An in vitro-in silico workflow for predicting renal clearance of PFAS

Abstract

Per- and poly-fluoroalkyl substances (PFAS) have a wide range of elimination half-lives (days to years) in humans, thought to be in part due to variation in proximal tubule reabsorption. While human biomonitoring studies provide important data for some PFAS, renal clearance $\left({\text{CL}}_{\text{renal}}\right)$ predictions for hundreds of PFAS in commerce requires experimental studies with in vitro models and physiologically-based in vitro-to-in vivo extrapolation (IVIVE). Options for studying renal proximal tubule pharmacokinetics include cultures of renal proximal tubule epithelial cells (RPTECs) and/or microphysiological systems. This study aimed to compare ${\text{CL}}_{\text{renal}}$ predictions for PFAS using in vitro models of varying complexity (96-well plates, static 24-well Transwells and a fluidic microphysiological model, all using human telomerase reverse transcriptase-immortalized and OAT1-overexpressing RPTECs combined with in silico physiologically-based IVIVE. Three PFAS were tested: one with a long half-life (PFOS) and two with shorter half-lives (PFHxA and PFBS). PFAS were added either individually (5 μM) or as a mixture (2 μM of each substance) for 48 hours. Bayesian methods were used to fit concentrations measured in media and cells to a three-compartmental model to obtain the in vitro permeability rates, which were then used as inputs for a physiologically-based IVIVE model to estimate in vivo ${\text{CL}}_{\text{renal}}$. Our predictions for human ${\text{CL}}_{\text{renal}}$ of PFAS were highly concordant with available values from in vivo human studies. The relative values of ${\text{CL}}_{\text{renal}}$ between slow- and faster-clearance PFAS were most highly concordant between predictions from 2D culture and corresponding in vivo values. However, the predictions from the more complex model (with or without flow) exhibited greater concordance with absolute ${\text{CL}}_{\text{renal}}$. Overall, we conclude that a combined in vitro-in silico workflow can predict absolute ${\text{CL}}_{\text{renal}}$ values, and effectively distinguish between PFAS with slow and faster clearance, thereby allowing to prioritize PFAS with a greater potential for bioaccumulation in humans.

1. Introduction

Per- and poly-fluoroalkyl substances (PFAS), a large class of man-made fluorinated organic chemicals, have widespread uses in various industrial and commercial products. It is estimated that over 1,400 PFAS are used in over 200 industrial applications (Lu et al., 2024) and over 10,000 PFAS have been listed on various chemical inventories (OECD, 2018). Many of these chemicals are resistant to degradation and are both persistent in the environment and bioaccumulative in a variety of organisms; importantly, elimination half-lives of many PFAS differ widely among mammals with humans being the species with the slowest clearance (Pizzurro et al., 2019; Schulz et al., 2020). Additionally, Panieri et al. (2022) reviewed numerous studies highlighting the long elimination half-lives of PFAS in humans, leading to potentially increased toxicity and human health concerns.

Understanding the elimination of PFAS is essential for human health assessment of these chemicals. The main elimination routes for PFAS are through bile/feces and urine (Lu et al., 2024). Although, for some PFAS, renal clearance has been found to contribute to a smaller (8 – 46%) proportion of the total clearance (Harada et al., 2007; Fujii et al., 2015), it is a leading hypothesis that the long half-lives of some PFAS in humans are due to efficient reabsorption in the kidneys (Harada et al., 2005; Fujii et al., 2015; Pizzurro et al., 2019). Therefore, information on renal clearance can distinguish PFAS with slower or faster elimination, corresponding with higher or lower levels of potential health concerns, respectively. Moreover, active reabsorption of PFAS in the renal proximal tubule creates challenges when using traditional reverse toxicokinetics-based in vitro-to-in vivo extrapolation (IVIVE) that assume glomerular filtration as the primary source of renal clearance (Wilkinson and Shand, 1975; Pearce et al., 2017).

Because there are limited in vivo data on renal clearance for most environmental chemicals, including PFAS, in vitro testing followed by IVIVE is as a pragmatic approach to fill in existing data gaps (Scotcher et al., 2016a; Scotcher et al., 2016b; Chang et al., 2022). For instance, Neuhoff et al. (2013) proposed an IVIVE scaling approach, implemented in the commercial SimCYP package (Jamei et al., 2013), that begins with transporter activity data from transfected cell lines, and then uses a series of scaling factors. However, this approach has numerous uncertainties due to data limitations, such as relative abundances and membrane localizations of different transporters in vitro vs in vivo. A more empirical approach was proposed by Kunze et al. (2014) for estimating renal clearance based on IVIVE from bidirectional permeability studies in a Transwell-based culture using LLC-PK₁ cells and data from 20 compounds. However, this method’s potential application to PFAS is uncertain because the compounds studied by these authors had much shorter half-lives (hours to days) as compared to PFAS (weeks to months), and because drugs and PFAS have distinct physicochemical properties, such as presence of fluorine-carbon bonds in the latter.

In the context of the in vitro proximal tubule cell-based assays, the utilization of non-human (pig-derived) cell lines such as LLC-PK₁ has the potential to introduce disparities between predicted and observed human in vivo renal clearance due to concerns about species differences. The Tox21 report (Krewski et al., 2020) has advocated for the characterization of human renal cells (primary and immortalized) as models for evaluating the toxicological effects of environmental chemicals. One such widely used model (Simon et al., 2014; Mihevc et al., 2020) is human telomerase reverse transcriptase (hTERT)-immortalized human renal proximal tubule cells (RPTECs). Beyond consideration of cell types, the complexity of the in vitro models can also potentially impact the prediction of kinetic-related parameters through the application of IVIVE. Microphysiological systems, including tissue chips, have the potential to mitigate certain constraints inherent in traditional 2D cultures, attributed to the incorporation of fluid shear stress and the more faithful replication of the human physiological microenvironment (Marx et al., 2020). However, there is limited research examining the influence of varying complexities of in vitro assays (e.g., 2D vs 3D culture; single vs multi-chambered labware; static vs fluidic models) on the prediction of renal clearance.

