In this article, we review the past applications of in vitro models in

identifying human hepatotoxins and then focus on the use of

multiscale experimental models in drug development, including the
use of zebrafish and human cell-based, 3-dimensional (3D),
microfluidic systems of liver functions as key components in applying
Quantitative Systems Pharmacology (QSP). We have implemented QSP
as a platform to improve the rate of success in the process of drug
discovery and development of therapeutics.1,2 Our working definition
of QSP is “Determining the mechanism(s) of disease progression and
mechanism(s) of action of drugs on multiscale systems through
iterative and integrated computational and experimental methods to
optimize the development of therapeutic strategies” (Fig. 1).
Fig. 1.
QSP is an approach to drug discovery and development that applies iterative and
integrated computational and experimental methods to determine the mechanism(s) of
disease progression and mechanism(s) of action of drugs on multiscale systems. QSP
starts with patients and patient samples, applies computational and experimental
models, and ends with fundamental knowledge that optimizes therapeutic treatments
for patients. This article focuses on the use of multiscale experimental models for liver
toxicity and efficacy testing, especially phenotypic models.

The stakeholders involved in drug development from academia,

industry, and government agencies have long understood the need to
improve drug candidate selection by optimizing efficacy while
screening out potential toxins so as to concentrate efforts on
candidates with favorable chances for market approval. A survey of
the number of new drugs released between 2000 and 2009
demonstrated a 25-year low in drug approvals despite increases in
research and development (R&D) investment.3 Laverty and
colleagues4 reported that 66% of failed clinical trials were due to a
lack of efficacy and 21% for unacceptable drug toxicity. However,
Sacks and colleagues5 evaluated clinical drug trials between 2000 and
2012 using additional criteria to further refine the analysis and
reported that lack of efficacy alone accounted for only 41% of failures,
while the combination of poor efficacy and safety accounted for 35%,
and safety alone accounted for 19% of drug failures. Together,
cardiovascular and liver toxicity accounted for nearly 75% of all
postmarket drug withdrawals in the United States between 1975 and
2007.6 Although cardiotoxicity has recently surpassed hepatotoxicity
as the main organ toxicity ending clinical trials or causing postmarket
drug withdrawal, hepatotoxicity has been the most frequent cause of
drug product recalls between 1953 and 2014.7
Go to:

CURRENT STATUS OF DRUG HEPATOTOXICITY PREDICTION

USING MAMMALIAN IN VIVO MODELS

The liver is responsible for a wide range of functions, including

xenobiotic detoxification, protein synthesis, synthesis and storage of
glucose, production of the bile necessary for digestion, and regulation
of blood cholesterol and triglycerides. The organ is positioned
downstream of the gastrointestinal tract to enable “first-pass”
clearance of orally ingested drugs and toxins. The structural
organization of the liver sinusoidal space facilitates close contact
between circulating compounds and the transporter-rich hepatocyte
membrane proteins that allow for rapid and efficient transport of
drugs from the portal blood. The high capacity for biotransformation
in the hepatocyte also facilitates the generation of reactive
metabolites that can cause liver damage.8

The key data for assessing hepatotoxicity derives from drug safety
study protocols approved by regulatory agencies. Although evidence
suggests that preclinical animal studies can predict up to
approximately 70% of human toxicity, several problems are apparent
from this approach.6 First, the traditional animal studies clearly fail to
identify all possible adverse liver effects, because many compounds
pass safely through animal testing only to be found hepatotoxic in the
clinical trials or in the postmarketing. Second, the traditional drug
safety protocols were designed only to ask broad questions with a
simple yes or no answer; for example, is the compound hepatotoxic? Is
the compound a reproductive toxin? Traditional preclinical drug
safety assessments essentially ignored the why and the how of
toxicity.9 Historically, few efforts were made to link the observational
results from drug safety studies to molecular and cell level events,
mechanisms of toxicity (MOTs), or to interactions between tissues and
organs.

