need for evidence‐based guidance on medication use during lactation. This is particularly important for the development of new drugs, as it ensures that both maternal health needs and infant safety are adequately addressed.

In a drug development setting, data regarding the safety of medications during lactation are often limited or absent at the time of market approval. Historically, such data have been collected post‐approval, which poses significant challenges for healthcare providers and nursing mothers. Implementation of the Pregnancy and Lactation Labeling Rule (PLLR) in 2015 by the U.S. Food and Drug Administration (FDA) established upgraded criteria for incorporating data on the utilization of prescription medications in pregnant women or lactating mothers. The PLLR requires changes to the content and format of information presented in the drug labeling relating to the risks of the presence of the drug in breast milk and its potential effects on the breastfed child. PLLR supports safer prescribing practices and encourages the inclusion of pregnant and lactating women in clinical research, leading to more evidence‐based guidelines and recommendations. ³ , ⁴ A recent report presents the trends of issued lactation‐related Post‐Marketing Commitments (PMCs) and Post‐Marketing Requirements (PMRs) for 488 New Drug Applications (NDAs) approved by the FDA between 2000 and 2022. An increase in pregnancy and lactation‐related PMRs was observed following the enactment of the PLLR in 2015. Among the 59 NDAs which included PMRs, 18 (31%) had lactation‐related PMRs, of which 16 were issued after 2015. Notably, lactation‐related PMCs for new molecular entities have not been requested to date. ⁵

Another cross‐sectional study reviewed the labeling data for 290 therapeutic drugs under 19 drug categories, approved by the FDA between 2010 and 2019. Only 148 therapeutic products (51%) had any data associated with lactation, 143 (49.3%) originating from animal studies and 8 (2.8%) from human studies. ⁶ Based on the latest information from the clinical trial website, ⁷ only 18 lactation pharmacokinetics studies were conducted between 2010 and 2023. The paucity of drug safety data on breastfeeding women stems from the routine exclusion of this population from clinical trials due to scientific, ethical, regulatory, and legal concerns. ⁸ However, this is likely to change in the coming years; in addition to including more participants from underrepresented racial and ethnic populations in clinical trials within the US, the recent FDA diversity and inclusivity draft guidance and ICH E21 final concept paper encourage the inclusion of other underrepresented populations including pregnant and lactating women. ⁹ , ¹⁰ The gradual acceptance of clinical studies involving lactating women marks a positive shift toward the approach of prioritizing the collection of actual data over the perpetuation of knowledge gaps, leading to improved safety and efficacy profiles for medications used during lactation. Indeed, there are currently 24 clinical lactation pharmacokinetics studies ongoing, centering on determining drug exposure in lactating mothers and/or breastfeeding infants. ⁷ The increasing number of lactation studies suggests growing awareness of the importance of timely and robust data to support drug benefit–risk decision‐making for lactating women.

The use of model‐based approaches such as population pharmacokinetics (PopPK) and physiologically‐based pharmacokinetic (PBPK) modeling to predict relevant exposures of maternal medicines during breastfeeding and to inform decision‐making on providing active treatment in mothers is gaining traction. ¹¹ Given its mechanistic nature, PBPK modeling can be used to predict drug concentrations in breast milk and in infants during breastfeeding prior to the conduct of any clinical studies trials or indeed to supplement clinical lactation data which often reflects a small number of subjects, sub‐optimal scenarios (limited samples and variation in sampling times and milk volume), and untested scenarios (effects of genotypes or different dosage regimens). ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ User‐friendly commercially available platforms which integrate this complex methodology and associated data, include the Simcyp Population‐Based Simulator (Certara UK Ltd) (http://www.simcyp.com/), GastroPlus (SimulationsPlus, Lancaster, CA) (http://www.simulations‐plus.com/) and PK‐SIM (Bayer Technology Services, Germany) (http://www.systems‐biology.com/products/pk‐sim.html). This tutorial provides some background on key factors affecting the transfer of drug into milk during lactation and prediction methods to quantify this, followed by a comprehensive overview of the application of PBPK modeling in the context of lactation including breastfed infants, providing guidance on model development, verification, and interpretation.

QUANTITATIVE ASSESSMENT OF DRUG TRANSFER TO BREAST MILK

Drug transfer occurs from the maternal bloodstream into the mammary glands via the mammary epithelial barrier, where they can subsequently be secreted into breast milk. Given the challenges of conducting clinical lactation studies, several non‐clinical methods have been used to predict drug transfer into the milk. A recent review provides an extensive overview of the in vitro, in vivo, and in silico methods that are available to study the transfer of maternal medication into human breast milk. While several in vitro models (animal and human mammary epithelial cell lines) are available, cell culture model characterization, including quantitative transport across the in vitro blood‐milk barrier, remains limited. ¹⁷ Furthermore, animal in vivo models have also been studied and used in FDA‐approved product labeling. However, animal models typically do not mimic human physiology and have been shown to have significant differences in drug binding to milk proteins and composition of the milk itself. ¹⁸ A recent literature review on animal models highlights the significant anatomical and physiological similarities and differences in lactation and milk composition across commonly used species, emphasizing the complexity of selecting suitable models for in vivo lactation trials. The review noted that data for some species were either poor or missing, indicating a need for more comprehensive physiological studies on routinely used laboratory animals. ¹⁹ Due to such limitations, relying on quantitative animal data for FDA labeling without proper clinical context can be misleading. An FDA workshop summary on current data collection approaches for medications used during lactation also concluded that, while animals can show the presence of a drug in milk, they are inadequate for predicting human milk drug concentration profiles. Consequently, the FDA's PLLR advises against including animal data if human data are available, recommending that only the presence or absence of a drug in milk be noted when animal data are included. Nevertheless, animal data can be valuable in certain situations, and the PLLR should better clarify the expectations for generating and using animal data. ²⁰ Among the different in vitro models of mammary epithelial cells (MECs), primary cell cultures offer the opportunity to study the factors that regulate physiologically relevant development of normal mammary epithelial cells under defined conditions. ²¹ Despite this, there remains the scope to further develop, optimize, and validate, mainly human MEC based models, which may provide the opportunity to better understand these mechanisms.

