Integrating In Vitro Testing and Physiologically-Based Pharmacokinetic
(PBPK) Modelling for Chemical Liver Toxicity Assessment – a Case
Study of Troglitazone

Integrating In Vitro Testing and Physiologically-Based Pharmacokinetic (PBPK)
Modelling for Chemical Liver Toxicity Assessment – a Case Study of Troglitazone
Lin Yu1,2, Hequn Li3

1 Academy of Military Medicine, Academy of Military Sciences, 27 Taiping Road,
Beijing 100850, PR China
2
Institute of Disease Control and Prevention, People’s Liberation Army, 20 Dongda
Street, Beijing 100071, PR China
3 Unilever Safety and Environmental Assurance Center, Colworth Science Park,
Sharnbrook, Bedfordshire MK44 1LQ, UK
4 Department of Environmental Health, Rollins School of Public Health, Emory
University, Atlanta, GA 30322, USA
To whom correspondence should be addressed. E-mail: [email protected] Journal Pre-proof
2
Highlights:
 A PBPK model was developed to predicting pharmacokinetics of troglitazone in human.
 Mitochondria-mediated toxicity endpoints were used to derive BMDLs.
 The performance of the animal-free approach was preliminarily evaluated in human.
 The BMDLs based Cmax metric matched well with clinical hepatotoxic context.
Abstract
In vitro to in vivo extrapolation (IVIVE) for next-generation risk assessment (NGRA) of
chemicals requires computational modeling and faces unique challenges. Using mitochondria￾related toxicity data of troglitazone (TGZ), a prototype drug known for liver toxicity, from
HepaRG, HepG2, HC-04, and primary human hepatocytes, we explored inherent uncertainties
in IVIVE, including cell models, cellular response endpoints, and dose metrics. A human
population physiologically-based pharmacokinetic (PBPK) model for TGZ was developed to
predict in vivo doses from in vitro point-of-departure (POD) concentrations. Compared to the
200-800 mg/d dose range of TGZ where liver injury was observed clinically, the predicted
POD doses for the mean and top one percentile of the PBPK population were 28-372 and 15-
178 mg/d respectively based on Cmax dosimetry, and 185-2552 and 83-1010 mg/d respectively
based on AUC. In conclusion, although with many uncertainties, integrating in vitro assays and
PBPK modeling is promising in informing liver toxicity-inducing TGZ doses.
Keywords: troglitazone, PBPK modeling, reverse dosimetry, PODs, toxicity testing
alternativesJournal Pre-proof
3
1. Introduction
To set “safe” levels of chemical exposure for humans, an enormous number of experimental
animals (more than 100 million and worth €2 billion) are currently used in toxicological studies
every year (Hartung, 2009). However, this testing method is costly and time-consuming and
the relevance and accuracy of extrapolating findings in animal studies to human health risk has
been questioned (Hartung, 2008). In particular, traditional animal-based testing has not utilized
many of the advances in toxicology regarding the mechanisms of chemical perturbation of
biological systems. Using animals for toxicity testing nowadays is therefore facing enormous
challenges for scientific, ethical, economic and legislative constraints. As a result, toxicity
testing of chemicals is evolving towards innovative, animal-free, in vitro and in silico
approaches (Thiel et al., 2017).
In vitro to in vivo extrapolation (IVIVE) is an integrated process of translating in vitro
toxicity data to in vivo health risk predictions, containing both toxicokinetic and toxicodynamic
extrapolations. The IVIVE process faces many challenges and uncertainties, requiring a
number of in silico approaches to bridge the data gaps (NRC, 2017; Zhang et al., 2018). To this
end, physiologically-based pharmacokinetic/toxicokinetic (PBPK/PBTK) models are
especially useful as “bottom-up” tools that can quantitatively characterize the dynamic changes
in the concentrations of a chemical and its metabolite in plasma and organs by considering the
chemical’s physiochemical properties, the absorption, distribution, metabolism and excretion
(ADME) characteristics and related physiological processes (Rowland et al., 2011). While they
are commonly used in a forward manner to predict plasma/tissue dosimetry given an external
exposure, PBPK models are also used in a “reverse” manner to predict, from a tissue
concentration, the corresponding external dose levels in a given exposure scenario. This reverse
dosimetry approach can be applied in the setting of IVIVE, where the concentrations of the test
Journal Pre-proof
4
chemical in a cell assay causing a predefined in vitro points-of-departure (PODs) are used to
predict the in vivo exposure levels that can result in tissue or plasma concentrations similar to
the in vitro POD concentration (Louisse et al., 2015). The predicted exposure levels can then
be compared with the expected or actual exposure levels in humans, if available, to assess
margin of safety.
Although still in its infancy, a number of toxicokinetic IVIVE studies have been
published integrating in vitro testing and extrapolation modeling to predict the risks of organ
toxicity induced by various chemicals. For instance, the developmental toxicity caused by
tebuconazole (Li et al., 2017), glycol ethers (Louisse et al., 2010), phenols (Strikwold et al.,
2017), retinoid (Strikwold et al., 2013), and all-trans-retinoic acid (Louisse et al., 2015) were
predicted for human and animals based on in vitro toxicity data and in silico kinetic modelling.
The kidney toxicity induced by aristolochic acid I (Abdullah et al., 2016), liver injury caused
by hepatotoxicant azathioprine (Thiel et al., 2017), and reproductive toxicity (anti-androgenic
effect) of a diverse group of chemicals (Dent et al., 2018) were also predicted with the IVIVE
approach. Despite these progresses, which have mainly used one cellular biomarker to estimate
toxicity, more studies are required to gain further experience and knowledge to understand the
uncertainties involved in predicting in vivo toxicity in humans before the in vitro and in silico
approach could be reliably put into practice for human health risk assessment for a broader
range of compounds. These uncertainties include selection of cells, dosimetry, selection of
cellular biomarkers and time of observation, POD levels, “averageness” of selected cell models
and inter-individual variability, etc (Zhang et al., 2018). One particular challenge is the
determination of an in vitro POD which can correspond to the in vivo POD for a given apical
endpoint. Traditional approaches have used NOEL, LOEL, and even IC50/EC50 to define the
cutoff values. More recently benchmark dose modelling (BMD) has been used more frequently
Journal Pre-proof
5
as it considers the entire concentration- or dose-response curve and the uncertainty inherent to
the experimental data (Filipsson et al., 2003). Since future in vitro assays will be designed
around the toxicity pathways a test chemical perturbs, identifying the appropriate POD for the
involved toxicity pathways will play an important role in the success of the new approach
method (NRC, 2007). In addition, which in vitro concentration metrics to use, such as maximal
nominal concentration or area under the curve (AUC), both of which describe some aspects of
the cellular exposure to the test chemical, is important, too (Groothuis et al., 2015). In this
study, we aimed to use a prototype compound, troglitazone (TGZ), a drug known to induce
liver toxicity in humans, to explore some of these in vitro assay parameters and understand
how in vitro assays, combined with PBPK modelling-based reverse dosimetry, can be better
used through reducing the uncertainties to improve the IVIVE approach.