Adherence to the regulatory agencies’ drug safety protocols also failed

to account for the poor concordance between animal and human
organ toxicity (Table 1). The concordance can be as low as 40% for the
liver to better than 90% for drugs with hematological liability.10 A
good demonstration of how animal testing failed to identify clinical
hepatotoxic drugs is exemplified by a series of structurally similar and
marketed drug pairs for the same therapeutic indication (Table 2). In
each case, one of the drugs in the pair exhibited no hepatotoxicity
during preclinical and clinical trials or in postmarket surveillance, but
the other was “silent” during the animal studies, yet induced
hepatotoxicity in clinical trials or in the postmarket release. This
discordance in the liver findings has been attributed to differences in
the metabolism and metabolic clearance pathways between man and
animal test species.11

Table 1
Concordance of animal and human organ toxicity

Target Organ Concordance, %

Liver 40–5410,52
Cutaneous/Ophthalmic 36
Endocrine 60
Urinary tract 64
Neurologic 70
Cardiovascular 80
Gastrointestinal 85
Hematologic 91
Open in a separate window
Data from Olson H, Betton G, Robinson D, et al. Concordance of the toxicity of
pharmaceuticals in humans and in animals. Regul Toxicol Pharmacol 2000;32(1):56–67.

Table 2
Clinical liver effects of structurally similar drugs found safe in animals

Human Liver Human Liver

Toxic Drug Structure Safe Drug Structure
Nefazodone Trazodone

Bromfenac Diclofenac

Alpidem Zolpidem
Human Liver Human Liver
Toxic Drug Structure Safe Drug Structure
Trovafloxacin Moxifloxacin

Ibufenac Ibuprofen

Tolcapone Entacapone

Open in a separate window

After expensive postmarket drug recalls for Troglitazone and

Bromfenac and the restrictive labeling for Trovafloxacin and
Tolcapone, the pharmaceutical industry initiated strategies to
prescreen compounds for liver toxicity. In one example, the industry
successfully implemented in vitro and specialized in vivo
pharmacokinetic (PK) screening early in the lead optimization
process. Although in 1993 nearly 39% of compounds failed in clinical
trials due to poor PK, this fell dramatically to 7% by 2003 following
implementation of early drug discovery PK profiling.12 However,
during the same period of time, the rate of compounds failing in the
clinic due to toxicity rose from 10% to 16%.12 Based on the success in
PK profiling, it was projected that a strategy similar to early PK
screening also would be successful in identifying and eliminating
potential hepatotoxins before proceeding into preclinical testing.
Go to:

IN VITRO MODELS FOR PREDICTING DRUG-INDUCED

HEPATOTOXICITY

Many commonly used in vitro hepatotoxicity assays rely on

subcellular liver fractions, established hepatoma cell lines, primary
animal and human hepatocytes, liver slices, and whole perfused livers.
The use of in vitro data from microsomes, primary hepatocytes and S9
subcellular fractions to predict in vivo drug clearance is generally a
well-accepted procedure.13,14 Whole perfused livers and liver slices on
the other hand are low throughput, require continual usage of
animals, are costly, and still suffer the lack of concordance with human
toxic liabilities. These models have been reviewed elsewhere.11

Rodent and human primary hepatocytes have become a mainstay of

hepatotoxicity testing in the in vitro laboratory.11 Large numbers of
healthy hepatocytes can be isolated from a single rat or from a human
liver resection or autopsy. This has allowed moderate to high-
throughput screening to identify potential hepatotoxins, while at the
same time reducing animal use, amount of test agent, cost per
compound tested, and the time required to make a toxic liability
decision. The primary hepatocyte has been a convenient model for
investigating MOT, pharmacokinetics, identification of metabolites,
and dose response toxicity. However, an important limitation in the
use of isolated primary hepatocytes is a reduction in function and
differentiation after 24 to 48 hours in culture.15 To avoid this issue,
many in vitro hepatotoxicity tests have been conducted using
established human hepatoma cell lines such as HepG2 and HepaRG or
with immortalized primary hepatocytes such the Corning HepatoCell
(Corning, NY).16–18 The limitation to these cell types, however, is low or
absent biotransformation for many of the important cytochrome P450
enzymes and phase II conjugation reactions involved in drug
clearance.11,19 The advantages and disadvantages of the more
commonly used single-cell and cell fraction in vitro models are
compared in Table 3.