Drug transfer into breast milk is determined by drug factors such as ionization, plasma protein binding, molecular weight, lipophilicity, biochemical characteristics of milk, and drug pharmacokinetics in the mother. ²² Various mathematical methods have been reported for predicting the transfer of drugs into human milk (i.e., to estimate the milk to plasma (M/P) ratio) based on drug physicochemical and milk properties. ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ Most of the models assume that the transfer of drugs from maternal plasma to breast milk across the mammary epithelial membrane is a passive diffusion process (Figure 1).

FIGURE 1.

Lactation PBPK model scheme: During lactation, the mammary epithelium is organized into lobes containing numerous lobules and connected to lactiferous ducts, which drain milk toward the nipple (bottom‐left). Drugs enter through mammary epithelium and are secreted into milk by passive diffusion and/or active transport as the paracellular pathway is blocked during lactation. An integrated lactation model algorithm (e.g., phase‐distribution model equation) predicts the milk to plasma ratio (M/P) by considering the drug's physicochemical properties, its binding affinity to proteins in plasma and milk and milk properties (bottom‐right). The maternal PBPK model, along with the predicted or clinical M/P ratio, gives drug exposure in maternal plasma and milk. The milk exposure information, combined with infant milk intake, provides an estimate of the absolute and relative infant daily dose (IDD and RIDD) of the maternal ingested drug via breastfeeding (top‐left to right).

The two most widely used mathematical models for predicting M/P ratios include the phase distribution model (Equation 1) ²⁵ , ²⁶ and the log‐transformed phase distribution model (Equation 2). ²⁵ Both models utilize the same physicochemical and milk physiological parameters, that is, degree of free drug ionization in maternal plasma $\left(f_{\mathrm{p}}^{\mathrm{un}}\right)$ and milk $\left(f_{\mathrm{mk}}^{\mathrm{un}}\right)$, fraction unbound in maternal plasma $\left({\mathit{fu}}_{\mathrm{p}}\right)$ and milk $\left({f\mathit{u}}_{\mathrm{mk}}\right)$ fat component in milk $\left(f_{\mathrm{fat}}\right)$ and milk lipid‐to‐ultrafiltrate partition coefficient $\left(P_{\mathrm{mk}}\right)$. To improve the prediction performance, the phase distribution model was logarithmically transformed to obtain the regression coefficients for individual components, forming log‐transformed models for basic and acidic drugs separately.

$$\begin{array}{c}\frac{\mathrm{M}}{\mathrm{P}}=\frac{{f\mathit{u}}_{\mathrm{p}}·f_{\mathrm{p}}^{\mathrm{un}}}{f_{\mathrm{mk}}^{\mathrm{un}}}·\frac{1}{{f\mathrm{u}}_{\mathrm{mk}}·\left(\frac{1}{1+f_{\mathrm{fat}}·\left({f\mathit{u}}_{\mathrm{mk}}·P_{\mathrm{mk}}-1\right)}\right)}\hfill \end{array}$$ (1)

$$\begin{array}{c}\mathrm{Ln}\frac{\mathrm{M}}{\mathrm{P}}=a+b\ln \left(\frac{f_{\mathrm{p}}^{\mathrm{un}}}{f_{\mathrm{mk}}^{\mathrm{un}}}\right)+c\ln {f\mathit{u}}_{\mathrm{p}}+d\ln \left[\left(\frac{1-f_{\mathrm{fat}}}{{\mathit{fu}}_{\mathrm{mk}}}\right)+f_{\mathrm{fat}}·P_{\mathrm{mk}}\right]\hfill \end{array}$$ (2)

where, a = 0.025, b = 2.28, c = 0.89, d = 0.51 for basic drugs; and a = −0.405, b = 9.36, c = −0.69, d = −1.54 for acidic drugs. Drug ${\mathit{fu}}_{\mathrm{mk}}$ (Equation 3) and $P_{\mathrm{mk}}$ (Equation 3) are predicted as follows:

$${\mathit{fu}}_{\mathrm{mk}}=\frac{{{\mathit{fu}}_{\mathrm{p}}}^{0.448}}{{0.000694}^{0.448}+{{\mathit{fu}}_{\mathrm{p}}}^{0.448}}$$ (3)

$$P_{\mathrm{mk}}={10}^{\left(-0.88+1.29{\mathrm{Log}D}_{\mathrm{pH},\mathrm{mk}}\right)}$$ (4)

It should be noted that the prediction equation of ${\mathit{fu}}_{\mathrm{mk}}$ (Equation 3) was derived from experimentally measured unbound fractions in plasma and skimmed milk of only 14 drugs (with ${\mathit{fu}}_{\mathrm{p}}$ ≥ 0.3 (n = 7), ${\mathit{fu}}_{\mathrm{p}}$ = 0.01 to 0.1 (n = 6), ${\mathit{fu}}_{\mathrm{p}}$ = 0.001 (n = 1)). ²⁹ More recently, Yang et al. ³⁰ developed another ${\mathit{fu}}_{\mathrm{mk}}$ equation based on measured ${\mathit{fu}}_{\mathrm{p}}$ and PSA for a more diverse set of 39 compounds, (${\mathit{fu}}_{\mathrm{p}}$ ranging from 0.013 to 1.00), which integrated within an IVIVE model led to improved predictions of M/P ratios. However, in vitro derived ${\mathit{fu}}_{\mathrm{mk}}$ for a dataset of 30 drugs indicated highly variable ${\mathit{fu}}_{\mathrm{mk}}$ for highly bound drugs, suggesting potential challenges in accurate prediction and the need for mechanistic prediction. ²¹ Similarly, the $P_{\mathrm{mk}}$ prediction equation (Equation 4) was derived using measured apparent octanol to water coefficient (${\mathrm{Log}D}_{\mathrm{pH},7.2}$) and milk lipid‐to‐ultrafiltrate partition coefficient (at pH 7.2) data from another similar set of 14 drugs (six acidic and eight basic drugs with ${\mathrm{Log}D}_{\mathrm{pH},7.2}$ values ranging from −0.47 to 2.92). ²⁵ The milk lipid‐to‐ultrafiltrate partition coefficient refers to the ratio of the concentration of a drug in breast milk lipid to its concentration in the ultrafiltrate portion (aqueous) of breast milk, which provides an indication of the extent to which a drug partitions into the lipid component of breast milk. Thus, lipophilic drugs which tend to partition more into the milk fat may be more likely to have higher concentrations in breast milk.