TGZ was the first thiazolidinediones (TZDs) used to treat type 2 diabetes by improving
insulin resistance (Plosker and Faulds, 1999). It was withdrawn from the market due to a
significant increase in the risk of hepatotoxicity developed after 3 months of use with no
additional clinical benefit compared to other available TZDs (Babai et al., 2018; Kung and
Henry, 2012). While multiple mechanisms have been proposed for the hepatotoxicity of TGZ
(Yokoi, 2010), perturbation of the mitochondrion appears to be a major toxicity pathway
involved despite that the molecular initiating event (MIE) is still not clear. TGZ is believed to
inhibit the activities of mitochondrial respiratory chain (MRC) complex enzymes, increase
reactive oxygen species (ROS), decrease mitochondrial membrane potential (MMP) and
prevent ATP synthesis (Hu et al., 2015). Besides, TGZ selectively stimulates the degradation
of PGC-1α protein, and suppresses the expression of mitochondrial biogenesis regulator Tfam
and oxidative phosphorylation enzymes (ATP5b and cytochrome C oxidase) (Liao et al., 2010).
ATP decrease can lead to the inhibition of the bile salt export pump (BSEP), whose main role
Journal Pre-proof
6
is to pump bile salt molecules out of the hepatocyte (Kullak-Ublick et al., 2000). Bile salt
accumulation in hepatocytes can induce further mitochondrial dysfunction and even cell death
because of its intrinsic detergent property (Delzenne et al., 1992; Gores et al., 1998).
Therefore, in the present study we focus on using in vitro hepatocyte assays to
interrogate mitochondrial toxicity. Various cell types have been used in vitro to study TGZ
toxicity, including primary human hepatocytes (PHH) and several liver-derived cell lines:
HepG2, HC-04, and HepaRG. These studies provide us an opportunity to compare these cell
models as applied to liver toxicity prediction. In the commonly used HepG2 cells, it was
observed that (i) TGZ induced apoptosis (Yamamoto et al., 2001), which may involve c-Jun
N-terminal protein kinase activation (Bae and Song, 2003) and upregulating apoptotic genes
(Guo et al., 2006), (ii) TGZ induced decreases in MMP followed by cell death (Tirmenstein et
al., 2002) or a rise of intracellular calcium and activation of caspase 3 (Bova et al., 2005), (iii)
TGZ promoted the degradation of PGC-1α protein (Liao et al., 2010), and (iv) TGZ induced
toxicity through involving chaperone proteins (Maniratanachote et al., 2005). In HC-04 cells,
TGZ caused intramitochondrial oxidative stress which in turn led to activation of Ask1-
dependent cell death signaling pathways (Lim et al., 2008). In PHHs, it was observed that TGZ
caused decreases in the oxygen consumption rate (Goda et al., 2016), mitochondrial DNA
damages and dysfunction and cell death (Rachek et al., 2009).
HepRG cells are a recently developed hepatocyte model believed to better resemble
PHHs, especially in metabolic capacity (Guillouzo et al., 2007). They have been used to
characterize TGZ-induced cytotoxicity and mitochondrial toxicity either in 2D (Hu et al., 2015,
Bell et al., 2017) or 3D (Gunness et al., 2013; Hendriks et al., 2016; Ramaiahgari et al., 2017)
cultures. Yet in these studies, the concentration-response data of mitochondrial toxic and
Journal Pre-proof
7
cytotoxic endpoints either did not cover enough concentration points or did not meet other
criteria to allow us to identify POD using the BMD approach. To fill this data gap, we generated
our own in vitro concentration-response data using HepaRG cells in the present study. The
generated HepaRG data and those from other cell models reported in the literature were used
together to identify PODs with the BMD modelling method. We then developed and validated
a human population PBPK model for TGZ and used the model to conduct IVIVE that predicted
the in vivo dose from the in vitro POD TGZ concentration metrics. By comparing the predicted
in vivo dose to clinical dose range where liver injury was observed, our study demonstrated
that the PBPK modelling-based IVIVE approach – when appropriate cells, in vitro biomarker,
POD level, and dose metrics are determined – is a promising alternative to animal toxicity
testing.
2. Materials and Methods
2.1. Cell culture
The human hepatoma HepaRG cells (Biopredict International, Saint Gregory, France) were
maintained according to the supplier’s recommendations with minor modifications. Cells were
thawed and seeded using Williams’ E medium (Gibco, Grand Island, USA) supplemented with
10% fetal bovine serum (FBS, Biological Industries, Kibbutz Beit-Haemek, Israel), 100 IU/mL
penicillin (North China Pharmaceutical Co., Ltd, Shijiazhuang, China), 100 μg/mL
streptomycin (Lukang Pharmaceutical Co., Ltd, Jining, China), 2 mM GlutaMax (Gibco, Grand
Island, USA), 5 μg/mL insulin (National Institutes for Food and Drug Control, Beijing, China),
and 0.5 mM hydrocortisone hemisuccinate (Solarbio, Beijing, China). Cultures were
maintained in a sterile humidified incubator at 37 °C and 5% CO2. Passaging was performed
every 2-3 days. Specifically, when cultures reached 80-90% confluence, the well-grown cells
were rinsed with PBS, digested with 0.25% trypsin-0.02% EDTA for 2 min and observed for
Journal Pre-proof
8
cell morphology under microscope. The trypsin solution was discarded to stop the digestion
when significant cell shrinkage occurred, and then fresh serum-containing cell culture medium
was added immediately. Cells were blown down gently, the cell suspension was centrifugated
for 3 min (1000 rpm), and the supernatant was discarded. Cells were resuspended with culture
medium, counted, and inoculated according to 1 to 4 volume ratio. Besides passaging, the
digested cells were also used for toxicity assays.
2.2. Measurement of cell viability
Cell viability was measured using Alamar Blue (Life Technologies, Eugene, USA) according
to the supplier’s protocol. 7, 000 cells per well were seeded in 96-well plates in 100 µL
Williams’ E medium. TGZ (Batch NO.: 4B/197127, Tocris Bioscience, Bristol, UK) was
dissolved with DMSO at 100 mM, and diluted with the culture medium to reach final
concentrations of 0, 3.12, 6.25, 12.5, 25, 50 µM, respectively, at a volume of 100 µL. Cells
were exposed to different concentrations of TGZ as above for 24 h. 10 μL Alamar Blue was
then added to each well, followed by 2-h incubation at 37 ºC. The fluorescence intensity was
read on a fluorescence spectrophotometer (SpectraMax i3x, Molecular Devices, San Jose, USA)
at excitation and emission wavelengths of 530 nm and 590 nm respectively.
2.3. Determination of mitochondrial superoxide and mitochondrial mass
A fluorescein-staining cocktail of 100 nM Hoechst 33342 (Life Technologies), 5 µM
MitoSOXTM Red (Life Technologies) and 100 nM MitoTracker® Green (Life Technologies)
was prepared with 1 x HBSS (Gibco) at 37 oC. MitoSOXTM Red is a fluorescence dye
selectively targeting mitochondria and labeling superoxide anion. MitoTracker® Green stains
mitochondria independent of membrane potential and is used to detect mitochondrial mass. 7,
000 cells per well were seeded in 96-well plate in 100 µL Williams’ E medium. Cells were
Journal Pre-proof
9
treated with 100 µL of TGZ medium at 0, 1.56, 3.12, 6.25, 12.5, and 25 µM for 24 h.
