You are on page 1of 11

Reproductive Toxicology 19 (2005) 327–337

Review

Dose–response modeling in reproductive toxicology in the


systems biology era
Melvin E. Andersen∗ , Russell S. Thomas, Kevin W. Gaido, Rory B. Conolly
CIIT Centers for Health Research, Research Triangle Park, NC 27709-2137, USA

Received 15 October 2004; received in revised form 1 December 2004; accepted 3 December 2004

Abstract

Systems biology approaches for modeling cellular signaling networks affected by chemical exposures should soon produce integrated
methodologies capable of predicting dose–response relationships for developmental toxicants and for other toxic responses. This paper out-
lines an emerging strategy for systems biology approaches in dose–response modeling. Genome-wide functional screens, bioinformatic tools,
and network mapping technologies together can provide directed graph representations of the cellular signaling networks. The graphical rep-
resentations can be converted into mathematical models that permit predicting the shapes of dose–response curves for altered cell signaling by
test compounds during development. Systems biology approaches require interdisciplinary teams with expertise in reproduction, cell biology,
signal transduction, mathematical/biomedical modeling, and risk assessment. In addition to outlining a systems approach for dose–response
research, this paper discusses initial stages of application of this strategy to examine inhibition of steroidogenesis in testes by phthalate
esters.
© 2004 Published by Elsevier Inc.

Keywords: System biology; Risk assessment; Signal transduction; Biological circuits; Cellular networks; Phthalates; Steroidogenesis; Testes malformations

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
2. Systems approaches in pharmacokinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
3. Linking doses and response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
4. The arrival of systems biology in dose–response assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
5. Cell signaling motifs as targets for developmental toxicants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
6. Phthalates and male reproductive tract development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
7. Assessing components of signaling pathways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
8. Creating mathematical models of function and perturbations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
9. How will these technologies be used in reproductive toxicology?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

1. Introduction

∗ Corresponding author. Tel.: +1 919 558 1205; fax: +1 919 558 1300. A major practical goal of toxicology research with en-
E-mail address: MAndersen@ciit.og (M.E. Andersen). vironmental chemicals and of safety research with drugs in

0890-6238/$ – see front matter © 2004 Published by Elsevier Inc.


doi:10.1016/j.reprotox.2004.12.004
328 M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337

the pharmaceutical industry is to understand the biological to as ‘omics,’ including components of genes, gene products,
factors that determine the shape of dose–response curves for proteins and various small molecules. This diverse array of
adverse responses at low levels of incidence, i.e., in regions data, however, needs to be organized quantitatively to create
of the dose–response curve where the probability of response information, i.e., to discover how the parts of the system are
in a population is small. When the quantitative relationships organized and controlled both in time and in space to give rise
among these factors are known with confidence, incidence to biological functions. We can define ‘systems approaches’
can be predicted (calculated) for various exposure situations as the attempt to quantitatively integrate information over
and these estimates of incidence can be used to predict ex- multiple levels of biological organization to understand how
pected risks to exposed populations. Together with social or biological functions are produced from simpler molecular,
political policies that prescribe tolerable risks for specific cellular and organ level interactions This paper outlines the
human populations, these predictions of risk serve multiple emerging contributions of systems biology in shaping new
functions, such as establishing exposure guidelines or sup- directions for PBPK, PBPD, and dose–response modeling in
porting product registrations. In the absence of knowledge quantitative human health risk assessment and reproductive
about the shape of dose–response curves in regions of low toxicology.
incidence, a variety of defaults are used in risk and safety as-
sessments. These defaults take conservative positions about
likely risks in exposed humans in an attempt to insure ade- 2. Systems approaches in pharmacokinetics
quate protection in the absence of knowledge.
The safety/risk assessment process requires both qualita- In physiologically based pharmacokinetic (PBPK) model-
tive and quantitative inputs. Qualitative studies provide infor- ing, compartments correspond to discrete tissues or to group-
mation about: (1) hazard, the possible effects of a compound ings of tissues with appropriate volumes, blood flows and
irrespective of exposure considerations; (2) mode-of-action, pathways for metabolism of test chemicals [1,2]. PBPK mod-
the manner in which chemical interacts with targets in bio- els include pertinent biochemical and physicochemical con-
logical systems to cause perturbations that lead to adverse stants for metabolism and solubility in each compartment.
responses; (3) progression, the steps connecting these inter- Routes of dosing (i.e., routes of administration) are included
actions through to impaired function; and (4) susceptibility, in their proper relationship to the overall physiology. The
the factors that may predispose an individual to an adverse equations that form the basis of the PBPK model also ac-
response. Susceptibility factors include: gender, age, genet- count for the time sequence of dose input into experimental
ics, co-exposure to other compounds, pre-existing disease, animals or test subjects and permit input by multiple routes if
or lifestyle. A full risk assessment process requires organiz- necessary for specific exposure situations. In pregnancy, fetal
ing this information quantitatively to make predictions of the tissues are also incorporated along with uterus and placenta
shape of the dose–response curve in regions of low incidence (see Fig. 1), as in a recent model for isopropanol and acetone
in a diverse human population. [3]. Each compartment in the model is described with a mass-
One challenge in quantitative health risk assessment arises balance differential equation (MBDE) whose terms mathe-
from the need to have models for dosimetry, i.e., the deliv- matically represent biological processes. The set of equations
ery of active forms of the test molecules/metabolites to tar- is solved by numerical integration to simulate tissue time-
get tissues, and for responses, i.e., the manner in which the course concentrations of chemicals and their metabolites.
molecular and cellular interactions of toxic compounds cause PBPK models focusing on developmental periods contain
perturbations that are sufficiently large and sufficiently pro- descriptions of changing organ volumes during gestation and
longed to lead to an adverse response. The models for dose during the post-partum period in the female [4,5]. Descrip-
and response are conveniently dichotomized as pharmacoki- tions of dosimetry for the developing organism usually con-
netic (PK) models (for dose) and pharmacodynamic (PD) sider both fetal exposure by transport across the placenta and
models (for response). Pharmacokinetics is generally defined neonatal exposure by transfer of compounds from the lac-
as what the body does to the compound; pharmacodynamics tating mother to the newborn in milk. Some PBPK models
is what the compound does to the body. To have confidence account for interactions of circulating compounds with spe-
in predictions from PK and PD models, these models need cific target receptors or for covalent interactions of chemicals
to be biologically realistic. Physiologically based (PB) mod- with tissue constituents. Modeling these reversible and irre-
els, both PBPK and PBPD models, tend to be more firmly versible molecular interactions with cell constituents is the
grounded in principles of biology and biochemistry. One risk initial step in developing PBPD models for effects of chem-
assessment goal is co-ordination of PBPK/PD models to cre- icals on biological processes.
ate biologically based dose–response (BBDR) models that While not encompassing current systems approaches that
predict expected incidence of adverse responses for varied include high throughput ‘omic’ methodologies, PBPK mod-
exposure situations. eling is an example of a systems approach at the level of cells,
Today, we have the ability to assess the components of cel- organs and organisms to ascertain the mechanisms of distri-
lular, organ, and organism-level processes with astonishing bution and interactions of drug and other chemicals in the
detail using a variety of relatively new technologies, referred body. Modern day PBPK modeling was originally developed
M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337 329

