Ontogeny of Scaling Factors for Pediatric Physiologically Based Pharmacokinetic Modeling and Simulation: Cytosolic Protein Per Gram of Liver

Abstract

Scaling factors are necessary for translating in vitro drug biotransformation data to in vivo clearance values in physiologically-based pharmacokinetic modeling and simulation. Values for microsomal protein per gram of liver are available from several sources for use as a scaling factor to estimate hepatic clearance from microsomal drug biotransformation data. However, data regarding the distribution of cytosolic protein per gram of liver (CPPGL) values across the lifespan are limited, and sparse pediatric data have been published to date. Thus, CPPGL was determined in 160 liver samples from pediatric (n = 129) and adult (n = 31) donors obtained from multiple sources: the University of Maryland Brain and Tissue Bank, tissue retrieval services at the University of Minnesota and University of Pittsburgh, and Sekisui-XenoTech. Tissues were homogenized and subjected to differential centrifugation to isolate cytosolic fractions. Cytosolic protein content was determined by BCA assay. CPPGL varied from two- to sixfold within each age group/developmental stage. Tissue source and sex did not contribute substantially to variability in protein content. Regression analyses revealed minimal change in CPPGL over the first two decades of life (logCPPGL increases 0.1 mg/g per decade). A mean ± S.D. CPPGL value of 44.4 ± 17.4 mg/g or median 41.0 mg/g is representative of values observed between birth and early adulthood (0–18 years, n = 129).

SIGNIFICANCE STATEMENT

Cytosolic protein per gram of liver (CPPGL) is a scaling factor required for physiologically based pharmacokinetic modeling and simulation of drug biotransformation by cytosolic enzymes, but pediatric data are limited. Although CPPGL varies from two- to sixfold within developmental stages, a value of 44.4 ± 17.4 mg/g (mean ± S.D.) is representative of the pediatric period (0–18 years, n = 129).

Introduction

Physiologically based pharmacokinetic (PBPK) modeling and simulation has emerged as an important tool for drug development, facilitating the combination of preclinical and clinical data to predict drug disposition in the target population. Indeed, over the past decade there has been a steady increase in the number of new drug applications that use PBPK modeling (Grimstein et al., 2019). In these applications, PBPK modeling has been employed to understand drug-drug interaction potential and drug disposition and aid in dose selection and exposure optimization (Grimstein et al., 2019; Zhang et al., 2020). In 2019, nearly half (45%) of small molecule new drug approvals contained PBPK modeling, a twofold increase from 2014 (Zhang et al., 2020). The predictive accuracy of PBPK models depends, in part, on understanding the physiologic characteristics of the target population. Scaling factors are often used to address special considerations within a population (e.g., allometric scaling for pediatric patients) to more accurately model drug clearance.

In the context of PBPK modeling and simulation, common scaling factors include microsomal protein per gram of liver (MPPGL), hepatocellularity, and liver mass for bottom-up extrapolation of in vitro drug biotransformation using subcellular fractions or cellular systems to organ and whole-body clearance. Models often employ a single value representative across a population of virtual patients, and sample sizes are generally insufficient to assess the potential impact of biologic influences like age, sex, and disease state that may influence values for key scaling factors. To address this challenge, researchers have attempted to generate more robust estimations, such as meta-analysis of published MPPGL values from different studies (Barter et al., 2007). Recently, our group reported the developmental trajectory of MPPGL between birth and 18 years of age in a set of pediatric liver microsomes (Leeder et al., 2022). Overall, the study concluded that there was minimal change in logMPPGL values over this age range and revealed the limitations inherent when tissues are obtained from multiple sources with different procurement and processing procedures. Addressing important knowledge deficits, such as variability in MPPGL across the pediatric age range, can be expected to improve the performance of PBPK modeling and simulation, specifically for drug clearance that involves microsomal enzymes, e.g., cytochrome P450 (P450) and UGT biotransformation. Opportunities remain to increase the quantity and quality of data supporting PBPK modeling and simulation for drugs that undergo clearance by alternative pathways.