Table 3
Commonly used cell and cellular fraction in vitro models for absorption, distribution,
metabolism, excretion, and toxicity (ADMET)

Model Application Advantage Disadvantage Reference

Hepatocyte Liver Fast, inexpensive, Overestimates in 14,53
cell fractions clearance, individual variation can vivo metabolism;
microsomes metabolite ID be studied only CYP and
UGT enzymes
Model Application Advantage Disadvantage Reference
S9 Liver Phase I and phase II Lower enzyme 14
clearance, activity activity in the S9
metabolite ID fraction, may miss
low level
metabolites
1° hepatocyte Liver In vivo levels of drug 4-h time limit 19
suspensions clearance, metabolism and
metabolite ID transport proteins,
cryopreservation
Transgenic Metabolite ID Single enzyme reactions Overestimate 14
cell lines for generate high levels of involvement of one
PK metabolites for enzyme species
structural ID
Established Toxicity Established cell line, Absence or low 14,19
liver cells testing, MOT, inexpensive expression of most
lines HepG2 induction phase I and phase
II enzymes
HepaRG Toxicity CYP1A2 and 3A4 High CYP3A4, 19,54
testing, MOT, inducible,established Cyp 7A1
induction, cell line expression, but low
metabolite ID in all other CYP
levels compared
with primary
hepatocytes
Primary Heps PK, toxicity Well characterized, Decline in 14
testing, intact metabolism, intact differentiated
metabolite ID transporters, functions; no
cryopreservation immune or fibrosis
cells; single donor
variation
Spheroids: PK, toxicity Cell functions 55–57
Extend differentiated
established testing, lower than primary
cell functions from days
cell lines metabolite ID hepatocytes, low
to weeks, 3D, improved
HepG2, urea, albumin
metabolism
HepaRG production
Open in a separate window
Abbreviations: CYP, cytochrome p450; Heps, hepatocytes; MOT, mechanisms of toxicity;
PK, pharmacokinetic; UGT, udp-glucuronosyltransferase

The primary hepatocyte and immortalized hepatocytes have been

used in 2D monolayers, 2D co-cultures, 3D single-cell type, and 3D co-
cultures depending on the questions posed. Two-dimensional
monolayer assays have been applied to collect simple cell death end
points to triage large numbers of compounds.11 In recent years, high-
throughput screening (HTS) assays have been used to measure MOTs
known to be relevant to mechanisms of clinical hepatotoxicity. The
latter assay type, which is often referred to as “fit-for-purpose,” tests
compounds in a model designed solely to identify a specific
mechanism of toxicity. Examples of “fit-for-purpose” hepatocyte
assays include mitochondrial dysfunction, oxidative stress, bile salt
exporter protein inhibition, covalent binding, and pregnane X receptor
nuclear receptor modulation. These 5 mechanisms have a
demonstrable association to increased risk for clinical
hepatotoxicity.20–24

Although preclinical animal testing remains critical for Investigational

New Drug and New Drug Application approval, a significant shift to
alternative approaches using quantitative structure-activity
relationships (QSAR) computational models, simple in vitro
cytotoxicity and “fit-for-purpose” HTS assays have been promoted by
governmental agencies such as the Environmental Protection Agency,
the National Center for the Advancement of Translational Sciences
(NCATS), and the National Toxicology Program. These initiatives have
resulted in the development of a number of databases (eg, Tox21,
ToxCast) and models for prioritizing compounds based on HTS assay
hepatotoxicity, as well as other organ toxicities.25 In addition to the
government initiatives, most R&D organizations in academia and the
pharmaceutical industry have a slate of in vitro toxicity assays and
computational approaches designed to eliminate unfavorable
compounds early in the drug discovery process.26
Go to:

PAST EXPERIENCES WITH IN VITRO LIVER ABSORPTION,

DISTRIBUTION, METABOLISM, EXCRETION, AND TOXICITY
MODELS PREDICTING HUMAN CLINICAL TOXICITY

The concordance of in vitro toxicity testing with clinical hepatotoxicity

varies from only 25% for the simple in vitro assays to nearly 80% for
the more complex assays and analyses.17,27 Table 4 presents a subset of
computational approaches: in vitro 2D human cell-based HTS assays;
a human cell co-culture model; and one in vitro covalent binding assay
selected from a joint Food and Drug Administration (FDA) National
Center for Toxicology Research, California Institute of Technology, and
Hannover Medical School report which cross-validated the
concordance of these methods to human drug-induced liver injury
(DILI).27 Three interesting findings are noteworthy from the results:
(1) multiparametric cell-based models performed better than the
computational or cell-free covalent binding assays; (2) the predictive
result for DILI-negative drugs (assay specificity) was always higher
than the predictive result for DILI-positive drugs (assay sensitivity);
and (3) an apparent upper limit to the predicted DILI nearing 75% to
80% was reached with the existing in vitro cell-based systems.27 A
likely explanation for the plateau of predictive concordance of 80%
with existing in vitro cell models is the reductionist strategy used to
simplify the organized, multicellular complexity of the liver
microenvironment to single-cell or even 2-cell type testing assays.

Table 4
Concordance of large-scale in vitro screening to human hepatotoxicitya

# True True
Name/Ty Compou Overallb Concord Positiv Negativ Refere
Model pe nds ance, % es, % es, % nce
In silico
models
Structural
alert DEREKc 623 56 46 73 58
4
QSAR/Mole commerci
cular al
descriptors programsd ∼1600 63 39 87 59
2D 382 76 76 75 60
molecular
descriptor
PaDel 1087 69 67 70 27,61
molecular
descriptor
e

ECF 6 295 59 53 65 62
molecular
descriptor
sf
 # True True
Name/Ty Compou Overallb Concord Positiv Negativ Refere
Model pe nds ance, % es, % es, % nce
2D in vitro cell
models
6
63
HepG2 endpointg 102 70 40 100
11
HepG2 endpointh 136 76 N/Ai N/A 17
1° Human 4
hepatocytes endpointj 344 75–80 50–60 95–100 64
Co-cultured Micropatt 45 78 90 65
1° human ern
66
hepatocytes, surface 4
stromal cells endpointk
In vitro cell- Covalent 223 68 45 90 66
free assays binding
assay-
glutathion
e adduct
formation
Open in a separate window
Abbreviations: QSAR, quantitative structure-activity relationship; 2D, 2-dimensional.
aCross-validated results.

bConcordance = True positives (sensitivity) + true negatives (specificity)/2.

cDEREK - toxicity prediction software (Lhasa, Leeds, UK)

dMC4PC, MDL-QSAR, BioEpisteme, Predictive Data Miner.

ePaDel open source software for calculating molecular descriptors.

fECF (extended connectivity functional fingerprints).

gCell counts, nuclear area, plasma membrane integrity, lysosomal activity, mitochondrial

membrane potential and mitochondrial area.

hCell count, DNA degradation, nuclear size, cytoskeletal disruption, DNA damage

response, oxidative stress, mitosis, stress kinase, mitochondrial membrane potential

and area, cell cycle arrest. Results not cross validated.
iNot available.

jCell count, mitochondrial damage, oxidative stress, and intracellular glutathione.

kGlutathione levels, ATP levels, albumin, and urea secretion.

Data from Chen M, Bisgin H, Tong L, et al. Toward predictive models for drug-induced
liver injury in humans: arewe there yet? Biomark Med 2014;8(2):201–13.

Go to:
ZEBRAFISH LARVAE AS A LOW-COST, MEDIUM-
THROUGHPUT, WHOLE-ORGANISM PLATFORM TO PREDICT
DRUG HEPATOTOXICITY

The use of animals for experimentation, especially warm-blooded

species, presents ethical concerns, and governing bodies now strive to
implement the reduction, refinement, and replacement (“3R”) of
animals strategy in research.28 The European Union Directive
2010/63/EU on the protection of animals used for scientific purposes
requires the use of species with the lowest capacity to experience
pain, suffering, and distress, and mandates that the smallest number
of animals be used to obtain scientifically valid results. Studies in
rodents are further limited by the high costs for acquisition and
maintenance.