Fig. 1. Systems approaches to disposition of chemicals within the body. These two groupings of tissues represent a schematic of a PBPK model for isopropanol
(IPA) and its major metabolite, acetone (ACE), describing tissue dosimetry of both compounds in the developing fetus. These models have tissue compartments
described with volumes, blood flows (designated by Q), and tissue:blood partition coefficients. The transport across the placenta is diffusion-limited, described
by a permeation coefficient-area cross-product called PAF. Tissue volumes (weights) of many organs, and especially the fetus and placenta, vary during
pregnancy. Metabolism occurs in liver converting IPA to ACE with Michaelis–Menten kinetic constants, Vmaxc and Km . ACE is eliminated from the body by
hepatic metabolism with constants Vmax1c and Km1 . The IPA model allows input by inhalation, dermal contact and oral intake. Both compounds can be excreted
by metabolism or by exhalation. Predicted concentrations are determined by equations that integrate information on physiology, anatomy, and biochemistry,
producing predictions about the response of the entire system. Details of the routes of exposure (inhalation with wash-in wash-out characteristics in the mucous
phase of the respiratory tract) and the two-compartment representation for oral uptake (with pdose, stomach, and duodenum) are more fully described in the
original publication [3].

by two chemical engineers, Drs. Kenneth Bischoff and Robert the adduction of proteins or DNA by reactive parent chem-
Dedrick, who incorporated engineering principles, physiol- icals or their metabolites in various tissues (e.g., ethylene
ogy, chemistry and biochemistry into a computer modeling oxide [9] or acrylonitrile [10]). In these cases, the models
platform to predict kinetics [6,7]. Many PBPK models in- integrate molecular, cellular, organ level, and organism-level
tegrate information across multiple levels of organization, processes at a physiological level of organization to account
especially when describing interactions of compounds with for the time courses of chemicals, metabolites, and bound
molecular targets, such as reversible binding of ligands to complexes within multiple organs in the body. These PBPK
specific receptors (e.g., methotrexate [6] or dioxin [8]), or models attempt to account for major determinants of the dis-
330 M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337

tribution and elimination of compounds without describing these model structures, adverse endpoints occur due to al-
every physical chemical process involved in transport and terations in cell replication, apoptosis, and mutation rates;
storage of chemical in various tissues. More detail can be in- however, the cellular processes that are affected leading to
cluded in these models as more information becomes avail- alterations in these macroscopic cellular behaviors have not
able on chemical disposition from specific experiments. yet been described with respect to the effects of chemicals on
cellular signaling pathways or the interactions among signal-
ing pathways. Perhaps, the best example of a BBDR model
3. Linking doses and response for a developmental endpoint is the model for 5-flurouracil,
which links tissue concentrations with enzyme inhibition, im-
Over the past 15 years, toxicology and risk assessment paired nucleotide synthesis, altered DNA synthesis and de-
have emphasized the exposure dose–response relationship velopmental anomalies [15,16]. The challenge with the 5-FU
linking the presentation of chemical to an organism with spe- model or any other quantitative mathematical model for a
cific toxicological or pathological responses (Fig. 2). PBPK developmental endpoint is the difficulty in first providing an
models provide more detail on the sequence of steps through adequate description of the underlying biology that is being
tissue interactions, including binding of reactive molecules affected by the compound. The problem in describing biol-
with cellular constituents or recognition of chemical struc- ogy is a particular issue for development where processes
tures by reversible binding to cellular receptors involved in and structures are changing rapidly and consecutive devel-
regulating cell signaling pathways. For many risk assess- opmental landmarks are critically dependent on completion
ments, measures of tissue dose, sometimes called dose met- of earlier steps. Kavlock and Setzer [17] have provided an
rics [11] (i.e., measures of tissue dose believed to be closely excellent review of progress in constructing BBDR for de-
associated with the adverse responses) are used as the basis velopmental processes up to 1996.
for the dose–response portion of chemical risk assessment.
The specific steps that lead from these dose metrics to re-
sponse have usually been considered part of the pharmaco- 4. The arrival of systems biology in dose–response
dynamics (PD) of the response to chemical exposures. These assessment
dose metrics are the equivalent of a ‘biologically equivalent
dose’ (BED): these dose metrics link delivery of active forms While PBPK modeling represents a systems approach on
of the chemical to a response via the mode of action for the the physiological and biochemical levels, the ability to apply
compound in causing a particular toxic response. more integrated systems approaches to pharmacodynamic
In general, PD models used in modeling pharmaceutical modeling was simply not possible before the expansion of
responses have been more empirical, utilizing simple effect methods in molecular biology and various ‘omics’ technolo-
compartments correlated with blood or tissue concentrations gies of the past 10 years. New, high throughput, broad cov-
of active chemical. In risk assessment applications, there erage technologies – genomics, transcriptomics, proteomics,
have been attempts to link tissue dose metrics with more and metabonomics – provide the molecular ‘parts list’ for
integrated cellular level responses. Other BBDR approaches, cells, tissues, organs, and eventually organisms. Systems bi-
i.e., two-stage clonal growth models for carcinogenesis and ology attempts to integrate this diverse body of data to pro-
cell growth-based models for developmental processes, show duce knowledge about how the organization of these parts
some potential for modeling carcinogenesis [12] and limited leads to specific biological functions. In a review in Science
characterization of cell dynamics in development [13,14]. In in 2000, Lander and Weinberg [18] discussed the implications