Since only unbound drugs can permeate the mammary epithelial barrier, drugs that have a high degree of maternal protein binding in the plasma, generally have lower partitioning into milk. However, it should be recognized that drug binding also occurs in milk with milk proteins. While milk composition data and information on fat components are readily available in the literature, data on the binding of drugs in milk and its partitioning to milk lipids are limited and often historical. Therefore, both mathematical models utilize relationships derived from regression analyses based on experimental measurements for a limited number of drugs. While these equations may provide accurate predictions for drugs with similar physicochemical properties utilized in the datasets, they can also be mispredicted, particularly for drugs that fall outside the physicochemical space considered for the development of these equations. For instance, neither prediction method for ${\mathit{fu}}_{\mathrm{mk}}$ has included drugs (except for atenolol with ${\mathit{fu}}_{\mathrm{p}}$ of 0.001) with a high degree of protein binding in plasma (i.e., very low ${\mathit{fu}}_{\mathrm{p}}$). Additionally, the prediction method for $P_{\mathrm{mk}}$ does not include drugs with very high lipophilicity (${\mathrm{Log}D}_{\mathrm{pH},7.2}$ > 2.92). Given that many new drugs in development are metabolically stable and tend to have high plasma protein binding and high lipophilicity values, this could present problems in terms of prediction accuracy. Furthermore, many drugs of interest in the global health space exhibit similar characteristics (e.g., Atovaquone, Lumefantrine). Including a more diverse set of compounds in generating these predictive algorithms or utilizing a more mechanistic approach based on specific milk proteins are necessary to gain more confidence in the predictions. With the growing emphasis on lactation research and the increasing availability of data, there are numerous ongoing efforts aimed at improving and refining these relationships.

MILK COMPOSITION DYNAMICS AND ITS INFLUENCE ON DRUG TRANSFER

Drug distribution into the milk can be affected by the dynamic changes in composition and the biochemical properties of breast milk throughout the stages of lactation (i.e., colostrum‐first few days postpartum (PP), transitional milk: ~6–15 days PP, and mature milk‐ 15 days PP). Measurements conducted in milk indicate that the fat content increases from 3.4%–4.3% in colostrum to 4.5%–6.1% in mature milk. ³¹ A more recent study conducted in mature milk from 1 month to over 24 months showed a gradual increase, with the mean fat content increasing from 2.8%–3.9% up to 12 months reaching its peak of 6.4%–9.4% in mature milk of >24 months postpartum. ³² It is also important to note that there is high variability in the reported data for fat content, which may be partly attributed to the use of different measurement techniques. ³³ Variation in milk pH has also been observed, which may be due to changes in milk composition throughout lactation. Colostrum typically exhibits a slightly higher pH of around 7.4, followed by a gradual decline over time to a minimum value of 7.0, which then stabilizes as the milk matures, slowly increasing to around a pH of around 7.2. ³⁴ , ³⁵ The milk pH also exhibits significant variability across different studies and reports in the literature. This variability can be attributed to several factors, such as timing of milk collection, processing and its storage, or the methods used for measuring pH. ³⁶ , ³⁷ Longitudinal studies with standardized protocols can help provide more accurate and comparable data.

Changes in these milk parameters and/or drug properties can have a significant impact on the prediction of drug transfer, depending on the drug characteristics. In general, as milk contains substantially more lipid and less protein than plasma and is slightly more acidic, drugs which tend to concentrate in milk are weak bases, with low plasma protein binding (${\mathit{fu}}_{\mathrm{p}}$) and high lipid solubility ($\mathrm{Log}P$). The pH gradient between maternal serum and breast milk can be a major determinant of the quantity of drug excreted into the milk. Figure 2 depicts the impact of drug physicochemical properties and colostrum versus mature milk pH differences on the drug transfer to breast milk for acids and bases. For example, for bases, breast milk pH is thought to be the most sensitive parameter which can greatly influence the M/P ratio prediction. In general, basic drugs will be more unionized in the plasma (at pH 7.4), diffuse through the mammary gland barrier and become more ionized in the mature milk (at pH ~7.0 to 7.2), which favors their concentration in breast milk due to an ion‐trapping phenomenon. As a comparison, the M/P ratio of such drugs would be higher in mature milk than colostrum. For example, tramadol, a basic drug with a pKa of 9.41, has an observed M/P ratio of 1.85 at 2–4 days postpartum, which was predicted in the model as 1.93, assuming a milk pH of 7.24 in the absence of measured milk pH information. ¹³ Utilizing this verified model for tramadol and examining the impact of varying pH (i.e., transition to mature milk), the predicted M/P ratios were 2.4, 4.0, and 6.8, representing a 1.2 to 3.5‐fold increase at milk pH of 7.2, 7.1, and 7.0, respectively. In contrast, acidic drugs will show the opposite trend, resulting in lower M/P ratios; and in comparison to colostrum, these drugs will have a lower M/P ratio in mature milk (ex. thiopental, a weak acid). ³⁸

FIGURE 2.