Although microsomal P450 enzymes are responsible for the majority of drug biotransformation reactions, cytosolic enzymes contribute to a variety of biotransformation pathways for both exogenous and endogenous substrates. In recent years, there has been growing emphasis on the development of drug candidates that undergo non-CYP–mediated metabolism to reduce drug-drug interaction potential. One study evaluating new drug approvals by the United States Food and Drug Administration found that approximately 30% of new drugs approved between 2006 and 2015 involved non-P450–mediated biotransformation (Cerny, 2016). Many non-P450 drug-metabolizing enzymes are present in the cytosol, such as sulfotransferases, carbonyl reductases, methyltransferases, N-acetyltransferases, esterases, epoxide hydrolases, and aldehyde oxidase (Saravanakumar et al., 2019; Xie and Xie, 2020). As such, cytosolic enzymes are involved in the clearance of many drugs currently on the market, such as abacavir, bupropion, hydroxyzine, naltrexone, oxcarbazepine, doxorubicin, and daunorubicin (Porter et al., 2000; Bains et al., 2010; Bamfo et al., 2023). For PBPK modeling and simulation using these nonmicrosomal pathways, robust estimates of relevant scaling factors are essential for optimal predictive accuracy of the models (Doerksen et al., 2021).

The evidence base for cytosolic protein per gram of liver (CPPGL) in humans is growing but essentially limited to values derived from adults. To date, investigations into CPPGL have been in a small number of adult livers (n = 4 to n = 23) from various groups (Pacifici et al., 1988; Wynne et al., 1992; Boogaard et al., 1996; Gibbs et al., 1998; Mutch et al., 2007; Doerksen et al., 2021). Some have evaluated the impact of liver disease (e.g., cirrhosis, cancer) on CPPGL in relation to healthy tissue (El-Khateeb et al., 2020; Vasilogianni et al., 2021). Renwick et al. (2002) reported mean CPPGL values from three liver donors aged 10, 11, and 50 years without mention of the range or individual values for the two pediatric samples. Thus, sparse pediatric data are available. The purpose of this investigation was to characterize the ontogeny of CPPGL for use in PBPK modeling and simulation of non-P450–mediated biotransformation in pediatric populations.

Methods

Materials and Reagents

All chemicals were of reagent grade and purchased from either Sigma Millapore Chemical Co. or Thermo-Fisher Scientific (Fairlawn, NJ).

Liver Samples

This study involved a total of 160 liver samples (n = 129 pediatric and n = 35 adult) obtained from various sources. The primary sources of tissue were the Eunice Kennedy Shriver National Institute of Child Health and Human Development–supported tissue retrieval program at the Brain and Tissue Bank for Developmental Disorders at the University of Maryland (Baltimore, MD; now the University of Maryland Brain and Tissue Bank and a member of the National Institutes of Health NeuroBioBank network; n = 61, including n = 53 pediatric, and n = 8 adult samples) and the National Institutes of Health–supported Liver Tissue Cell Distribution System, with sites at the University of Minnesota (n = 34 pediatric) and the University of Pittsburgh (n = 37, comprised of n = 14 pediatric and n = 23 adult samples). Additional pediatric liver samples were obtained from XenoTech, LLC (Kansas City, KS; n = 23), Vitron (Tucson, AZ; n = 4), the University of Miami (n = 1), and the Association of Human Tissue Users (n = 1). Supplemental Table 1 provides a list of liver samples by age and source. Supplemental Table 2 provides demographic data, including reported cause of death (when available) and individual donor CPPGL values. Samples ranged in age from birth to 79 years of age; 57/160 (36%) were female, 102 (64%) were male, and sex was unknown for one sample. Reported race was as follows: African American n = 40 (25%), Caucasian n = 76 (48%), Hispanic n = 7 (4%), Native American n = 1, and Pacific Islander n = 1; race was not reported for 35 samples (22%), including all samples from the University of Minnesota (n = 34). Use of the tissue samples was classified as nonhuman subjects research by the Children’s Mercy Pediatric Institutional Review Board. Tissues were stored at or below −70°C.