Fig. 2. The exposure dose–response continuum. Developed as a pictorial representation of the steps involved in moving from exposure to response in the late
1980s, this popular linear representation focuses on toxicity, i.e., the perturbation caused by chemical treatment. This schematic shows the correspondence of
regions of the continuum with PBPK and PBPD processes, with the definition of the dose metric serving as the dividing line between the two model structures.
PBPK models predict dose metrics; PBPD models account for the consequences of that particular dose metric within tissues.
M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337 331

of the information generated by ‘genomics’ technologies for well understood prior to their broad use on more applied
an understanding of biology, noting that: problems. The same is true for toxicology where the linear
paradigm, employed to show the relationship of exposure,
“The long-term goal is to use this information to recon-
dose and response for integrating studies of toxicity (Fig. 2),
struct the complex molecular circuitry that operates within
now needs to focus much more heavily on investigating the
the cell—to map out the network of interacting proteins that
underlying biology prior to evaluating the perturbation of
determines the underlying logic of various cellular biologi-
the system following chemical exposure (Fig. 3). In this re-
cal functions including cell proliferation, responses to phys-
casting of the exposure dose–response linkage, toxicity and
iologic stresses, and acquisition and maintenance of tissue-
efficacy are defined by the intersection of chemical action
specific differentiation functions. A longer term goal, whose
with the biological system. Toxicology and pharmacology
feasibility remains unclear, is to create mathematical models
are disciplines at the interface of chemistry/pharmacokinetics
of these biological circuits and thereby predict these various
(primarily embedded in the vertical component) and biol-
types of cell biological behavior.”
ogy/pharmacodynamics (primarily captured by the horizon-
From our perspective, a defining characteristic of current tal chain). The next generation of systems approaches in PK
initiatives in systems biology is the iterative quantitative eval- and PD modeling will include increasingly detailed descrip-
uation, through laboratory experiments and computer mod- tions of biology afforded by new technologies and the expan-
eling, of the manner in which the components of biological sion of modeling tools available for describing the biological
systems are organized together to give rise to these circuit signaling processes affected by chemical exposures and drug
elements that control biological function. The aspect of com- treatments.
putational biology has become an intense area of activity for
understanding cell behavior as highlighted in the focused se-
ries of papers in Nature in 2002 [19,20]. Perturbations of these 5. Cell signaling motifs as targets for developmental
biological processes by environmental stressors, including toxicants
chemicals and drugs, can lead to adverse responses (toxic-
ity), restoration of normal function to a compromised tissue Considerable work has appeared in recent years focusing
(efficacy), or control of biological processes, such as occurs on cells as small machines conducting specific processes.
with use of hormonal therapies for birth control purposes in In addition to maintenance of these functions, the cell con-
women. sists of multiple interacting modules, with various feedback
The greatest progress in any discipline has always been controllers that regulate these behaviors. Many of these con-
achieved when the underlying principles of the system are trol modules have been examined in prokaryote cell systems

Fig. 3. Toxicity in a systems biology context. Toxic responses arise from perturbations in a system that converts various inputs through a series of biological
steps to produce specific functions necessary for health. The goals of a systems biology approach to dose–response assessment are (1) to define the nature of
perturbations associated with chemical exposures, (2) to map the networks of interacting proteins in which the site of perturbation is embedded, and (3) to
determine when these perturbations are likely to be sufficiently great to cause adverse health consequences in exposed animals or exposed people. Here the
pharmacodynamic process is the normal biological system and its responses to perturbations, not simply the perturbation itself. Individual pathways will have
inputs from other signaling networks, designated as secondary effector pathways, and perturbations may elicit adaptive changes that will influence dose–response
behaviors for altered signaling.
332 M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337