Impact of drug physicochemical properties and milk pH on drug transfer to breast milk: The schematic illustrates example drugs with varying physicochemical properties and their partition into breast milk. Assuming an equal proportion of unbound unionized drug (i.e., 1) crosses the mammary epithelium barrier for all drugs (except for an acid with a pKa of 2), changes in milk pH and their relative impact on drug ionization and M/P ratio (with respect to unbound amount) are depicted. Considering the average milk pH for colostrum as 7.4 (same as plasma) and mature milk pH as 7.2, for a basic drug with a pKa of 2, all of the drug in colostrum and mature milk will be unionized (unionized to ionized unbound drug amount = 1:0), resulting in the same M/P ratio for both colostrum and mature milk. For a base with a pKa of 8, approximately 80% of the drug in plasma or colostrum will be ionized (1:4). However, in mature milk, where there is more ionization (~86%), a higher amount (1:6.3) of the drug will reside in mature milk (M/P = 1.5) compared to colostrum (M/P = 1). This indicates that such a drug will be most sensitive to slight changes in milk pH. In contrast, an acidic drug with a pKa of 8 will be more ionized (~20%) in colostrum (1:0.25) compared to mature milk (~14% ionized—1:0.16), resulting in slightly less drug residing in mature milk (M/P = 0.93) compared to colostrum (M/P = 1). In the case of an acid with a pKa of 2, all of the drug is expected to be ionized, leading to negligible penetration into the milk.

The variation in fat content is further complicated by the stage of a breastfeed (foremilk versus hindmilk). A report on the measured milk fat content in foremilk (milk collected at the start of a feed) and hindmilk (milk collected at the end of a feed) suggested significant differences in fat content, with a higher fat percentage (2.2 to 4.5‐fold) observed in hindmilk compared to foremilk. ³⁹ Changes in milk fat content may affect the exposure of highly lipophilic drugs in breast milk. One such example is the lipophilic drug sertraline and its metabolite desmethyl sertraline, which have $\mathrm{Log}P$ values of 5.15 and 4.72, respectively. The breast milk concentrations of sertraline and desmethylsertraline were found to be the lowest in the first 10–20 mL of breast milk, and the highest in the more lipophilic hindmilk, with approximately 2‐fold higher concentrations relative to those in the foremilk. ⁴⁰ Thus, it is recommended to collect the entire milk volume for clinical lactation studies. ⁴ However, there are challenges associated with clinically collecting whole breast milk. Currently, lactation study designs are not consistent and are likely to be heterogenous in terms of sampling times; hence, information on milk type sampling is often lacking, which can further contribute to variability in milk exposure for different clinical scenarios. To account for this variability in model predictions, real‐world data, if available, can be used to capture changes in fat composition between foremilk and hindmilk at different stages of lactation. While the burden of additional measurements can't be underestimated, especially in early postpartum days, when nursing is not well established and mother and infant are still adapting to breastfeeding, sensitivity analysis around the model predictions for the M/P ratio can offer valuable insights in the absence of such detailed information. This can help elucidate the effects of changes in milk fat, pH, and other variables, providing a better understanding without the need for extensive additional measurements.

CLINICAL LACTATION STUDIES AND ESTIMATE OF INFANT DAILY DOSE DURING BREASTFEEDING

A guidance document released by the FDA in 2019 provided recommendations for pharmaceutical companies on how to address the potential impact of maternal drug exposure. ⁴ For evaluation of the safety of the drug in breastfeeding infants, it was stated that data from clinical lactation studies, supported by other relevant information including drug physicochemical properties, mechanism of drug entry into breast milk, data from nonclinical studies, and infant factors, can be used to provide recommendations to minimize infant exposure. The standard method of quantifying drug passage into breast milk is to administer a drug to a nursing mother, either because she is taking the drug therapeutically or for the purpose of the study. Sufficient drug concentrations in milk should be measured to allow the calculation of an area under the milk concentration–time curve ($\mathrm{AUC}$), an average milk concentration ($\mathrm{AUC}$ /milk sampling duration) and ideally, maximum concentration in milk $\left(C_{\max }\right)$. Typically, a M/P ratio would then be calculated using these data. Furthermore, milk exposures can be utilized to estimate the absolute infant daily dose (IDD) (Equation 5) and relative infant doses (RID) (Equation 6) based on the approach first published by the WHO, which provides an estimate of the amount of drug that an infant is likely to be exposed to via breastfeeding. ⁴¹

$$\begin{array}{c}\text{Absolute infant daily dose}\left(\mathrm{mg}/\mathrm{kg}/\mathrm{day}\right)\hfill \\ =\text{Milk}{C}_{\left(\mathrm{avg}\text{or}\max \right)}\left(\mathrm{mg}/\mathrm{L}\right)\hfill \\ \times \text{Daily Milk Intake}\left(\left(\mathrm{L}/\mathrm{kg}\right)/\mathrm{day}\right)\hfill \end{array}$$ (5)

$$\begin{array}{c}\text{Relative infant dose}\left(\%\right)\hfill \\ =\frac{\text{Absolute infant dose}\left(\mathrm{mg}/\mathrm{kg}/\mathrm{day}\right)}{\text{Maternal dose}\left(\mathrm{mg}/\mathrm{kg}\right)}\times 100\hfill \end{array}$$ (6)

Although this approach typically utilizes the average concentration ($C_{\mathrm{avg}}$) in the milk, the maximum drug concentration $\left(C_{\max }\right)$ can also be used to assess the worst‐case scenario. An average daily milk intake of 150 mL per kg of infant body weight which is based on the typical milk intake observed in breastfeeding infants is applied. Based on this assumption, an infant with a body weight of 4 kg will consume the total estimated daily milk intake of 600 mL. In clinical practice, it is often challenging to precisely determine the exact milk intake of each breastfeeding infant. Thus, average milk intake values are commonly used as a pragmatic approach. However, it's important to note that the individual variations in milk intake and infant body weight may occur, and that actual milk intake can vary among infants and over time. ⁴² Additionally, factors such as feeding frequency, infant age, and maternal drug metabolism can influence the actual drug exposure experienced by nursing infants. ⁴³ In the absence of this detailed information, such scenarios can be considered using PBPK modeling as described in later sections.