Subsequently, each well was washed with 1 x HBSS, and 100 µL staining cocktail was added
and incubated for 40 min in the dark. Prior to image acquisition 100 µL HBSS was used for a
final rinse. Assay plates were analyzed using an Operetta High-Content Imaging System
(PerkinElmer, Waltham, USA), with a 10X Plan Fluor objective. Nine digital images were
captured in each well and analyzed using Harmony® 4.1 High Content Imaging and Analysis
Software (PerkinElmer, Waltham, USA).
2.4. Statistical analysis
Data were normalized to control (without TGZ treatment) which was considered as 100%
response and expressed as mean ± SD. All statistical analyses were conducted using one-way
ANOVA in Origin 9.0 (Originlab, Northampton, USA) for evaluating differences between
groups of interest and control. P < 0.05 was considered as statistically significant.
2.5. Literature review for TGZ-induced in vitro cytotoxicity and mitochondrial toxicity studies
The terms “troglitazone”, “liver injury”, “hepatotoxicity”, combined in Boolean logic
“troglitazone and (liver injury or hepatotoxicity)”, were searched in the PubMed database. In
addition, the submission of Rezulin (TGZ tablet) to FDA (FDA, 1999) and relevant references
therein were also examined. An article was accepted in our analysis if it simultaneously met
the following inclusion criteria. (1) Experiments only used human hepatic cell models; (2) Cells
were treated with TGZ for ≤ 24 h. (3) For the purpose of BMD modelling, there were at least
three concentration groups including control, and the mean ± SD (or stand error, SE) was
available when individual data was not reported (EPA, 2016).
2.6. Human PBPK model development and validation
Journal Pre-proof
10
PBPK modelling was performed using the GastroPlusTM 9.6 software (Simulations Plus Inc.,
Lancaster, CA, USA). In brief, the model contains 14 organ/tissue compartments including the
lung, liver, spleen, gut, adipose tissue, muscle, heart, brain, kidney, skin, reproductive organ,
red marrow, yellow marrow, and rest of the body, which were connected by the venous and
arterial blood circulation. A series of differential equations were used to quantitatively simulate
the ADME processes. As TGZ was a small lipophilic molecule with good cell membrane
permeability, perfusion-limited tissue distribution was assumed and the Lukacova (Rodgers￾Single) method was used for calculating the tissue/plasma partition coefficients (Kp) of TGZ
(Zhuang and Lu, 2016). The liver was considered to be the main site of TGZ elimination since
it eliminates 84.5% of TGZ after oral administration in human (Loi et al., 1999b). PBPK
parameters attributed to physiochemical and ADME properties of TGZ are listed in Table 1,
including molecular weight, LogP, acid dissociation constant (pKa), solubility, mean
precipitation time, diffusion coefficient, drug particle density, effective particle radius,
blood/plasma concentration ratio (Rbp), unbound fraction of drug in plasma (Fup%) and
clearance.
Table 1. Troglitazone-related parameters for the construction of PBPK models using
GastroPlusTM
Parameter Value Methods/reference
molecular weight (g/mol) 441.545 DrugBank Plus
logP 3.6 DrugBank Plus
pKa 7.77 Predicted-ADMET
solubility (mg/mL at pH 5.98) 0.0287 Predicted-ADMET
mean precipitation time (sec) 900 Predicted-ADMET
diffusion coefficient (cm2
/s*105
) 0.61 Predicted-ADMET
drug particle density (g/ml) 1.2 Predicted-ADMET
effective particle radius (μm) 25 Predicted-ADMET
Journal Pre-proof
11
Rbp 0.55 Cubitt et al., 2011
Fup% 1 Cubitt et al., 2011
clearance (CL, mL/min/kg) 2.5 Cubitt et al., 2011
To evaluate the accuracy of the PBPK model in predicting TGZ pharmacokinetics in
human, the blood/plasma concentrations predicted by the model with default parameter values
were compared with observed ones in volunteers given multiple oral administrations (P.O.) of
200, 400, 600 mg TGZ tablets daily for 7 consecutive days (Loi et al., 1999a), as well as single
P.O. of 400 mg TGZ tablet (Young et al., 1998), and these studies were not used in model
building. Table 2 shows the volunteer demographics, doses, exposure routes and durations
applied in PBPK model evaluation.
Table 2. Baseline demographic characteristics in the troglitazone pharmacokinetics

N.A.- no subject information available; P.O.- oral administration.
2.7. Sensitivity analysis
Parameter sensitivity analysis (PSA) was performed with GastroPlusTM for oral TGZ tablet
under fed condition, to assess the importance of the selected parameters in predicting Cmax and
AUC. Only one parameter was changed at a time gradually from one-tenth to ten-fold of its

PSA was performed for parameters related to drug physiochemical and ADME properties and
their interactive nature which may influence Cmax and AUC, including small intestine transit
time, stomach transit time, solubility, LogP, molar radius, effective permeability (Peff),
diffusion coefficient, dose, particle radius, particle density, first pass extract from gut, first pass
extract from liver, Rbp, fraction unbound in plasma (Fup%), Kps in organs, and hepatic
clearance.
2.8. Population PBPK simulation
A virtual PBPK population was developed in GastroPlusTM to simulate potential inter￾individual viability in the pharmacokinetics of TGZ. All of the relevant physiological and
pharmacokinetic parameters were randomly sampled from pre-defined distributions for each
individual. Some PBPK parameters, which are dependent on race, gender and age, have
distributions defined by the GastroPlusTM built-in Population Estimates for Age-Related
Physiology (PEAR Physiology) generator. For this study, a virtual population of 100
Caucasians with 50 males and 50 females, age ranging between 35-95 years, body weight
between 60-120 kg and BMI between 25-40, was generated. During population simulation,
mean and highest values of Cmax, and AUC(0-t) after oral administration of TGZ tablet in the
virtual population were calculated.
2.9. Derivation of BMCL5 on in vitro toxicity data
Journal Pre-proof
13
BMD modeling was performed according to the guidance by the US Environmental Protection
Agency (EPA)’s BMD Software (BMDS) version 2.7 (https://www.epa.gov/bmds) to
determine in vitro PODs for TGZ cytotoxicity and mitochondrial toxicity. In vitro data obtained
from the present study and the literature were modeled using BMDS to obtain the benchmark
concentration lower bound (BMCL) using all continuous models provided in BMDS. A 5%
change from the control level was defined as the benchmark response (BMR) in our study. The
requirements for acceptance of a model are: p-value Test 1 < 0.05, p-value Test 2 > 0.05, p￾value Test 3 > 0.05, p-value Test 4 > 0.05, and the absolute value of scaled residual ((observed
minus predicted response)/SE) of the concentration group closest to the BMD ≤ 2. All models
that met the requirements were considered for the determination of BMCL5, but as a
conservative measure, only the lowest BMCL5 was used as the in vitro POD for IVIVE in the
present study.