and mathematical models developed to describe the behavior


of the control networks. In some cases, simple prokaryotic
cells have been genetically engineered to contain specific cir-
cuit elements, i.e., biological oscillators, switches, amplifiers,
etc., and these constructs have been examined by laboratory
experiments and by computation [21–25]. These computa-
tional models evaluate the protein networks within cells, the
genetic control of these networks and the logic of cellular
responses affected by these networks [26–28]. These control
processes include activation of latent networks, silencing of
active networks, and modulation of networks. Control of the
circuitry occurs by various input signals to the cells. Input
signals, including various hormones, may activate cell sig-
naling networks to modulate both control circuitry and affect
downstream genetic networks that control new cellular func-
tions. In eukaryotic cells, a considerable body of work has
focused on modeling cell cycle control [29,30]. Progress in
understanding a variety of common signaling themes [31,32]
has led to increasing appreciation of the manner in which Fig. 4. Platelet derived growth factor (PDGF) signals through a receptor
tyrosine kinase signaling motif. This circuit diagram, redrawn from [34],
these modular components are used to achieve and control
served as the hypothesis in developing a mathematical model of the signaling
various cellular functions. network for platelet derived growth factor (PDGF) in vitro. Abbreviations:
From a dose–response modeling perspective, many xeno- PDGF-R, platelet-derived growth factor receptor; PKC, protein kinase C;
biotics interact with cellular receptors of various kinds to AA, arachidonic acid; PP2A, protein phosphatase 2A; MEK, MAPK or ERK
perturb cell circuits and modulate cellular functions. These kinase; MAPK, mitogen-activated protein kinase 1, 2; cPLA2, cytoplasmic
phospholipase A2; MKP, MAP kinase phosphatase; DAG, diacylglycerol.
interactions are especially relevant for so-called receptor-
SHC, src homology adapter protein; Grb2, growth factor receptor-bound pro-
mediated toxicants that are recognized in the body by specific tein 2; Ras, ras superfamily of monomeric GTPases; Raf, mitogen-activated
endogenous receptors. These compounds have the potential protein kinase in the Ras signaling pathway. The dark blue arrow from MAPK
to cause toxicity since they are mimics of endogenous signal- (lacking in the original) represents downstream protein targets that might in-
ing molecules leading to over-stimulation of natural circuits. clude transcription factors and other cell regulatory proteins. The activation
of the cascade and involvement of multiple downstream targets combine to
Alternatively, compounds may cause toxicity because they
create the switch for cellular responses as described in [34].
are competitive inhibitors. Competitive inhibitors will bind
the active sites of receptors, but fail to activate important cir-
cuits in the presence of normal signals. These compounds on intracellular reactions without necessarily increasing the
lead to diseases associated with deficiencies of signaling numbers of parameters that have to be estimated for specific
molecules and circuit activation. The National Academy of reactions [37] Model validation processes for these cellular
Sciences report, Scientific Frontiers of Developmental Toxi- pathways require a variety of studies conducted in vitro with
city and Risk Assessment [33], highlights the suite of known specific cell systems.
intercellular signaling motifs and their roles in development. While there are many signaling modules within any cell,
These signaling pathways are potential targets for toxic re- initial progress in understanding cellular behaviors is likely to
sponses for compounds that interfere with their function, focus on dissecting and modeling individual signaling mod-
cause over-stimulation of pathways at inappropriate times ules, giving rise eventually to a modular approach to under-
during development, or lead to aberrant signaling by activat- standing more integrated biological pathways [38]. Individ-
ing circuits inappropriately. ual modules that serve as primary targets of toxic compounds
Simulation models of several signaling pathways and drugs as well as the circuitry that controls these target
have been developed for mammalian cells in recent modules are likely to be the primary focus of initial systems
years—including ones for the platelet derived growth fac- biology approaches with these exogenous compounds.
tor (PDGF) pathway [34] in mouse 3T3 cells (Fig. 4) and for An interesting aspect of the PDGF signaling [34] is adapta-
tumor necrosis factor-alpha (TNF-␣) signaling through the tion to persistence of PDGF signaling through transcriptional
nuclear factor kappa B (NF-␬B) pathway in mouse fibroblasts control of a phosphatase (MKP) that blocks the usual activa-
[35,36]. These pathway models are similar to the sets of ordi- tion of mitogen-activated protein kinase (MAPK) responses
nary differential equations used in most PBPK models. They in the signaling network (Fig. 4). Such adaptive responses to
include many reactions within a cell and require evaluation toxicants that mimic signaling molecules would have at least
of multiple parameters for all the biochemical steps embed- two consequences: (1) modulation of dose–response curves
ded within the signaling networks. Models based on multiple by altering pathway sensitivity to bioactive small molecules
differential equations are being augmented by Boolean ap- and (2) desensitization of critical biological processes leading
proaches using on–off logic to increase the model coverage to failure of endogenous signaling and toxic consequences of
M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337 333

this failed signaling. These adaptive processes are likely to (MBP). Treatment of pregnant rats with high doses of DBP
be important in establishing shapes of dose–response curves leads to reproductive tract malformations in male pups. The
in regions of low incidence—the regions of most interest for broad biochemical mode of action for phthalates was identi-
risk and safety assessment. A more appropriate schematic of fied by the biology of the malformations, suggesting failure
the exposure dose–response paradigm of the past 15 years of normal androgen signaling [39] and then by direct mea-
should include adaptation and repair (Fig. 3) as an important surements of the reduction in testes testosterone [40]. Sub-
processes in regulating dose–response behaviors. sequently, studies have categorized the changes in gene tran-
script levels and proteins that occur in fetal testes exposed
to phthalates during late gestation. Changes were noted in
6. Phthalates and male reproductive tract many genes required for steroid transport/synthesis and lipid
development metabolism [41,42]. These studies provide a catalog of the al-
terations in expression of genes and proteins in the testes and
As yet, there are no examples of comprehensive systems provide a representation of the time sequence of changes af-
biology approaches to provide high-fidelity dose–response ter treatment with the phthalates. In general, gene expression
assessments for perturbations of critical signaling pathways studies and time-sequence analysis of alterations in gene ex-
during development, or, for that matter, for any other toxic pression in vivo, as conducted in the fetal testes with DBP or
response. However, the components required to pursue a sys- in uteri with estrogen treatment [43], could suggest hypothe-
tems biology approach for dose–response modeling can be ses regarding targets of toxic compounds and for cascades
enumerated and the laboratory studies and associated math- of signaling that ensue from interactions with these targets,
ematical modeling analysis required for such an approach an important aspect of cataloging in vivo effects of develop-
can be outlined. The final section of this review discusses mental toxicants (see left side of panel in Fig. 5).
the structure of a systems biology approach in relation to a In the adult testes, steroidogenesis requires G-protein cou-
specific developmental endpoint: reproductive tract malfor- pled signaling by luteininzing hormone (LH) interacting with
mations in male pups associated with treatment of pregnant LH receptors and simultaneous activation of arachidonic acid
rats with high doses of phthalate esters. Several staff mem- (AA) signaling pathways [44,45] associated with release of
bers at our home institution are now involved in creating a AA from membrane phospholipids during LH-stimulated
systems biology approach for developmental responses of the signaling. The AA is believed to be produced by activa-
male reproductive tract to phthalates. tion of an enzyme, cytoplasmic phospholipase A2 (cPLA2),
Phthalate esters, such as di(n-butyl)phthalate (DBP), are which releases AA from the 2-position of membrane phos-
metabolized to mono-esters, i.e., mono-n-butyl phthalate phatidyl choline. The central feedback loop with LH and