The RID calculation provides a standardized method of relating the infant's dose to the maternal dose. According to the WHO, an RID value of less than 10% is generally considered to be safe, an RID of 10%–25% should be used with caution, and those drugs with an RID >25% are likely to be unsafe. Of course, these values have to be put into context using the totality of evidence for the drug. Indeed, the specific threshold may vary based on the drug itself and the clinical context. For example, the risk may vary depending on the infant age, metabolism, and pharmacokinetic, or the drug‐specific parameters. In this respect, it should also be noted that neonates may have been exposed in utero to drugs taken by their mothers, and that in utero exposure may be an order of magnitude greater than that received via breast milk. This may be particularly prominent for drugs with long half‐lives (such as lamotrigine or dolutegravir), which have the potential for substantial in utero exposure and may pose a greater risk, especially in the first 1–2 weeks after birth. ⁴⁴ , ⁴⁵ , ⁴⁶

It is important to recognize that the RID provides an estimate of relative dose, not the actual exposure of the drug in the breastfeeding infant. Nor does it account for time‐variant physiological changes that occur in the infants or mothers during the breastfeeding period which can affect the ADME and ultimately, the PK of the drug. This is where PBPK modeling comes into play, utilizing the infant daily dose to input into the pediatric PBPK model to predict drug exposure, simulate different scenarios, and assess the risk in breastfed infants/neonates (Figure 3).

FIGURE 3.

Modeling approach for predicting drug exposure in breastfeeding neonates/infants: A clinically observed, or model predicted IDD, together with number of feeds given to the neonate/infant per day, provides the dose input for the pediatric/preterm PBPK model. Infant exposure considering different feeding patterns (standard feeding pattern versus worst‐case scenarios) can be performed as part of an infant risk assessment. Additionally, the impact of various ‘what if’ scenarios on exposure in neonate/infant can be assessed utilizing pediatric or preterm model capabilities.

DEVELOPEMENT, VERIFICATION, AND BEST PRACTICES FOR APPLICATION OF PBPK MODELING IN LACTATION

In the absence of any clinical lactation data, PBPK modeling can be used to predict milk concentration‐time profiles in breastfeeding mothers, which then provide an estimate of an IDD in the breastfed infant (Figure 1). The latter is used as daily dose in PBPK simulations to generate plasma exposures in the infant (Figure 3). Alternatively, if a clinical lactation study has been performed, the observed IDD can be used in simulations, and untested scenarios, such as different feeding regimens, can be simulated (Figure 3). Prior to these steps, a robust drug model is required to have confidence in these predictions. Figure 4 outlines the best practice approach for lactation PBPK modeling, delineating the following steps.

FIGURE 4.

A best practice approach for lactation PBPK modeling and its diverse applications: A workflow describing base PBPK model development and verification in adult and pediatric (blue), extending it to lactation modeling to estimate IDD (with predicted or clinical M/P), applying it to pediatric/preterm PBPK model and predict infants/neonates exposure (orange), application to explore multiple ‘What‐if’ scenarios (green) and how these efforts can feed into the ultimate questions around the risk: Benefit for both mother and breastfed infants/neonates (gray). Typically, a model is first developed to capture the drug exposure/Pharmacokinetic (PK) in healthy adults. The performance of the drug model alongside algorithms describing physiological and demographical data is then assessed compared to clinically observed data (verification). When clinical data are available describing PK in children and lactating mothers as well as adults, it builds confidence in the performance of the model. The verified model can then be applied, in conjunction with infant daily doses (calculated from observed or predicted M/P), to predict infant/neonate exposure, which can further be extended to assess the impact of different scenarios in breastfed infants/neonates.

Base model development and verification

The development of a robust PBPK model begins with a detailed characterization of the drug physiochemical properties and the physiological attributes relevant, absorption, distribution, metabolism, and excretion. ⁴⁷ This base model is then verified against clinically observed pharmacokinetics (PK) data in adults and may be optimized through iterative processes involving the comparison of simulated outcomes with observed clinical data.

Verification of the model in adult populations is an essential step to ensure that it can accurately predict concentration‐time profiles for a given compound. Once a satisfactory level of confidence is achieved in the adult model, attention shifts toward pediatric populations. The pediatric PBPK model accounts for the physiological and biochemical changes during children development. Specifically, the age‐related changes in tissue volumes, tissue composition, tissue blood flows, hematocrit, plasma binding protein, glomerular filtration rate (GFR), gastrointestinal physiology, ontogeny of drug‐metabolizing enzymes' abundance and time varying physiology have been incorporated in the PBPK model. ⁴⁸ , ⁴⁹ , ⁵⁰ The established base model should be able to recover the observed PK profiles against pediatric populations. The ethnicities difference, disease conditions, and malnutrition in children should all be considered during base model verification.

Lactation model development and verification

When the base model is verified in the corresponding adult and pediatric populations, the model undergoes further specialization in the context of lactation. The initial verification is focused on maternal PK profiles. The verified model can then be used in simulation of maternal PK profiles, which are critical for understanding drug transfer into breast milk. Further modification of the virtual population may be required if the specific physiological changes during lactation influence drug exposure significantly.

The next step is to determine drug excretion in the milk. Incorporating a reported milk to plasma (M/P) ratio from a clinical study may serve as a more practical approach. Alternatively, if maternal and breast milk data are not available, predicted M/P ratios can be used with caution, as there are uncertainties around the predicted values based on established methods (see earlier sections for limitations). The M/P ratio, combined with the maternal plasma concentration‐time profile, is used to estimate the IDD likely to be consumed via breastfeeding. As described previously, there are several methods available for estimating an IDD, and use of $C_{\max }$ as a representative milk concentration, is typically used to account for the worst‐case scenario.