2.10. Translation of in vitro POD to in vivo POD and evaluation of the IVIVE approach
To translate in vitro POD to in vivo POD, two concentration metrics were used: Cmax and AUC.
Given that TGZ is hepatoxic, the liver concentration was deemed to be the most relevant metric
to use for IVIVE of the in vitro POD concentration. When using the Cmax metric, the lowest
BMCL5, derived from the suite of fitted BMD models for each in vitro endpoint, was aimed at
by adjusting the oral dose of TGZ tablet for the PBPK model such that the predicted liver Cmax
is equal to the lowest BMCL5. When using the AUC metric, the mathematical product of the
lowest BMCL5 and 24-h was aimed at by the PBPK model to achieve an equivalent predicted
AUC of the liver concentration (Daston et al., 2010). Finally, these predicted in vivo POD doses
were compared with the clinical oral doses of TGZ where liver toxicity has been observed to
evaluate the prediction performance of the IVIVE approach.
3. Results
Journal Pre-proof
14
3.1. TGZ induced concentration-dependent decreases in cell viability
The viability of HepaRG cells after exposure to TGZ at concentrations of 0-50 µM for 24 h
was evaluated. Compared with cells not treated with TGZ (control), cell viability decreased as
the concentration of TGZ increased (Fig. 1A). Cell viability for the 12.5, 25, and 50 μM TGZ
groups decreased significantly to 85.4%, 76.4%, and 47.7% of the control level, respectively.
3.2. TGZ induced mitochondrial superoxide accumulation and mitochondrial mass decline
After HepaRG cells were exposed to TGZ for 24 h, mitochondrial superoxide levels, as
monitored by MitoSox Red, were elevated and the 13.7% increase at 3.125 μM and 32.2%
increase at 25 μM were significant compared to control (Fig. 1B). Mitochondrial mass, as
measured with MitoTracker® Green, significantly declined concentration-dependently to 87.5%
of the control level at 25 μM TGZ (Fig. 1C).
3.3. Human PBPK model validation
The PBPK model was validated with clinical plasma concentration-time data. The PBPK
model-simulated TGZ plasma concentration-time profiles (at default parameter values) and the
clinical data are shown in Fig. 2. There was a 0.7-1.3 fold error between the Cmax predicted and
observed, and 0.6-1.1 fold error between the AUC predicted and observed, under
multiple/single dose or fed/fasted conditions (Table 3). Because the prediction errors for all
conditions are within 0.5-2-fold, the PBPK model developed is considered reasonably reliable
to predict the plasma and organ concentrations of TGZ in human (Jiang et al., 2013).
Table 3. The mean observed and predicted dose metrics of TGZ in human under
different dosing regimens
Dose Plasma Cmax (µg/mL) Plasma AUC0-t (µg×h/mL) Reference
Journal Pre-proof
15

3.4. Sensitivity analysis
PSA was performed for oral TGZ tablet under fed condition to determine important parameters
that have large influence on the model. The results show that Cmax is more sensitive to changes
in solubility, Peff, dose, mean drug particle radius, drug particle density, Fup% and hepatic
clearance (Fig. 3A), with lower sensitivity to changes in other parameters such as the transit
time in stomach and small intestine, Rbp, and Kps in organs (results not shown). Compared
with the results for Cmax, AUC(0-t) is sensitive to fewer parameters including dose, Fup% and
hepatic clearance (Fig. 3B).
3.5. Derivation of BMCL5 from in vitro data
The concentration-response data of cytotoxicity and mitochondrial toxicity induced by TGZ,
both from our own experiments and literature, were shown in Fig. 4. The derived BMCL5 of
TGZ ranges between 0.3-6.4 μM for HepaRG, HC-04, PHH and HepG2 cells exposed to TGZ
for 24 h (Table 4). For each cell model and endpoint combination, the BMCL5 values derived
from different BMD models differ by about 2-fold or less and only the lowest was used for
subsequent IVIVE as a conservative measure.
Table 4. BMCL5 derived from the concentration-response curves in Fig. 4
In vitro endpoint Cell model BMCL5 (µM) Studies
Journal Pre-proof
16

Note: the ranges of BMCL5 values represent those derived from different fitted BMD models.
3.6. Translation of in vitro BMCL5 to in vivo POD doses and evaluation of the predictive
performance of the IVIVE approach
Two concentration metrics, liver Cmax and AUC, were used to derive in vivo POD doses for
TGZ from the in vitro BMCL5 by conducting reverse dosimetry via the population PBPK
model developed. There are two alternative underlying assumptions: in a human individual (1)
if the liver Cmax reaches or surpasses BMCL5, liver toxicity may occur; (2) if the liver AUC
reaches or surpassed BMCL5 × 24 h, liver toxicity may occur. For the first case, a single oral
dose of TGZ, i.e., the input to the PBPK model, was varied until the mean liver Cmax predicted
by the population PBPK model equaled BMCL5 and this oral dose was designated as the mean
in vivo POD dose for the Cmax metric. Using BMCL5 for various cell models and cellular
toxicity endpoints, the mean POD doses ranged between 28-372 mg/d, varying by about 13-
fold (Table 5). For the second case, a single oral dose of TGZ was similarly varied until the
mean AUC of liver TGZ concentration in the first 24 h predicted by the population PBPK
model equaled BMCL5 × 24 h and this oral dose was designated as the mean in vivo POD dose
for the AUC metric. Using BMCL5 × 24 h for various cell models and cellular toxicity
Journal Pre-proof
17
endpoints, the POD doses derived ranged between 185-2552 mg/d, varying by about 14-fold
(Table 5). This range of POD doses was about 7-fold higher than that when the Cmaxmetric was
used.
While the above IVIVE analysis was based on the population means of Cmax and AUC,
as a conservative approach we also examined the situation when the Cmax and AUC in the top
one percentile of the virtual population (i.e., the most sensitive subpopulation which has the
highest Cmax or AUC) reached BMCL5 and BMCL5 × 24 h respectively. For the case of using
Cmax as the metric, the POD doses ranged between 15-178 mg/d, varying by about 12-fold
between combinations of various cell models and cellular toxicity endpoints (Table 5). This
range of POD doses is about half of that when population mean Cmax was used. For the case of
using AUC as the metric, the POD doses ranged between 83-1010 mg/d, varying by about 12-
fold between combinations of various cell models and cellular toxicity endpoints (Table 5).
This range of POD doses is slightly more than half of that when population mean AUC was
used.