Fig. 5. Systems biology approaches for dose–response assessment for developmental toxicity have a series of components. The components of a systems
biology approach relying on current genomic technologies include expansion of conventional hazard identification studies to include microarray studies to
assess altered gene expression in target tissues in the affected tissues. These studies can provide hypotheses regarding molecular targets of action if they are
unknown. On the right side, more mechanistically oriented studies can be pursued after establishing appropriate cell systems for close examination. These
in vitro and ex vivo systems can be examined using genome-wide microarray expression and LOF and GOF screens to identify likely targets for chemical
perturbations and mapping the affected pathways. These genomic tools provide the information necessary to create representations of pathways for cellular
signaling (as in Fig. 4) leading to construction of computational dose–response models for the in vitro/ex vivo responses of cells or groups of cells. These
computational models together with the empirical dose–response model from the intact organism are brought together in the in vivo dose–response modeling.
334 M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337

steroidogenesis does not operate in the fetal testes and the screen is more targeted to identifying genes that play a func-
exact control of fetal steroidogenesis remains unknown. One tional role in the specific process. In the second approach, the
hypothesis for phthalate mode of action under evaluation is design of the screen is altered to identify molecular targets
that the monoalkylphthalate metabolites may be inhibiting of a specific molecule. Cells would be treated with the toxic
steroidogenesis by interfering with the downstream AA sig- compound causing partial inhibition of function. Individual
naling required for steroidogenesis in the testes. This portion cDNAs would then be screened to identify which genes when
of the steroidogenesis pathway may be common to both fe- overexpressed can complement the inhibitory effects of the
tal and adult Leydig cells even though LH does not serve compound. For identifying chemical targets in AA-mediated
as a stimulatory hormone in the fetus. Using a systems bi- processes, a GOE screen might be pursued using a known
ology approach, we are evaluating the normal cell circuitry inhibitor of cPLA2 and AA release such as chloroquin. The
associated with AA-mediated steroidogenesis by addressing GOE screen with transfected cDNAs would evaluate the suite
several key questions. They are: how will available labora- of genes that restore function and allow visualization of the
tory and software-based computational research tools be em- AA-mediated pathways of steroidogenesis.
ployed (1) to create an understanding of the normal circuitry When applied on a broad scale, optimally by testing with
controlling steroidogenesis, (2) to evaluate the sites where siRNAs and cDNAs representing the majority of genes in the
phthalates might alter signaling pathways, and (3) to create genome, these combined technologies provide a map of the
quantitative mathematical models of the alterations in these molecular components of cell signaling that control cellu-
signaling pathways? lar functions in a cell specific context. Successful pursuit of
these technologies today depends on robotic technologies for
screening libraries of genes and co-ordinate development of a
7. Assessing components of signaling pathways bioinformatics infrastructure to store and manipulate results
of the screens.
Estimating consequences of a perturbation rests on our A secondary benefit of a systems approach to understand-
ability to understand the ‘dose–response’ relationships for ing signaling networks and their perturbation by exposures
control of normal biological functions. For instance, post- to chemicals will be an enhanced ability to identify determi-
natal steroidogenesis in the testes requires appropriate con- nants of susceptibility. As full signaling pathways are elu-
centrations of LH, LHR, AA and other components of an cidated by genomic methodologies and interactions among
intact steroidogenesis pathway. The components of the sig- pathways become clarified, the biological contributions to
naling pathway have to be functionally identified, organized susceptibility, especially with regard to contributions from
into their appropriate networks (as was done in Fig. 4 for secondary signaling pathways (the secondary effector path-
PDGF signaling), and modeled computationally. To dissect ways in Fig. 3) and from crosstalk among signaling modules,
these networks, a number of research groups have begun will be more easily appreciated. With this information avail-
to apply large-scale reverse genetic screens to systemati- able, strategies can be developed to more fully assess the
cally identify genes that play a functional role in specific genetic contributors to susceptibility in human populations.
signaling pathways [46–51]. In these screens, cell-based as-
says are constructed with various cellular endpoints [48,49]
or reporter genes that indicate activation of a specific path- 8. Creating mathematical models of function and
way [46,47,50]. Full-length cDNAs or short inhibitory RNAs perturbations
(siRNAs) are introduced into the cells to provide a gain-of-
expression (GOE) or loss-of-expression (LOE) form of ge- The results from the GOE and LOE screens are stored
netic manipulation, respectively. The genes identified using in relational databases and the experimental data are sup-
these approaches are assumed to play a functional role in the plemented with known biomolecular interactions (e.g.,
specific pathway of interest. protein–protein interactions). Currently, the supplemental in-
The following paragraph describes work that is in progress formation is derived from both public databases (e.g., BIND
in our laboratory to use similar functional genomic ap- and DIP) or mined from the literature using either expert cu-
proaches to determine the signaling pathways involved rated datasets or natural language processing algorithms. All
gonadotropin-stimulated steroidogenesis in the adult mouse of these biological tools provide knowledge about the cir-
testes. These studies are planned, but have not been com- cuitry that controls cellular responses to chemical stressors
pleted at this point, and are provided as an example of the and the consequences of perturbations by toxic compounds.
overall approaches noted above. In LOE screens, cells would The diverse datasets are then synthesized and bioinformatic
be stimulated by LH and the expression of individual genes tools convert the experimental functional genomic data into
are knocked down using siRNAs. Genes altering steroidoge- directed graph representations of the intercellular signaling
nesis would be noted and cataloged. With GOE screens, two events. Based on the graphical representation of the system,
approaches could be used. First, cells with minimal LH stim- mechanistic dose–response models would then be developed
ulation would be transfected with individual cDNAs to see using ordinary differential equations (ODEs) or Boolean rep-
which genes produce enhanced steroidogenesis. The GOE resentations [37,51] to quantitatively describe the circuitry in-
M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337 335