In this study, we aimed to enhance prediction of PFAS renal clearance by combining human cell-based in vitro testing of varying complexity with a three-compartment in silico model for the in vitro system and physiologically-based in silico modeling for kidney clearance. We used the TERT1-OAT1 cell line of RPTEC cells grown in 96-well plates, as well as in static or fluidic Transwells to measure renal uptake, secretion, and retention of three PFAS. Concentration-time course data obtained from these in vitro studies were fitted with a chip-based compartmental model using the Bayesian Markov chain Monte Carlo algorithm to determine in vitro permeability across the cell and lumen/blood compartments. Using these data, we developed a physiologically-based kidney model to extrapolate in vitro transport and uptake parameters to predictions of in vivo renal clearance.

2. Materials and Methods

The overall study workflow is presented in Fig. 1. There were two phases of the workflow: (1) in vitro transport experiments using OAT1-overexpressing RPTECs cultured in platforms of varying complexity, and (2) in silico modeling of the experimental data and physiologically-based IVIVE to renal clearance. Experimental details are provided as follows.

Figure 1.

Schematic of the overall in vitro-in silico workflow for predicting renal clearance.

2.1. Materials

2.1.1. Tested chemicals

Perfluorohexanoic acid (PFHxA; Cat# U0067, CASRN: 307-24-4) and perfluoro-butanesulfonic acid (PFBS; Cat# N0709, CASRN: 375-73-5) were purchased from TCI America (Portland, OR). Perfluorooctanesulfonic acid (PFOS; Cat# 6164-3-08, CASRN: 1763-23-1) was purchased from Synquest Labs (Alachua, FL). C¹³-PFOA was purchased from Wellington Lab (Cat# MPFOA; Guelph, Ontario, Canada), and used as an internal standard for all PFAS compounds.

2.1.2. Cell culture and Exposure

The human RPTEC TERT1-OAT1 was obtained from ATCC (Cat# CRL-4031-OAT1, Manassas, VA). Cells were cultured in DMEM:F12 (Cat# 30-2006, ATCC) supplemented with the hTERT Immortalized RPTEC Growth Kit (Cat# ACS-4007, ATCC), and a final concentration of 0.1 mg/mL G418 (Geneticin^™ G418 Sulfate; Cat# 10131035, Gibco, Waltham, MA). This cell line is stably expressing OAT1 from the vendor, and pure populations were maintained through the use of G418 to ensure only the transfected population is present. RPTECs were cultured at 37°C and 5% CO₂ in Tissue Culture-treated 75 cm² flasks (Cat# CLS430641U, Corning, Corning, NY) prior to seeding onto Transwells. On the day of seeding, 24-well Transwells (Cat# 3413, Corning) were inverted, and the bottom surface was coated with 90 μL of 100 μg/mL fibronectin (Cat# F1141-5MG, Sigma-Aldrich, St. Louis, MO) in phosphate buffered saline (PBS; Cat# 20012027, Gibco) for 1 hour at 37°C. After coating, excess solution was removed and cells were seeded onto the bottom surface at a density of 625,000 cells/mL in cell culture media (50,000 cells in 80 μL), and allowed to attach for 4 hours at 37°C. After incubation, Transwells were re-inverted into the culture plates, and medium was added (750 μL to the bottom, and 150 μL to the top well). Cell-seeded Transwells were incubated overnight, then either left in the plates as-is (static condition) or transferred to the T12 Barrier plate (fluidic condition, 2 μL/s media flow in the bottom chamber) on the PhysioMimix^® organ-on-chip platform (both from CNBio, Cambridge, UK). Cell culture media were exchanged in both static and fluidic cultures every 48 hours. Cells were cultured under static or fluidic conditions for 7 days prior to PFAS exposure, and barrier formation was confirmed via Transepithelial Electrical Resistance (TEER) testing to ensure monolayer formation using the EVOM2 epithelial voltammeter (Cat# STX2, World Precision Instruments, Sarasota, FL). Cultures with TEER less than 65 Ω‧cm2 were excluded from testing. As a 2D comparator, TERT1-OAT1 cells were also seeded in a 96-well tissue culture-treated plates (100,000 cells/cm²; Cat# 3603, Corning), and cultured for 48h prior to chemical exposures.

2.2. Bidirectional transport studies

To evaluate transport and cellular uptake, PFHxA, PFBS, and PFOS were added to cell culture media in either a single (5 μM) or mixture (2 μM each, 6 μM total) settings. PFAS-containing medium was either added to the bottom well (750 μL, apical surface, A-to-B) or top well (200 μL, basolateral surface, B-to-A) of the Transwell cultures. Fresh culture medium was added to the opposite (recipient) well in either treatment configuration. After chemical addition, cultures were returned to the incubator and maintained in either static or fluidic (PhysioMimix^® T12) conditions. Media were sampled (50 μL) at 2 and 4 hours; at 4 hours, all media were removed, and cells were re-treated with chemical-containing fresh media due to the low volumes in the top (basolateral) wells. After re-treatment, media were again sampled (50 μL) at 24- and 48-hours post-exposure. After the final timepoint, static and fluidic cultures were removed from their respective platforms, rinsed once with PBS and lysed with M-PER^™ Mammalian Protein Extraction Reagent (60 μL/well; Cat# 78503, ThermoFisher Scientific, Waltham, MA) for 10 minutes at room temperature. Due to the limited throughput and expense of the fluidic model, cell lysates were collected only at the final (48h) timepoint as this was a destructive assay. Media and cell lysate samples were stored at −20°C until chemical analysis. In the 96-well cultures (2D comparator), only cell lysates were collected, as this model did not allow for bidirectional transport between multiple compartments.

2.3. Analytical Methods

For chemical analysis, media and cell lysate samples were extracted using liquid/liquid extraction with protein precipitation. Briefly, 50 μL of media or lysate sample was added to 100 μL of chilled acetonitrile (Cat# A998-4, Fisher Scientific, Waltham, MA) containing an internal standard (0.1 μM C¹³-PFOA), then vortexed before centrifuging and transferring supernatant to new 1.5 mL microcentrifuge tube. The supernatant was then evaporated to dryness using a Savant^™ SpeedVac^™ (ThermoFisher Scientific, Cat# SPD1010) and reconstituted with 50 μL of mobile phase A (see below). Samples were then transferred to autosampler vials containing 200 μL fused inserts (Cat# A998-4, Ibis Scientific, Las Vegas, NV) and stored at −20° C until analysis.