In recent years, there has been an increased recognition that in vitro

phenotypic experimental cell models, as well as small multicellular
organisms, can be used in the multiscale approach described as part of
QSP.1 Zebrafish in particular have attracted attention not only as a
model for drug discovery, but also as a preclinical model for toxicity
assessments.29–32 Their prospective position in drug discovery and
toxicity assessment is envisioned to be a bridge between simple cell-
based and the still mandated mammalian testing.33

Zebrafish are uniquely positioned for large-scale experimentation.

Zebrafish are vertebrate animals with high similarity to mammals,
both organotypically and physiologically. They have a tractable,
diploid genome that is 70% to 80% similar to humans and that is
amenable to both forward and reverse genetics. Because of their small
size, zebrafish, at the larval stage, are compatible with multiwell plate
formats used in HTS, requiring only small amounts of
compounds/drugs. Their high fecundity makes it possible to obtain
large numbers of specimens for experimentation, dramatically
reducing cost compared with rodent models. The zebrafish embryo
therefore provides a cost-effective opportunity to discover potential
drug liabilities using functional assays in a living animal as a
complement to the emerging human tissue models.
The zebrafish embryonic liver is completely developed and functional
by 72 hours post fertilization (hpf), as judged by organ appearance
and functional markers, such as phase 1 and phase 2
biotransformation capabilities, serum protein secretion, glycogen
storage, and lipogenesis.34–36 Importantly, transgenic zebrafish larvae
expressing human Cyp3A4 have been developed and will find use in
PK and toxicity testing.37

Assays for zebrafish hepatotoxicity have thus far mainly been

observational. In zebrafish larvae, necrotic cells can be visually
identified by a change in appearance from translucency to opaque
black.31 A major shortcoming of macroscopic cell death assays is that
they are not sensitive enough to detect early toxicity.38 Nonetheless, a
variant of this methodology using liver degeneration, changes in size,
and yolk sac retention as endpoints has recently been published and
shown to predict 8 of 8 known hepatotoxicants.39 Changes in liver
appearance, for example, cellular organization, interactions, and
shape, also can be detected by histopathology from tissue slices,
although this method is time-consuming, requires a trained
pathologist, and therefore is usually reserved to validate observations
by other measurements. Last, hepatotoxicity can be assessed in the
adult zebrafish using canonical liver enzyme assays (eg, alanine
aminotransferase), although the use of adults eliminates the
convenience, ethical impact, and high-throughput compatibility that
embryos offer.40 Our own data suggest that even gross organism
toxicity, assessed by visual inspection of morphologic changes in 72
hpf larvae (ie, bent tails, distended peritoneum and edema, pericardial
congestion) can distinguish hepatotoxic from nontoxic agents (Table
5).

Table 5
Gross morphologic observations in zebrafish larvae identify known toxicants including
hepatotoxic amiodarone

Compound/Dose Gross Concentration,

Range Morphology μM

Dose range 200 66 20 6.6 2 0.66 DMSO

Compound/Dose Gross Concentration,
Range Morphology μM

Menadione Live/deada 0/4 0/4 0/4 4/0 4/0 4/0 4/0

Visual toxicityb 1/4 2/4 0/4 0/4

Amiodarone Live/dead 0/4 0/4 4/0 4/0 4/0 4/0 4/0

Visual toxicity 2/4 3/4 0/4 1/4 0/4

Dose range 1000 300 100 30 10 3 DMSO

Caffeine Live/dead 4/0 4/0 4/0 4/0 4/0 4/0 4/0

Visual toxicity 3/4 4/4 0/4 0/4 0/4 0/4 0/4

Dose range 20 6.6 2 0.66 0.2 0.07 DMSO

CCCP Live/dead 1/4 4/0 4/0 4/0 4/0 4/0 4/0

Visual toxicity 0/1 2/4 3/4 3/4 4/4 2/4 0/4

Rotenone Live/dead 1/4 3/4 4/0 4/0 4/0 4/0 4/0

Visual toxicity 1/1 0/3 1/4 2/4 3/4 2/4 0/4
Open in a separate window
Abbreviations: CCCP, Carbonyl cyandide m-chorophenyl hydrazone; DMSO, dimethyl
sulfoxide.
aTranslucent (live), Opaque (dead).