volved in these responses (the intrinsic biology) and the site pression in affected tissues and examination of cells in vitro
of action(s) of the stressors (the perturbation). Experience with functional genomic screens are important techniques
in describing the functions of these circuits in simple cell for providing ‘high coverage’ of possible targets and draw-
systems has highlighted the non-linearity of many biological ing inferences about affected circuitry. As more is known
signaling networks [52]. As modeling strategies evolve, the about targets and affected circuitry, the data needed for these
more complex interactions among multiple signaling path- evaluations would be correspondingly reduced.
ways [53] will need to be included to evaluate the processes A significant challenge in the emerging systems biology
by which complex interrelationships among multiple path- approaches described here for reproductive and develop-
ways lead to unexpected or ‘emergent’ behaviors. mental endpoints is the difficulty of finding in vitro/ex
The aim of these quantitative simulation models for our vivo cell models that will permit high throughput studies.
purposes here is to describe the biology in sufficient detail Reproduction and development are highly dependent on
to understand the biological factors that control the shape of the interactions and subtle timing of interactions among
the dose–response curve. To accomplish this end, not every many cell types. Identifying affected targets and evaluating
detail of the signaling pathway would have to be included. these targets in single cell types in vitro will give only
These dose–response models will likely have some connec- a fragmentary picture of the full details of perturbations
tions in the circuit described as simply on-or-off with dose, occurring in the developing fetus. Nonetheless, strategies
while other steps will be described with greater biological de- of producing a detailed quantitative understanding of
tail. In general, critical components of the affected circuitry the control of pathways critical for proper development
will require a greater level of detail in the descriptions. The from iterative approaches using in vitro models, whole
application of these systems biology, dose–response models animal studies, and computational analysis of interact-
to safety and risk assessments is the primary goal, ultimately ing cellular networks appear to us to be the ones most
permitting prediction of the shape of the dose–response curve likely to provide the basic information needed to draw
based on knowledge of the basic cellular circuitry, and the mechanistic inferences of the shapes of dose–response
sites and intensity of perturbations by chemical stressors. curves in regions of low incidence for these responses
(Fig. 5). Initiatives soliciting proposals for multi-scale
modeling (www.nsf.gov/pubs/2004/nsf04607/nsf04607.htm)
9. How will these technologies be used in and for creating Centers for Biomedical Comput-
reproductive toxicology? ing (grants1.nih.gov/grants/guide/rfa-files/RFA-RM-04-
022.html) and Systems Biology (grants.nih.gov/grants/
The technologies, by helping to determine the biologi- guide/rfa-files/RFA-GM-05-010.html) organized around
cal determinants of the shape of the dose–response curve for driving biological projects at the National Institutes of
adverse responses, are intended to address several important Health are also strong indications that these integrated
risk assessment uncertainties. These uncertainties include the approaches across multiple levels of biological organization
shape of the dose–response curve at low doses, the risks posed are regarded as essential components in understanding how
by low level exposures to the human population, and the rel- the parts of biological system contribute to the control of
evance of observations of high dose toxicity in rodents for specific biological functions.
human risk assessment. Some rodent responses to phthalates, The task of creating these quantitative dose–response
primarily hepatocarcinogenesis, are not believed to be rele- models for endpoints related to reproductive toxicology ap-
vant markers for risks of these responses in human popula- pears daunting. However, the experience in high through-
tions, even though the PPAR signaling pathways affected by put biology, bioinformatics, and computation over the past
the phthalates do appear to be present in both rodents and decade indicates that progress in these areas has become as-
people [54]. The consequences of activation of this receptor tonishingly rapid. The primary focus initially with these sys-
for liver neoplasia differ between rodents and humans. It is tems approaches for dose–response modeling in reproduc-
not presently known if humans would show similar sensitiv- tive toxicity is to ask the right questions, design appropriate
ity to high-dose exposures to phthalate esters as do the male high throughput studies in cell models, and develop quanti-
rat pups. tative models of affected pathways. Initial results with sev-
The manner in which these systems approaches are applied eral chemicals should provide insights into the shape of the
with any specific toxic response is dependent on the level dose–response curve for perturbations of specific cell sig-
of prior knowledge concerning the molecular targets of the naling networks involved in organism development and then
chemicals, the extent of previous research on altered patterns subsequent research can be targeted at more clearly estab-
of gene expression, and the level of understanding of the path- lishing how alterations in primary target pathways produce
way and circuits based on available literature. When knowl- dose–response relationships for more integrated biological
edge about mode of action and targets is meager, the use of responses. These initial prototype assessments could focus
genomic methods is especially important in order to achieve on important classes of reactive or receptor-mediated toxi-
‘high coverage’ of possible targets rather than a one-by-one cants and be useful for improving the design and efficiency
examination of specific hypotheses. Evaluation of gene ex- of experiments to create dose–response models in the future.
336 M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337