Preparation of Human Liver Cytosol

Human liver microsomes and cytosols were prepared by differential centrifugation as described by Lu and Levin (1972) and as previously reported (Leeder et al., 2022). Briefly, preweighed, frozen liver samples were placed in homogenizing buffer (∼3 mL/g liver; 50 mM Tris.HCl, pH 7.4, at 4°C, containing 150 mM KCl and 2 mM EDTA) and allowed to thaw at 4°C for 15–20 minutes. Liver samples were quickly minced with dissecting scissors on ice, placed in Potter-Elvehjem–type glass mortars (round bottom) and homogenized on ice with a Polytron tissue homogenizer (Kinematica USA, Bohemia, NY) using 3- to 4-second bursts of grinding for 1 to 2 passes. Liver samples were subjected to further homogenization (3 to 4 strokes on ice, 2 to 3 passes) in the glass mortars with Teflon pestles utilizing a motor-driven tissue homogenizer (Caframo Model BDC-3030, Wiarton, ON, Canada). Homogenates were placed into low-speed centrifuge tubes, filled with homogenization buffer, briefly mixed, an aliquot of homogenate removed, and the volume recorded. Subsequently, nuclei and lysosomes were removed from the homogenate by centrifugation (800 g_max for 15 minutes at 4°C). The resulting supernatant was further centrifuged (12,000 g_max for 20 minutes at 4°C), and the supernatant fraction was subjected to ultracentrifugation (105,000 g_max for 70 minutes at 4°C). The resulting supernatant (cytosol) was stored at −80°C for future cytosolic assays. Protein concentrations were determined with a Micro BCA Protein Assay kit (Pierce Chemical Co., Rockford, IL) using bovine serum albumin (Sigma, St. Louis, MO, USA) as the standard.

Estimation of Liver Cytosolic Protein Content

CPPGL was calculated using the following equations:

where,

Removal of the subcellular fractions as described above from the homogenate volume was reflected in Volume_Cyt and accounted for ≤1% of total volume.

Data and Statistical Analyses

To characterize the relationship between CPPGL and developmental stage, liver samples were stratified according to recommendations of the Eunice Kennedy Shriver National Institute of Child Health and Human Development (Williams et al., 2012). Specifically, samples were categorized as neonatal (birth to 28 days; n = 3), infant (29 days to 12 months; n = 17), toddler (13 months to 2 years; n = 9), early childhood (2–5 years; n = 20), middle childhood (6–11 years; n = 32), early adolescence (12–18 years; n = 47), and late adolescence (19–21 years; n = 0). Adult samples were arbitrarily divided into two additional categories of younger adult (21–50 years; n = 16) and older adult (>50 years; n = 15). ANOVA was used to compare CPPGL values across age groups, with Tukey’s honestly significant different test used for post hoc analysis.

In a multivariate analysis, CPPGL was modeled as a function of age, adjusting for sex, by fitting a linear mixed model in SAS 9.4. Because most sources included no samples beyond early adolescence, the analysis was limited to samples with an age of 0–18. CPPGL was log transformed for modeling to yield residuals consistent with the mixed model assumption of normal errors. A random source intercept was included to adjust for clustering of samples within sources. Between-within degrees of freedom were specified to adjust for the small number of clusters.

A subset of data was also used to evaluate CPPGL values as a function of age across the lifespan. The dataset used for this analysis, hereafter referred to as the “Pittsburgh” dataset, included samples from donors aged 5–79 years (n = 37) from a single source (Pittsburgh). All the liver donors >18 years of age in our study were either from this source or the University of Maryland Brain and Tissue Bank for Developmental Disorders, and the latter source included only two samples from donors with an age >25 years, making its data unsuitable for modeling CPPGL across the lifespan. CPPGL was modeled as a function of sex and age in three models: one with a linear effect for age and two allowing for a nonlinear effect of age with a one- or two-knot spline. The three models were compared using tenfold crossvalidation to identify the model with the strongest predictive performance for utility in PBPK modeling. Crossvalidation provides an estimate of a model’s ability to predict new data (i.e., data not used in the original model fit) and protects against over-fitting. The model with the lowest mean squared error across crossvalidation folds was chosen as the final model. CPPGL was not log transformed for this analysis as residuals for the final model were roughly normal.

All statistical analyses were conducted in JMP Pro version 14.2.0, SPSS version 24 (IBM, Armonk, NY), SAS version 9.4 (SAS Institute Inc., Cary, NC), or R (R Core Team, Vienna, Austria).

Results

General Distribution of CPPGL Values

Over the combined cohort (n = 160), CPPGL values varied 6.6-fold (16.6 mg/g to 109 mg/g). Mean CPPGL over the entire age range (n = 160, 0–79 years) was 50.2 ± 20.8 mg/g, with a median of 46.6 mg/g and interquartile range of 33.5–64.5 mg/g.