IDD represents the total dose per day that an exclusively breastfed infant could consume from breast milk. However, the percentage of intake from breast milk may differ from 100% to less than 25% in reality. In addition, the breastfeed frequency per day may vary. ⁵¹ Typically, a baby over the first few weeks and months will breastfeed about 8 to 12 times in 24 h. From 6 to 12 months, the breastfeed frequency will be reduced to 6–7 times daily as infants start eating semi‐solid and solid foods. Thereafter, the frequency will be further reduced to twice a day. Therefore, the IDD should be divided by the typical number of daily feeds to determine the dose per feed according to the age. This data is then applied in the pediatric PBPK model and verified against PK data in breastfed infants (if available).

Exploration of what‐if scenarios and sensitivity analysis in PBPK lactation modeling

Prospective application simulations

Upon successful verification of the lactation model with maternal and infant PK profiles (if available), the stage is set for advanced simulations and sensitivity analyses. The prospective applications of the model can predict outcomes under varying conditions and to understand the impact of certain key parameters, particularly those associated with uncertainty. Sensitivity analysis plays a crucial role in identifying which parameters have a significant impact on the model's outcomes.

Maturation and ontogeny

When extending the application of the model to neonates, due to the sparse data on the ontogeny function of metabolizing enzymes and the maturation of biliary and renal excretion, there can be substantial uncertainty. When any of these pathways are predominant for the elimination of tested drugs, an inadequate understanding of their development in neonates can introduce significant uncertainty in the model predictions. To address this, sensitivity analyses may involve simulating different ontogeny scenarios, ranging from no ontogeny (no activity difference to adults) to slow ontogeny, to assess their potential impact on the drug exposure in neonates. ¹⁴ , ⁵²

Genetic polymorphisms

Genetic polymorphisms in drug‐metabolizing enzymes are another critical factor influencing drug metabolism and excretion. In lactation modeling, simulations can be extended to consider the effects of genetic variations within both mothers and infants. By simulating the IDD from mothers with various genetic polymorphisms and analyzing the corresponding drug concentrations in breastfed infants with the same or different genetic polymorphisms, we can better understand how composite maternal/infant genotypes may influence drug exposure. ¹⁵

Milk composition

The M/P ratio is sensitive to both the physicochemical characteristics of the drug and the dynamic composition of breast milk, which includes variations in aqueous, lipid, protein content, and pH levels. These changes in milk composition can significantly affect the solubility, ionization, and stability of drugs within the milk. Consequently, this impacts the M/P ratio, the estimated IDD, and the infant's drug exposure. Conducting sensitivity analysis across the spectrum of milk composition during various postpartum stages enables researchers to gain insights into how these variables can affect drug kinetics in breastfeeding scenarios.

Assessing worst‐case scenarios

The culmination of these simulations and sensitivity analyses is to evaluate the worst‐case scenarios to determine the maximum potential exposure of the drug in infants. By considering extremes, such as the highest possible drug concentrations in breast milk and the most unfavorable genetic polymorphisms, the model can help assess potentially the highest risk to the neonate. This approach ensures that the safety margin for drug use during lactation is established based on a conservative assessment.

Risk and benefit evaluation

The ultimate objective of the lactation PBPK model is to facilitate a comprehensive risk/benefit evaluation. The model allows prediction of infant drug exposure through breast milk consumption, balancing the necessity of maternal medication with the safety of the breastfeeding infant. By simulating various clinical and what‐if scenarios, the model aids in making informed decisions about drug use during lactation. When put in context of the safety margins of the drug and used alongside other available sources of lactation‐related data, PBPK modeling can provide valuable insight into the risk/benefits of drug exposure in breastfeeding infants.

EXAMPLES OF APPLICATIONS OF PBPK MODELING TO PREDICT CLINICALLY UNTESTED SCENARIOS

Case study 1: Gaining confidence with available clinical data, predicting IDD, and assessing preterm neonatal exposure scenarios

This example demonstrates a potential application of PBPK modeling to predict theophylline/caffeine exposures in preterm neonates during the first 2 weeks of life with different gestational ages. ¹⁶ Theophylline and caffeine are commonly used medications in preterm neonates, primarily for the treatment of apnea of prematurity. Theophylline undergoes an additional metabolic process in preterm neonates (not in adults) resulting in the formation of caffeine via CYP1A2 and CYP2E1.

Available clinical data for theophylline in maternal plasma and breast milk, as well as theophylline and caffeine data in neonates, were utilized to develop and verify the models in Simcyp simulator version 21. The neonate model utilized dynamic changes in growth physiology and ontogeny functions to describe the maturation of CYP1A2 and CYP2E1 enzymes and their impact on exposure.

The predicted M/P ratios were within the range reported in the literature (0.57–0.89), with the phase distribution model predicting 0.49 and the log‐transformed model predicting 0.87. As a pragmatic approach, the average M/P ratio of the two values, i.e., 0.68 (5% CV), was used to calculate the dose in the neonatal PBPK simulations. The theophylline plasma levels in preterm (at 28 and 32 GW) and term neonates (at 38 GW) from birth to 2 weeks were simulated by redefining the subjects over time, considering time‐varying physiology for the entire simulation duration.

Neonate doses were derived from the predicted IDD based on milk $C_{\mathrm{avg},\mathrm{ss}}$, (0.94 mg/kg/day) and $C_{\max ,\mathrm{ss}}$, (1.4 mg/kg/day as a worst‐case scenario) assuming 150 mL/kg/day and 4‐h feeding frequency during the first 2 weeks of life for preterm or term neonates. An additional scenario was simulated to predict the impact of in utero exposure to theophylline by using a bolus dose to set the initial concentration in neonatal plasma to match the observed concentration in the cord blood at birth, thereby mimicking the real clinical situation. These scenarios were compared with the therapeutic window (5–12 mg/L) defined for apnea of prematurity, with the dosage regimen (5 mg/kg theophylline infused over 30 min followed by 1 h‐infusion of 1.1 mg/kg/12 h) given to preterm neonates.