Acknowledgments [14] Whitaker SY, Tran HT, Portier CJ. Development of a biologically-
based controlled growth and differentiation model for developmental
toxicology. J Math Biol 2003;46:1–16.
The authors are indebted to a group of colleagues at CIIT
[15] Setzer RW, Lau C, Mole ML, Copeland MF, Rogers JM, Kavlock RJ.
Centers for Health Research who contributed significantly Toward a biologically based doe-response model for developmental
to our ideas about systems approaches in toxicology and toxicity of 5-fluorouracil in the rat: a mathematical construct. Toxicol
dose–response assessment, including Drs. William Green- Sci 2001;59:49–58.
lee, Frederick Miller, Cecilia Tan, Li You, Kamin Johnson, [16] Lau C, Mole ML, Copeland MF, Rogers JM, Kavlock RJ, Shuey
DL, et al. Toward a biologically based doe-response model for de-
and David Dorman. M.E.A. also expresses appreciation to
velopmental toxicity of 5-fluorouracil in the rat: acquisition of ex-
Dr. Raymond Yang, Colorado State University, and Harvey perimental data. Toxicol Sci 2001;59:37–48.
Clewell, Health Sciences Institute, ENVIRON, for helpful [17] Kavlock RJ, Setzer RW. The road to embryologically based
discussions about systems approaches in PK and PD model- dose–response models. Environ Health Perspect 1996;104:107–
ing over many years. We also thank the Long-Range Research 21.
[18] Lander ES, Weinberg RA. Genomics: journey to the center of biol-
Initiative (LRI) Program of the American Chemistry Coun-
ogy. Science 2000;287:1777–82.
cil for support of the research outlined here and for support [19] Surridge C. Nature insight: computational biology. Nature
for organizing interdisciplinary teams to focus on research 2002;420:205.
to improve dose–response assessment methodologies for as- [20] Kitano H. Computational systems biology. Nature 2002;420:206–10.
sessing health risks of chemicals at environmentally relevant [21] Guet CC, Elowitz MB, Hsing W, Seibler S. Combinatorial synthesis
of genetic networks. Science 2002;296:1466–70.
levels of exposure.
[22] Hasty JF, et al. Designer gene networks: towards fundamental cel-
lular control. Chaos 2001;11:207–20.
[23] Hasty J, McMillen D, Collins JJ. Engineered gene circuits. Nature
2002;420:224–30.
[24] McMillen D, Kopell N, Hasty J, Collins JJ. Synchronizing genetic
References relaxation oscillators by intercell signalling. PNAS 2002;999:679–84.
[25] Tyson JJ, et al. Sniffers, buzzers, toggles and blinkers: dynamics of
[1] Bischoff KB, Brown RH. Drug distribution in mammals. Chem Eng regulatory and signaling pathways in the cell. Curr Opin Cell Biol
Prog Symp Ser 1966;62:33–45. 2003;15:221–31.
[2] Leung HW. Development and utilization of physiologically based [26] Alm E, Arkin AP. Biological networks. Curr Opin Struct Biol
pharmacokinetic models for toxicological applications. J Toxicol En- 2003;13:193–202.
viron Health 1991;32:247–67. [27] Davidson EH, Rast JP, Oliveri P, Ransick A, Calestani C, Yuh
[3] Gentry PR, Covington TR, Andersen ME, Clewell HJ. Applica- CH, et al. A genomic regulatory network for development. Science
tion of a physiologically-based pharmacokinetic model for iso- 2002;295:1669–78.
propanol in the derivation of an RfD/RfC. Regul Toxicol Pharmacol [28] Ferrell Jr JE. Self-perpetuating states in signal transduction: positive
2002;36:51–68. feedback, double-negative feedback and bistability. Curr Opin Chem
[4] Corley RA, Mast TJ, Carney EW, Rogers JM, Daston GP. Evalua- Biol 2002;6:140–8.
tion of physiologically based models of pregnancy and lactation for [29] Tyson JJ. Models of cell cycle control. J Biotechnol 1999;71:239–44.
their application in children’s health risk assessment. CRC Crit Rev [30] Novak B, Tyson JJ. Biochem Soc Trans 2003;31:1526–9.
Toxicol 2003;33:137–211. [31] Ray LB, Adler EM, Gough NR. Common signaling pathways. Sci-
[5] Gentry PR, Covington TR, Clewell III HJ. Evaluation of the po- ence 2004;306:1505.
tential impact of pharmacokinetic differences on tissue dosimetry in [32] Schlessinger J. Common and distinct elements in cellular signaling
offspring during pregnancy and lactation. Regul Toxicol Pharmacol via EGF and FGF receptors. Science 2004;306:1506–7.
2004;38:1–16. [33] Scientific Frontiers in Developmental Toxicology and Risk Assess-
[6] Bischoff KB, Dedrick RL, Zaharko DS, Longstreth JA. Methotrexate ment. National Research Council. National Academy Press, 2101
pharmacokinetics. J Pharm Sci 1971;60:1128–33. Constitution Ave., N.W., Washington, DC 20418; 2000.
[7] Dedrick RL. Animal scale-up. J Pharmacokinet Biopharm [34] Bhalla US, Ram PT, Iyengar R. MAP kinase phosphatase as a locus
1973;1:435–61. of flexibility in a mitogen-activated protein kinase signaling network.
[8] Kohn MC, Lucier GW, Clark GC, Sewall C, Tritscher A, Portier CJ. Science 2002;297:1018–23.
A mechanistic model of the effects of dioxin on gene expression in [35] Hoffmann A, Levchenko A, Scott ML, Baltimore D. The I␬B-NF-
the rat liver. Toxicol Appl Pharmacol 1993;120:138–54. ␬B signaling module: temporal control and selective gene activation.
[9] Krishnan K, Gargas ML, Fennell TR, Andersen ME. A Science 2002;298:1241–5.
physiologically-based description of ethylene oxide dosimetry in the [36] Cho K-H, Shin S-Y, Lee H-W, Wolkenhauer O. Investigations into
rat. J Toxicol Ind Health 1992;8:121–40. the analysis and modeling of the TNF-␣-mediated NF-␬B-signaling
[10] Gargas ML, Andersen ME, Teo SKO, Batra R, Fennel TR, Ked- pathway. Genome Res 2003;13:2412–22.
deris GL. A physiologically-based dosimetry description of acry- [37] Bolouri H, Davidson EH. Modeling transcriptional regulatory net-
lonitrile and cyanoethyleneoxide in the rat. Toxicol Appl Pharmacol works. Bioessays 2002;24:1118–29.
1995;134:185–94. [38] Hartwell LH, Hopfield JJ, Leibler S, Murray AW. From molecular
[11] Andersen ME. Tissue dosimetry in risk assessment: what’s the prob- to modular cell biology. Nature 1999;402:C47–52.
lem here any way? Drinking Water Health 1987;8:8–23. [39] Mylchreest E, Cattley RC, Foster PM. Male reproductive tract mal-
[12] Moolgavkar SH, Luebeck G. Two-event model for carcinogenesis: formations in rats following gestational and lactational exposures
biological, mathematical and statistical considerations. Risk Anal to di(n-butyl) phthalate: an antiandrogenic mechanism? Toxicol Sci
1990;10:323–41. 1998;43:47–60.
[13] Leroux BG, Leisenring WM, Moolgavkar SH, Faustman EM. A [40] Shultz VD, Phillips S, Sar M, Foster PM, Gaido KW. Altered
biologically-based dose–response model for developmental toxicol- gene profiles in fetal rat testes after in utero exposure to di(n-
ogy. Risk Anal 1996;16:449–58. butyl)phthalate. Toxicol Sci 2001;64:233–42.
M.E. Andersen et al. / Reproductive Toxicology 19 (2005) 327–337 337