Table 5. Predicted POD doses of TGZ using PBPK modelling-based IVIVE approach,
based on liver highest (top 1%) Cmax, mean Cmax, highest (top 1%) AUC and mean AUC

viability HepaRG 135 286 793 1878 present study
HepG2 178 372 1010 2552 Liao et al., 2010
HC-04 133 269 757 1783 Lim et al., 2008
PHH 48 90 268 599 Rachek et al., 2009
mito-mass HepaRG 20 42 129 288 present study
Journal Pre-proof
18
mito-ROS HepaRG 158 345 933 2303 present study
HepG2 39 77 230 510 Liao et al., 2010
ATP PHH 15 28 83 185 Rachek et al., 2009
Fold range – 11.9 13.3 12.2 13.8 -
To evaluate the performance of the PBPK modelling-based IVIVE approach, the
predicted POD doses were compared with the frequently used clinical dose range (200-800
mg/d) where liver adverse effects have been observed (Table 6 and references therein). The
results show that the POD doses of TGZ derived based on population mean Cmax were below
or within the clinical 200-800 mg/d dose range (Fig. 5A). However, all the POD doses derived
based on the top one percentile Cmax were below the clinical dose range, with the highest POD
dose 178 mg/d sitting right below the 200 mg/d mark (Fig. 5B). In contrast to using Cmax, using
AUC resulted in considerable overlaps of predicted POD dose range with the clinical 200-800
mg/d dose range (Fig. 6), especially for the r top one percentile AUC, while population mean
AUC seems to overshoot the range.
Table 6. Case reports of TGZ-induced liver dysfunction, injury and failure

q.d.- once daily, b.i.d.- twice daily, t.i.d.-three times daily. ALT-alanine aminotransferase test (normal
5~40 U/L), AST-aspartate aminotransferase test (normal 10~35 U/L).
4. Discussion
Journal Pre-proof
20
While traditional animal-based toxicity testing and risk assessment approach is associated with
uncertainties associated with inter-species extrapolation, inter-individual variabilities and other
uncertainty factors (Hartung, 2009), next-generation risk assessment using the PBPK
modelling-based IVIVE paradigm is associated with uncertainties arising from different
sources. These uncertainties include those relating to choice of cell type, in vitro kinetics and
dose metric, choice of in vitro biomarker(s)/endpoint(s), magnitude and duration of endpoint
departure and determination of inter-individual variability (Zhang et al., 2018). The present
study aimed to evaluate the fitness of the approach method under some of its uncertainties in
informaing chemical doses that may result in liver injury in humans. TGZ was used as a case
study due to its known hepatoxicity and availability ofin vitro assays and clinical data.
Choosing human cells and cellular biomarkers that are fit-for-purpose is important and
probably is the step of the in vitro assay-based IVIVE approach that introduces the most
uncertainty. To evaluate hepatotoxicity, here we used data from 4 different human liver cell
models: PHH and three cell lines, HepaRG, HC-04 and HepG2. Of the cell lines it is unclear
which one best represents the average human liver physiology in vivo, as the difference
between these cells can potentially affect both the toxicokinetics and toxicodynamics (Berger
et al., 2016). HepaRG cells are believed to better resemble primary human hepatocytes
compared to HepG2 and HC-04 cells especially in metabolic capacity, but its other
characteristics compared with the two cell lines are not clear (Guillouzo et al., 2007). In human,
TGZ has three main metabolites: a sulfate conjugate (M1) catalyzed by phenol sulfotransferase
(ST1A3), a glucuronide conjugate (M2) catalyzed by UDP-glucuronosyltransferase (UGT),
and a quinone metabolite (M3) catalyzed by CYPs (CYP3A4, CYP2C8, and CYP2C19) from
sulfation, glucuronidation, and oxidation reactions, respectively. TGZ itself, M1 and M3 are
responsible for the liver toxicity (Yokoi, 2010). The variations of metabolizing enzymes in the
Journal Pre-proof
21
different liver cell models used in this work would result in different amounts of parent and
toxic metabolites, and thus different toxic responses to TGZ.
We show here that the cytotoxicity of TGZ as evaluated by viability assays varied
slightly among the 4 cell models. In PHHs, TGZ treatment resulted in BMCL5 values ranging
between 1.1-2.3 µM, while the three cell lines have higher but similar BMCL5 levels ranging
between 3.1-6.4 µM (Table 4), suggesting that PHHs are more sensitive to TGZ. When using
the lowest cytotoxicity BMCL5 for each cell model and the population mean liver Cmax was
matched to the BMCL5, we found that the POD doses of TGZ predicted with HepaRG, HC-04,
and HepG2 cells fall within the dose range where liver injury has been observed clinically
(200-800 mg/d), while PHH produced a more protective (conservative) POD dose (Fig. 5A).
In comparison, among the POD doses derived from mitochondrial biomarkers, only the one
using HepaRG ROS is within the clinical dose range, while mitochondrial mass and ATP are
conservative as biomarkers for in vivo POD prediction using BMCL5 levels (Fig 5A). When
the liver Cmax of only the top 1% most sensitive virtual population was matched to the BMCL5,
the predicted in vivo POD doses from all cellular biomarkers are lower than the clinical dose
range (Fig. 5B), suggesting overly conservative predictions. Increasing in vitro BMR level
from 5% to 10% or higher for cellular responses may help to make the prediction less
conservative, but uncertainty may increase due to divergence in concentration-response among
different cell models and biomarkers. Moreover, the concentration-response of mitochondrial
mass in HepaRG cells plateaued around 90% of the control (Fig. 4B), therefore identifying
BMCL10 based on BMC10 will likely meet with increasing uncertainty with such data.
The in vivo POD doses predicted by the AUC approach consistently cover the 200-800
mg/d clinical dose range regardless of using the population mean or top one percentile AUC to
Journal Pre-proof
22
match the in vitro BMCL5 AUC, although half of the in vivo POD doses predicted with
population mean surpassed the clinical dose range (Fig. 6). The POD doses predicted by the
AUC metric are generally higher than those predicted by the Cmax metric by about 6-fold (Table
5). This less conservative prediction with AUC may be attributed to the fact that while the in
vitro kinetics of TGZ in the cell models is unknown, BMCL5 × 24 h is likely an overestimate
of the AUC that cells experienced in vitro because TGZ clearance in the medium is not
considered when using BMCL5 × 24 h. Had a lower in vitro AUC value been used due to TGZ
clearance, the derived in vivo POD doses will be correspondingly lower, thus closer to the
values predicted with Cmax.
In risk assessments using results from traditional animal-based toxicology data, unless
known otherwise, a default uncertainty factor of 10 is normally used to account for inter-species
difference (Renwick, 1993). In human cell-based in vitro approaches to risk assessment, this
inter-species uncertainty factor becomes irrelevant (Abdullah et al., 2016). However, risk
assessments using data from in vitro human cells brings in other uncertainties such as selection
of cell models and cellular biomarkers (Boonpawa et al., 2017). Ideally the human cells used
should represent an average response of the particular cell type involved in vivo, however this
can hardly be the case in practice, because many of the cells used as of today were originally
derived from human cancer cells. Even the use of PHH or human iPSC-derived differentiated
cells, which resemble normal cells more closely than cancer-derived cells, also need to be well
characterized for their “averageness” because cells from human individuals can behave very
differently due to inter-individual variability. In the present study, the cell viability
concentration-response diverged between the four in vitro hepatocyte models(Fig. 4). However,
because BMR5 is only a small departure from the baseline, the BMCL5 values for these cells
are still comparable except for PHH cells which as primary cells appear to be more sensitive
Journal Pre-proof
23
to TGZ. The in vivo POD doses derived from these cell models and the multiple cellular
endpoints differed by about one order of magnitude regardless of using the Cmax or AUC
approach (Table 5), suggesting a reasonably constrained uncertainty in this regard.