bBent tail, distended peritoneum, edema, pericardial congestion.

Zebrafish toxicity research is now shifting from observational to

mechanism-based toxicity assays. Mesens and colleagues41 explored a
molecular endpoint that captures effects on lipid metabolism because
liver injury is frequently associated with perturbations in lipid
metabolism. Hence, the group looked at expression of liver-specific
fatty acid binding protein 10a (L-FABP 10a) as a molecular biomarker
for hepatotoxicity and found that changes in expression were
predictive of specific mechanisms. A corollary example was recently
published by Verstraelen and colleagues,42 in which they evaluated the
expression of 5 liver-specific genes (including 2 apoptosis, and 2
metabolism-related) following exposure with 5 known toxicants.
Their results confirmed those of Mesens and colleagues41 with L-FABP
10a and further documented that biomarker responses are
compound-dependent, mechanism-dependent, and concentration-
dependent. At the present time, the utility of biomarkers for
prediction of toxicity appears to have potential in “fit-for-purpose”
studies. If highly predictive biomarkers can be found, the zebrafish
offers the opportunity to generate transgenic reporter lines that
would greatly increase throughput.

Our own work has embraced adaptation of in vitro, human

mechanism-based toxicity models and screening in zebrafish. In
addition to morphologic observations (see Table 5), this suite of
assays includes measurements of reactive oxygen species (ROS) and
mitochondrial membrane perturbations because they are very good
predictors of clinical toxicity.24 Fig. 2 illustrates the utility of ROS
measurements in zebrafish larvae using menadione. Menadione is a
naphthoquinone that generates ROS in cells through redox cycling.
Menadione caused time-dependent and dose-dependent generation of
ROS in zebrafish that correlated with embryonal toxicity and ROS
induction and death in cultured hepatocytes, although there were
quantitative differences between these types of models, likely due to
differences in drug uptake, glutathione levels, and possibly
metabolism. Additional development of these methods is required.

Fig. 2.
ROS generation correlates with death in zebrafish larvae and rat hepatocytes. (A)
Zebrafish larvae at 72 hpf were arrayed in 96-well microplates, loaded with
dihydroethidium (DHE) for 30 minutes, and treated with menadione (50 μM). At the
indicated time points, plates were scanned and analyzed for red oxyethidium
fluorescence on an ArrayScan VTi high-content reader (ThermoFisher, Waltham, MA,
USA). (B) Fluorescence micrographs of dose-dependent menadione-induced
oxyethidium generation at 48 hpf. (C) Zebrafish larvae at 72 hpf (closed circles) or
cultured rat hepatocytes (open circles) were treated in 96-well plates with various
concentrations of menadione and oxyethidium fluorescence quantified. Toxicity
correlated with production of ROS in both zebrafish embryos and hepatocytes.

Go to:

CASE STUDY: USING “FIT-FOR-PURPOSE” ASSAY

EVALUATIONS TO RANK-ORDER COMPOUNDS

Given the limits to predicting human hepatotoxicity from current in

vitro and in vivo methods, additional improvements to the test
systems and analytical methods are needed to select better
compounds for preclinical testing. A case study is presented to
illustrate one strategy used at the University of Pittsburgh Drug
Discovery Institute to rank-order compounds for hepatotoxicity risk.
Compounds are screened through zebrafish embryos and a set of in
vitro “fit-for-purpose” cytotoxicity and mechanism of toxicity assays
are then applied (Fig. 3). Rank ordering does not rely on any one
single assay, but as a profile of risk factors calculated from the safety
margin (the ratio of toxic level to therapeutic level) categorically
binned into high, moderate, or low risk. The development of this
approach was an extension of the time-dependent and concentration-
dependent multiplexed MOT endpoint assays for mitochondrial
function, oxidative stress, and cytotoxicity analyses developed and
validated in the CellCiphr HepG2 and primary hepatocyte toxicity
panels. Our use of the safety margin to classify risk, taken together
with a Pfizer study that reported an increase in concordance between
in vitro assays and clinical hepatotoxicity with drugs that induced 2 or
more MOTs, suggested that some improvements are possible.17,23,43 In
the case study described here, the chance of clinical hepatotoxicity
increases in compounds with more high and moderate risk factors.
Fig. 3.
Case use of multiple “fit-for-purpose” assays to determine hepatotoxicity risk. The
concentration of inhibitor in which the response is reduced by half (IC50) results from 6
different assays in zebrafish, HepG2 and primary hepatocytes are used to calculate the
safety margin, defined as the toxic IC50 response/Cmax blood concentration. The lower
the safety window the higher the risk for hepatotoxicity. To rank-order compounds, the
safety margin is categorized into high (red), moderate (yellow), or low (green) risk to
generate the heat map. The overall rank order is determined by the number of high and
moderate risks. a Acute lethal dose at which one-half of zebrafish embryos are dead by
24-hour exposure. b Drug quantity not sufficient (QNS) to repeat study to calculate
safety margin.

Go to:

EMERGENCE OF HUMAN TISSUE AND ORGAN MODELS

For the reasons presented previously, new strategies are required to

better identify hepatotoxic compounds, especially chronic toxicity, as
well as to develop better human efficacy and disease models. The
development of biomimetic, multicellular, 3D, microfluidic
microphysiology models of the human liver and other organs are in
development.44,45
Cellular responses to drugs in the intact human organ are more
accurately represented by 3D human cell cultures than the traditional
static 2D cell cultures, with additional advantages provided by the
inclusion of media perfusion to provide nutrients, oxygen, chemicals,
and remove waste products.46–48 Researchers are now capitalizing on
the increased availability of human primary, immortalized, or induced
pluripotent stem cell (iPSC)-derived hepatocytes, new bioengineering
materials, microfabrication techniques, and microfluidic devices to
construct reasonable representations of the adult human liver acinus
in 3D multicellular microphysiological systems (MPS).48,49 These MPS
can be maintained for a month or longer, allowing chronic, as well as
acute, responses to drug challenges. In our recent study, we
demonstrated acute and chronic drug effects, including the induction
of fibrosis by methotrexate and the induction of immune-mediated
hepatotoxicity.48 A comparison of some current static and perfused 3D,
multicell models are presented in Table 6.

Table 6
Examples of multicellular static and microfluidic liver models for absorption,
distribution, metabolism, excretion, and toxicity (ADMET) testing

Model Application Advantage Disadvantage References

Static models
Co-culture Stellate cell Long term 3D Specialized 67
spheroids activation culturing with plates to from
improved drug spheroids,
metabolism and specialized
output of albumin, culturing
urea techniques
Co-culture Hepatotoxicity 2D cultures, 65,68
Hepatocytes
micropatterned Metabolite ID specialized
maintain
primary Disease models plates
differentiated
hepatocytes with
function 2–3 wk
fibroblasts
Hepatotoxicity Hepatocytes Specialized 69
Four-cell Metabolite ID maintain culturing
spheroids primary differentiated techniques
hepatocytes, function >3 wk,
primary liver 3D, immune-
NPC mediated toxicity
3D microfluidic, multicellular liver models
Model Application Advantage Disadvantage References
Primary Hepatotoxicity Hepatocytes 70
hepatocytes, maintain Specialized
endothelial cells differentiated culturing
function >3 wk, techniques,
microfluidic perfusion system
improves function
Hepatocytes, PK, toxicity, Hepatocytes Specialized 48,51,71
endothelial cells, therapeutic maintain culturing
stellate cells, intervention, differentiated techniques,
Kupffer-like liver disease function >3 wk, perfusion system
immune cells model immune-mediated
toxicity, fibrosis
activation
microfluidic
improves function
Open in a separate window
Abbreviations: 2D, 2 dimensional; 3D, 3 dimensional; NPC, non parenchymal cells; PK,
pharmacokinetic.