[41] Thompson CJ, Ross SM, Gaido KW. Di(n-butyl) phthalate impairs [48] Berns K, Hijmans EM, Mullenders J, Brummelkamp TR, Velds A,
cholesterol transport and steroidogenesis in the fetal rat testis through Heimerikx M, et al. A large-scale RNAi screen in human cells iden-
a rapid and reversible mechanism. Endocrinology 2004;145:1227–37. tifies new components of the p53 pathway. Nature 2004;428:431–
[42] Lehmann KP, Phillips S, Sar M, Foster PMD, Gaido KW. Dose- 7.
dependent alterations in gene expression and testosterone synthesis in [49] Michiels F, van Es H, van Rompaey L, Merchiers P, Francken B,
the fetal testes of male rats exposed to di(n-butyl) phthalate. Toxicol Pittois K, et al. Arrayed adenoviral expression libraries for functional
Sci 2004;81:60–8. screening. Nat Biotechnol 2002;20:1154–7.
[43] Fertuck KC, et al. Identification of temporal patterns of gene ex- [50] Huang Q, Raya A, DeJesus P, Chao SH, Quon KC, Caldwell JS,
pression in the uteri of immature, oviarectomized mice following et al. Identification of p53 regulators by genome-wide functional
exposure to ethynylestradiol. Physiol Genom 2003;15:127–41. analysis. Proc Natl Acad Sci USA 2004;101:3456–61.
[44] Wang X, Walsh LP, Stocco DM. The role of arachidonic acid on LH- [51] Akutsu T, Kuhara S, Maruyama O, Miyano S. A system for iden-
stimulated steroidogenesis and steroidogenic acute regulatory pro- tifying genetic regulatory networks from gene expression patterns
tein accumulation in MA-10 mouse Leydig tumor cells. Endocrine produced by gene disruption and overexpressions. Genome Inform
1999;10:7–12. Ser Workshop Genome Inform 1998;9:151–60.
[45] Wang X, Walsh LP, Reinhart AJ, Stocco DM. The role of arachidonic [52] Andersen ME, Dennison JE, Thomas RS, Conolly RB. New direc-
acid in steroidogenesis and steroidogenic acute regulatory (StAR) tions in incidence-dose modeling. Trends Biotechnol, in press.
gene and protein expression. J Biol Chem 2000;275:20204–9. [53] Weng G, Bhalla US, Iyengar R. Complexity in biological signaling
[46] Chanda SK, White S, Orth AP, Reisdorph R, Miraglia L, Thomas systems. Science 1999;284:92–6.
RS, et al. Genome-scale functional profiling of the mammalian AP-1 [54] Chevalier S, Roberts RA. Perturbations of rodent hepatocyte growth
signaling pathway. Proc Natl Acad Sci USA 2003;100:12153–8. control by nongenotoxic hepatocarcinogens: mechanisms and lack
[47] Lum L, Yao S, Mozer B, Rovescalli A, Von Kessler D, Nirenberg of relevance for human health (review). Oncol Rep 1998;5:1319–
M, et al. Identification of Hedgehog pathway components by RNAi 27.
in Drosophila cultured cells. Science 2003;299:2039–45.

You might also like