Another default factor of 10 is also used to account for inter-individual difference in
traditional human health risk assessment (Renwick, 1993), which is an issue that the in vitro
cell assay-based approach also has to face. PBPK modeling-based IVIVE provides an
opportunity to reduce the uncertainty by explicitly accounting for the inter-individual
differences with chemical-specific adjustment factors (CSAF). Human population PBPK
models can take into considerations the variations in physiological parameters and chemical￾specific metabolism arising from genetic polymorphism to derive a CSAF to replace the default
factor of 3.16 for individual-to-individual differences in PK. In the present study, we used
human population PBPK models to explore the POD doses of the population mean as well as
the top one percentile sensitive population. The ratio between these two POD doses is close to
2.0 for each cell model and cellular endpoint combination when Cmax is used, and is close to
2.3 when AUC is used (Table 5). Conceivably the CSAF for the PK step, if calculated as the
ratio of the POD dose of the population mean over that of the top 5% of the population, will be
even smaller, resulting in reduced uncertainty with the IVIVE approach. Regarding the source
of PK variability, it was found that the donors of cryopreserved human hepatocytes which had
lower amounts of glutathione (GSH)-conjugated TGZ were more sensitive to TGZ-induced
hepatoxicity, suggesting differences in the cellular detoxification capability contribute to
human variability (Kostrubsky et al., 2000; Prabhu et al., 2002). Given the role of glutathione
S-transferases (GST) in catalyzing GSH to reactive chemicals or their metabolites, GST gene
differences have been considered as an important contributor to inter-individual variability in
TGZ-induced liver injury (Yokoi, 2010).
Journal Pre-proof
24
5. Conclusion
The present work shows that integrating in vitro assays and PBPK modelling-based reverse
dosimetry in the IVIVE approach provides a promising platform to inform the dose range of
TGZ that induces hepatotoxicity. Yet further optimization remains for such new approach
method before it can be reliably applied with confidence. For example, the free drug
concentration in the culture medium should be considered, which may better correlate with the
free tissue concentration. The predicted PODs vary with the in vitro cell models and endpoints
selected, which is another source of uncertainty in this approach that needs to be further reduced.
Whether the specific in vitro cellular POD used here can be generalized to the safety assessment
of other hepatotoxic chemicals remain to be seen. Journal Pre-proof
25
Conflicts of interest
The authors declare no conflict of interests.
Declaration of interests
☒ The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.
☐The authors declare the following financial interests/personal relationships which may be
considered as potential competing interests:
Acknowledgement
This work was supported by National Natural Science Foundation of China (81430090,
81470167); Beijing Nova Program (Z171100001117103); AMMS Innovative Foundation
(2017CXJJ13) and Unilever International Collaborative Project (MA-2015-00410). Special
thanks go to Alistair Middleton and Maria Baltazar for their great support while the IVIVE part
of the study was carried out at the Unilever R&D Colworth, UK.
Journal Pre-proof
26
Reference
Abdullah, R., Alhusainy, W., Woutersen, J., Rietjens, I.M.C.M., Punt, A., 2016. Predicting
points of departure for risk assessment based on in vitro cytotoxicity data and physiologically
based kinetic (PBK) modeling: The case of kidney toxicity induced by aristolochic acid I.
Food and Chemical Toxicology 92, 104-116.
Babai, S., Auclert, L., Le-Louet, H., 2018. Safety data and withdrawal of hepatotoxic drugs.
Therapie.
Bae, M.A., Song, B.J., 2003. Critical role of c-Jun N-terminal protein kinase activation in
troglitazone-induced apoptosis of human HepG2 hepatoma cells. Molecular pharmacology
63, 401-408.
Bell, C.C., Dankers, A.C.A., Lauschke, V.M., Sison-Young, R., Jenkins, R., Rowe, C.,
Goldring, C.E., Park, K., Regan, S.L., Walker, T., Schofield, C., Baze, A., Foster, A.J.,
Williams, D.P., van de Ven, A.W.M., Jacobs, F., Houdt, J.v., Lähteenmäki, T., Snoeys, J.,
Juhila, S., Richert, L., Ingelman-Sundberg, M., 2018. Comparison of Hepatic 2D Sandwich
Cultures and 3D Spheroids for Long-term Toxicity Applications: A Multicenter Study.
Toxicological Sciences 162, 655-666.
Bell, C.C., Lauschke, V.M., Vorrink, S.U., Palmgren, H., Duffin, R., Andersson, T.B.,
Ingelman-Sundberg, M., 2017. Transcriptional, Functional, and Mechanistic Comparisons of
Stem Cell–Derived Hepatocytes, HepaRG Cells, and Three-Dimensional Human Hepatocyte
Spheroids as Predictive In Vitro Systems for Drug-Induced Liver Injury. Drug Metabolism
and Disposition 45, 419-429.
Berger, B., Donzelli, M., Maseneni, S., Boess, F., Roth, A., Krahenbuhl, S., Haschke, M.,
2016. Comparison of Liver Cell Models Using the Basel Phenotyping Cocktail. Frontiers in
pharmacology 7, 443.
Boonpawa, R., Spenkelink, A., Punt, A., Rietjens, I., 2017. In vitro-in silico-based analysis of
the dose-dependent in vivo oestrogenicity of the soy phytoestrogen genistein in humans.
British journal of pharmacology 174, 2739-2757.
Bova, M.P., Tam, D., McMahon, G., Mattson, M.N., 2005. Troglitazone induces a rapid drop
of mitochondrial membrane potential in liver HepG2 cells. Toxicol Lett 155, 41-50.
Cubitt, H.E., Houston, J.B., Galetin, A., 2011. Prediction of human drug clearance by
multiple metabolic pathways: integration of hepatic and intestinal microsomal and cytosolic
data. Drug metabolism and disposition: the biological fate of chemicals 39, 864-873.
Daston, G., Chapin, R., Scialli, A., Piersma, A., Carney, E., Rogers, J., M Friedman, J., 2010.
A Different Approach to Validating Screening Assays for Developmental Toxicity.
Delzenne, N.M., Calderon, P.B., Taper, H.S., Roberfroid, M.B., 1992. Comparative
hepatotoxicity of cholic acid, deoxycholic acid and lithocholic acid in the rat: in vivo and in
vitro studies. Toxicol Lett 61, 291-304.
Dent, M.P., Li, H., Carmichael, P.L., Martin, F.L., 2018. Employing Dietary Comparators to
Perform Risk Assessments for Anti-Androgens Without Using Animal Data. Toxicological
Sciences 167, 375-384.
Journal Pre-proof
27
EPA, U., 2016. Benchmark Dose Software (BMDS) User Guide.
FDA, 1999. Rezulin (troglitazone) tablets.
https://www.accessdata.fda.gov/drugsatfda_docs/label/1999/20720s12lbl.pdf. Accessed in
November 9th 2017.
Filipsson, A.F., Sand, S., Nilsson, J., Victorin, K., 2003. The benchmark dose method–
review of available models, and recommendations for application in health risk assessment.
Critical reviews in toxicology 33, 505-542.