Variability in CPPGL: Impact of Age

The distribution of CPPGL as a function of age is illustrated in Fig. 1. CPPGL within age group/developmental stage is displayed in Fig. 1A and with age as a continuous variable in Fig. 1B. Due to skewness, CPPGL values are plotted on the log scale. Individual samples are color-coded by Eunice Kennedy Shriver National Institute of Child Health and Human Development age group. Because there were only three samples within the neonatal age range, they were combined with the “infant” group (n = 20) and designated with a white “x” within the corresponding symbol.

Fig. 1.

Relationship between CPPGL values and age. (A) Age is expressed as a categorical variable using National Institutes of Health–recommended age strata (Williams et al., 2012): group 1, infancy (28 days to 12 months of age); group 2, toddler (13 months to 2 years of age); group 3, early childhood (2–5 years of age); group 4, middle childhood (6–11 years of age); group 5, early adolescence (12–18 years of age). For reference, groups 6 and 7 represent younger adults (19–50 years of age) and older adults (>50 years of age), using an age of 50 years as an arbitrary cutoff. Box plots were constructed using the “outlier” format in JMP Pro 14.3. Boxes are defined by the first and third quartiles (25th and 75th quantiles, respectively), and the median is indicated by the horizontal line within the boxes; the sample mean is indicated by the dashed horizontal line. The whiskers extend from the ends of the box to the outermost datapoint that falls within the distance calculated as the third quartile + 1.5*(interquartile range) at the upper bound and the first quartile − 1.5*(interquartile range) at the lower bound; points extending beyond the whiskers are considered outliers. (B) Age is presented as a continuous variable; “B” indicates day of “birth.”

A summary of CPPGL distribution in each age group/developmental stage is also provided in Table 1. For the pediatric age range (0–18 years, n = 129), mean ± S.D. CPPGL values were 44.4 ± 17.4, with a median of 41.0 mg/g. For the adult age range (25–79 years, n = 31), mean ± S.D. CPPGL values were 74.3 ± 16.4, with a median of 73.0. The mean difference between pediatric and adult samples was −30.0 mg/g (95% confidence interval, −38.0 to −21.9; P < 0.001).

TABLE 1.

Distribution of CPPGL values within each age group or developmental stage according to Eunice Kennedy Shriver National Institute of Child Health and Human Development recommended age strata. The dataset included only three neonatal samples (birth to 28 days postnatal age) that were included in the “infant” group.

Group	Category	N	Mean	S.D.	Median	IQR	Minimum	Maximum	Fold Range
			(mg/g)	(mg/g)	(mg/g)	(mg/g)	(mg/g)	(mg/g)	(mg/g)
1	Infant	20	41.0	22.6	35.0	25.5–59.7	18.4	109.1	5.9
2	Toddler	9	42.0	14.2	40.4	30.2–54.4	26.5	65.1	2.5
3	Early childhood	21	43.0	14.1	46.9	30.1–51.6	16.6	77.1	4.6
4	Middle childhood	32	43.7	14.2	40.7	31.7–49.1	19.9	82.8	4.2
5	Early adolescent	47	47.4	18.8	42.5	32.6–60.5	18.1	98.6	5.5
6	Younger adult	16	74.6	12.4	73.2	67.7–81.4	53.2	100.3	1.9
7	Older adult	15	74.1	20.2	70.6	54.2–92.4	50.9	107.9	2.1
All		160	50.2	20.8	46.6	33.5–64.5	16.6	109.1	6.6

IQR, interquartile range.

The distribution of CPPGL labeled by tissue source across the sample age range is illustrated in Fig. 2. CPPGL within age group/developmental stage is displayed in Fig. 2A and with age as a continuous variable in Fig. 2B. Neonatal samples were combined with the “infant” group as described for Fig. 1.

Fig. 2.