The simulation results indicated that the duration for which the theophylline concentration remained within the therapeutic range varied with gestational age due to the immaturity of elimination mechanisms at birth, which differ with the gestational age of the neonates. In the worst‐case scenario, without considering in utero exposure, the theophylline concentrations were lower than the therapeutic range. However, when in utero exposure was considered, the theophylline concentrations remained within the therapeutic range for 2–4 days post‐birth, indicating a potential therapeutic benefit in treating apnea. Simulation results suggested that the predicted caffeine levels in neonatal plasma across all scenarios were too low to exert a significant pharmacological effect if compared to the caffeine concentrations achieved with therapeutic doses used for treating apnea. Consequently, suggesting that these low levels of caffeine are unlikely to contribute to the pharmacological effect of theophylline.

Case study 2: Predicting exposure in breastfed neonates to assess treatment options in breastfeeding mothers with Plasmodium vivax malaria

Here we describe an application of PBPK modeling to supplement available clinical data to support the radical cure P. vivax in breastfeeding women with children of all ages. ¹⁴ Primaquine is one of two 8‐aminoquinoline drugs available to cure this disease and prevent its characteristic repeated relapses. The drug is a lipophilic diprotic base that is readily absorbed and mainly metabolized by MAO‐A in adults (approximately 90%) with a small contribution by CYP2D6. ⁵³ Clinical lactation data available in 20 mother‐infant pairs indicate minimal drug distribution into milk and undetectable drug concentration in infants. ⁵⁴ Based on these findings, the authors recommended that primaquine should not be withheld from mothers breastfeeding infants or young children but stated that more information in neonates was required.

The PBPK model, which was developed and published previously, was verified in Simcyp simulator version 21 using these and other clinically available data in adults, and children. ¹⁵ , ⁵³ An IDD estimated from the primaquine exposure in breast milk was then used in simulations of virtual infants (>28 days) with time‐varying physiology, including the drug‐metabolizing enzyme ontogenies. Simulations of virtual neonates (<28 days) for which there are no clinical PK data were then performed to bridge a knowledge gap.

As the clinical study did not control or report infant feeding times, it was assumed that the IDD was consumed across six feeding times (4 h apart). The resulting simulations indicated that primaquine exposures in breasted infants (>28 days) and neonates (<28 days) were around 1000‐fold lower than in the mother.

A key assumption in this case is the pattern of MAO‐A ontogeny in the absence of literature data. In addition, two simulations assuming no ontogeny (i.e., the neonate has adult levels of enzyme) and a slow ontogeny (assuming it takes some time for MAO‐A to reach adult levels) were run, and the results were compared in the context of maternal exposure. Even assuming a slow ontogeny, primaquine exposure in neonates remained approximately 200‐fold lower than in the breastfeeding mothers. The impact of changes in breastmilk composition on the predictions was also assessed. Although varying breast milk fat content (3.5%–4.5%) had minimal impact on M/P, changes in breast milk pH (7.6 to 7.2) increased M/P more than 2‐fold. In summary, the simulations were able to show that the concentrations of primaquine in breast milk are very low irrespective of the circumstances and are therefore unlikely to cause adverse effects in the breastfeeding infant.

Case study 3: Predicting impact of maternal and pediatric CYP2B6 genotype on the exposure in breastfed children

In this example PBPK modeling is used to describe the impact of CYP2B6 genotype‐specific efavirenz exposure in lactating women on exposures in their breastfed babies. ¹⁵ Efavirenz, used to treat human immunodeficiency virus infection, is a lipophilic monoprotic acid that is metabolized by and induces CYP2B6 and CYP3A4, causing auto‐induction. As such, exposure of efavirenz in breastfed babies is subject to a complex interplay of auto‐induction, pharmacogenetics, and ontogeny of CYP2B6 and CYP3A4, and penetration of the drug into milk.

A verified PBPK model in Simcyp simulator version 21 for efavirenz describing the CYP3A4‐ and CYP2B6‐mediated auto‐induction during multiple dosing was able to accurately predict efavirenz pharmacokinetics in African mothers at post‐partum with different CYP2B6 genotypes (516GG, GT, and TT). A virtual African pediatric population was developed and verified. The efavirenz model was used in conjunction with the African pediatric population to simulate efavirenz exposures in infants aged 3 to 36 months. Intensive efavirenz PK data available in 47 efavirenz naïve infants aged 3 to 36 months from TB‐endemic countries carrying either CYP2B6 516GG/GT or 516TT genotypes (>75% are African) were used for verification. Thereafter, the exposure of efavirenz in neonates based on different combinations of maternal and pediatric CYP2B6 genotype was predicted using the calculated IDD, which was split equally over six feeds per day (every 4 h) in the absence of information in the clinical study. The estimated average RIDD from CYP2B6 516GG, GT, and TT nursing mothers was less than 10%. The simulated infant efavirenz plasma concentrations at mid‐dose interval associated with composite maternal/infant CYP2B6 genotypes capture the range of clinically reported individual concentrations. In summary, this case study demonstrates a powerful application of PBPK modeling for a complex scenario where rich clinical data may be required to understand the impact of multiple mechanisms and covariates but are not likely to be available for most drugs.