Fukano, M., Amano, S., Sato, J., Yamamoto, K., Adachi, H., Okabe, H., Fujiyama, Y.,
Bamba, T., 2000. Subacute hepatic failure associated with a new antidiabetic agent,
troglitazone: A case report with autopsy examination. Human Pathology 31, 250-253.
Gitlin, N., Julie, N.L., Spurr, C.L., Lim, K.N., Juarbe, H.M., 1998. Two Cases of Severe
Clinical and Histologic Hepatotoxicity Associated with Troglitazone. Ann. Int. Med. 129, 36-
38.
Goda, K., Takahashi, T., Kobayashi, A., Shoda, T., Kuno, H., Sugai, S., 2016. Usefulness of
in vitro combination assays of mitochondrial dysfunction and apoptosis for the estimation of
potential risk of idiosyncratic drug induced liver injury.
Gores, G.J., Miyoshi, H., Botla, R., Aguilar, H.I., Bronk, S.F., 1998. Induction of the
mitochondrial permeability transition as a mechanism of liver injury during cholestasis: a
potential role for mitochondrial proteases. Biochimica et biophysica acta 1366, 167-175.
Groothuis, F.A., Heringa, M.B., Nicol, B., Hermens, J.L., Blaauboer, B.J., Kramer, N.I.,
2015. Dose metric considerations in in vitro assays to improve quantitative in vitro-in vivo
dose extrapolations. Toxicology 332, 30-40.
Guillouzo, A., Corlu, A., Aninat, C., Glaise, D., Morel, F., Guguen-Guillouzo, C., 2007. The
human hepatoma HepaRG cells: A highly differentiated model for studies of liver
metabolism and toxicity of xenobiotics. Chemico-Biological Interactions 168, 66-73.
Gunness, P., Mueller, D., Shevchenko, V., Heinzle, E., Ingelman-Sundberg, M., Noor, F.,
2013. 3D Organotypic Cultures of Human HepaRG Cells: A Tool for In Vitro Toxicity
Studies.
Guo, L., Zhang, L., Sun, Y., Muskhelishvili, L., Blann, E., Dial, S., Shi, L., Schroth, G.,
Dragan, Y.P., 2006. Differences in hepatotoxicity and gene expression profiles by anti￾diabetic PPAR gamma agonists on rat primary hepatocytes and human HepG2 cells.
Molecular diversity 10, 349-360.
Hartung, T., 2008. Thoughts on limitations of animal models. Parkinsonism & Related
Disorders 14, S81-S83.
Hartung, T., 2009. Toxicology for the twenty-first century. Nature 460, 208-212.
Hendriks, D.F.G., Fredriksson Puigvert, L., Messner, S., Mortiz, W., Ingelman-Sundberg, M.,
2016. Hepatic 3D spheroid models for the detection and study of compounds with cholestatic
liability. Scientific reports 6, 35434.
Journal Pre-proof
28
Hu, D., Wu, C.Q., Li, Z.J., Liu, Y., Fan, X., Wang, Q.J., Ding, R.G., 2015. Characterizing the
mechanism of thiazolidinedione-induced hepatotoxicity: An in vitro model in mitochondria.
Toxicology and applied pharmacology 284, 134-141.
Jiang, X.L., Zhao, P., Barrett, J.S., Lesko, L.J., Schmidt, S., 2013. Application of
physiologically based pharmacokinetic modeling to predict acetaminophen metabolism and
pharmacokinetics in children. CPT: pharmacometrics & systems pharmacology 2, e80.
Kohlroser, J., Mathai, J., Reichheld, J., Banner, B.F., Bonkovsky, H.L., 2000. Hepatotoxicity
due to troglitazone: report of two cases and review of adverse events reported to the United
States Food and Drug Administration. The American Journal of Gastroenterology 95, 272-
276.
Kostrubsky, V.E., Sinclair, J.F., Ramachandran, V., Venkataramanan, R., Wen, Y.H., Kindt,
E., Galchev, V., Rose, K., Sinz, M., Strom, S.C., 2000. The role of conjugation in
hepatotoxicity of troglitazone in human and porcine hepatocyte cultures. Drug metabolism
and disposition: the biological fate of chemicals 28, 1192-1197.
Kullak-Ublick, G.A., Beuers, U., Paumgartner, G., 2000. Hepatobiliary transport. Journal of
Hepatology 32, 3-18.
Kung, J., Henry, R.R., 2012. Thiazolidinedione safety. Expert opinion on drug safety 11,
565-579.
Li, H., Heller, D.S., Leevy, C.B., Zierer, K.G., Klein, K.M., 2000. Troglitazone-induced
fulminant hepatitis: Report of a case with autopsy findings. Journal of Diabetes and its
Complications 14, 175-177.
Li, H., Zhang, M., Vervoort, J., Rietjens, I.M., van Ravenzwaay, B., Louisse, J., 2017. Use of
physiologically based kinetic modeling-facilitated reverse dosimetry of in vitro toxicity data
for prediction of in vivo developmental toxicity of tebuconazole in rats. Toxicol Lett 266, 85-
93.
Liao, X., Wang, Y., Wong, C.W., 2010. Troglitazone induces cytotoxicity in part by
promoting the degradation of peroxisome proliferator-activated receptor gamma co-activator-
1alpha protein. British journal of pharmacology 161, 771-781.
Lim, P.L., Liu, J., Go, M.L., Boelsterli, U.A., 2008. The mitochondrial
superoxide/thioredoxin-2/Ask1 signaling pathway is critically involved in troglitazone￾induced cell injury to human hepatocytes. Toxicological sciences : an official journal of the
Society of Toxicology 101, 341-349.
Loi, C.M., Alvey, C.W., Vassos, A.B., Randinitis, E.J., Sedman, A.J., Koup, J.R., 1999a.
Steady-state pharmacokinetics and dose proportionality of troglitazone and its metabolites.
Journal of clinical pharmacology 39, 920-926.
Loi, C.M., Young, M., Randinitis, E., Vassos, A., Koup, J.R., 1999b. Clinical
pharmacokinetics of troglitazone. Clin Pharmacokinet 37, 91-104.
Louisse, J., Bosgra, S., Blaauboer, B.J., Rietjens, I.M.C.M., Verwei, M., 2015. Prediction of
in vivo developmental toxicity of all-trans-retinoic acid based on in vitro toxicity data and in
silico physiologically based kinetic modeling. Archives of toxicology 89, 1135-1148.
Journal Pre-proof
29
Louisse, J., de Jong, E., van de Sandt, J.J.M., Blaauboer, B.J., Woutersen, R.A., Piersma,
A.H., Rietjens, I.M.C.M., Verwei, M., 2010. The Use of In Vitro Toxicity Data and
Physiologically Based Kinetic Modeling to Predict Dose-Response Curves for In Vivo
Developmental Toxicity of Glycol Ethers in Rat and Man. Toxicological Sciences 118, 470-
484.
Maniratanachote, R., Minami, K., Katoh, M., Nakajima, M., Yokoi, T., 2005. Chaperone
proteins involved in troglitazone-induced toxicity in human hepatoma cell lines.
Toxicological sciences : an official journal of the Society of Toxicology 83, 293-302.