CPPGL values as a function of age with distribution of tissue source. (A) Age is expressed as a categorical variable using National Institutes of Health–recommended age strata (same as Fig. 1A). Box plots were constructed using the “outlier” format in JMP Pro 14.3. Boxes are defined by the first and third quartiles (25th and 75th quantiles, respectively), and the median is indicated by the horizontal line within the boxes; the sample mean is indicated by the dashed horizontal line. The whiskers extend from the ends of the box to the outermost datapoint that falls within the distance calculated as the third quartile + 1.5*(interquartile range) at the upper bound and the first quartile − 1.5*(interquartile range) at the lower bound; points extending beyond the whiskers are considered outliers. (B) Age is presented as a continuous variable. For both (A) and (B), data points are colored according to source of tissues: University of Maryland Brain and Tissue Bank for Developmental Disorders (UMBTB) (green), University of Minnesota (red), University of Pittsburgh (blue), Sekisui-Xenotech (gray), Vitron (yellow), and other (University of Miami and Association of Human Tissue Users, one sample each). The three points with a black “x” in the center (all from UMBTB; green) represent neonatal samples from the day of birth to 28 days postnatal age. B, day of birth.

Variability in CPPGL as a Function of Age and Source: Mixed Model

A linear mixed model was fit to the data from postnatal donors aged 0–18 from all sources to examine source and age as contributors to variability in logCPPGL (Table 2). Random source intercept variance was near zero [intraclass correlation coefficient (ICC) < 0.0001], suggesting little evidence for source effects. On average, each decade within the first two decades of life was associated with a small increase in logCPPGL (0.1 mg/g or 0.27 S.D.).

TABLE 2.

Results of linear mixed model for logCPPGL as a function of age and sex for samples from donors aged 0–18 years.

	Coefficient (95% CI)	Scaled Coefficienta (95% CI)	P Value
Intercept	3.6 (3.5, 3.8)	3.4 (3.0, 3.8)
Age (decades)	0.1 (0.0, 0.2)	0.3 (0.0, 0.6)	0.079
Female sex	0.1 (0.0, 0.2)	0.3 (−0.1, 0.6)	0.165

CI, confidence interval.

^aLogCPPGL scaled to have S.D. 1.

Variability in CPPGL across the Lifespan: Predictive Model Exploration

Using the Pittsburgh dataset comprised of samples from donors aged 5–79 years (n = 37), we evaluated the predictive power of linear and spline models for the association between age and CPPGL. The one-knot spline model (Fig. 3) demonstrated the strongest predictive performance based on mean squared error across the 10 crossvalidation folds. Fit to the entire sample, the model yielded R² = 0.52. Sex was included as a covariate but did not contribute meaningfully [female: β (95% confidence interval) = −1.0 (−12.0, 9.9); P = 0.851]. The fitted model equation was CPPGL-hat = 34.4 − 1.0*Female + 71.8*Spline basis variable 1 + 3.6*Spline basis variable 2, where the spline basis was computed using age in decades with a single knot at the median age of 41 years.

Fig. 3.

The impact of age across the lifespan on CPPGL. In an exploratory subanalysis using the Pittsburgh dataset (n = 37, age 0–79 years), a one-knot spline model was used to describe age trajectory.

Discussion

CPPGL is a scaling factor critical for extrapolating drug biotransformation by cytosolic enzymes to hepatic clearance in PBPK modeling and simulation. There are limited data regarding CPPGL values across the lifespan, particularly in pediatrics. We report CPPGL values in a relatively large cohort of samples across the lifespan and describe the impact of age, sex, and tissue source.

CPPGL values varied 5.9-fold in the first year of life and 5.5-fold in the adolescent period. Variability in CPPGL appeared to decrease with age, with twofold variability detected in adulthood. The range of CPPGL values reported in our study (16.6–109.1 mg/g liver tissue) is consistent with the ranges reported by other groups, 24.8–134 mg/g (Cubitt et al., 2011; El-Khateeb et al., 2020; Doerksen et al., 2021; Vasilogianni et al., 2021). Representative CPPGL values in our pediatric samples (median, 41.0 mg/g; mean, 44.4 mg/g in n = 129; age range, 0–18 years) are lower than those reported in healthy adult samples by El-Khateeb et al. (2020) (median, 75.4 mg/g in n = 13; age range, 36–83 years), Vasilogianni et al. (2021) (mean, 56.2 mg/g in n = 16; age range, 34–85 y), and Cubitt et al. (2011) (weighted mean, 79.8 mg/g in n = 49 from four published sources; known age range, 36–88 years) and lower than our own adult samples (median, 73.0 mg/g; mean, 74.3 mg/g in n = 31; age range, 25–79 years). Scaling factors most relevant to the target population of interest should be used, and the pediatric CPPGL values reported here will enable pediatric-specific in vitro–in vivo extrapolation.