Quantitative analyses were performed on a triple quadrupole mass spectrometer (Cat# 6470, Agilent, Santa Clara, CA) using negative ion mode electrospray ionization source. Capillary voltage, sheath gas temperature, and sheath gas pressure were set to 4500V, 300°C and 50 psi respectively. All samples were analyzed using this method. Briefly, extracted samples (20 μL) were auto-injected onto a ZORBAX SSHD Eclipse Plus C18 column (3.0 × 50 mm, 1.8 μm; Cat# 959757-302, Agilent) with a guard column (2.1 × 5 mm, 1.8 μm; Cat# 821725-901, Agilent) using a 1290 Infinity II LC (Agilent). The column temperature was set to 40°C and flow rate to 0.4 mL/min. Initial chromatographic conditions were 90% mobile phase A (HPLC-grade water containing 5 mM ammonium acetate) and 10% mobile phase B (95% MeOH with 5 mM ammonium acetate). At two minutes, mobile phase B increases to 30% until minute 14 when this increases to 95%. At 14.5 minutes, this was changed to 100% mobile phase B before returning to 10% at 15.5 minutes. This condition remained for the rest of the method which ended at 17 minutes. Samples were normalized to internal standard and reported in units of concentration (μM), with a limit of quantitation of 0.0048 μM.

2.4. In vitro-to-in vivo extrapolation method for predicting renal clearance

The approach for extrapolating the in vitro data obtained from cell-based experiments to predict in vivo human renal clearance of PFAS involved (1) assessing in vitro permeability with respect to secretion and reabsorption based on in vitro flux data, and (2) incorporating these in vitro permeability data into our newly developed physiologically-based kidney model to predict renal clearance.

2.4.1. Utilizing a Bayesian approach to determine in vitro permeability

We first constructed an in silico model (Nagayasu et al., 2019) based on the layout of the Transwells to describe the time-dependent PFAS flux between the two sides of the cell monolayer, as well as cell lysates (Fig. 1). This model consists of three compartments: an apical compartment (lumen), a basolateral compartment (blood), and a cell compartment that accounts for cellular accumulation. The differential equations for describing time-dependent concentrations in respective compartments were as follows:

$$\frac{d{A}_{blood,i}}{dt}=Area\cdot \left(P_1\cdot C_{cell,i}-P_2\cdot C_{blood,i}\right)$$ (1)

$$\frac{d{A}_{cell,i}}{dt}=Area\cdot \left(P_2\cdot C_{blood,i}+P_3\cdot C_{lumen,i}-\left(P_1+P_4\right)\cdot C_{cell,i}\right)$$ (2)

$$\frac{d{A}_{lumen,i}}{dt}=Area\cdot \left(P_4\cdot C_{cell,i}-P_3\cdot C_{lumen,i}\right)$$ (3)

where $A_j$ is the amount (moles) of PFAS in compartment $j$; $Area$ is the cell layer surface area, $C_j$ is the concentration (moles/mL) of PFAS in compartment $j$; $P_1$ and $P_2$ are the permeability (cm/hr) between blood and cell compartments; and $P_3$ and $P_4$ are the permeability (cm/hr) between lumen and cell compartments. As shown in Fig. 1, $P_1$ and $P_3$ are related to tubular reabsorption, while $P_2$ and $P_4$ are related to tubular secretion. Lastly, the $i$ label refers different experiments with different initial conditions, i.e., adding chemicals to the basolateral (B-to-A) or apical (A-to-B) compartment; for each type of system (static versus fluidic Transwell). Experimental dataset comprising four time points of concentrations in the recipient side and one point of cell concentration were fit using this system of differential equations, including equations 1 – 3, along with the initial assumption that the amount of chemical in the recipient side and cell ($\text{t=}0$) is 0. The data with different initial conditions were fitted by a single common set of kinetic parameters (i.e., $P_1$ to $P_4$) for each PFAS, system, and exposure.

2.4.1.1. Assigning prior distributions of parameters

In the absence of pre-existing information for $P_1$ to $P_4$, their priors were designated as noninformative lognormal distributions with $\text{GM=}{10}^{-3}$ and $\text{GSD=}{10}^5$, anticipating that these distributions would encompass the fitted values. Furthermore, we assumed a log-normal distribution for the estimation of likelihood; the geometric standard deviation of residual error was given a noninformative prior consisting of a log-uniform distribution, with a lower bound of 1.1 (10% error) and upper bound of 10 (order of magnitude error, indicating potentially poor model fit). This wide range was selected so as to be unbiased a priori as to whether the model has a good or poor fit. Detailed information and model codes are available at https://github.com/hsingchiehlin/PFAS_KidneyChip_model.

2.4.1.2. Convergence testing

To evaluate model convergence, four independent Markov chains were run. The initial segment of iterations constituted the burn-in phase, while the concluding portion of iterations served as output for assessing convergence and estimating the posterior distributions of pertinent parameters. The precise number of iterations and the segment designated for burn-in were determined through convergence diagnosis, specifically focusing on achieving an estimated scale reduction $\left(\widehat{\text{R}}\right)$ of lower than 1.2, as proposed by Gelman and Rubin (1992). Five thousand iterations were saved from each chain for posterior inference.

2.4.1.3. Model performance: comparing observed and simulated concentrations

In addition to evaluating model convergence, it was essential to assess the performance of our chip compartmental model using updated posterior permeability values. We conducted simulations of time-dependent concentrations in wells and cell lysates, presenting these values as medians and 90% confidence intervals. Subsequently, we compared these simulated concentrations with observed values through visualization.