FUTURE DIRECTIONS

The push from the clinicians and regulators to generate more information in breastfeeding mothers is likely to require increasing support from model‐informed efforts. Modeling approaches are being increasingly applied to predict exposure levels in infants as this is crucial for assessing the safety and potential effects of drugs transmitted through breast milk. By their very nature, PBPK models offer to provide such predictions, particularly in these under‐served populations. In this tutorial, we have demonstrated how prediction algorithms for M/P ratios embedded within a PBPK model framework, can be used for scenarios where clinical lactation data are not available or indeed to support untested clinical scenarios where such a study has been performed. While the choice of examples and case studies was mainly driven by the availability of clinical lactation data, which are more readily available in the global health scenarios, the findings are relevant to new drug development. The relevance of these case studies extends beyond their immediate context and contributes to a broader understanding of drug development in real‐world setting.

In this tutorial, two models for prediction of M/P ratios were discussed and applied in our case studies. Based on the information available, it appears that the phase distribution model tends to overestimate M/P ratios for acidic drugs and underestimate M/P ratios for basic and neutral drugs. Additionally, the log‐transformed model shows poor predictions of M/P ratios for acidic drugs. ³⁰ A more recent IVIVE‐based model proposed by Yang et al. ³⁰ offers an alternative prediction method and importantly, shows applicability for predicting the M/P ratio for drugs that are actively transported. Based on preliminary research, various uptake (OATP, OCT, OCTN, ENT, GLUT, MRP, MCT, LAT, NTCP, PEPT), and efflux transporters (MRP1, MRP5, P‐gp, BCRP) have been detected in human mammary epithelial cells. While the excretion of the vast majority of drugs is explained with a passive diffusion model, active transport into breast milk is implicated for some drugs (e.g., acyclovir, cimetidine, methotrexate, nifedipine, nitrofurantoin, atenolol, and metformin). A recent report from Alshogran and colleagues provided a comprehensive list of thirty‐one commonly used drug in lactation, of which thirteen drugs were identified as substrates for various transporters that are also expressed in the mammary gland. Among these, BCRP (ABCG2) is the most studied transporter in the mammary gland and may play the most significant role as it is highly expressed in the lactating mammary gland. ¹⁷ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ Additionally, P‐gp expression in the mammary gland is also believed to be a protective mechanism to limit the transfer of potentially harmful substances into breast milk, thereby protecting the nursing infant from exposure to certain drugs and toxins. Attempts have been made to derive models that also apply to actively transported drugs by either using QSAR, ⁶² or in vitro caco‐2 permeability data (P‐gp and BCRP substrates) ³⁰ , ⁶³ or human MECs (P‐gp substrates). ⁶⁴ Recently, a mechanistic PBPK modeling approach incorporating BCRP‐mediated transport kinetics for five BCRP drug substrates was reported. ⁶⁵ However, addressing the uncertainty in estimating the mammary gland surface area and transporter in vitro‐in vivo extrapolation (IVIVE) is warranted to gain more confidence in the model. Transfer experiments using human MECs may be promising for studying the effect of transporters on drug exposure in milk and potentially unravel the mechanisms involved. Furthermore, when combined with PBPK modeling, they could support quantitative predictions of medication concentrations in human milk via active transport.

As a best practice, most PBPK models first verify performance in non‐pregnant individuals before applying them to postpartum or lactating women. However, changes in postpartum physiology can significantly affect pharmacokinetics in postpartum women. These changes may be small or large depending on the drug, and inaccuracies in model predictions can be indicative of these physiological alterations. Any physiological changes during pregnancy take some time to return to non‐pregnant levels after delivery. These changes are more prominent in the immediate and early postpartum period and need to be considered in predictions in this period. Few reports have examined changes in postpartum physiology, but data on the immediate postpartum period is still sparse. ⁶⁶ , ⁶⁷ The most prominent changes include decreases in tissue volume, blood flow, and renal function, which are elevated during pregnancy. Conversely, there is an increase in binding proteins as they return to non‐pregnant levels. The timing of these changes is crucial, and quantitative consideration is required. This has potential for future incorporation into PBPK models to improve predictions.

It also needs to be recognized that limited availability of robust data from longitudinal studies and variability in measurements from the actual clinical lactation studies present problems. Specifically, the dynamic nature of breastfeeding with variations in milk composition between foremilk and hindmilk, as well as changes in infant feeding patterns and dynamic milk intake add to the complexity. However, with the burden of additional measurements, there is an existing gap in understanding the impact of colostrum, foremilk, and milk pH in clinical practices and their implications in trials. These aspects are often overlooked and require further exploration as more clinical data becomes available. In order to be confident in the predictions, there has been to be some level of confidence in the clinical data being generated and an improved understanding of the impact of the aforementioned issues on drug exposures in milk. While the PBPK model flexibility allows the incorporation of available data at various stages of lactation, integrating longitudinal data for dynamic changes in milk parameters can facilitate the development of dynamic models to simulate the effect of corresponding changes in drug disposition. It is envisaged that the increasing number of lactation studies being conducted will ultimately lead to improved understanding of drug transport into milk but also provide more information on the systems data required to drive the PBPK models. More clinical data will allow more verification of the PBPK models to be performed and build more confidence in this area or at least identify knowledge gaps.

Addressing these challenges requires interdisciplinary collaboration and innovative approaches. Advancements in PBPK modeling techniques, coupled with ongoing research on lactation physiology and drug disposition, hold promise for further enhancing our understanding of drug transfer during lactation.

FUNDING INFORMATION

No funding was received for this work.

CONFLICT OF INTEREST STATEMENT

A.P., X.P., L.M.A., and K.R.Y. are employees of Certara UK Limited (Predicted Technologies Division) and may hold shares in Certara.

DISCLAIMER

As Deputy Editor in Chief of CPT: Pharmacometrics & Systems Pharmacology, Karen Rowland Yeo was not involved in the review or decision process for this paper.

Pansari A, Pan X, Almond LM, Rowland‐Yeo K. A tutorial on physiologically based pharmacokinetic approaches in lactation research. CPT Pharmacometrics Syst Pharmacol. 2024;13:1841‐1855. doi: 10.1002/psp4.13232