We evaluated the impact of tissue source on CPPGL values and found little evidence for a source effect of the same magnitude as we reported for MPPGL values despite the fact that the cytosolic fractions were obtained from the ultracentrifugation process and from the same liver donor cohort as the previously reported MPPGL results (Leeder et al., 2022). The lack of apparent source effect on CPPGL values is not immediately evident but may be related to the differences in how MPPGL and CPPGL are calculated. Calculation of MPPGL includes a measure of enzyme activity, such as the cytochrome P450 oxidoreductase (POR) activity (Leeder et al., 2022), to account for microsomal recovery. In the case of POR, enzyme activity appeared to be influenced by tissue source as a proxy for differences in procurement as POR activity in liver homogenates, and POR activity and protein content in microsomes were significantly lower in autopsy samples with post mortem intervals of up to 15 hours compared with liver tissues flash frozen within 1 hour or perfused with buffer after procurement. Furthermore, POR activity per nmol of POR protein was also lower in autopsy samples compared with flash-frozen or perfused liver (Leeder et al., 2022). In contrast to potential loss of microsomes during processing, recovery of cytosolic protein is assumed to be nearly complete, and CPPGL is reported without correction for recovery and thus does not include any assessment of protein activity. Limited attempts have been made to correct for cytosol recovery using a marker of cytosol enzyme activity. Scotcher et al. (2017) used glutathione S-transferase activity to characterize cytosol activity in human kidney. However, this marker was not specific to cytosol as activity was detectable in the microsomal fraction as well. This resulted in overestimation in nearly 20% of their sample (i.e., protein concentrations in cytosolic and microsomal fractions exceeded total protein concentration in homogenate). Other attempts at determining CPPGL did not describe correction for cytosolic recovery (Doerksen et al., 2021).

Part of the utility of our findings lies in the potential of model-based prediction of CPPGL values in a new sample. We explored the predictive performance of various models and found that a one-knot spline model performed best, explaining half of the variability in CPPGL values (R² = 0.52). Although a linear model was examined, it had inferior predictive power and did not visually fit the data. We recognize that different investigators may approach the same data from different perspectives and underlying assumptions, and therefore, individual sample level demographic data and CPPGL values can be found in Supplemental Table 1 for additional independent analyses. Corresponding MPPGL values can be obtained in the supplemental data accompanying Leeder et al. (2022).

Our findings advance the potential of PBPK modeling and simulation by contributing data regarding the ontogeny of the cytosolic scaling factor, CPPGL, to accompany previously published data on the developmental trajectory of MPPGL derived from the same tissues. Although CPPGL values per se did not appear to be influenced by tissue procurement and processing in this study, we have observed source effects analogous to those described for microsomal POR activity and protein in the measurement of naltrexol formation (attributed primarily to aldoketoreductase family 1, member C4) in this set of liver cytosols (Stancil et al., 2022). Indeed, when we calculated predicted liver clearance from CPPGL values reported here and naltrexol biotransformation rates in individual donors reported in Stancil et al. (2022), the role of tissue source becomes apparent when cytosolic enzyme activity is part of the calculation (Supplemental Fig. 1). Given that cytosolic enzyme activities, such as naltrexol formation, represent the initial in vitro data to simulate naltrexone clearance in vivo by PBPK modeling, tissue procurement and processing remain important considerations for generating accurate and reliable pediatric data. Therefore, we join Doerksen et al. (2021) in encouraging standardization of protocols for procurement, processing, and storage of pediatric tissues as a relatively rare, but valuable, resource to promote accurate modeling and simulation for pediatric applications.

Abbreviations

CPPGL

cytosolic protein per gram liver

MPPGL

microsomal protein per gram liver

P450

cytochrome P450

PBPK

physiologically based pharmacokinetic

POR

cytochrome P450 oxidoreductase

Authorship Contributions

Participated in research design: Stancil, Pearce, Leeder.

Conducted experiments: Pearce.

Performed data analysis: Stancil, Staggs, Leeder.

Wrote or contributed to the writing of the manuscript: Stancil, Pearce, Staggs, Leeder.