2.4.2. Physiologically-based kidney model

To incorporate in vitro permeability into the prediction of in vivo renal clearance, we developed a physiologically-based kidney model to depict the chemical kinetics within the proximal tubule lumen, cells, and the surrounding blood vessels, as illustrated in the lower-left panel of Fig. 1. Guided by the physiological attributes of the kidney, as the renal artery blood circulates through the glomerulus, a fraction of the blood undergoes filtration into the proximal tubule lumen based on the glomerular filtration rate (GFR). Simultaneously, the chemical in its free form is distributed within both the blood and the lumen. Therefore, based on the principle of chemical mass balance, the equation can be expressed as:

$$Q_K\cdot C_{A,free}=Q_{PT}\cdot C_{Lumen}+\left(Q_K-Q_{PT}\right)\cdot C_{KB,free}$$ (5)

where $Q_{\text{K}}$ and $Q_{\text{PT}}$ are flow rate in kidney arterioles and proximal tubule; $C_{\text{A,free}}$, and $C_{\text{KB,free}}$ are free PFAS concentration in kidney arterioles and unfiltered fraction of blood, respectively. ${\text{C}}_{\text{Lumen}}$ are PFAS concentration in proximal tubule lumen. Regarding the flux between the cell and blood/tubule lumen, we assumed that the concentration within the cell would be in a steady state. Consequently, the mass balance equations are delineated as follows:

$$\left\{\begin{array}{l}{C}_{Cell}\cdot Area\cdot P_1=C_{KB,free}\cdot Area\cdot P_2\hfill \\ C_{Lumen}\cdot Area\cdot P_3=C_{Cell}\cdot Area\cdot P_4\hfill \end{array}\right.$$ (6)

Which can be simplified to:

$$C_{Lumen}=C_{KB,free}\cdot \frac{P_2\cdot P_4}{P_1\cdot P_3}$$ (7)

Based on Eq. (7), we assigned $\frac{P_2\cdot P_4}{P_1\cdot P_3}$ as the permeability ratio (P ratio). Here, we assumed that in vivo ${\text{P}}_1$ to ${\text{P}}_4$ are same as ones predicted by utilizing the Bayesian approach and in vitro models. In the case of 2D cell culture, where the basolateral side of the cell is attached to the bottom of wells, we made the assumption that $P_1$ was equal to $P_2$. As a result, the P ratio can be expressed as:

$$Prati{o}_{2D}=\frac{P_4}{P_3}=\frac{C_{Lumen}}{C_{Cell}}$$ (8)

B combining Eqs. (5) and (7), the equation for PFAS concentration in the proximal tubule lumen can be expressed as:

$$C_{Lumen}=\frac{Q_K\cdot C_{A,free}}{\left(Q_{PT}+\frac{\left(Q_K-Q_{PT}\right)}{Pratio}\right)}$$ (9)

Subsequently, we utilized the following equation to calculate the extraction rate (ER) and Eq. (9) to derive an equation for calculating the renal clearance of free PFAS $\left(C{L}_{renal,free}\right)$:

$$ER=Q_{PT}\cdot C_{Lumen}=C{L}_{renal,free}\cdot C_{A,free}$$ (10)

$$C{L}_{renal,free}=\frac{Q_K\cdot Q_{PT}}{\left(Q_{PT}+\frac{\left(Q_K-Q_{PT}\right)}{Pratio}\right)}$$ (11)

To calculate the renal clearance of total PFAS $\left(C{L}_{renal,total}\right)$, Eq. (11) should be adjusted with the fraction unbound in blood $\left(f_{ub}\right)$:

$$C{L}_{renal,total}=\frac{C{L}_{renal,free}}{f_{ub}}$$ (12)

The fraction unbound in blood, a chemical-specific parameter, was obtained from the HTTK database (Pearce et al., 2017). The values for PFOS, PFBS, and PFHxA were 0.00558, 0.0228, and 0.0198, respectively. Regarding parameters related to human kidney physiological features, $Q_K$ was set to 29.6 L/hour (7.1×10⁶ mL/day), assuming a bodyweight of 70 kg (Chou and Lin, 2019). Given that proximal tubules contribute to 70% of water reabsorption, $Q_{PT}$ was determined by multiplying the GFR by 0.3. This adjustment accounts for the substantial reduction in volume occurring between the glomerular filtrate and the filtrate exiting the proximal tubule (i.e., 6.77×0.3 = 2.03 L/hour, or 4.9×10⁴ mL/day) (Chou and Lin, 2019; Feraille et al., 2022).

Subsequently, we compared the predicted renal clearance to the in vivo human renal clearance obtained from published studies. We compared our results with reported human in vivo renal clearance, predictions from RTK-based IVIVE (as used in high-throughput toxicokinetics - HTTK; $\text{renal}\text{clearance}=\text{GFR}\cdot f_{ub}$), and the in vitro extrapolation method proposed by Kunze et al. (2014), as these use similar input data as our method. The reported values of human in vivo renal clearances and half-lives for PFAS and their sources are compiled summarized in Table 1 (details in Tables S1-S2). Additionally, considering variations in the calculation method of Kunze et al. (2014) based on the concentration in the recipient wells at different exposure time points, we calculated renal clearance at 2, 4, 24, and 48 hours of exposure. Using this method, no significant difference between renal clearance values calculated at different time points was found (Fig. S1).

Table 1.

List of PFAS used in this study.

Full name (abbreviation)	CASRN	$\text{C}{\text{L}}_{\text{renal}}\left(\text{mL}/\text{Kg}/\text{day}\right)$ a	T1/2 (years)b
Perfluorooctanesulfonic acid (PFOS)	1763-23-1	0.008 – 0.022	2.89 – 21.6
Perfluorohexanoic acid (PFHxA)	307-24-4	8.21 – 174	0.09 – 1.63
Perfluorobutanesulfonic acid (PFBS)	375-73-5	7.97 – 17.8	0.07 – 0.12

^aRange of median values reported by different studies; see details in Table S1.

^bRange of median values reported by different studies; see details in Table S2.

2.5. Computational software

The compartmental modeling with MCMC algorithm was conducted using GNU MCSim v6.1.0 in the R language platform (Bois, 2009). The following data analysis and visualization were also performed within R (version 4.1.2) using RStudio and packages, including rstan (version 2.21.7), tidyverse (version 1.3.2), dplyr (version 1.0.10), ggplot2 (version 3.4.2), ggh4x (version 0.2.4), ggridges (version 0.5.4), and ggpattern (version 1.0.1). All raw data and computational executing codes for modeling and data analysis are available in Supplemental Materials and the GitHub repository: https://github.com/hsingchiehlin/PFAS_KidneyChip_model.