Neuschwander-Tetri, B.A., Isley, W.L., Oki, J.C., Ramrakhiani, S., Quiason, S.G., Phillips,
N.J., Brunt, E.M., 1998. Troglitazone-induced hepatic failure leading to liver transplantation.
A case report. Annals of internal medicine 129, 38-41.
NRC, 2007. Toxicity Testing in the 21st Century: A Vision and a Strategy. The National
Academies Press, Washington, DC.
NRC, 2017. Using 21st Century Science to Improve Risk-Related Evaluations, Using 21st
Century Science to Improve Risk-Related Evaluations, pp. 1-182.
Plosker, G.L., Faulds, D., 1999. Troglitazone: a review of its use in the management of type 2
diabetes mellitus. Drugs 57, 409-438.
Prabhu, S., Fackett, A., Lloyd, S., McClellan, H.A., Terrell, C.M., Silber, P.M., Li, A.P.,
2002. Identification of glutathione conjugates of troglitazone in human hepatocytes.
Chemico-Biological Interactions 142, 83-97.
Rachek, L.I., Yuzefovych, L.V., Ledoux, S.P., Julie, N.L., Wilson, G.L., 2009. Troglitazone,
but not rosiglitazone, damages mitochondrial DNA and induces mitochondrial dysfunction
and cell death in human hepatocytes. Toxicology and applied pharmacology 240, 348-354.
Ramaiahgari, S.C., Waidyanatha, S., Dixon, D., DeVito, M.J., Paules, R.S., Ferguson, S.S.,
2017. From the Cover: Three-Dimensional (3D) HepaRG Spheroid Model With
Physiologically Relevant Xenobiotic Metabolism Competence and Hepatocyte Functionality
for Liver Toxicity Screening. Toxicological sciences : an official journal of the Society of
Toxicology 159, 124-136.
Renwick, A.G.,1993. Data‐ derived safety factors for the evaluation of food additives and
environmental contaminants. Food Additives & Contaminants 10, 275-305.
Rowland, M., Peck, C., Tucker, G., 2011. Physiologically-based pharmacokinetics in drug
development and regulatory science. Annual review of pharmacology and toxicology 51, 45-
73.
Shibuya, A., Watanabe, M., Fujita, Y., Saigenji, K., Kuwao, S., Takahashi, H., Takeuchi, H.,
1998. An Autopsy Case of Troglitazone-Induced Fulminant Hepatitis. Diabetes Care 21,
2140-2143.
Strikwold, M., Spenkelink, B., de Haan, L.H.J., Woutersen, R.A., Punt, A., Rietjens, I., 2017.
Integrating in vitro data and physiologically based kinetic (PBK) modelling to assess the in
vivo potential developmental toxicity of a series of phenols. Archives of toxicology 91, 2119-
2133.
30
Strikwold, M., Spenkelink, B., Woutersen, R.A., Rietjens, I.M.C.M., Punt, A., 2013.
Combining in vitro embryotoxicity data with physiologically based kinetic (PBK) modelling
to define in vivo dose–response curves for developmental toxicity of phenol in rat and human.
Archives of toxicology 87, 1709-1723.
Thiel, C., Cordes, H., Conde, I., Castell, J.V., Blank, L.M., Kuepfer, L., 2017. Model-based
contextualization of in vitro toxicity data quantitatively predicts in vivo drug response in
patients. Archives of toxicology 91, 865-883.
Tirmenstein, M.A., Hu, C.X., Gales, T.L., Maleeff, B.E., Narayanan, P.K., Kurali, E., Hart,
T.K., Thomas, H.C., Schwartz, L.W., 2002. Effects of Troglitazone on HepG2 Viability and
Mitochondrial Function. Toxicological Sciences 69, 131-138.
Vella, A., de Groen, P.C., Dinneen, S.F., 1998. Fatal hepatotoxicity associated with
troglitazone. Annals of internal medicine 129, 1080.
Yamamoto, Y., Nakajima, M., Yamazaki, H., Yokoi, T., 2001. Cytotoxicity and apoptosis
produced by troglitazone in human hepatoma cells. Life sciences 70, 471-482.
Yokoi, T., 2010. Troglitazone. Handbook of experimental pharmacology, 419-435.
Young, M.A., Lettis, S., Eastmond, R., 1998. Improvement in the gastrointestinal absorption
of troglitazone when taken with, or shortly after, food. British journal of clinical
pharmacology 45, 31-35.
Zhang, Q., Li, J., Middleton, A., Bhattacharya, S., Conolly, R.B., 2018. Bridging the Data
Gap From in vitro Toxicity Testing to Chemical Safety Assessment Through Computational
Modeling. Frontiers in public health 6, 261.
Zhuang, X., Lu, C., 2016. PBPK modeling and simulation in drug research and development.
Acta pharmaceutica Sinica. B 6, 430-440. Journal Pre-proof
31
Figure Legend
Fig. 1. TGZ-induced cytotoxicity and mitochondrial toxicity in HepaRG cells. Cells were
treated with TGZ at various concentrations for 24 h. Cell viability (A) was measured using
Alamar Blue assay. Mitochondrial ROS (B) and mitochondrial mass (C) were determined by
high content analysis. Data are presented as means ± SD from 3 independent experiments. *
P<0.05 compared with control. Journal Pre-proof
32
Fig. 2. Mean observed vs. predicted plasma concentration profiles of TGZ in human for model
validation following various oral doses. Symbols represent mean observed data digitized from
published studies indicated. P.O., oral administration. Journal Pre-proof
33
Fig. 3. Parameter sensitivity analysis for oral administration of TGZ tablet under fed condition.
The Y-axis is simulated Cmax (A) and AUC (B). The center of the X-axis for each parameter
tested represents the default value that was used in the simulations shown in Fig. 2. Each of the
X-axis scales shows the lower and higher bounds of the parameter values for PSA. RefSol
(reference solubility), Peff (effective permeability), RadPart (drug particle radius), Fup
(fraction unbound in plasma), CL-liver (liver clearance).
Fig. 4. In vitro concentration-response curves of toxicity endpoints for TGZ in human liver cell
models. Data are presented as mean ± SD. Journal Pre-proof
34
Fig. 5. Predicted POD single oral dose of TGZ in human by using PBPK modelling-based
IVIVE approach with liver Cmax as the surrogate metric. (A) Using mean Cmax of the virtual
population, (B) using top one percentile Cmax of the virtual population. Each diamond symbol
denotes the POD dose derived from using in vitro BMCL5 (the lowest predicted by different
BMD models) of a particular combination of cell model and in vitro endpoint. The two vertical
dashed lines denote the clinical dose range (200-800 mg/d).
Journal Pre-proof
35
Fig. 6. Predicted POD single oral dose of TGZ in human by using PBPK modelling-based
IVIVE approach with liver AUC as the surrogate metric. (A) Using mean AUC of the virtual
population, (B) using top one percentile AUC of the Troglitazone virtual population. Each diamond symbol
denotes the POD dose derived from using in vitro BMCL5 (the lowest predicted by different
BMD models) of a particular combination of cell model and in vitro endpoint. The two vertical
dashed lines denote the clinical dose range (200-800 mg/d). Journal Pre-proof