Of particular interest to those who study hepatotoxicity is the creation

of a “liver on a chip” with iPSC-derived adult hepatocytes from
patients who have susceptibility to DILI events or other defined
genetic and disease backgrounds. This would place liver MPS
platforms at the center of personalized medicine and in the continuum
of the QSP approach to drug discovery and evaluation of disease
progression.1,2 The potential ramifications and promises of this new
paradigm for drug discovery, disease progression, and toxicity
assessments are discussed in more detail in the Prospectus.
Go to:

PROSPECTUS: MOVING TO THE FUTURE: INTEGRATING THE

HUMAN LIVER ON A CHIP, COMPUTATIONAL MODELS, AND
QUANTITATIVE SYSTEMS PHARMACOLOGY

Collectively, the limited concordance of laboratory animal drug safety

testing with human safety, the apparent 80% limit of success of
human-based 2D in vitro models and the 65% to 75% rate of success
with computational models to predict drug-induced clinical
hepatotoxicity has shifted the focus to the creation of human, 3D,
microfluidic systems, referred to as MPS. One such platform has been
developed at the University of Pittsburgh Drug Discovery Institute and
integrates an MPS liver model using 4 human liver cell types
organized into a microfluidic, 3D, sinusoidal complex, with the
capacity for live cell monitoring of MOT using fluorescence-based
biosensors over a period of several weeks. Secreted proteins,
cytokines, and metabolites collected from the efflux are analyzed by
biochemical assays and mass spectrometry along with the results
from imaging biosensors for parameters such as apoptosis, ROS
production, and free calcium levels. All of the data are linked in a
database designed to collect, manage, and model the data (Fig. 4).
Integrated human MPS platforms that are biomimetics of normal
organ structure and function have the potential to improve on the
current predictive limit (approximately 80%) and diminish the odds
of “silent” human hepatotoxins from being introduced into clinical
trials or the market.

Fig. 4.
Overview of the Human Liver Microphysiology Platform for studying human liver
physiology, disease models and drug safety testing. The platform is composed of the
following: (A) the Sequentially Layered, Self-Assembly Liver model (SQL-SAL)
constructed from a microfluidic device and 4 human cell types, a fraction of which are
“sentinel” cells expressing fluorescence-based biosensors, and that can include disease-
specific cells, such as cancer cells. Data are collected from the model via (B) high-
content imaging readouts of transmitted light contrast and fluorescence; and (C)
biochemical and mass spectrometry readouts.48,51 (D) The multiplexed data are uploaded
into the Microphysiology Systems Database (MPS-Db) to manage data, associate
external data sources, and build predictive models of human efficacy and toxicity.50

Continued improvements to the MPS liver models will include the

application of renewable cells (eg, human iPSC-derived adult
hepatocytes, as well as the nonparenchymal cells) to permit the
investigation of the heterogeneous human genetic backgrounds, as
well as specific diseases (eg, nonalcoholic fatty liver disease,
hepatocarcinoma, and rare childhood liver diseases). Further
advances also will include liver metabolic zonation and higher
throughput arrays of MPS. The microphysiology database also will
continue to evolve as a tool to manage, mine, and model the
experimental data, as well as public sources of preclinical and clinical
findings, expert-based drug knowledge, physical properties,
pharmacology targets, adverse event reporting, and large datasets
from “omics,” including toxicogenomics, metabolomics, proteomics,
reactive metabolite proteomics, and transcriptomics.50 Furthermore,
QSP is expected to increase our understanding of the integrated and
interacting cellular, tissue, and organ networks; genes; proteins; and
metabolic processes that give rise to liver disease progression,
therapeutic efficacy, and drug-induced hepatotoxicity.1,50