3. Results

3.1. Uptake of PFAS by TERT1-OAT1 cells cultured in 96-well plates

In the most basic in vitro model (Fig. 1), TERT1-OAT1 cells were cultured in 96-well plates and exposed to single PFAS or their mixture for 48 hours to match the lysate collection times of the Transwell and fluidic models, and allow for sufficient intracellular accumulation for detection by LCMS/MS. The cumulative amount of PFAS in the cell lysate was then measured at the end of the experiment. Fig. 2A shows the intracellular accumulation of PFAS represented as the percentage taken up from the media. Additionally, Fig. 2B displays the PFAS concentrations (μM) present in cell lysates after 48 hours of exposure. The raw amount and concentration data for each replicate are also listed in Table S3. We found that PFOS exhibited a much (orders of magnitude) higher uptake by cells as compared to PFHxA or PFBS, commensurate with the differences in their half-lives in humans (Table 1).

Figure 2.

Mean values with standard deviations of (A) the percentage of amount uptaken by TERT1-OAT1 cells cultured in 96-well plates with single (black bars) and mixture (hatched bars) exposure at 48 hours, and (B) their calculated cellular concentrations by the amount data in Table S3 and the cell volume. The cell volume is based on formula for the volume of a sphere: V = 4/3 πr³. The radius of TERT1-OAT1 is assumed to be similar with HEK293 cells (6.5 μm), and the number of cells seeded in each well is 50,000. Shown are mean ± standard deviation values for each group (n = 3–4 replicates per group). Asterisks (****) denote statistically significant (p<0.0001) differences between groups (as indicated by the brackets) using 2-way ANOVA followed by Sidak’s multiple comparisons test.

Significant differences were found in the percentage uptake and cellular concentrations between single and mixture exposures for PFOS (p<0.0001), but not for PFHxA or PFBS (Fig. 2). The percentage of PFOS taken up by cells in the single exposure (47.5 ± 1.9% when treated at 5 μM) was notably lower than the percentage uptake by cells in the mixture exposure (87.6 ± 6.0% when treated at 2 μM).

3.2. Bidirectional flux of PFAS across TERT1-OAT1 cells in Transwells

We also tested PFAS transport in two more complex in vitro models (Fig. 1) where TERT1-OAT1 cells were cultured in Transwells with no media flow (static condition) or in Transwells with recirculating flow across the apical side of cells (fluidic condition) and exposed to single PFAS or their mixture for 48 hours. To confirm the barrier function of TERT1-OAT1 cells cultured on Transwells in either static or fluidic conditions, TEER values were measured at the beginning and the end of exposures (Fig. S2). The average TEER values were in line with previously reported values for TERT-OAT1 RPTECs (Wieser et al., 2008; Secker et al., 2019). PFAS exposures did not affect barrier function, indicating that no cytotoxicity occurred at these exposures. In fact, TEER continued to increase, indicating that the chosen treatments were indeed non-toxic for the cells. This slight TEER increase can likely be associated with continued cell growth on the Transwells. However, the cell vendor (ATCC) reports a doubling time of 70–96 hours, so cell growth was likely minimal and was not considered in this model.

Fig. 3 illustrates the comparisons of bidirectional flux for three PFAS across various culture and exposure conditions, presented as time-dependent net reuptake rates. Among the three tested PFAS, the net reuptake rates of PFOS were higher than those of PFHxA or PFBS. Generally, the maximal net reuptake rates for PFOS and PFHxA were observed after 2- or 4-hour exposure, followed by a gradual transition to a steady state. However, for PFBS, the flux at the first two time points with mixture exposure exhibited contrasting results, indicating possible ‘tubular’ secretion. Additionally, based on two-way ANOVA and Tukey HSD post-hoc testing, net flux under static conditions tended to be equal to or slightly higher than fluidic conditions for PFOS, and equal or lower for PFHxA or PFBS; between single and mixture conditions, and net flux tended to be equal to or higher under single compound testing for all three compounds (Statistical differences in flux are reported in Table S4).

Figure 3.

Comparisons of bidirectional flux and net reuptake rates of PFOS, PFHxA, and PFBS across TERT1-OAT1 cell layers in Transwell-based static and fluidic cultures exposed to either (A) single, or (B) mixture of tested compounds (as indicated in the legend to the y-axis). Black lines indicate lumen-to-blood (circles) and blood-to-lumen (squares) transport rates. Dashed red line (triangles) indicates the net reuptake for each condition as shown. Statistical analysis results for net reuptake comparing static vs. fluidic and single chemical vs. mixtures testing using two-way ANOVA and Tukey HSD post-hoc tests are provided in Table S4.

3.3. In vitro data-based compartmental modeling

Next, using in vitro data detailed above, fitting of the compartmental model was performed for the three PFAS in single and mixture exposures and both static and fluidic culture conditions. For each PFAS exposure condition, a common set of permeability parameters was fitted to simultaneously capture the concentrations from the A-to-B (Lumen to Blood) and B-to-A (Blood to Lumen) assays. Thus, for each PFAS, system, and exposure, there were 8 time points for “recipient-side” concentrations (4 each for A-to-B and B-to-A, at 2h to 48h) and two time points for cell concentration (1 each for A-to-B and B-to-A, at 48 hr), with three replicates each. The raw concentration data for each replicate is presented in Table S5. Permeability estimates, meaning $P_1$ to $P_4$, for all three PFAS under the various study designs converged at 300,000 iterations of each MCMC sampling chain, with every 30 iterations saved. Specifically, after discarding the initial one-fifth of iterations for each chain as burn-in, the $\widehat{\text{R}}$ values were all lower than 1.2, confirming convergence (Table S6). The final posterior distributions with four chains and detailed $\widehat{\text{R}}$ values for parameters are presented in Fig. S3 and Table S6, along with statistical summaries. We note that the parameters $P_1$ to $P_4$ are not individually fully identifiable given the available data, and hence there is some multimodality in the posterior distributions for some of the individual permeability parameters, especially for PFOS. However, this challenge does not impact the final prediction of renal clearance because the final input of the physiological-based kidney model is the P ratio, which shows a narrow, single-modal posterior distribution (Fig. S3B). In terms of model performance, Fig. 4 shows comparisons between simulated and measured concentrations in recipient wells and within cells, revealing highly consistent results. Detailed time-dependent comparisons are also shown in Fig. S4.

Figure 4.

Comparisons of model-fitted PFOS, PFHxA and PFBS concentrations and measured data in the cell and recipient well (i.e., blood or lumen compartment) in the static/fluidic Transwell-based cultures of TERT1-OAT1 cells exposed to (A) single or (B) mixture of PFAS compounds. The simulation results were obtained by using Bayesian Markov Chain Monte Carlo simulation to fit the time-courses of in vitro concentrations to a compartmental model (see Fig. 1). Median simulation values are shown, with individual time-courses and posterior prediction intervals provided in Figure S4.

In addition, intracellular concentrations from 2D cell exposures (96 well plates) were compared with those from static and fluidic Transwells using the A-to-B exposure configuration, as depicted in Fig. S5. This comparison was conducted considering that media were added to the apical surface of cells seeded in the 96-well plate. The intracellular concentrations were found to be within an order of magnitude regardless of experimental platform (96-well plates or static and fluidic Transwells). Additionally, the intracellular concentrations of PFOS were approximately 100 times higher than those of PFBS and PFHxA, regardless of the experimental model. In bidirectional transport studies with Transwells, intracellular PFAS concentrations were consistently higher when treatments were added to the apical side as compared to the basolateral side. This suggests that the cellular uptake transporters involved in PFAS accumulation were likely polarized to the apical surface (Figs. S4 and S5).

3.4. Predictions of total renal clearance using IVIVE

The permeability ratio (P ratio) is a crucial parameter for determining renal clearance and is defined as the ratio of secretion to reabsorption. Fig. 5 illustrates the predicted permeability ratios for the three tested PFAS in 96-well plate cell culture, as well as in static and fluidic Transwells. The permeability ratios from 96-well plate cell cultures were calculated based on raw concentrations in the cell lysates and media (Table S3). The permeability ratios from studies in Transwells were derived from the posteriors of permeability obtained from the compartmental modeling. Regardless of culture condition, the permeability ratios for PFOS were consistently less than 1 (Fig. 5). This suggests that the transport direction of PFOS in TERT1-OAT1 cells favors reabsorption rather than secretion. The permeability ratios for PFHxA and PFBS consistently ranged either just under or around 1. This suggests a more balanced, passive transport as compared to PFOS. Among the in vitro models, the permeability ratios derived for 96-well plate cultures were consistently lower than those for Transwells under either static or fluid conditions.

Figure 5.

Predictions for permeability (P) ratios for PFOS, PFHxA, and PFBS under study designs with the 96-well plates (2D), static/fluidic Transwell-based cultures exposed to single or mixture of PFAS compounds. The P ratio for 2D culture was calculated by using raw data on intracellular and media concentrations (Table S3), and those for static and fluidic Transwells were posterior predictions from Bayesian compartmental modeling (see Figures 1 and 4). The bars represent median; the boxes represent the range of 25% and 75%-tile; the whiskers represent the 5% and 95%-tile. The vertical red dashed line at a P ratio = 10⁰ = 1 indicates no net secretion or reabsorption.

Next, using the permeability ratios as input, we ran our physiologically-based model to predict renal clearance for each PFAS under different culture conditions (Table S7). Fig. 6 presents a summary comparison between our predictions and the renal clearance values from studies in humans as reported in the literature (Tables 1 and S1); for reference, we also include the results for two previous IVIVE approaches, based on httk (Pearce et al., 2017) and Kunze et al. (2014). For PFHxA and PFBS, httk, Kunze et al. (2014), and both Transwell-based cell models, all predict renal clearance values within an order of magnitude of the reported huma in vivo values. The 96-well plate culture results for these two compounds tended to underpredict renal clearance, though the deviation was in the health-protective direction. For PFOS, on the other hand, httk and Kunze et al. (2014) severely overpredicted renal clearance, which is not health protective and could lead to an underestimation of the body burden. Both Transwell-based cell models also over-predicted renal clearance by about an order of magnitude, albeit the lower confidence bounds were close to the reported human in vivo values. The PFOS predictions from 96-well plate experiments, on the other hand, underpredicted renal clearance. Moreover, in this 2D culture, the clearance predictions for PFOS in the mixture setting (at 2 uM) were smaller by an order of magnitude than those in the single chemical setting (at 5 uM), suggesting possible saturation of re-absorption at the higher concentration.

Figure 6.

Comparison of reported and predicted PFOS, PFHxA and PFBS renal clearance. Two previous IVIVE methods (top left – httk and Kunze et al., 2014), or our in vitro-in silico workflow for single (circles) and mixture (triangles) exposure based on TERT1-OAT1 cell data from either 96-well plate cultures (top right), or static and fluidic Transwell-based cultures (bottom). The x-error bars represent the range of medians (thicker bars) or the range of individual estimates (thinner bars) for human in vivo renal clearance as reported in literature (details in Table S1). The y-error bars represent the 90% posterior prediction intervals from the Bayesian compartmental modeling. The dashed line represents equality and shaded region is a 10-fold interval in either direction.

Thus, overall, the Transwell-based models, regardless of whether the flow/shear stress were present, had the best human in vivo concordance in terms of predicting absolute values for renal clearance, with estimates within about 10-fold. Interestingly, similar results were obtained when single chemicals or their mixtures were tested. However, one potential limitation is that while higher clearance compounds appeared biased low, the lower clearance compound appeared to be biased high. Still, although predictions using 96-well plate cell culture resulted in absolute renal clearances lower than in vivo values by an order of magnitude, the relative values of renal clearance between lower- and higher-clearance PFAS were more discernible compared to those calculated using the Transwell-based models. Additionally, these relative values matched well with corresponding in vivo relative values. However, there were some differences in 2D culture between single and mixture testing of PFOS that are suggestive of possible saturation effects.

4. Discussion

The intricate mechanisms of renal clearance for a given chemical involve processes such as glomerular filtration, tubular passive diffusion, and tubular active secretion and reabsorption; however, few experimental and/or computational approaches exist for quantitative prediction of renal clearance (Scotcher et al., 2016a; Scotcher et al., 2016b; Chang et al., 2022). The most common assumption is that renal clearance consists of only passive filtration of the free (unbound) compound in plasma; e.g., this is the basis of U.S. EPA’s high throughput toxicokinetics-based approach to IVIVE (Pearce et al., 2017). However, for chemicals such as PFAS, active transport, i.e., against-concentration gradient transport, in the kidney is thought to play a critical role in the determining overall elimination rates, which for this class of substances can range over orders of magnitude from days to years, and for which there can be very large species differences (Harada et al., 2007; Zhang et al., 2015). To address this challenge, on the one hand, top-down approaches such as QSAR have been attempted (Dawson et al., 2023); however, due to data sparsity, this approach has only been able to categorize PFAS into 4 bins based on half-life (Bin 1: <12 hour; Bin 2: <1 week; Bin 3: <2 months; Bin 4: >2 months), rather than providing a continuum of quantitative predictions. Thus, even though our in vitro-in silico models have errors up to 10-fold, they still represent an improvement over existing approaches for PFAS. On the other hand, while a number of bottom-up approaches using in vitro data have been proposed for drugs (Neuhoff et al., 2013; Kunze et al., 2014; Kumar et al., 2020), their utility to PFAS is unclear because of the different physical-chemical properties, the vast inter-species differences in PFAS clearance which would limit the utility of the data from non-human cells, and the fact that existing models have not been confirmed with extremely long half-life compounds. Therefore, there is a critical need to evaluate the predictive power of in vitro approaches, specifically for PFAS.

There are a number of study and analysis design considerations when examining in vitro test systems for potential use in the context of determining renal clearance of PFAS. First, there are large species differences in PFAS elimination (Pizzurro et al., 2019); therefore, it is critical to use cells of human origin. In this study, we used TERT1-immortalized RPTECs isolated from primary human tissue and overexpressing OAT1, making it an accessible cell-based tool for this assessment (Simon et al., 2014). We selected these OAT1 overexpressing cells due to their extensive use in our previous work, demonstrating stability and the ability to form robust, reproducible barriers on Transwell culture platforms. However, primary human cells, or cells that over-express other more relevant transporters, can be used to further improve human relevance and also probe inter-individual variability. Second, due to the sheer number (hundreds to thousands) of PFAS for which renal clearance data are needed, it is critical that any in vitro model system has a reasonable throughput so that at least dozens, if not hundreds, of compounds can be tested. Therefore, we compared several model designs, from 96-well plate cultures to Transwell-based models without and with media flow. Ultimately, we tested both static and fluidic Transwells, the latter enabled by the PhysioMimix^™ TC12 system. Additionally, we tested PFAS both individually as well as in a mixture. Not only does this design enable multiplexed testing, but it also represents a more environmentally-relevant approach to human exposures, which invariably involves multiple simultaneous PFAS.

Overall, our results highlight the enhanced capability of the combined in vitro-in silico workflow as a sensible method for distinguishing between lower- and higher-clearance PFAS. It is also a method that produced more human-relevant results with PFAS when compared to existing in silico approaches for estimating renal clearance. In particular, for all three experimental test systems used herein, we found that the transport of PFOS is notably skewed towards reabsorption as opposed to secretion when compared to PFHxA and PFBS, which aligns qualitatively with the relative amounts of human renal clearance reported in the literature (Harada et al., 2005; Zhang et al., 2013; Zhou et al., 2014; Gao et al., 2015; Fu et al., 2016). Our findings suggest that, compared to existing IVIVE methods for renal clearance, a simple, 96-well plate-based cell culture model may be sufficient to differentiate between PFAS with slow and fast renal clearance on a relative basis. At the same time, the Transwell-based in vitro models exhibited greater accuracy for predictions on an absolute basis. We also observed that introducing fluid shear stress to the Transwells using the PhysioMimix^™ system did not confer noticeable advantages over the simpler, static Transwell model in the context of this bidirectional transport study. Despite numerous studies demonstrating the enhancement of cell polarity and function of some xenobiotic transporters with fluid shear stress in RPTEC cultures (Duan et al., 2008; Jang et al., 2013; Ross et al., 2021), it may not be necessary for the regulation of the transport pathways involved in PFAS reuptake. Lastly, our results indicate that renal clearance predictions from mixture exposures are generally similar to those from single exposures. This suggests that future high-throughput PFAS screenings may be designed with chemical mixtures, significantly cutting down on the number of replicates and cell requirements. However, the apparent differences in intracellular PFOS concentrations and renal clearance estimates between single (5 μM) and mixture (2 μM each) exposures in the 96-well plate-based system imply that saturation of transport pathways may occur. This observation underscores the critical need for careful consideration of exposure concentrations in these mixture settings to account for potential saturation, particularly with compounds exhibiting higher transporter affinity.

These findings also align with mechanistic studies conducted by Zhao et al. utilizing another RPTEC line, HEK293 derived from human embryonic kidneys (Zhao et al., 2015; Zhao et al., 2017). According to these studies, apical sodium-dependent bile salt transporter (ASBT) and organic anion transporting polypeptide (OATP) 2B1, typically situated on the apical site of proximal tubule cells, exhibit a higher affinity for PFOS compared to PFBS. The outcomes of these investigations also substantiate the presence of higher PFOS concentrations within RPTEC lysates when contrasted with PFBS. Moreover, Yang et al. (2010) demonstrated that PFAS with chain lengths ranging from C8 to C11 are likely substrates for the OAT4 apical kidney transporter. More recently, Louisse et al. (2023) conducted a comparative study on the uptake of seven PFAS compounds: PFHpA (Perfluoroheptanoic acid), PFOA (Perfluorooctanoic Acid), PFNA (Perfluorononanoic acid), PFDA (Perfluorodecanoic acid), PFBS (Perfluorobutane sulfonate), PFHxS (Perfluorohexane sulfonate), and PFOS (Perfluorooctanesulfonic acid) in OAT4-transfected HEK cells. Their findings clearly indicated the uptake of all PFAS compounds except for PFBS. This study underscored the involvement of OAT4 in the reuptake of PFAS in proximal tubule cells and revealed a high correlation between human half-lives and intracellular PFAS concentration in RPTECs cultured in vitro. The data from these transporter studies collectively support the participation of several apical renal transporters in the reuptake of PFAS compounds.

While the Transwell-based system improved the accuracy of absolute predicted renal clearance in our in vitro-in silico approach, certain limitations still remain and need additional consideration. First, we have only examined three PFAS; although they appear to represent a reasonably wide range in terms of elimination half-lives, additional testing is needed to confirm the predictivity of our approach. Another challenge is the lack of human data on elimination rates, specifically for renal clearance. Furthermore, because both renal and biliary/fecal clearance mechanisms contribute to PFAS elimination, the relative importance of which may dominate will depend on the PFAS and the net kinetics of all involved transporters; and due to uncertainties in the volume of distribution for PFAS, the relationship between half-lives (about which more is known) and renal clearance is not always straight-forward. Thus, the incorporation of these data into either compartmental of physiologically-based pharmacokinetic models will require additional data or predictions as to the extent of other clearance mechanisms beyond the kidney.

Second, our in silico approach depended on time-course data for transport kinetics, meaning that additional time sampling points for measuring concentrations of cell, donor and receiver compartments may be needed in future studies. This additional time and experimental cost may pose challenges for high-throughput screening experiments. Moreover, our chosen sampling time points starting at 2 hr may have contributed to the imprecision of our measurements. Thus, it may be possible to optimize this aspect of the design further to only one or two time points. On the one hand shorter time periods, such 15min used for membrane permeability experiments, could be used to estimate net flux in the “linear” range, though such designs are typically only used in two-compartment models (A vs. B) where cellular accumulation is not considered. Moreover, for slowly clearance compounds, analytical detection limits may be a limiting factor a shorter time points. Nagayasu et al. (2019) employed a three-compartment model to predict the efflux ratio of drugs (the ratio of efflux to influx permeability) for P-glycoprotein in Caco-2 cells and proposed the “single-point sampling approach” as a promising method for predicting the efflux ratio, offering convenience for high-throughput screening. This approach involved sampling concentrations only at the initial time in the donor compartment and the final time-point (2 hours) in the intracellular and receiver compartments in bidirectional transporter studies. Although such an approach may enhance the feasibility of high-throughput screening, careful consideration should be given to determining the choice for the final time-point, especially for PFAS with longer half-lives compared to the drugs tested by Nagayasu et al. (2019) which had half-lives ranging from 3 to 48 hours.

Furthermore, it should be noted all the in vitro models utilize a single cell line, TERT1-OAT1 RPTECs. While there is limited evidence indicating that OAT1 is a significant transporter for PFAS compounds, these cells are highly standardized and easy to culture. This makes them an attractive choice for renal in vitro studies and well-suited for this “proof of concept” study. To accurately identify transporter involvement in PFAS uptake and accumulation, future studies will need to include basal lines as well as lines overexpressing other transporters. Consequently, additional cell lines and human primary cells may be necessary to fully understand the interactions of these transporters. While our Transwell experiments did not suggest the need for a scaling factor to adjust for the expression profile of cell line utilized to in vivo transport, additional data are needed to confirm this observation. Our results from 2D culture suggested a systematic underprediction of clearance under this protocol, which could warrant development of a scaling factor. Additionally, the use of a single cell line does not incorporate any population variation, a consideration critical in decision-making (Rusyn et al., 2022). Thus, differences in our results with respect to the reported population median values may be due, in part, to inter-individual variation. Indeed, the reported human renal clearance values for both PFOS and PFBS have extremely high variation at the individual level, both across studies and within studies. This may be, in part, due to true population variation; however, measurement discrepancies likely also contribute to the observed variance, as pharmacokinetic modeling-based studies of overall half-life for several PFAS have indicated population geometric standard deviations of only around 1.6-fold. Nonetheless, it can be difficult to ascertain the extent to the differences between our in vitro results and the reported in vivo clearance values are due to experimental uncertainties or true variability. Addressing inter-individual variation would require a population-based in vitro model, ideally one that is more reproducible, for example iPSC-derived cells (Burnett et al., 2021).

In conclusion, our in vitro-in silico workflow, that may utilize experimental data from human RPTECs cultured on either 96-well plates or Transwells in static cultures, appears to be effective in distinguishing PFAS with lower and higher clearances. Our findings suggest that PFAS can be effectively screened in mixture settings, and a simple 96-well plate cell culture may be sufficient for prioritizing relative values of clearance; this approach holds potential for implementing our method in high-throughput screening. The utilization of Transwell-based cell culture models can further enhance the predictive accuracy for absolute renal clearance, with the more straightforward static cultures appearing to be sufficient. Still, the errors in absolute clearance predictions still range up to an order of magnitude, so improvements in predictive accuracy should still be pursued to be made more reliable. Moreover, given the current limitations in human in vivo renal clearance data, we emphasize the need for larger-scale studies to screen PFAS in both 96-well plate cell culture and in Transwell-based bidirectional transporter assays, further evaluating the feasibility and robustness of our new IVIVE method. Ultimately, we anticipate that our in vitro-in silico workflow can effectively contribute to filling a critical data gap in characterizing health risks associated with PFAS compounds.

Highlights.

• Some PFAS can bioaccumulate in humans, in part due to slow renal clearance ${\text{CL}}_{\text{renal}}$

• We test use of 2D culture and MPS models to measure ${\text{CL}}_{\text{renal}}$ for three PFAS

• Combined with physiologically-based IVIVE, we can accurately predict ${\text{CL}}_{\text{renal}}$

• Our in vitro-in silico workflow can prioritize PFAS that bioaccumulate in humans